perf!: use 64k insteadn of 8k buffers for blob_io and ephemeral_file

.envrc

@@ -1 +0,0 @@
-use flake . --impure

.github/workflows/build_and_test.yml

@@ -477,7 +477,6 @@ jobs:
           BUILD_TAG: ${{ needs.tag.outputs.build-tag }}
           PAGESERVER_VIRTUAL_FILE_IO_ENGINE: tokio-epoll-uring
           PAGESERVER_GET_VECTORED_IMPL: vectored
-          PAGESERVER_GET_IMPL: vectored
 
       # Temporary disable this step until we figure out why it's so flaky
       # Ref https://github.com/neondatabase/neon/issues/4540
@@ -736,7 +735,7 @@ jobs:
         run: |
           mkdir -p .docker-custom
           echo DOCKER_CONFIG=$(pwd)/.docker-custom >> $GITHUB_ENV
-      - uses: docker/setup-buildx-action@v2
+      - uses: docker/setup-buildx-action@v3
 
       - uses: docker/login-action@v3
         with:
@@ -793,7 +792,7 @@ jobs:
         run: |
           mkdir -p .docker-custom
           echo DOCKER_CONFIG=$(pwd)/.docker-custom >> $GITHUB_ENV
-      - uses: docker/setup-buildx-action@v2
+      - uses: docker/setup-buildx-action@v3
         with:
           # Disable parallelism for docker buildkit.
           # As we already build everything with `make -j$(nproc)`, running it in additional level of parallelisam blows up the Runner.
@@ -866,7 +865,7 @@ jobs:
       run:
         shell: sh -eu {0}
     env:
-      VM_BUILDER_VERSION: v0.28.1
+      VM_BUILDER_VERSION: v0.23.2
 
     steps:
       - name: Checkout

.gitignore

@@ -23,7 +23,3 @@ compaction-suite-results.*
 
 # pgindent typedef lists
 *.list
-
-# nix dev env
-.direnv
-.devenv

Cargo.lock

@@ -599,7 +599,7 @@ dependencies = [
  "once_cell",
  "pin-project-lite",
  "pin-utils",
- "rustls 0.21.11",
+ "rustls 0.21.9",
  "tokio",
  "tracing",
 ]
@@ -2519,7 +2519,7 @@ dependencies = [
  "http 0.2.9",
  "hyper 0.14.26",
  "log",
- "rustls 0.21.11",
+ "rustls 0.21.9",
  "rustls-native-certs 0.6.2",
  "tokio",
  "tokio-rustls 0.24.0",
@@ -3658,7 +3658,6 @@ dependencies = [
  "tokio-util",
  "toml_edit",
  "tracing",
- "twox-hash",
  "url",
  "utils",
  "walkdir",
@@ -4060,7 +4059,7 @@ dependencies = [
  "futures",
  "once_cell",
  "pq_proto",
- "rustls 0.22.4",
+ "rustls 0.22.2",
  "rustls-pemfile 2.1.1",
  "serde",
  "thiserror",
@@ -4351,7 +4350,7 @@ dependencies = [
  "routerify",
  "rstest",
  "rustc-hash",
- "rustls 0.22.4",
+ "rustls 0.22.2",
  "rustls-pemfile 2.1.1",
  "scopeguard",
  "serde",
@@ -4543,7 +4542,7 @@ dependencies = [
  "itoa",
  "percent-encoding",
  "pin-project-lite",
- "rustls 0.22.4",
+ "rustls 0.22.2",
  "rustls-native-certs 0.7.0",
  "rustls-pemfile 2.1.1",
  "rustls-pki-types",
@@ -4697,7 +4696,7 @@ dependencies = [
  "once_cell",
  "percent-encoding",
  "pin-project-lite",
- "rustls 0.21.11",
+ "rustls 0.21.9",
  "rustls-pemfile 1.0.2",
  "serde",
  "serde_json",
@@ -4957,9 +4956,9 @@ dependencies = [
 
 [[package]]
 name = "rustls"
-version = "0.21.11"
+version = "0.21.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7fecbfb7b1444f477b345853b1fce097a2c6fb637b2bfb87e6bc5db0f043fae4"
+checksum = "629648aced5775d558af50b2b4c7b02983a04b312126d45eeead26e7caa498b9"
 dependencies = [
  "log",
  "ring 0.17.6",
@@ -4969,9 +4968,9 @@ dependencies = [
 
 [[package]]
 name = "rustls"
-version = "0.22.4"
+version = "0.22.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "bf4ef73721ac7bcd79b2b315da7779d8fc09718c6b3d2d1b2d94850eb8c18432"
+checksum = "e87c9956bd9807afa1f77e0f7594af32566e830e088a5576d27c5b6f30f49d41"
 dependencies = [
  "log",
  "ring 0.17.6",
@@ -5085,7 +5084,6 @@ dependencies = [
  "aws-smithy-async",
  "bincode",
  "bytes",
- "camino",
  "chrono",
  "clap",
  "crc32c",
@@ -5284,7 +5282,7 @@ checksum = "2e95efd0cefa32028cdb9766c96de71d96671072f9fb494dc9fb84c0ef93e52b"
 dependencies = [
  "httpdate",
  "reqwest",
- "rustls 0.21.11",
+ "rustls 0.21.9",
  "sentry-backtrace",
  "sentry-contexts",
  "sentry-core",
@@ -5832,7 +5830,8 @@ checksum = "81cdd64d312baedb58e21336b31bc043b77e01cc99033ce76ef539f78e965ebc"
 [[package]]
 name = "svg_fmt"
 version = "0.4.2"
-source = "git+https://github.com/neondatabase/fork--nical--rust_debug?branch=neon#b9501105e746629004bc6d0473639320939dbe10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f83ba502a3265efb76efb89b0a2f7782ad6f2675015d4ce37e4b547dda42b499"
 
 [[package]]
 name = "syn"
@@ -6194,7 +6193,7 @@ checksum = "0ea13f22eda7127c827983bdaf0d7fff9df21c8817bab02815ac277a21143677"
 dependencies = [
  "futures",
  "ring 0.17.6",
- "rustls 0.22.4",
+ "rustls 0.22.2",
  "tokio",
  "tokio-postgres",
  "tokio-rustls 0.25.0",
@@ -6207,7 +6206,7 @@ version = "0.24.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e0d409377ff5b1e3ca6437aa86c1eb7d40c134bfec254e44c830defa92669db5"
 dependencies = [
- "rustls 0.21.11",
+ "rustls 0.21.9",
  "tokio",
 ]
 
@@ -6217,7 +6216,7 @@ version = "0.25.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "775e0c0f0adb3a2f22a00c4745d728b479985fc15ee7ca6a2608388c5569860f"
 dependencies = [
- "rustls 0.22.4",
+ "rustls 0.22.2",
  "rustls-pki-types",
  "tokio",
 ]
@@ -6678,7 +6677,7 @@ dependencies = [
  "base64 0.21.1",
  "log",
  "once_cell",
- "rustls 0.21.11",
+ "rustls 0.21.9",
  "rustls-webpki 0.100.2",
  "url",
  "webpki-roots 0.23.1",
@@ -7355,7 +7354,7 @@ dependencies = [
  "regex-automata 0.4.3",
  "regex-syntax 0.8.2",
  "reqwest",
- "rustls 0.21.11",
+ "rustls 0.21.9",
  "scopeguard",
  "serde",
  "serde_json",

Cargo.toml

@@ -157,8 +157,7 @@ socket2 = "0.5"
 strum = "0.24"
 strum_macros = "0.24"
 "subtle"  = "2.5.0"
-# https://github.com/nical/rust_debug/pull/4
-svg_fmt = { git = "https://github.com/neondatabase/fork--nical--rust_debug", branch = "neon" }
+svg_fmt = "0.4.1"
 sync_wrapper = "0.1.2"
 tar = "0.4"
 task-local-extensions = "0.1.4"

Makefile

@@ -25,16 +25,14 @@ ifeq ($(UNAME_S),Linux)
 	# Seccomp BPF is only available for Linux
 	PG_CONFIGURE_OPTS += --with-libseccomp
 else ifeq ($(UNAME_S),Darwin)
-	ifndef DISABLE_HOMEBREW
-		# macOS with brew-installed openssl requires explicit paths
-		# It can be configured with OPENSSL_PREFIX variable
-		OPENSSL_PREFIX ?= $(shell brew --prefix openssl@3)
-		PG_CONFIGURE_OPTS += --with-includes=$(OPENSSL_PREFIX)/include --with-libraries=$(OPENSSL_PREFIX)/lib
-		PG_CONFIGURE_OPTS += PKG_CONFIG_PATH=$(shell brew --prefix icu4c)/lib/pkgconfig
-		# macOS already has bison and flex in the system, but they are old and result in postgres-v14 target failure
-		# brew formulae are keg-only and not symlinked into HOMEBREW_PREFIX, force their usage
-		EXTRA_PATH_OVERRIDES += $(shell brew --prefix bison)/bin/:$(shell brew --prefix flex)/bin/:
-	endif
+	# macOS with brew-installed openssl requires explicit paths
+	# It can be configured with OPENSSL_PREFIX variable
+	OPENSSL_PREFIX ?= $(shell brew --prefix openssl@3)
+	PG_CONFIGURE_OPTS += --with-includes=$(OPENSSL_PREFIX)/include --with-libraries=$(OPENSSL_PREFIX)/lib
+	PG_CONFIGURE_OPTS += PKG_CONFIG_PATH=$(shell brew --prefix icu4c)/lib/pkgconfig
+	# macOS already has bison and flex in the system, but they are old and result in postgres-v14 target failure
+	# brew formulae are keg-only and not symlinked into HOMEBREW_PREFIX, force their usage
+	EXTRA_PATH_OVERRIDES += $(shell brew --prefix bison)/bin/:$(shell brew --prefix flex)/bin/:
 endif
 
 # Use -C option so that when PostgreSQL "make install" installs the

control_plane/src/bin/neon_local.rs

@@ -417,54 +417,6 @@ async fn handle_tenant(
                 println!("{} {:?}", t.id, t.state);
             }
         }
-        Some(("import", import_match)) => {
-            let tenant_id = parse_tenant_id(import_match)?.unwrap_or_else(TenantId::generate);
-
-            let storage_controller = StorageController::from_env(env);
-            let create_response = storage_controller.tenant_import(tenant_id).await?;
-
-            let shard_zero = create_response
-                .shards
-                .first()
-                .expect("Import response omitted shards");
-
-            let attached_pageserver_id = shard_zero.node_id;
-            let pageserver =
-                PageServerNode::from_env(env, env.get_pageserver_conf(attached_pageserver_id)?);
-
-            println!(
-                "Imported tenant {tenant_id}, attached to pageserver {attached_pageserver_id}"
-            );
-
-            let timelines = pageserver
-                .http_client
-                .list_timelines(shard_zero.shard_id)
-                .await?;
-
-            // Pick a 'main' timeline that has no ancestors, the rest will get arbitrary names
-            let main_timeline = timelines
-                .iter()
-                .find(|t| t.ancestor_timeline_id.is_none())
-                .expect("No timelines found")
-                .timeline_id;
-
-            let mut branch_i = 0;
-            for timeline in timelines.iter() {
-                let branch_name = if timeline.timeline_id == main_timeline {
-                    "main".to_string()
-                } else {
-                    branch_i += 1;
-                    format!("branch_{branch_i}")
-                };
-
-                println!(
-                    "Importing timeline {tenant_id}/{} as branch {branch_name}",
-                    timeline.timeline_id
-                );
-
-                env.register_branch_mapping(branch_name, tenant_id, timeline.timeline_id)?;
-            }
-        }
         Some(("create", create_match)) => {
             let tenant_conf: HashMap<_, _> = create_match
                 .get_many::<String>("config")
@@ -1528,8 +1480,6 @@ fn cli() -> Command {
             .subcommand(Command::new("config")
                 .arg(tenant_id_arg.clone())
                 .arg(Arg::new("config").short('c').num_args(1).action(ArgAction::Append).required(false)))
-            .subcommand(Command::new("import").arg(tenant_id_arg.clone().required(true))
-                .about("Import a tenant that is present in remote storage, and create branches for its timelines"))
         )
         .subcommand(
             Command::new("pageserver")

control_plane/src/local_env.rs

@@ -129,7 +129,6 @@ pub struct PageServerConf {
 
     pub(crate) virtual_file_io_engine: Option<String>,
     pub(crate) get_vectored_impl: Option<String>,
-    pub(crate) get_impl: Option<String>,
 }
 
 impl Default for PageServerConf {
@@ -142,7 +141,6 @@ impl Default for PageServerConf {
             http_auth_type: AuthType::Trust,
             virtual_file_io_engine: None,
             get_vectored_impl: None,
-            get_impl: None,
         }
     }
 }

control_plane/src/pageserver.rs

@@ -92,7 +92,6 @@ impl PageServerNode {
             http_auth_type,
             virtual_file_io_engine,
             get_vectored_impl,
-            get_impl,
         } = &self.conf;
 
         let id = format!("id={}", id);
@@ -112,11 +111,6 @@ impl PageServerNode {
         } else {
             String::new()
         };
-        let get_impl = if let Some(get_impl) = get_impl {
-            format!("get_impl='{get_impl}'")
-        } else {
-            String::new()
-        };
 
         let broker_endpoint_param = format!("broker_endpoint='{}'", self.env.broker.client_url());
 
@@ -130,7 +124,6 @@ impl PageServerNode {
             broker_endpoint_param,
             virtual_file_io_engine,
             get_vectored_impl,
-            get_impl,
         ];
 
         if let Some(control_plane_api) = &self.env.control_plane_api {
@@ -441,11 +434,6 @@ impl PageServerNode {
                 .map(serde_json::from_str)
                 .transpose()
                 .context("parse `timeline_get_throttle` from json")?,
-            switch_to_aux_file_v2: settings
-                .remove("switch_to_aux_file_v2")
-                .map(|x| x.parse::<bool>())
-                .transpose()
-                .context("Failed to parse 'switch_to_aux_file_v2' as bool")?,
         };
         if !settings.is_empty() {
             bail!("Unrecognized tenant settings: {settings:?}")
@@ -564,11 +552,6 @@ impl PageServerNode {
                     .map(serde_json::from_str)
                     .transpose()
                     .context("parse `timeline_get_throttle` from json")?,
-                switch_to_aux_file_v2: settings
-                    .remove("switch_to_aux_file_v2")
-                    .map(|x| x.parse::<bool>())
-                    .transpose()
-                    .context("Failed to parse 'switch_to_aux_file_v2' as bool")?,
             }
         };
 

control_plane/src/storage_controller.rs

@@ -472,16 +472,6 @@ impl StorageController {
             .await
     }
 
-    #[instrument(skip(self))]
-    pub async fn tenant_import(&self, tenant_id: TenantId) -> anyhow::Result<TenantCreateResponse> {
-        self.dispatch::<(), TenantCreateResponse>(
-            Method::POST,
-            format!("debug/v1/tenant/{tenant_id}/import"),
-            None,
-        )
-        .await
-    }
-
     #[instrument(skip(self))]
     pub async fn tenant_locate(&self, tenant_id: TenantId) -> anyhow::Result<TenantLocateResponse> {
         self.dispatch::<(), _>(

control_plane/storcon_cli/src/main.rs

@@ -1,15 +1,15 @@
-use std::{collections::HashMap, str::FromStr, time::Duration};
+use std::{collections::HashMap, str::FromStr};
 
 use clap::{Parser, Subcommand};
-use hyper::{Method, StatusCode};
+use hyper::Method;
 use pageserver_api::{
     controller_api::{
         NodeAvailabilityWrapper, NodeDescribeResponse, ShardSchedulingPolicy,
         TenantDescribeResponse, TenantPolicyRequest,
     },
     models::{
-        LocationConfigSecondary, ShardParameters, TenantConfig, TenantConfigRequest,
-        TenantCreateRequest, TenantShardSplitRequest, TenantShardSplitResponse,
+        ShardParameters, TenantConfig, TenantConfigRequest, TenantCreateRequest,
+        TenantShardSplitRequest, TenantShardSplitResponse,
     },
     shard::{ShardStripeSize, TenantShardId},
 };
@@ -120,12 +120,6 @@ enum Command {
         #[arg(long)]
         tenant_id: TenantId,
     },
-    /// For a tenant which hasn't been onboarded to the storage controller yet, add it in secondary
-    /// mode so that it can warm up content on a pageserver.
-    TenantWarmup {
-        #[arg(long)]
-        tenant_id: TenantId,
-    },
 }
 
 #[derive(Parser)]
@@ -587,94 +581,6 @@ async fn main() -> anyhow::Result<()> {
             }
             println!("{table}");
         }
-        Command::TenantWarmup { tenant_id } => {
-            let describe_response = storcon_client
-                .dispatch::<(), TenantDescribeResponse>(
-                    Method::GET,
-                    format!("control/v1/tenant/{tenant_id}"),
-                    None,
-                )
-                .await;
-            match describe_response {
-                Ok(describe) => {
-                    if matches!(describe.policy, PlacementPolicy::Secondary) {
-                        // Fine: it's already known to controller in secondary mode: calling
-                        // again to put it into secondary mode won't cause problems.
-                    } else {
-                        anyhow::bail!("Tenant already present with policy {:?}", describe.policy);
-                    }
-                }
-                Err(mgmt_api::Error::ApiError(StatusCode::NOT_FOUND, _)) => {
-                    // Fine: this tenant isn't know to the storage controller yet.
-                }
-                Err(e) => {
-                    // Unexpected API error
-                    return Err(e.into());
-                }
-            }
-
-            vps_client
-                .location_config(
-                    TenantShardId::unsharded(tenant_id),
-                    pageserver_api::models::LocationConfig {
-                        mode: pageserver_api::models::LocationConfigMode::Secondary,
-                        generation: None,
-                        secondary_conf: Some(LocationConfigSecondary { warm: true }),
-                        shard_number: 0,
-                        shard_count: 0,
-                        shard_stripe_size: ShardParameters::DEFAULT_STRIPE_SIZE.0,
-                        tenant_conf: TenantConfig::default(),
-                    },
-                    None,
-                    true,
-                )
-                .await?;
-
-            let describe_response = storcon_client
-                .dispatch::<(), TenantDescribeResponse>(
-                    Method::GET,
-                    format!("control/v1/tenant/{tenant_id}"),
-                    None,
-                )
-                .await?;
-
-            let secondary_ps_id = describe_response
-                .shards
-                .first()
-                .unwrap()
-                .node_secondary
-                .first()
-                .unwrap();
-
-            println!("Tenant {tenant_id} warming up on pageserver {secondary_ps_id}");
-            loop {
-                let (status, progress) = vps_client
-                    .tenant_secondary_download(
-                        TenantShardId::unsharded(tenant_id),
-                        Some(Duration::from_secs(10)),
-                    )
-                    .await?;
-                println!(
-                    "Progress: {}/{} layers, {}/{} bytes",
-                    progress.layers_downloaded,
-                    progress.layers_total,
-                    progress.bytes_downloaded,
-                    progress.bytes_total
-                );
-                match status {
-                    StatusCode::OK => {
-                        println!("Download complete");
-                        break;
-                    }
-                    StatusCode::ACCEPTED => {
-                        // Loop
-                    }
-                    _ => {
-                        anyhow::bail!("Unexpected download status: {status}");
-                    }
-                }
-            }
-        }
     }
 
     Ok(())

docs/storage_controller.md

@@ -1,150 +0,0 @@
-# Storage Controller
-
-## Concepts
-
-The storage controller sits between administrative API clients and pageservers, and handles the details of mapping tenants to pageserver tenant shards. For example, creating a tenant is one API call to the storage controller,
-which is mapped into many API calls to many pageservers (for multiple shards, and for secondary locations).
-
-It implements a pageserver-compatible API that may be used for CRUD operations on tenants and timelines, translating these requests into appropriate operations on the shards within a tenant, which may be on many different pageservers. Using this API, the storage controller may be used in the same way as the pageserver's administrative HTTP API, hiding
-the underlying details of how data is spread across multiple nodes.
-
-The storage controller also manages generations, high availability (via secondary locations) and live migrations for tenants under its management. This is done with a reconciliation loop pattern, where tenants have an “intent” state and a “reconcile” task that tries to make the outside world match the intent.
-
-## APIs
-
-The storage controller’s HTTP server implements four logically separate APIs:
-
-- `/v1/...` path is the pageserver-compatible API. This has to be at the path root because that’s where clients expect to find it on a pageserver.
-- `/control/v1/...` path is the storage controller’s API, which enables operations such as registering and management pageservers, or executing shard splits.
-- `/debug/v1/...` path contains endpoints which are either exclusively used in tests, or are for use by engineers when supporting a deployed system.
-- `/upcall/v1/...` path contains endpoints that are called by pageservers. This includes the `/re-attach` and `/validate` APIs used by pageservers
-  to ensure data safety with generation numbers.
-
-The API is authenticated with a JWT token, and tokens must have scope `pageserverapi` (i.e. the same scope as pageservers’ APIs).
-
-See the `http.rs` file in the source for where the HTTP APIs are implemented.
-
-## Database
-
-The storage controller uses a postgres database to persist a subset of its state. Note that the storage controller does _not_ keep all its state in the database: this is a design choice to enable most operations to be done efficiently in memory, rather than having to read from the database. See `persistence.rs` for a more comprehensive comment explaining what we do and do not persist: a useful metaphor is that we persist objects like tenants and nodes, but we do not
-persist the _relationships_ between them: the attachment state of a tenant's shards to nodes is kept in memory and
-rebuilt on startup.
-
-The file `[persistence.rs](http://persistence.rs)` contains all the code for accessing the database, and has a large doc comment that goes into more detail about exactly what we persist and why.
-
-The `diesel` crate is used for defining models & migrations.
-
-Running a local cluster with `cargo neon` automatically starts a vanilla postgress process to host the storage controller’s database.
-
-### Diesel tip: migrations
-
-If you need to modify the database schema, here’s how to create a migration:
-
-- Install the diesel CLI with `cargo install diesel_cli`
-- Use `diesel migration generate <name>` to create a new migration
-- Populate the SQL files in the `migrations/` subdirectory
-- Use `DATABASE_URL=... diesel migration run` to apply the migration you just wrote: this will update the `[schema.rs](http://schema.rs)` file automatically.
-  - This requires a running database: the easiest way to do that is to just run `cargo neon init ; cargo neon start`, which will leave a database available at `postgresql://localhost:1235/attachment_service`
-- Commit the migration files and the changes to schema.rs
-- If you need to iterate, you can rewind migrations with `diesel migration revert -a` and then `diesel migration run` again.
-- The migrations are build into the storage controller binary, and automatically run at startup after it is deployed, so once you’ve committed a migration no further steps are needed.
-
-## storcon_cli
-
-The `storcon_cli` tool enables interactive management of the storage controller. This is usually
-only necessary for debug, but may also be used to manage nodes (e.g. marking a node as offline).
-
-`storcon_cli --help` includes details on commands.
-
-# Deploying
-
-This section is aimed at engineers deploying the storage controller outside of Neon's cloud platform, as
-part of a self-hosted system.
-
-_General note: since the default `neon_local` environment includes a storage controller, this is a useful
-reference when figuring out deployment._
-
-## Database
-
-It is **essential** that the database used by the storage controller is durable (**do not store it on ephemeral
-local disk**). This database contains pageserver generation numbers, which are essential to data safety on the pageserver.
-
-The resource requirements for the database are very low: a single CPU core and 1GiB of memory should work well for most deployments. The physical size of the database is typically under a gigabyte.
-
-Set the URL to the database using the `--database-url` CLI option.
-
-There is no need to run migrations manually: the storage controller automatically applies migrations
-when it starts up.
-
-## Configure pageservers to use the storage controller
-
-1. The pageserver `control_plane_api` and `control_plane_api_token` should be set in the `pageserver.toml` file. The API setting should
-   point to the "upcall" prefix, for example `http://127.0.0.1:1234/upcall/v1/` is used in neon_local clusters.
-2. Create a `metadata.json` file in the same directory as `pageserver.toml`: this enables the pageserver to automatically register itself
-   with the storage controller when it starts up. See the example below for the format of this file.
-
-### Example `metadata.json`
-
-```
-{"host":"acmehost.localdomain","http_host":"acmehost.localdomain","http_port":9898,"port":64000}
-```
-
-- `port` and `host` refer to the _postgres_ port and host, and these must be accessible from wherever
-  postgres runs.
-- `http_port` and `http_host` refer to the pageserver's HTTP api, this must be accessible from where
-  the storage controller runs.
-
-## Handle compute notifications.
-
-The storage controller independently moves tenant attachments between pageservers in response to
-changes such as a pageserver node becoming unavailable, or the tenant's shard count changing. To enable
-postgres clients to handle such changes, the storage controller calls an API hook when a tenant's pageserver
-location changes.
-
-The hook is configured using the storage controller's `--compute-hook-url` CLI option. If the hook requires
-JWT auth, the token may be provided with `--control-plane-jwt-token`. The hook will be invoked with a `PUT` request.
-
-In the Neon cloud service, this hook is implemented by Neon's internal cloud control plane. In `neon_local` systems
-the storage controller integrates directly with neon_local to reconfigure local postgres processes instead of calling
-the compute hook.
-
-When implementing an on-premise Neon deployment, you must implement a service that handles the compute hook. This is not complicated:
-the request body has format of the `ComputeHookNotifyRequest` structure, provided below for convenience.
-
-```
-struct ComputeHookNotifyRequestShard {
-    node_id: NodeId,
-    shard_number: ShardNumber,
-}
-
-struct ComputeHookNotifyRequest {
-    tenant_id: TenantId,
-    stripe_size: Option<ShardStripeSize>,
-    shards: Vec<ComputeHookNotifyRequestShard>,
-}
-```
-
-When a notification is received:
-
-1. Modify postgres configuration for this tenant:
-
-   - set `neon.pageserver_connstr` to a comma-separated list of postgres connection strings to pageservers according to the `shards` list. The
-     shards identified by `NodeId` must be converted to the address+port of the node.
-   - if stripe_size is not None, set `neon.stripe_size` to this value
-
-2. Send SIGHUP to postgres to reload configuration
-3. Respond with 200 to the notification request. Do not return success if postgres was not updated: if an error is returned, the controller
-   will retry the notification until it succeeds..
-
-### Example notification body
-
-```
-{
-  "tenant_id": "1f359dd625e519a1a4e8d7509690f6fc",
-  "stripe_size": 32768,
-  "shards": [
-      {"node_id": 344, "shard_number": 0},
-      {"node_id": 722, "shard_number": 1},
-  ],
-}
-```

flake.lock

@@ -1,477 +0,0 @@
-{
-  "nodes": {
-    "cachix": {
-      "inputs": {
-        "devenv": "devenv_2",
-        "flake-compat": [
-          "devenv",
-          "flake-compat"
-        ],
-        "nixpkgs": [
-          "devenv",
-          "nixpkgs"
-        ],
-        "pre-commit-hooks": [
-          "devenv",
-          "pre-commit-hooks"
-        ]
-      },
-      "locked": {
-        "lastModified": 1712055811,
-        "narHash": "sha256-7FcfMm5A/f02yyzuavJe06zLa9hcMHsagE28ADcmQvk=",
-        "owner": "cachix",
-        "repo": "cachix",
-        "rev": "02e38da89851ec7fec3356a5c04bc8349cae0e30",
-        "type": "github"
-      },
-      "original": {
-        "owner": "cachix",
-        "repo": "cachix",
-        "type": "github"
-      }
-    },
-    "devenv": {
-      "inputs": {
-        "cachix": "cachix",
-        "flake-compat": "flake-compat_2",
-        "nix": "nix_2",
-        "nixpkgs": "nixpkgs_2",
-        "pre-commit-hooks": "pre-commit-hooks"
-      },
-      "locked": {
-        "lastModified": 1714390914,
-        "narHash": "sha256-W5DFIifCjGYJXJzLU3RpqBeqes4zrf0Sr/6rwzTygPU=",
-        "owner": "cachix",
-        "repo": "devenv",
-        "rev": "34e6461fd76b5f51ad5f8214f5cf22c4cd7a196e",
-        "type": "github"
-      },
-      "original": {
-        "owner": "cachix",
-        "repo": "devenv",
-        "type": "github"
-      }
-    },
-    "devenv_2": {
-      "inputs": {
-        "flake-compat": [
-          "devenv",
-          "cachix",
-          "flake-compat"
-        ],
-        "nix": "nix",
-        "nixpkgs": "nixpkgs",
-        "poetry2nix": "poetry2nix",
-        "pre-commit-hooks": [
-          "devenv",
-          "cachix",
-          "pre-commit-hooks"
-        ]
-      },
-      "locked": {
-        "lastModified": 1708704632,
-        "narHash": "sha256-w+dOIW60FKMaHI1q5714CSibk99JfYxm0CzTinYWr+Q=",
-        "owner": "cachix",
-        "repo": "devenv",
-        "rev": "2ee4450b0f4b95a1b90f2eb5ffea98b90e48c196",
-        "type": "github"
-      },
-      "original": {
-        "owner": "cachix",
-        "ref": "python-rewrite",
-        "repo": "devenv",
-        "type": "github"
-      }
-    },
-    "flake-compat": {
-      "flake": false,
-      "locked": {
-        "lastModified": 1673956053,
-        "narHash": "sha256-4gtG9iQuiKITOjNQQeQIpoIB6b16fm+504Ch3sNKLd8=",
-        "owner": "edolstra",
-        "repo": "flake-compat",
-        "rev": "35bb57c0c8d8b62bbfd284272c928ceb64ddbde9",
-        "type": "github"
-      },
-      "original": {
-        "owner": "edolstra",
-        "repo": "flake-compat",
-        "type": "github"
-      }
-    },
-    "flake-compat_2": {
-      "flake": false,
-      "locked": {
-        "lastModified": 1696426674,
-        "narHash": "sha256-kvjfFW7WAETZlt09AgDn1MrtKzP7t90Vf7vypd3OL1U=",
-        "owner": "edolstra",
-        "repo": "flake-compat",
-        "rev": "0f9255e01c2351cc7d116c072cb317785dd33b33",
-        "type": "github"
-      },
-      "original": {
-        "owner": "edolstra",
-        "repo": "flake-compat",
-        "type": "github"
-      }
-    },
-    "flake-parts": {
-      "inputs": {
-        "nixpkgs-lib": "nixpkgs-lib"
-      },
-      "locked": {
-        "lastModified": 1712014858,
-        "narHash": "sha256-sB4SWl2lX95bExY2gMFG5HIzvva5AVMJd4Igm+GpZNw=",
-        "owner": "hercules-ci",
-        "repo": "flake-parts",
-        "rev": "9126214d0a59633752a136528f5f3b9aa8565b7d",
-        "type": "github"
-      },
-      "original": {
-        "id": "flake-parts",
-        "type": "indirect"
-      }
-    },
-    "flake-utils": {
-      "inputs": {
-        "systems": "systems"
-      },
-      "locked": {
-        "lastModified": 1689068808,
-        "narHash": "sha256-6ixXo3wt24N/melDWjq70UuHQLxGV8jZvooRanIHXw0=",
-        "owner": "numtide",
-        "repo": "flake-utils",
-        "rev": "919d646de7be200f3bf08cb76ae1f09402b6f9b4",
-        "type": "github"
-      },
-      "original": {
-        "owner": "numtide",
-        "repo": "flake-utils",
-        "type": "github"
-      }
-    },
-    "flake-utils_2": {
-      "inputs": {
-        "systems": "systems_2"
-      },
-      "locked": {
-        "lastModified": 1710146030,
-        "narHash": "sha256-SZ5L6eA7HJ/nmkzGG7/ISclqe6oZdOZTNoesiInkXPQ=",
-        "owner": "numtide",
-        "repo": "flake-utils",
-        "rev": "b1d9ab70662946ef0850d488da1c9019f3a9752a",
-        "type": "github"
-      },
-      "original": {
-        "owner": "numtide",
-        "repo": "flake-utils",
-        "type": "github"
-      }
-    },
-    "gitignore": {
-      "inputs": {
-        "nixpkgs": [
-          "devenv",
-          "pre-commit-hooks",
-          "nixpkgs"
-        ]
-      },
-      "locked": {
-        "lastModified": 1709087332,
-        "narHash": "sha256-HG2cCnktfHsKV0s4XW83gU3F57gaTljL9KNSuG6bnQs=",
-        "owner": "hercules-ci",
-        "repo": "gitignore.nix",
-        "rev": "637db329424fd7e46cf4185293b9cc8c88c95394",
-        "type": "github"
-      },
-      "original": {
-        "owner": "hercules-ci",
-        "repo": "gitignore.nix",
-        "type": "github"
-      }
-    },
-    "nix": {
-      "inputs": {
-        "flake-compat": "flake-compat",
-        "nixpkgs": [
-          "devenv",
-          "cachix",
-          "devenv",
-          "nixpkgs"
-        ],
-        "nixpkgs-regression": "nixpkgs-regression"
-      },
-      "locked": {
-        "lastModified": 1712911606,
-        "narHash": "sha256-BGvBhepCufsjcUkXnEEXhEVjwdJAwPglCC2+bInc794=",
-        "owner": "domenkozar",
-        "repo": "nix",
-        "rev": "b24a9318ea3f3600c1e24b4a00691ee912d4de12",
-        "type": "github"
-      },
-      "original": {
-        "owner": "domenkozar",
-        "ref": "devenv-2.21",
-        "repo": "nix",
-        "type": "github"
-      }
-    },
-    "nix-github-actions": {
-      "inputs": {
-        "nixpkgs": [
-          "devenv",
-          "cachix",
-          "devenv",
-          "poetry2nix",
-          "nixpkgs"
-        ]
-      },
-      "locked": {
-        "lastModified": 1688870561,
-        "narHash": "sha256-4UYkifnPEw1nAzqqPOTL2MvWtm3sNGw1UTYTalkTcGY=",
-        "owner": "nix-community",
-        "repo": "nix-github-actions",
-        "rev": "165b1650b753316aa7f1787f3005a8d2da0f5301",
-        "type": "github"
-      },
-      "original": {
-        "owner": "nix-community",
-        "repo": "nix-github-actions",
-        "type": "github"
-      }
-    },
-    "nix_2": {
-      "inputs": {
-        "flake-compat": [
-          "devenv",
-          "flake-compat"
-        ],
-        "nixpkgs": [
-          "devenv",
-          "nixpkgs"
-        ],
-        "nixpkgs-regression": "nixpkgs-regression_2"
-      },
-      "locked": {
-        "lastModified": 1712911606,
-        "narHash": "sha256-BGvBhepCufsjcUkXnEEXhEVjwdJAwPglCC2+bInc794=",
-        "owner": "domenkozar",
-        "repo": "nix",
-        "rev": "b24a9318ea3f3600c1e24b4a00691ee912d4de12",
-        "type": "github"
-      },
-      "original": {
-        "owner": "domenkozar",
-        "ref": "devenv-2.21",
-        "repo": "nix",
-        "type": "github"
-      }
-    },
-    "nixpkgs": {
-      "locked": {
-        "lastModified": 1692808169,
-        "narHash": "sha256-x9Opq06rIiwdwGeK2Ykj69dNc2IvUH1fY55Wm7atwrE=",
-        "owner": "NixOS",
-        "repo": "nixpkgs",
-        "rev": "9201b5ff357e781bf014d0330d18555695df7ba8",
-        "type": "github"
-      },
-      "original": {
-        "owner": "NixOS",
-        "ref": "nixpkgs-unstable",
-        "repo": "nixpkgs",
-        "type": "github"
-      }
-    },
-    "nixpkgs-lib": {
-      "locked": {
-        "dir": "lib",
-        "lastModified": 1711703276,
-        "narHash": "sha256-iMUFArF0WCatKK6RzfUJknjem0H9m4KgorO/p3Dopkk=",
-        "owner": "NixOS",
-        "repo": "nixpkgs",
-        "rev": "d8fe5e6c92d0d190646fb9f1056741a229980089",
-        "type": "github"
-      },
-      "original": {
-        "dir": "lib",
-        "owner": "NixOS",
-        "ref": "nixos-unstable",
-        "repo": "nixpkgs",
-        "type": "github"
-      }
-    },
-    "nixpkgs-regression": {
-      "locked": {
-        "lastModified": 1643052045,
-        "narHash": "sha256-uGJ0VXIhWKGXxkeNnq4TvV3CIOkUJ3PAoLZ3HMzNVMw=",
-        "owner": "NixOS",
-        "repo": "nixpkgs",
-        "rev": "215d4d0fd80ca5163643b03a33fde804a29cc1e2",
-        "type": "github"
-      },
-      "original": {
-        "owner": "NixOS",
-        "repo": "nixpkgs",
-        "rev": "215d4d0fd80ca5163643b03a33fde804a29cc1e2",
-        "type": "github"
-      }
-    },
-    "nixpkgs-regression_2": {
-      "locked": {
-        "lastModified": 1643052045,
-        "narHash": "sha256-uGJ0VXIhWKGXxkeNnq4TvV3CIOkUJ3PAoLZ3HMzNVMw=",
-        "owner": "NixOS",
-        "repo": "nixpkgs",
-        "rev": "215d4d0fd80ca5163643b03a33fde804a29cc1e2",
-        "type": "github"
-      },
-      "original": {
-        "owner": "NixOS",
-        "repo": "nixpkgs",
-        "rev": "215d4d0fd80ca5163643b03a33fde804a29cc1e2",
-        "type": "github"
-      }
-    },
-    "nixpkgs-stable": {
-      "locked": {
-        "lastModified": 1710695816,
-        "narHash": "sha256-3Eh7fhEID17pv9ZxrPwCLfqXnYP006RKzSs0JptsN84=",
-        "owner": "NixOS",
-        "repo": "nixpkgs",
-        "rev": "614b4613980a522ba49f0d194531beddbb7220d3",
-        "type": "github"
-      },
-      "original": {
-        "owner": "NixOS",
-        "ref": "nixos-23.11",
-        "repo": "nixpkgs",
-        "type": "github"
-      }
-    },
-    "nixpkgs_2": {
-      "locked": {
-        "lastModified": 1713361204,
-        "narHash": "sha256-TA6EDunWTkc5FvDCqU3W2T3SFn0gRZqh6D/hJnM02MM=",
-        "owner": "cachix",
-        "repo": "devenv-nixpkgs",
-        "rev": "285676e87ad9f0ca23d8714a6ab61e7e027020c6",
-        "type": "github"
-      },
-      "original": {
-        "owner": "cachix",
-        "ref": "rolling",
-        "repo": "devenv-nixpkgs",
-        "type": "github"
-      }
-    },
-    "nixpkgs_3": {
-      "locked": {
-        "lastModified": 1714314149,
-        "narHash": "sha256-yNAevSKF4krRWacmLUsLK7D7PlfuY3zF0lYnGYNi9vQ=",
-        "owner": "nixos",
-        "repo": "nixpkgs",
-        "rev": "cf8cc1201be8bc71b7cbbbdaf349b22f4f99c7ae",
-        "type": "github"
-      },
-      "original": {
-        "owner": "nixos",
-        "ref": "nixpkgs-unstable",
-        "repo": "nixpkgs",
-        "type": "github"
-      }
-    },
-    "poetry2nix": {
-      "inputs": {
-        "flake-utils": "flake-utils",
-        "nix-github-actions": "nix-github-actions",
-        "nixpkgs": [
-          "devenv",
-          "cachix",
-          "devenv",
-          "nixpkgs"
-        ]
-      },
-      "locked": {
-        "lastModified": 1692876271,
-        "narHash": "sha256-IXfZEkI0Mal5y1jr6IRWMqK8GW2/f28xJenZIPQqkY0=",
-        "owner": "nix-community",
-        "repo": "poetry2nix",
-        "rev": "d5006be9c2c2417dafb2e2e5034d83fabd207ee3",
-        "type": "github"
-      },
-      "original": {
-        "owner": "nix-community",
-        "repo": "poetry2nix",
-        "type": "github"
-      }
-    },
-    "pre-commit-hooks": {
-      "inputs": {
-        "flake-compat": [
-          "devenv",
-          "flake-compat"
-        ],
-        "flake-utils": "flake-utils_2",
-        "gitignore": "gitignore",
-        "nixpkgs": [
-          "devenv",
-          "nixpkgs"
-        ],
-        "nixpkgs-stable": "nixpkgs-stable"
-      },
-      "locked": {
-        "lastModified": 1713775815,
-        "narHash": "sha256-Wu9cdYTnGQQwtT20QQMg7jzkANKQjwBD9iccfGKkfls=",
-        "owner": "cachix",
-        "repo": "pre-commit-hooks.nix",
-        "rev": "2ac4dcbf55ed43f3be0bae15e181f08a57af24a4",
-        "type": "github"
-      },
-      "original": {
-        "owner": "cachix",
-        "repo": "pre-commit-hooks.nix",
-        "type": "github"
-      }
-    },
-    "root": {
-      "inputs": {
-        "devenv": "devenv",
-        "flake-parts": "flake-parts",
-        "nixpkgs": "nixpkgs_3"
-      }
-    },
-    "systems": {
-      "locked": {
-        "lastModified": 1681028828,
-        "narHash": "sha256-Vy1rq5AaRuLzOxct8nz4T6wlgyUR7zLU309k9mBC768=",
-        "owner": "nix-systems",
-        "repo": "default",
-        "rev": "da67096a3b9bf56a91d16901293e51ba5b49a27e",
-        "type": "github"
-      },
-      "original": {
-        "owner": "nix-systems",
-        "repo": "default",
-        "type": "github"
-      }
-    },
-    "systems_2": {
-      "locked": {
-        "lastModified": 1681028828,
-        "narHash": "sha256-Vy1rq5AaRuLzOxct8nz4T6wlgyUR7zLU309k9mBC768=",
-        "owner": "nix-systems",
-        "repo": "default",
-        "rev": "da67096a3b9bf56a91d16901293e51ba5b49a27e",
-        "type": "github"
-      },
-      "original": {
-        "owner": "nix-systems",
-        "repo": "default",
-        "type": "github"
-      }
-    }
-  },
-  "root": "root",
-  "version": 7
-}

flake.nix

@@ -1,58 +0,0 @@
-{
-  inputs = {
-    nixpkgs.url = "github:nixos/nixpkgs/nixpkgs-unstable";
-    devenv.url = "github:cachix/devenv";
-  };
-
-  outputs = inputs @ {
-    flake-parts,
-    nixpkgs,
-    ...
-  }:
-    flake-parts.lib.mkFlake {inherit inputs;} {
-      imports = [
-        inputs.devenv.flakeModule
-      ];
-      systems = nixpkgs.lib.systems.flakeExposed;
-
-      perSystem = {
-        config,
-        self',
-        inputs',
-        pkgs,
-        system,
-        lib,
-        ...
-      }: {
-        devenv.shells.default = {
-          packages = with pkgs; [
-            rustup
-            postgresql_16
-            protobuf_26
-            icu74
-            pkg-config
-            bison
-            flex
-            openssl
-            libiconv
-            readline
-            zlib
-            curl
-          ];
-
-          env.DISABLE_HOMEBREW = "1";
-
-          scripts = {
-            neonmake = {
-              description = "Build Neon";
-
-              exec =
-                if (pkgs.stdenv.isDarwin)
-                then "make -j`sysctl -n hw.logicalcpu` -s"
-                else "make -j`nproc` -s";
-            };
-          };
-        };
-      };
-    };
-}

libs/metrics/src/lib.rs

@@ -256,16 +256,7 @@ fn update_rusage_metrics() {
     DISK_IO_BYTES
         .with_label_values(&["write"])
         .set(rusage_stats.ru_oublock * BYTES_IN_BLOCK);
-
-    // On macOS, the unit of maxrss is bytes; on Linux, it's kilobytes. https://stackoverflow.com/a/59915669
-    #[cfg(target_os = "macos")]
-    {
-        MAXRSS_KB.set(rusage_stats.ru_maxrss / 1024);
-    }
-    #[cfg(not(target_os = "macos"))]
-    {
-        MAXRSS_KB.set(rusage_stats.ru_maxrss);
-    }
+    MAXRSS_KB.set(rusage_stats.ru_maxrss);
 }
 
 fn get_rusage_stats() -> libc::rusage {

libs/pageserver_api/src/key.rs

@@ -1,6 +1,5 @@
 use anyhow::{bail, Result};
 use byteorder::{ByteOrder, BE};
-use bytes::BufMut;
 use postgres_ffi::relfile_utils::{FSM_FORKNUM, VISIBILITYMAP_FORKNUM};
 use postgres_ffi::{Oid, TransactionId};
 use serde::{Deserialize, Serialize};
@@ -22,107 +21,15 @@ pub struct Key {
     pub field6: u32,
 }
 
-/// The storage key size.
 pub const KEY_SIZE: usize = 18;
 
-/// The metadata key size. 2B fewer than the storage key size because field2 is not fully utilized.
-/// See [`Key::to_i128`] for more information on the encoding.
-pub const METADATA_KEY_SIZE: usize = 16;
-
-/// The key prefix start range for the metadata keys. All keys with the first byte >= 0x40 is a metadata key.
-pub const METADATA_KEY_BEGIN_PREFIX: u8 = 0x60;
-pub const METADATA_KEY_END_PREFIX: u8 = 0x7F;
-
-/// The (reserved) key prefix of relation sizes.
-pub const RELATION_SIZE_PREFIX: u8 = 0x61;
-
-/// The key prefix of AUX file keys.
-pub const AUX_KEY_PREFIX: u8 = 0x62;
-
-/// Check if the key falls in the range of metadata keys.
-pub const fn is_metadata_key_slice(key: &[u8]) -> bool {
-    key[0] >= METADATA_KEY_BEGIN_PREFIX && key[0] < METADATA_KEY_END_PREFIX
-}
-
 impl Key {
-    /// Check if the key falls in the range of metadata keys.
-    pub const fn is_metadata_key(&self) -> bool {
-        self.field1 >= METADATA_KEY_BEGIN_PREFIX && self.field1 < METADATA_KEY_END_PREFIX
-    }
-
-    /// Encode a metadata key to a storage key.
-    pub fn from_metadata_key_fixed_size(key: &[u8; METADATA_KEY_SIZE]) -> Self {
-        assert!(is_metadata_key_slice(key), "key not in metadata key range");
-        Key {
-            field1: key[0],
-            field2: u16::from_be_bytes(key[1..3].try_into().unwrap()) as u32,
-            field3: u32::from_be_bytes(key[3..7].try_into().unwrap()),
-            field4: u32::from_be_bytes(key[7..11].try_into().unwrap()),
-            field5: key[11],
-            field6: u32::from_be_bytes(key[12..16].try_into().unwrap()),
-        }
-    }
-
-    /// Encode a metadata key to a storage key.
-    pub fn from_metadata_key(key: &[u8]) -> Self {
-        Self::from_metadata_key_fixed_size(key.try_into().expect("expect 16 byte metadata key"))
-    }
-
-    /// Extract a metadata key to a writer. The result should always be 16 bytes.
-    pub fn extract_metadata_key_to_writer(&self, mut writer: impl BufMut) {
-        writer.put_u8(self.field1);
-        assert!(self.field2 <= 0xFFFF);
-        writer.put_u16(self.field2 as u16);
-        writer.put_u32(self.field3);
-        writer.put_u32(self.field4);
-        writer.put_u8(self.field5);
-        writer.put_u32(self.field6);
-    }
-
-    /// Get the range of metadata keys.
-    pub fn metadata_key_range() -> Range<Self> {
-        Key {
-            field1: METADATA_KEY_BEGIN_PREFIX,
-            field2: 0,
-            field3: 0,
-            field4: 0,
-            field5: 0,
-            field6: 0,
-        }..Key {
-            field1: METADATA_KEY_END_PREFIX,
-            field2: 0,
-            field3: 0,
-            field4: 0,
-            field5: 0,
-            field6: 0,
-        }
-    }
-
-    /// Get the range of aux keys.
-    pub fn metadata_aux_key_range() -> Range<Self> {
-        Key {
-            field1: AUX_KEY_PREFIX,
-            field2: 0,
-            field3: 0,
-            field4: 0,
-            field5: 0,
-            field6: 0,
-        }..Key {
-            field1: AUX_KEY_PREFIX + 1,
-            field2: 0,
-            field3: 0,
-            field4: 0,
-            field5: 0,
-            field6: 0,
-        }
-    }
-
     /// 'field2' is used to store tablespaceid for relations and small enum numbers for other relish.
     /// As long as Neon does not support tablespace (because of lack of access to local file system),
     /// we can assume that only some predefined namespace OIDs are used which can fit in u16
     pub fn to_i128(&self) -> i128 {
         assert!(self.field2 < 0xFFFF || self.field2 == 0xFFFFFFFF || self.field2 == 0x22222222);
-        (((self.field1 & 0x7F) as i128) << 120)
+        (((self.field1 & 0xf) as i128) << 120)
             | (((self.field2 & 0xFFFF) as i128) << 104)
             | ((self.field3 as i128) << 72)
             | ((self.field4 as i128) << 40)
@@ -132,7 +39,7 @@ impl Key {
 
     pub const fn from_i128(x: i128) -> Self {
         Key {
-            field1: ((x >> 120) & 0x7F) as u8,
+            field1: ((x >> 120) & 0xf) as u8,
             field2: ((x >> 104) & 0xFFFF) as u32,
             field3: (x >> 72) as u32,
             field4: (x >> 40) as u32,
@@ -141,11 +48,11 @@ impl Key {
         }
     }
 
-    pub const fn next(&self) -> Key {
+    pub fn next(&self) -> Key {
         self.add(1)
     }
 
-    pub const fn add(&self, x: u32) -> Key {
+    pub fn add(&self, x: u32) -> Key {
         let mut key = *self;
 
         let r = key.field6.overflowing_add(x);
@@ -174,8 +81,6 @@ impl Key {
         key
     }
 
-    /// Convert a 18B slice to a key. This function should not be used for metadata keys because field2 is handled differently.
-    /// Use [`Key::from_metadata_key`] instead.
     pub fn from_slice(b: &[u8]) -> Self {
         Key {
             field1: b[0],
@@ -187,8 +92,6 @@ impl Key {
         }
     }
 
-    /// Convert a key to a 18B slice. This function should not be used for metadata keys because field2 is handled differently.
-    /// Use [`Key::extract_metadata_key_to_writer`] instead.
     pub fn write_to_byte_slice(&self, buf: &mut [u8]) {
         buf[0] = self.field1;
         BE::write_u32(&mut buf[1..5], self.field2);
@@ -572,14 +475,12 @@ pub const AUX_FILES_KEY: Key = Key {
 // Reverse mappings for a few Keys.
 // These are needed by WAL redo manager.
 
-pub const NON_INHERITED_RANGE: Range<Key> = AUX_FILES_KEY..AUX_FILES_KEY.next();
-
 // AUX_FILES currently stores only data for logical replication (slots etc), and
 // we don't preserve these on a branch because safekeepers can't follow timeline
 // switch (and generally it likely should be optional), so ignore these.
 #[inline(always)]
 pub fn is_inherited_key(key: Key) -> bool {
-    !NON_INHERITED_RANGE.contains(&key)
+    key != AUX_FILES_KEY
 }
 
 #[inline(always)]
@@ -655,14 +556,11 @@ impl std::str::FromStr for Key {
 mod tests {
     use std::str::FromStr;
 
-    use crate::key::is_metadata_key_slice;
     use crate::key::Key;
 
     use rand::Rng;
     use rand::SeedableRng;
 
-    use super::AUX_KEY_PREFIX;
-
     #[test]
     fn display_fromstr_bijection() {
         let mut rng = rand::rngs::StdRng::seed_from_u64(42);
@@ -678,16 +576,4 @@ mod tests {
 
         assert_eq!(key, Key::from_str(&format!("{key}")).unwrap());
     }
-
-    #[test]
-    fn test_metadata_keys() {
-        let mut metadata_key = vec![AUX_KEY_PREFIX];
-        metadata_key.extend_from_slice(&[0xFF; 15]);
-        let encoded_key = Key::from_metadata_key(&metadata_key);
-        let mut output_key = Vec::new();
-        encoded_key.extract_metadata_key_to_writer(&mut output_key);
-        assert_eq!(metadata_key, output_key);
-        assert!(encoded_key.is_metadata_key());
-        assert!(is_metadata_key_slice(&metadata_key));
-    }
 }

libs/pageserver_api/src/keyspace.rs

@@ -15,13 +15,7 @@ pub struct KeySpace {
 }
 
 impl KeySpace {
-    /// Create a key space with a single range.
-    pub fn single(key_range: Range<Key>) -> Self {
-        Self {
-            ranges: vec![key_range],
-        }
-    }
-
+    ///
     /// Partition a key space into roughly chunks of roughly 'target_size' bytes
     /// in each partition.
     ///
@@ -70,10 +64,6 @@ impl KeySpace {
         KeyPartitioning { parts }
     }
 
-    pub fn is_empty(&self) -> bool {
-        self.total_size() == 0
-    }
-
     /// Merge another keyspace into the current one.
     /// Note: the keyspaces must not ovelap (enforced via assertions)
     pub fn merge(&mut self, other: &KeySpace) {
@@ -104,13 +94,12 @@ impl KeySpace {
 
     /// Remove all keys in `other` from `self`.
     /// This can involve splitting or removing of existing ranges.
-    /// Returns the removed keyspace
-    pub fn remove_overlapping_with(&mut self, other: &KeySpace) -> KeySpace {
+    pub fn remove_overlapping_with(&mut self, other: &KeySpace) {
         let (self_start, self_end) = match (self.start(), self.end()) {
             (Some(start), Some(end)) => (start, end),
             _ => {
                 // self is empty
-                return KeySpace::default();
+                return;
             }
         };
 
@@ -123,37 +112,30 @@ impl KeySpace {
             .skip_while(|range| self_start >= range.end)
             .take_while(|range| self_end > range.start);
 
-        let mut removed_accum = KeySpaceRandomAccum::new();
         for range in other_ranges {
             while let Some(overlap_at) = self.overlaps_at(range) {
                 let overlapped = self.ranges[overlap_at].clone();
 
                 if overlapped.start < range.start && overlapped.end <= range.end {
                     // Higher part of the range is completely overlapped.
-                    removed_accum.add_range(range.start..self.ranges[overlap_at].end);
                     self.ranges[overlap_at].end = range.start;
                 }
                 if overlapped.start >= range.start && overlapped.end > range.end {
                     // Lower part of the range is completely overlapped.
-                    removed_accum.add_range(self.ranges[overlap_at].start..range.end);
                     self.ranges[overlap_at].start = range.end;
                 }
                 if overlapped.start < range.start && overlapped.end > range.end {
                     // Middle part of the range is overlapped.
-                    removed_accum.add_range(range.clone());
                     self.ranges[overlap_at].end = range.start;
                     self.ranges
                         .insert(overlap_at + 1, range.end..overlapped.end);
                 }
                 if overlapped.start >= range.start && overlapped.end <= range.end {
                     // Whole range is overlapped
-                    removed_accum.add_range(self.ranges[overlap_at].clone());
                     self.ranges.remove(overlap_at);
                 }
             }
         }
-
-        removed_accum.to_keyspace()
     }
 
     pub fn start(&self) -> Option<Key> {
@@ -188,11 +170,6 @@ impl KeySpace {
     pub fn overlaps(&self, range: &Range<Key>) -> bool {
         self.overlaps_at(range).is_some()
     }
-
-    /// Check if the keyspace contains a key
-    pub fn contains(&self, key: &Key) -> bool {
-        self.overlaps(&(*key..key.next()))
-    }
 }
 
 ///
@@ -576,16 +553,7 @@ mod tests {
                 Key::from_i128(11)..Key::from_i128(13),
             ],
         };
-        let removed = key_space1.remove_overlapping_with(&key_space2);
-        let removed_expected = KeySpace {
-            ranges: vec![
-                Key::from_i128(2)..Key::from_i128(3),
-                Key::from_i128(6)..Key::from_i128(7),
-                Key::from_i128(11)..Key::from_i128(12),
-            ],
-        };
-        assert_eq!(removed, removed_expected);
-
+        key_space1.remove_overlapping_with(&key_space2);
         assert_eq!(
             key_space1.ranges,
             vec![
@@ -615,17 +583,7 @@ mod tests {
                 Key::from_i128(14)..Key::from_i128(17),
             ],
         };
-
-        let removed = key_space1.remove_overlapping_with(&key_space2);
-        let removed_expected = KeySpace {
-            ranges: vec![
-                Key::from_i128(3)..Key::from_i128(5),
-                Key::from_i128(8)..Key::from_i128(10),
-                Key::from_i128(14)..Key::from_i128(15),
-            ],
-        };
-        assert_eq!(removed, removed_expected);
-
+        key_space1.remove_overlapping_with(&key_space2);
         assert_eq!(
             key_space1.ranges,
             vec![
@@ -652,11 +610,7 @@ mod tests {
                 Key::from_i128(15)..Key::from_i128(17),
             ],
         };
-
-        let removed = key_space1.remove_overlapping_with(&key_space2);
-        let removed_expected = KeySpace::default();
-        assert_eq!(removed, removed_expected);
-
+        key_space1.remove_overlapping_with(&key_space2);
         assert_eq!(
             key_space1.ranges,
             vec![
@@ -683,17 +637,7 @@ mod tests {
         let key_space2 = KeySpace {
             ranges: vec![Key::from_i128(9)..Key::from_i128(19)],
         };
-
-        let removed = key_space1.remove_overlapping_with(&key_space2);
-        let removed_expected = KeySpace {
-            ranges: vec![
-                Key::from_i128(9)..Key::from_i128(10),
-                Key::from_i128(12)..Key::from_i128(15),
-                Key::from_i128(17)..Key::from_i128(19),
-            ],
-        };
-        assert_eq!(removed, removed_expected);
-
+        key_space1.remove_overlapping_with(&key_space2);
         assert_eq!(
             key_space1.ranges,
             vec![

libs/pageserver_api/src/models.rs

@@ -303,7 +303,6 @@ pub struct TenantConfig {
     pub lazy_slru_download: Option<bool>,
     pub timeline_get_throttle: Option<ThrottleConfig>,
     pub image_layer_creation_check_threshold: Option<u8>,
-    pub switch_to_aux_file_v2: Option<bool>,
 }
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
@@ -430,7 +429,6 @@ pub struct StatusResponse {
 #[derive(Serialize, Deserialize, Debug)]
 #[serde(deny_unknown_fields)]
 pub struct TenantLocationConfigRequest {
-    #[serde(skip_serializing_if = "Option::is_none")]
     pub tenant_id: Option<TenantShardId>,
     #[serde(flatten)]
     pub config: LocationConfig, // as we have a flattened field, we should reject all unknown fields in it
@@ -782,17 +780,6 @@ pub struct SecondaryProgress {
     pub bytes_total: u64,
 }
 
-#[derive(Serialize, Deserialize, Debug)]
-pub struct TenantScanRemoteStorageShard {
-    pub tenant_shard_id: TenantShardId,
-    pub generation: Option<u32>,
-}
-
-#[derive(Serialize, Deserialize, Debug, Default)]
-pub struct TenantScanRemoteStorageResponse {
-    pub shards: Vec<TenantScanRemoteStorageShard>,
-}
-
 pub mod virtual_file {
     #[derive(
         Copy,
@@ -860,72 +847,39 @@ impl TryFrom<u8> for PagestreamBeMessageTag {
     }
 }
 
-// In the V2 protocol version, a GetPage request contains two LSN values:
-//
-// request_lsn: Get the page version at this point in time.  Lsn::Max is a special value that means
-// "get the latest version present". It's used by the primary server, which knows that no one else
-// is writing WAL. 'not_modified_since' must be set to a proper value even if request_lsn is
-// Lsn::Max. Standby servers use the current replay LSN as the request LSN.
-//
-// not_modified_since: Hint to the pageserver that the client knows that the page has not been
-// modified between 'not_modified_since' and the request LSN. It's always correct to set
-// 'not_modified_since equal' to 'request_lsn' (unless Lsn::Max is used as the 'request_lsn'), but
-// passing an earlier LSN can speed up the request, by allowing the pageserver to process the
-// request without waiting for 'request_lsn' to arrive.
-//
-// The legacy V1 interface contained only one LSN, and a boolean 'latest' flag. The V1 interface was
-// sufficient for the primary; the 'lsn' was equivalent to the 'not_modified_since' value, and
-// 'latest' was set to true. The V2 interface was added because there was no correct way for a
-// standby to request a page at a particular non-latest LSN, and also include the
-// 'not_modified_since' hint. That led to an awkward choice of either using an old LSN in the
-// request, if the standby knows that the page hasn't been modified since, and risk getting an error
-// if that LSN has fallen behind the GC horizon, or requesting the current replay LSN, which could
-// require the pageserver unnecessarily to wait for the WAL to arrive up to that point. The new V2
-// interface allows sending both LSNs, and let the pageserver do the right thing. There is no
-// difference in the responses between V1 and V2.
-//
-// The Request structs below reflect the V2 interface. If V1 is used, the parse function
-// maps the old format requests to the new format.
-//
-#[derive(Clone, Copy)]
-pub enum PagestreamProtocolVersion {
-    V1,
-    V2,
-}
-
 #[derive(Debug, PartialEq, Eq)]
 pub struct PagestreamExistsRequest {
-    pub request_lsn: Lsn,
-    pub not_modified_since: Lsn,
+    pub latest: bool,
+    pub lsn: Lsn,
     pub rel: RelTag,
 }
 
 #[derive(Debug, PartialEq, Eq)]
 pub struct PagestreamNblocksRequest {
-    pub request_lsn: Lsn,
-    pub not_modified_since: Lsn,
+    pub latest: bool,
+    pub lsn: Lsn,
     pub rel: RelTag,
 }
 
 #[derive(Debug, PartialEq, Eq)]
 pub struct PagestreamGetPageRequest {
-    pub request_lsn: Lsn,
-    pub not_modified_since: Lsn,
+    pub latest: bool,
+    pub lsn: Lsn,
     pub rel: RelTag,
     pub blkno: u32,
 }
 
 #[derive(Debug, PartialEq, Eq)]
 pub struct PagestreamDbSizeRequest {
-    pub request_lsn: Lsn,
-    pub not_modified_since: Lsn,
+    pub latest: bool,
+    pub lsn: Lsn,
     pub dbnode: u32,
 }
 
 #[derive(Debug, PartialEq, Eq)]
 pub struct PagestreamGetSlruSegmentRequest {
-    pub request_lsn: Lsn,
-    pub not_modified_since: Lsn,
+    pub latest: bool,
+    pub lsn: Lsn,
     pub kind: u8,
     pub segno: u32,
 }
@@ -972,16 +926,14 @@ pub struct TenantHistorySize {
 }
 
 impl PagestreamFeMessage {
-    /// Serialize a compute -> pageserver message. This is currently only used in testing
-    /// tools. Always uses protocol version 2.
     pub fn serialize(&self) -> Bytes {
         let mut bytes = BytesMut::new();
 
         match self {
             Self::Exists(req) => {
                 bytes.put_u8(0);
-                bytes.put_u64(req.request_lsn.0);
-                bytes.put_u64(req.not_modified_since.0);
+                bytes.put_u8(u8::from(req.latest));
+                bytes.put_u64(req.lsn.0);
                 bytes.put_u32(req.rel.spcnode);
                 bytes.put_u32(req.rel.dbnode);
                 bytes.put_u32(req.rel.relnode);
@@ -990,8 +942,8 @@ impl PagestreamFeMessage {
 
             Self::Nblocks(req) => {
                 bytes.put_u8(1);
-                bytes.put_u64(req.request_lsn.0);
-                bytes.put_u64(req.not_modified_since.0);
+                bytes.put_u8(u8::from(req.latest));
+                bytes.put_u64(req.lsn.0);
                 bytes.put_u32(req.rel.spcnode);
                 bytes.put_u32(req.rel.dbnode);
                 bytes.put_u32(req.rel.relnode);
@@ -1000,8 +952,8 @@ impl PagestreamFeMessage {
 
             Self::GetPage(req) => {
                 bytes.put_u8(2);
-                bytes.put_u64(req.request_lsn.0);
-                bytes.put_u64(req.not_modified_since.0);
+                bytes.put_u8(u8::from(req.latest));
+                bytes.put_u64(req.lsn.0);
                 bytes.put_u32(req.rel.spcnode);
                 bytes.put_u32(req.rel.dbnode);
                 bytes.put_u32(req.rel.relnode);
@@ -1011,15 +963,15 @@ impl PagestreamFeMessage {
 
             Self::DbSize(req) => {
                 bytes.put_u8(3);
-                bytes.put_u64(req.request_lsn.0);
-                bytes.put_u64(req.not_modified_since.0);
+                bytes.put_u8(u8::from(req.latest));
+                bytes.put_u64(req.lsn.0);
                 bytes.put_u32(req.dbnode);
             }
 
             Self::GetSlruSegment(req) => {
                 bytes.put_u8(4);
-                bytes.put_u64(req.request_lsn.0);
-                bytes.put_u64(req.not_modified_since.0);
+                bytes.put_u8(u8::from(req.latest));
+                bytes.put_u64(req.lsn.0);
                 bytes.put_u8(req.kind);
                 bytes.put_u32(req.segno);
             }
@@ -1028,40 +980,18 @@ impl PagestreamFeMessage {
         bytes.into()
     }
 
-    pub fn parse<R: std::io::Read>(
-        body: &mut R,
-        protocol_version: PagestreamProtocolVersion,
-    ) -> anyhow::Result<PagestreamFeMessage> {
+    pub fn parse<R: std::io::Read>(body: &mut R) -> anyhow::Result<PagestreamFeMessage> {
+        // TODO these gets can fail
+
         // these correspond to the NeonMessageTag enum in pagestore_client.h
         //
         // TODO: consider using protobuf or serde bincode for less error prone
         // serialization.
         let msg_tag = body.read_u8()?;
-
-        let (request_lsn, not_modified_since) = match protocol_version {
-            PagestreamProtocolVersion::V2 => (
-                Lsn::from(body.read_u64::<BigEndian>()?),
-                Lsn::from(body.read_u64::<BigEndian>()?),
-            ),
-            PagestreamProtocolVersion::V1 => {
-                // In the old protocol, each message starts with a boolean 'latest' flag,
-                // followed by 'lsn'. Convert that to the two LSNs, 'request_lsn' and
-                // 'not_modified_since', used in the new protocol version.
-                let latest = body.read_u8()? != 0;
-                let request_lsn = Lsn::from(body.read_u64::<BigEndian>()?);
-                if latest {
-                    (Lsn::MAX, request_lsn) // get latest version
-                } else {
-                    (request_lsn, request_lsn) // get version at specified LSN
-                }
-            }
-        };
-
-        // The rest of the messages are the same between V1 and V2
         match msg_tag {
             0 => Ok(PagestreamFeMessage::Exists(PagestreamExistsRequest {
-                request_lsn,
-                not_modified_since,
+                latest: body.read_u8()? != 0,
+                lsn: Lsn::from(body.read_u64::<BigEndian>()?),
                 rel: RelTag {
                     spcnode: body.read_u32::<BigEndian>()?,
                     dbnode: body.read_u32::<BigEndian>()?,
@@ -1070,8 +1000,8 @@ impl PagestreamFeMessage {
                 },
             })),
             1 => Ok(PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
-                request_lsn,
-                not_modified_since,
+                latest: body.read_u8()? != 0,
+                lsn: Lsn::from(body.read_u64::<BigEndian>()?),
                 rel: RelTag {
                     spcnode: body.read_u32::<BigEndian>()?,
                     dbnode: body.read_u32::<BigEndian>()?,
@@ -1080,8 +1010,8 @@ impl PagestreamFeMessage {
                 },
             })),
             2 => Ok(PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
-                request_lsn,
-                not_modified_since,
+                latest: body.read_u8()? != 0,
+                lsn: Lsn::from(body.read_u64::<BigEndian>()?),
                 rel: RelTag {
                     spcnode: body.read_u32::<BigEndian>()?,
                     dbnode: body.read_u32::<BigEndian>()?,
@@ -1091,14 +1021,14 @@ impl PagestreamFeMessage {
                 blkno: body.read_u32::<BigEndian>()?,
             })),
             3 => Ok(PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
-                request_lsn,
-                not_modified_since,
+                latest: body.read_u8()? != 0,
+                lsn: Lsn::from(body.read_u64::<BigEndian>()?),
                 dbnode: body.read_u32::<BigEndian>()?,
             })),
             4 => Ok(PagestreamFeMessage::GetSlruSegment(
                 PagestreamGetSlruSegmentRequest {
-                    request_lsn,
-                    not_modified_since,
+                    latest: body.read_u8()? != 0,
+                    lsn: Lsn::from(body.read_u64::<BigEndian>()?),
                     kind: body.read_u8()?,
                     segno: body.read_u32::<BigEndian>()?,
                 },
@@ -1226,8 +1156,8 @@ mod tests {
         // Test serialization/deserialization of PagestreamFeMessage
         let messages = vec![
             PagestreamFeMessage::Exists(PagestreamExistsRequest {
-                request_lsn: Lsn(4),
-                not_modified_since: Lsn(3),
+                latest: true,
+                lsn: Lsn(4),
                 rel: RelTag {
                     forknum: 1,
                     spcnode: 2,
@@ -1236,8 +1166,8 @@ mod tests {
                 },
             }),
             PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
-                request_lsn: Lsn(4),
-                not_modified_since: Lsn(4),
+                latest: false,
+                lsn: Lsn(4),
                 rel: RelTag {
                     forknum: 1,
                     spcnode: 2,
@@ -1246,8 +1176,8 @@ mod tests {
                 },
             }),
             PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
-                request_lsn: Lsn(4),
-                not_modified_since: Lsn(3),
+                latest: true,
+                lsn: Lsn(4),
                 rel: RelTag {
                     forknum: 1,
                     spcnode: 2,
@@ -1257,16 +1187,14 @@ mod tests {
                 blkno: 7,
             }),
             PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
-                request_lsn: Lsn(4),
-                not_modified_since: Lsn(3),
+                latest: true,
+                lsn: Lsn(4),
                 dbnode: 7,
             }),
         ];
         for msg in messages {
             let bytes = msg.serialize();
-            let reconstructed =
-                PagestreamFeMessage::parse(&mut bytes.reader(), PagestreamProtocolVersion::V2)
-                    .unwrap();
+            let reconstructed = PagestreamFeMessage::parse(&mut bytes.reader()).unwrap();
             assert!(msg == reconstructed);
         }
     }

libs/pageserver_api/src/shard.rs

@@ -5,7 +5,6 @@ use crate::{
     models::ShardParameters,
 };
 use hex::FromHex;
-use postgres_ffi::relfile_utils::INIT_FORKNUM;
 use serde::{Deserialize, Serialize};
 use utils::id::TenantId;
 
@@ -538,24 +537,6 @@ impl ShardIdentity {
         }
     }
 
-    /// Special case for issue `<https://github.com/neondatabase/neon/issues/7451>`
-    ///
-    /// When we fail to read a forknum block, this function tells us whether we may ignore the error
-    /// as a symptom of that issue.
-    pub fn is_key_buggy_forknum(&self, key: &Key) -> bool {
-        if !is_rel_block_key(key) || key.field5 != INIT_FORKNUM {
-            return false;
-        }
-
-        let mut hash = murmurhash32(key.field4);
-        hash = hash_combine(hash, murmurhash32(key.field6 / self.stripe_size.0));
-        let mapped_shard = ShardNumber((hash % self.count.0 as u32) as u8);
-
-        // The key may be affected by issue #7454: it is an initfork and it would not
-        // have mapped to shard 0 until we fixed that issue.
-        mapped_shard != ShardNumber(0)
-    }
-
     /// Return true if the key should be discarded if found in this shard's
     /// data store, e.g. during compaction after a split.
     ///
@@ -668,13 +649,7 @@ fn key_is_shard0(key: &Key) -> bool {
     // relation pages are distributed to shards other than shard zero. Everything else gets
     // stored on shard 0.  This guarantees that shard 0 can independently serve basebackup
     // requests, and any request other than those for particular blocks in relations.
-    //
-    // The only exception to this rule is "initfork" data -- this relates to postgres's UNLOGGED table
-    // type. These are special relations, usually with only 0 or 1 blocks, and we store them on shard 0
-    // because they must be included in basebackups.
-    let is_initfork = key.field5 == INIT_FORKNUM;
-
-    !is_rel_block_key(key) || is_initfork
+    !is_rel_block_key(key)
 }
 
 /// Provide the same result as the function in postgres `hashfn.h` with the same name

libs/postgres_ffi/src/lib.rs

@@ -118,9 +118,7 @@ pub use v14::bindings::{TimeLineID, TimestampTz, XLogRecPtr, XLogSegNo};
 // Likewise for these, although the assumption that these don't change is a little more iffy.
 pub use v14::bindings::{MultiXactOffset, MultiXactStatus};
 pub use v14::bindings::{PageHeaderData, XLogRecord};
-pub use v14::xlog_utils::{
-    XLOG_SIZE_OF_XLOG_LONG_PHD, XLOG_SIZE_OF_XLOG_RECORD, XLOG_SIZE_OF_XLOG_SHORT_PHD,
-};
+pub use v14::xlog_utils::{XLOG_SIZE_OF_XLOG_RECORD, XLOG_SIZE_OF_XLOG_SHORT_PHD};
 
 pub use v14::bindings::{CheckPoint, ControlFileData};
 

libs/postgres_ffi/wal_craft/src/lib.rs

@@ -4,9 +4,7 @@ use log::*;
 use postgres::types::PgLsn;
 use postgres::Client;
 use postgres_ffi::{WAL_SEGMENT_SIZE, XLOG_BLCKSZ};
-use postgres_ffi::{
-    XLOG_SIZE_OF_XLOG_LONG_PHD, XLOG_SIZE_OF_XLOG_RECORD, XLOG_SIZE_OF_XLOG_SHORT_PHD,
-};
+use postgres_ffi::{XLOG_SIZE_OF_XLOG_RECORD, XLOG_SIZE_OF_XLOG_SHORT_PHD};
 use std::path::{Path, PathBuf};
 use std::process::Command;
 use std::time::{Duration, Instant};
@@ -264,21 +262,11 @@ fn craft_internal<C: postgres::GenericClient>(
         intermediate_lsns.insert(0, initial_lsn);
     }
 
-    // Some records may be not flushed, e.g. non-transactional logical messages. Flush now.
+    // Some records may be not flushed, e.g. non-transactional logical messages.
     //
-    // If the previous WAL record ended exactly at page boundary, pg_current_wal_insert_lsn
-    // returns the position just after the page header on the next page. That's where the next
-    // record will be inserted. But the page header hasn't actually been written to the WAL
-    // yet, and if you try to flush it, you get a "request to flush past end of generated WAL"
-    // error. Because of that, if the insert location is just after a page header, back off to
-    // previous page boundary.
-    let mut lsn = u64::from(client.pg_current_wal_insert_lsn()?);
-    if lsn % WAL_SEGMENT_SIZE as u64 == XLOG_SIZE_OF_XLOG_LONG_PHD as u64 {
-        lsn -= XLOG_SIZE_OF_XLOG_LONG_PHD as u64;
-    } else if lsn % XLOG_BLCKSZ as u64 == XLOG_SIZE_OF_XLOG_SHORT_PHD as u64 {
-        lsn -= XLOG_SIZE_OF_XLOG_SHORT_PHD as u64;
-    }
-    client.execute("select neon_xlogflush($1)", &[&PgLsn::from(lsn)])?;
+    // Note: this is broken if pg_current_wal_insert_lsn is at page boundary
+    // because pg_current_wal_insert_lsn skips page headers.
+    client.execute("select neon_xlogflush(pg_current_wal_insert_lsn())", &[])?;
     Ok(intermediate_lsns)
 }
 
@@ -332,49 +320,38 @@ impl Crafter for LastWalRecordXlogSwitchEndsOnPageBoundary {
 
         client.execute("CREATE table t(x int)", &[])?;
 
-        // Add padding so the XLOG_SWITCH record ends exactly on XLOG_BLCKSZ boundary.  We
-        // will use carefully-sized logical messages to advance WAL insert location such
-        // that there is just enough space on the page for the XLOG_SWITCH record.
-        loop {
-            // We start with measuring how much WAL it takes for one logical message,
-            // considering all alignments and headers.
+        // Add padding so the XLOG_SWITCH record ends exactly on XLOG_BLCKSZ boundary.
+        // We will use logical message as the padding. We start with detecting how much WAL
+        // it takes for one logical message, considering all alignments and headers.
+        let base_wal_advance = {
             let before_lsn = client.pg_current_wal_insert_lsn()?;
+            // Small non-empty message bigger than few bytes is more likely than an empty
+            // message to have the same format as the big padding message.
             client.execute(
                 "SELECT pg_logical_emit_message(false, 'swch', REPEAT('a', 10))",
                 &[],
             )?;
-            let after_lsn = client.pg_current_wal_insert_lsn()?;
-
-            // Did the record cross a page boundary? If it did, start over. Crossing a
-            // page boundary adds to the apparent size of the record because of the page
-            // header, which throws off the calculation.
-            if u64::from(before_lsn) / XLOG_BLCKSZ as u64
-                != u64::from(after_lsn) / XLOG_BLCKSZ as u64
-            {
-                continue;
-            }
-            // base_size is the size of a logical message without the payload
-            let base_size = u64::from(after_lsn) - u64::from(before_lsn) - 10;
-
-            // Is there enough space on the page for another logical message and an
-            // XLOG_SWITCH? If not, start over.
-            let page_remain = XLOG_BLCKSZ as u64 - u64::from(after_lsn) % XLOG_BLCKSZ as u64;
-            if page_remain < base_size - XLOG_SIZE_OF_XLOG_RECORD as u64 {
-                continue;
-            }
-
-            // We will write another logical message, such that after the logical message
-            // record, there will be space for exactly one XLOG_SWITCH. How large should
-            // the logical message's payload be? An XLOG_SWITCH record has no data => its
-            // size is exactly XLOG_SIZE_OF_XLOG_RECORD.
-            let repeats = page_remain - base_size - XLOG_SIZE_OF_XLOG_RECORD as u64;
-
-            client.execute(
-                "SELECT pg_logical_emit_message(false, 'swch', REPEAT('a', $1))",
-                &[&(repeats as i32)],
-            )?;
-            break;
+            // The XLOG_SWITCH record has no data => its size is exactly XLOG_SIZE_OF_XLOG_RECORD.
+            (u64::from(client.pg_current_wal_insert_lsn()?) - u64::from(before_lsn)) as usize
+                + XLOG_SIZE_OF_XLOG_RECORD
+        };
+        let mut remaining_lsn =
+            XLOG_BLCKSZ - u64::from(client.pg_current_wal_insert_lsn()?) as usize % XLOG_BLCKSZ;
+        if remaining_lsn < base_wal_advance {
+            remaining_lsn += XLOG_BLCKSZ;
         }
+        let repeats = 10 + remaining_lsn - base_wal_advance;
+        info!(
+            "current_wal_insert_lsn={}, remaining_lsn={}, base_wal_advance={}, repeats={}",
+            client.pg_current_wal_insert_lsn()?,
+            remaining_lsn,
+            base_wal_advance,
+            repeats
+        );
+        client.execute(
+            "SELECT pg_logical_emit_message(false, 'swch', REPEAT('a', $1))",
+            &[&(repeats as i32)],
+        )?;
         info!(
             "current_wal_insert_lsn={}, XLOG_SIZE_OF_XLOG_RECORD={}",
             client.pg_current_wal_insert_lsn()?,

libs/remote_storage/src/lib.rs

@@ -21,13 +21,11 @@ use std::{
     fmt::Debug,
     num::{NonZeroU32, NonZeroUsize},
     pin::Pin,
-    str::FromStr,
     sync::Arc,
     time::{Duration, SystemTime},
 };
 
 use anyhow::{bail, Context};
-use aws_sdk_s3::types::StorageClass;
 use camino::{Utf8Path, Utf8PathBuf};
 
 use bytes::Bytes;
@@ -136,11 +134,6 @@ impl RemotePath {
     pub fn strip_prefix(&self, p: &RemotePath) -> Result<&Utf8Path, std::path::StripPrefixError> {
         self.0.strip_prefix(&p.0)
     }
-
-    pub fn add_trailing_slash(&self) -> Self {
-        // Unwrap safety inputs are guararnteed to be valid UTF-8
-        Self(format!("{}/", self.0).try_into().unwrap())
-    }
 }
 
 /// We don't need callers to be able to pass arbitrary delimiters: just control
@@ -164,21 +157,47 @@ pub struct Listing {
 /// providing basic CRUD operations for storage files.
 #[allow(async_fn_in_trait)]
 pub trait RemoteStorage: Send + Sync + 'static {
-    /// List objects in remote storage, with semantics matching AWS S3's ListObjectsV2.
-    /// (see `<https://docs.aws.amazon.com/AmazonS3/latest/API/API_ListObjectsV2.html>`)
-    ///
-    /// Note that the prefix is relative to any `prefix_in_bucket` configured for the client, not
-    /// from the absolute root of the bucket.
-    ///
-    /// `mode` configures whether to use a delimiter.  Without a delimiter all keys
-    /// within the prefix are listed in the `keys` of the result.  With a delimiter, any "directories" at the top level of
-    /// the prefix are returned in the `prefixes` of the result, and keys in the top level of the prefix are
-    /// returned in `keys` ().
-    ///
-    /// `max_keys` controls the maximum number of keys that will be returned.  If this is None, this function
-    /// will iteratively call listobjects until it runs out of keys.  Note that this is not safe to use on
-    /// unlimted size buckets, as the full list of objects is allocated into a monolithic data structure.
+    /// Lists all top level subdirectories for a given prefix
+    /// Note: here we assume that if the prefix is passed it was obtained via remote_object_id
+    /// which already takes into account any kind of global prefix (prefix_in_bucket for S3 or storage_root for LocalFS)
+    /// so this method doesnt need to.
+    async fn list_prefixes(
+        &self,
+        prefix: Option<&RemotePath>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        let result = self
+            .list(prefix, ListingMode::WithDelimiter, None, cancel)
+            .await?
+            .prefixes;
+        Ok(result)
+    }
+    /// Lists all files in directory "recursively"
+    /// (not really recursively, because AWS has a flat namespace)
+    /// Note: This is subtely different than list_prefixes,
+    /// because it is for listing files instead of listing
+    /// names sharing common prefixes.
+    /// For example,
+    /// list_files("foo/bar") = ["foo/bar/cat123.txt",
+    /// "foo/bar/cat567.txt", "foo/bar/dog123.txt", "foo/bar/dog456.txt"]
+    /// whereas,
+    /// list_prefixes("foo/bar/") = ["cat", "dog"]
+    /// See `test_real_s3.rs` for more details.
     ///
+    /// max_keys limits max number of keys returned; None means unlimited.
+    async fn list_files(
+        &self,
+        prefix: Option<&RemotePath>,
+        max_keys: Option<NonZeroU32>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        let result = self
+            .list(prefix, ListingMode::NoDelimiter, max_keys, cancel)
+            .await?
+            .keys;
+        Ok(result)
+    }
+
     async fn list(
         &self,
         prefix: Option<&RemotePath>,
@@ -317,6 +336,41 @@ impl<Other: RemoteStorage> GenericRemoteStorage<Arc<Other>> {
         }
     }
 
+    // A function for listing all the files in a "directory"
+    // Example:
+    // list_files("foo/bar") = ["foo/bar/a.txt", "foo/bar/b.txt"]
+    //
+    // max_keys limits max number of keys returned; None means unlimited.
+    pub async fn list_files(
+        &self,
+        folder: Option<&RemotePath>,
+        max_keys: Option<NonZeroU32>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        match self {
+            Self::LocalFs(s) => s.list_files(folder, max_keys, cancel).await,
+            Self::AwsS3(s) => s.list_files(folder, max_keys, cancel).await,
+            Self::AzureBlob(s) => s.list_files(folder, max_keys, cancel).await,
+            Self::Unreliable(s) => s.list_files(folder, max_keys, cancel).await,
+        }
+    }
+
+    // lists common *prefixes*, if any of files
+    // Example:
+    // list_prefixes("foo123","foo567","bar123","bar432") = ["foo", "bar"]
+    pub async fn list_prefixes(
+        &self,
+        prefix: Option<&RemotePath>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        match self {
+            Self::LocalFs(s) => s.list_prefixes(prefix, cancel).await,
+            Self::AwsS3(s) => s.list_prefixes(prefix, cancel).await,
+            Self::AzureBlob(s) => s.list_prefixes(prefix, cancel).await,
+            Self::Unreliable(s) => s.list_prefixes(prefix, cancel).await,
+        }
+    }
+
     /// See [`RemoteStorage::upload`]
     pub async fn upload(
         &self,
@@ -565,7 +619,6 @@ pub struct S3Config {
     /// See [`DEFAULT_REMOTE_STORAGE_S3_CONCURRENCY_LIMIT`] for more details.
     pub concurrency_limit: NonZeroUsize,
     pub max_keys_per_list_response: Option<i32>,
-    pub upload_storage_class: Option<StorageClass>,
 }
 
 impl Debug for S3Config {
@@ -694,18 +747,6 @@ impl RemoteStorageConfig {
                     endpoint,
                     concurrency_limit,
                     max_keys_per_list_response,
-                    upload_storage_class: toml
-                        .get("upload_storage_class")
-                        .map(|prefix_in_bucket| -> anyhow::Result<_> {
-                            let s = parse_toml_string("upload_storage_class", prefix_in_bucket)?;
-                            let storage_class = StorageClass::from_str(&s).expect("infallible");
-                            #[allow(deprecated)]
-                            if matches!(storage_class, StorageClass::Unknown(_)) {
-                                bail!("Specified storage class unknown to SDK: '{s}'. Allowed values: {:?}", StorageClass::values());
-                            }
-                            Ok(storage_class)
-                        })
-                        .transpose()?,
                 })
             }
             (_, _, _, Some(_), None) => {

libs/remote_storage/src/local_fs.rs

@@ -5,9 +5,11 @@
 //! volume is mounted to the local FS.
 
 use std::{
-    collections::HashSet,
+    borrow::Cow,
+    future::Future,
     io::ErrorKind,
     num::NonZeroU32,
+    pin::Pin,
     time::{Duration, SystemTime, UNIX_EPOCH},
 };
 
@@ -20,11 +22,11 @@ use tokio::{
     io::{self, AsyncReadExt, AsyncSeekExt, AsyncWriteExt},
 };
 use tokio_util::{io::ReaderStream, sync::CancellationToken};
-use utils::crashsafe::path_with_suffix_extension;
+use tracing::*;
+use utils::{crashsafe::path_with_suffix_extension, fs_ext::is_directory_empty};
 
 use crate::{
     Download, DownloadError, Listing, ListingMode, RemotePath, TimeTravelError, TimeoutOrCancel,
-    REMOTE_STORAGE_PREFIX_SEPARATOR,
 };
 
 use super::{RemoteStorage, StorageMetadata};
@@ -91,47 +93,7 @@ impl LocalFs {
 
     #[cfg(test)]
     async fn list_all(&self) -> anyhow::Result<Vec<RemotePath>> {
-        use std::{future::Future, pin::Pin};
-        fn get_all_files<'a, P>(
-            directory_path: P,
-        ) -> Pin<Box<dyn Future<Output = anyhow::Result<Vec<Utf8PathBuf>>> + Send + Sync + 'a>>
-        where
-            P: AsRef<Utf8Path> + Send + Sync + 'a,
-        {
-            Box::pin(async move {
-                let directory_path = directory_path.as_ref();
-                if directory_path.exists() {
-                    if directory_path.is_dir() {
-                        let mut paths = Vec::new();
-                        let mut dir_contents = fs::read_dir(directory_path).await?;
-                        while let Some(dir_entry) = dir_contents.next_entry().await? {
-                            let file_type = dir_entry.file_type().await?;
-                            let entry_path =
-                                Utf8PathBuf::from_path_buf(dir_entry.path()).map_err(|pb| {
-                                    anyhow::Error::msg(format!(
-                                        "non-Unicode path: {}",
-                                        pb.to_string_lossy()
-                                    ))
-                                })?;
-                            if file_type.is_symlink() {
-                                tracing::debug!("{entry_path:?} is a symlink, skipping")
-                            } else if file_type.is_dir() {
-                                paths.extend(get_all_files(&entry_path).await?.into_iter())
-                            } else {
-                                paths.push(entry_path);
-                            }
-                        }
-                        Ok(paths)
-                    } else {
-                        bail!("Path {directory_path:?} is not a directory")
-                    }
-                } else {
-                    Ok(Vec::new())
-                }
-            })
-        }
-
-        Ok(get_all_files(&self.storage_root)
+        Ok(get_all_files(&self.storage_root, true)
             .await?
             .into_iter()
             .map(|path| {
@@ -158,14 +120,6 @@ impl LocalFs {
         // S3 object list prefixes can be arbitrary strings, but when reading
         // the local filesystem we need a directory to start calling read_dir on.
         let mut initial_dir = full_path.clone();
-
-        // If there's no trailing slash, we have to start looking from one above: even if
-        // `initial_dir` is a directory, we should still list any prefixes in the parent
-        // that start with the same string.
-        if !full_path.to_string().ends_with('/') {
-            initial_dir.pop();
-        }
-
         loop {
             // Did we make it to the root?
             if initial_dir.parent().is_none() {
@@ -341,66 +295,61 @@ impl RemoteStorage for LocalFs {
         let op = async {
             let mut result = Listing::default();
 
-            // Filter out directories: in S3 directories don't exist, only the keys within them do.
-            let keys = self
-                .list_recursive(prefix)
+            if let ListingMode::NoDelimiter = mode {
+                let keys = self
+                    .list_recursive(prefix)
+                    .await
+                    .map_err(DownloadError::Other)?;
+
+                result.keys = keys
+                    .into_iter()
+                    .filter(|k| {
+                        let path = k.with_base(&self.storage_root);
+                        !path.is_dir()
+                    })
+                    .collect();
+
+                if let Some(max_keys) = max_keys {
+                    result.keys.truncate(max_keys.get() as usize);
+                }
+
+                return Ok(result);
+            }
+
+            let path = match prefix {
+                Some(prefix) => Cow::Owned(prefix.with_base(&self.storage_root)),
+                None => Cow::Borrowed(&self.storage_root),
+            };
+
+            let prefixes_to_filter = get_all_files(path.as_ref(), false)
                 .await
                 .map_err(DownloadError::Other)?;
-            let keys = keys
-                .into_iter()
-                .filter(|k| {
-                    let path = k.with_base(&self.storage_root);
-                    !path.is_dir()
-                })
-                .collect();
 
-            if let ListingMode::NoDelimiter = mode {
-                result.keys = keys;
-            } else {
-                let mut prefixes = HashSet::new();
-                for key in keys {
-                    // If the part after the prefix includes a "/", take only the first part and put it in `prefixes`.
-                    let relative_key = if let Some(prefix) = prefix {
-                        let mut prefix = prefix.clone();
-                        // We only strip the dirname of the prefix, so that when we strip it from the start of keys we
-                        // end up with full file/dir names.
-                        let prefix_full_local_path = prefix.with_base(&self.storage_root);
-                        let has_slash = prefix.0.to_string().ends_with('/');
-                        let strip_prefix = if prefix_full_local_path.is_dir() && has_slash {
-                            prefix
-                        } else {
-                            prefix.0.pop();
-                            prefix
-                        };
-
-                        RemotePath::new(key.strip_prefix(&strip_prefix).unwrap()).unwrap()
-                    } else {
-                        key
-                    };
-
-                    let relative_key = format!("{}", relative_key);
-                    if relative_key.contains(REMOTE_STORAGE_PREFIX_SEPARATOR) {
-                        let first_part = relative_key
-                            .split(REMOTE_STORAGE_PREFIX_SEPARATOR)
-                            .next()
-                            .unwrap()
-                            .to_owned();
-                        prefixes.insert(first_part);
-                    } else {
-                        result
-                            .keys
-                            .push(RemotePath::from_string(&relative_key).unwrap());
-                    }
+            // filter out empty directories to mirror s3 behavior.
+            for prefix in prefixes_to_filter {
+                if prefix.is_dir()
+                    && is_directory_empty(&prefix)
+                        .await
+                        .map_err(DownloadError::Other)?
+                {
+                    continue;
+                }
+
+                let stripped = prefix
+                    .strip_prefix(&self.storage_root)
+                    .context("Failed to strip prefix")
+                    .and_then(RemotePath::new)
+                    .expect(
+                        "We list files for storage root, hence should be able to remote the prefix",
+                    );
+
+                if prefix.is_dir() {
+                    result.prefixes.push(stripped);
+                } else {
+                    result.keys.push(stripped);
                 }
-                result.prefixes = prefixes
-                    .into_iter()
-                    .map(|s| RemotePath::from_string(&s).unwrap())
-                    .collect();
             }
 
-            if let Some(max_keys) = max_keys {
-                result.keys.truncate(max_keys.get() as usize);
-            }
             Ok(result)
         };
 
@@ -611,6 +560,50 @@ fn storage_metadata_path(original_path: &Utf8Path) -> Utf8PathBuf {
     path_with_suffix_extension(original_path, "metadata")
 }
 
+fn get_all_files<'a, P>(
+    directory_path: P,
+    recursive: bool,
+) -> Pin<Box<dyn Future<Output = anyhow::Result<Vec<Utf8PathBuf>>> + Send + Sync + 'a>>
+where
+    P: AsRef<Utf8Path> + Send + Sync + 'a,
+{
+    Box::pin(async move {
+        let directory_path = directory_path.as_ref();
+        if directory_path.exists() {
+            if directory_path.is_dir() {
+                let mut paths = Vec::new();
+                let mut dir_contents = fs::read_dir(directory_path).await?;
+                while let Some(dir_entry) = dir_contents.next_entry().await? {
+                    let file_type = dir_entry.file_type().await?;
+                    let entry_path =
+                        Utf8PathBuf::from_path_buf(dir_entry.path()).map_err(|pb| {
+                            anyhow::Error::msg(format!(
+                                "non-Unicode path: {}",
+                                pb.to_string_lossy()
+                            ))
+                        })?;
+                    if file_type.is_symlink() {
+                        debug!("{entry_path:?} is a symlink, skipping")
+                    } else if file_type.is_dir() {
+                        if recursive {
+                            paths.extend(get_all_files(&entry_path, true).await?.into_iter())
+                        } else {
+                            paths.push(entry_path)
+                        }
+                    } else {
+                        paths.push(entry_path);
+                    }
+                }
+                Ok(paths)
+            } else {
+                bail!("Path {directory_path:?} is not a directory")
+            }
+        } else {
+            Ok(Vec::new())
+        }
+    })
+}
+
 async fn create_target_directory(target_file_path: &Utf8Path) -> anyhow::Result<()> {
     let target_dir = match target_file_path.parent() {
         Some(parent_dir) => parent_dir,
@@ -930,18 +923,13 @@ mod fs_tests {
         // No delimiter: should recursively list everything
         let (storage, cancel) = create_storage()?;
         let child = upload_dummy_file(&storage, "grandparent/parent/child", None, &cancel).await?;
-        let child_sibling =
-            upload_dummy_file(&storage, "grandparent/parent/child_sibling", None, &cancel).await?;
         let uncle = upload_dummy_file(&storage, "grandparent/uncle", None, &cancel).await?;
 
         let listing = storage
             .list(None, ListingMode::NoDelimiter, None, &cancel)
             .await?;
         assert!(listing.prefixes.is_empty());
-        assert_eq!(
-            listing.keys.into_iter().collect::<HashSet<_>>(),
-            HashSet::from([uncle.clone(), child.clone(), child_sibling.clone()])
-        );
+        assert_eq!(listing.keys, [uncle.clone(), child.clone()].to_vec());
 
         // Delimiter: should only go one deep
         let listing = storage
@@ -954,25 +942,7 @@ mod fs_tests {
         );
         assert!(listing.keys.is_empty());
 
-        // Delimiter & prefix with a trailing slash
-        let listing = storage
-            .list(
-                Some(&RemotePath::from_string("timelines/some_timeline/grandparent/").unwrap()),
-                ListingMode::WithDelimiter,
-                None,
-                &cancel,
-            )
-            .await?;
-        assert_eq!(
-            listing.keys,
-            [RemotePath::from_string("uncle").unwrap()].to_vec()
-        );
-        assert_eq!(
-            listing.prefixes,
-            [RemotePath::from_string("parent").unwrap()].to_vec()
-        );
-
-        // Delimiter and prefix without a trailing slash
+        // Delimiter & prefix
         let listing = storage
             .list(
                 Some(&RemotePath::from_string("timelines/some_timeline/grandparent").unwrap()),
@@ -981,66 +951,12 @@ mod fs_tests {
                 &cancel,
             )
             .await?;
-        assert_eq!(listing.keys, [].to_vec());
         assert_eq!(
             listing.prefixes,
-            [RemotePath::from_string("grandparent").unwrap()].to_vec()
-        );
-
-        // Delimiter and prefix that's partway through a path component
-        let listing = storage
-            .list(
-                Some(&RemotePath::from_string("timelines/some_timeline/grandp").unwrap()),
-                ListingMode::WithDelimiter,
-                None,
-                &cancel,
-            )
-            .await?;
-        assert_eq!(listing.keys, [].to_vec());
-        assert_eq!(
-            listing.prefixes,
-            [RemotePath::from_string("grandparent").unwrap()].to_vec()
-        );
-
-        Ok(())
-    }
-
-    #[tokio::test]
-    async fn list_part_component() -> anyhow::Result<()> {
-        // No delimiter: should recursively list everything
-        let (storage, cancel) = create_storage()?;
-
-        // Imitates what happens in a tenant path when we have an unsharded path and a sharded path, and do a listing
-        // of the unsharded path: although there is a "directory" at the unsharded path, it should be handled as
-        // a freeform prefix.
-        let _child_a =
-            upload_dummy_file(&storage, "grandparent/tenant-01/child", None, &cancel).await?;
-        let _child_b =
-            upload_dummy_file(&storage, "grandparent/tenant/child", None, &cancel).await?;
-
-        // Delimiter and prefix that's partway through a path component
-        let listing = storage
-            .list(
-                Some(
-                    &RemotePath::from_string("timelines/some_timeline/grandparent/tenant").unwrap(),
-                ),
-                ListingMode::WithDelimiter,
-                None,
-                &cancel,
-            )
-            .await?;
-        assert_eq!(listing.keys, [].to_vec());
-
-        let mut found_prefixes = listing.prefixes.clone();
-        found_prefixes.sort();
-        assert_eq!(
-            found_prefixes,
-            [
-                RemotePath::from_string("tenant").unwrap(),
-                RemotePath::from_string("tenant-01").unwrap(),
-            ]
-            .to_vec()
+            [RemotePath::from_string("timelines/some_timeline/grandparent/parent").unwrap()]
+                .to_vec()
         );
+        assert_eq!(listing.keys, [uncle.clone()].to_vec());
 
         Ok(())
     }

libs/remote_storage/src/s3_bucket.rs

@@ -30,7 +30,7 @@ use aws_sdk_s3::{
     config::{AsyncSleep, Builder, IdentityCache, Region, SharedAsyncSleep},
     error::SdkError,
     operation::get_object::GetObjectError,
-    types::{Delete, DeleteMarkerEntry, ObjectIdentifier, ObjectVersion, StorageClass},
+    types::{Delete, DeleteMarkerEntry, ObjectIdentifier, ObjectVersion},
     Client,
 };
 use aws_smithy_async::rt::sleep::TokioSleep;
@@ -62,7 +62,6 @@ pub struct S3Bucket {
     bucket_name: String,
     prefix_in_bucket: Option<String>,
     max_keys_per_list_response: Option<i32>,
-    upload_storage_class: Option<StorageClass>,
     concurrency_limiter: ConcurrencyLimiter,
     // Per-request timeout. Accessible for tests.
     pub timeout: Duration,
@@ -155,7 +154,6 @@ impl S3Bucket {
             max_keys_per_list_response: aws_config.max_keys_per_list_response,
             prefix_in_bucket,
             concurrency_limiter: ConcurrencyLimiter::new(aws_config.concurrency_limit.get()),
-            upload_storage_class: aws_config.upload_storage_class.clone(),
             timeout,
         })
     }
@@ -180,7 +178,10 @@ impl S3Bucket {
 
     pub fn relative_path_to_s3_object(&self, path: &RemotePath) -> String {
         assert_eq!(std::path::MAIN_SEPARATOR, REMOTE_STORAGE_PREFIX_SEPARATOR);
-        let path_string = path.get_path().as_str();
+        let path_string = path
+            .get_path()
+            .as_str()
+            .trim_end_matches(REMOTE_STORAGE_PREFIX_SEPARATOR);
         match &self.prefix_in_bucket {
             Some(prefix) => prefix.clone() + "/" + path_string,
             None => path_string.to_string(),
@@ -470,11 +471,16 @@ impl RemoteStorage for S3Bucket {
         // get the passed prefix or if it is not set use prefix_in_bucket value
         let list_prefix = prefix
             .map(|p| self.relative_path_to_s3_object(p))
-            .or_else(|| {
-                self.prefix_in_bucket.clone().map(|mut s| {
-                    s.push(REMOTE_STORAGE_PREFIX_SEPARATOR);
-                    s
-                })
+            .or_else(|| self.prefix_in_bucket.clone())
+            .map(|mut p| {
+                // required to end with a separator
+                // otherwise request will return only the entry of a prefix
+                if matches!(mode, ListingMode::WithDelimiter)
+                    && !p.ends_with(REMOTE_STORAGE_PREFIX_SEPARATOR)
+                {
+                    p.push(REMOTE_STORAGE_PREFIX_SEPARATOR);
+                }
+                p
             });
 
         let _permit = self.permit(kind, cancel).await?;
@@ -543,15 +549,11 @@ impl RemoteStorage for S3Bucket {
                 }
             }
 
-            // S3 gives us prefixes like "foo/", we return them like "foo"
-            result.prefixes.extend(prefixes.iter().filter_map(|o| {
-                Some(
-                    self.s3_object_to_relative_path(
-                        o.prefix()?
-                            .trim_end_matches(REMOTE_STORAGE_PREFIX_SEPARATOR),
-                    ),
-                )
-            }));
+            result.prefixes.extend(
+                prefixes
+                    .iter()
+                    .filter_map(|o| Some(self.s3_object_to_relative_path(o.prefix()?))),
+            );
 
             continuation_token = match response.next_continuation_token {
                 Some(new_token) => Some(new_token),
@@ -584,7 +586,6 @@ impl RemoteStorage for S3Bucket {
             .bucket(self.bucket_name.clone())
             .key(self.relative_path_to_s3_object(to))
             .set_metadata(metadata.map(|m| m.0))
-            .set_storage_class(self.upload_storage_class.clone())
             .content_length(from_size_bytes.try_into()?)
             .body(bytes_stream)
             .send();
@@ -636,7 +637,6 @@ impl RemoteStorage for S3Bucket {
             .copy_object()
             .bucket(self.bucket_name.clone())
             .key(self.relative_path_to_s3_object(to))
-            .set_storage_class(self.upload_storage_class.clone())
             .copy_source(copy_source)
             .send();
 
@@ -894,7 +894,6 @@ impl RemoteStorage for S3Bucket {
                                     .copy_object()
                                     .bucket(self.bucket_name.clone())
                                     .key(key)
-                                    .set_storage_class(self.upload_storage_class.clone())
                                     .copy_source(&source_id)
                                     .send();
 
@@ -1051,22 +1050,22 @@ mod tests {
             Some("/test/prefix/"),
         ];
         let expected_outputs = [
-            vec!["", "some/path", "some/path/"],
-            vec!["/", "/some/path", "/some/path/"],
+            vec!["", "some/path", "some/path"],
+            vec!["/", "/some/path", "/some/path"],
             vec![
                 "test/prefix/",
                 "test/prefix/some/path",
-                "test/prefix/some/path/",
+                "test/prefix/some/path",
             ],
             vec![
                 "test/prefix/",
                 "test/prefix/some/path",
-                "test/prefix/some/path/",
+                "test/prefix/some/path",
             ],
             vec![
                 "test/prefix/",
                 "test/prefix/some/path",
-                "test/prefix/some/path/",
+                "test/prefix/some/path",
             ],
         ];
 
@@ -1078,7 +1077,6 @@ mod tests {
                 endpoint: None,
                 concurrency_limit: NonZeroUsize::new(100).unwrap(),
                 max_keys_per_list_response: Some(5),
-                upload_storage_class: None,
             };
             let storage =
                 S3Bucket::new(&config, std::time::Duration::ZERO).expect("remote storage init");

libs/remote_storage/src/simulate_failures.rs

@@ -107,6 +107,27 @@ impl UnreliableWrapper {
 type VoidStorage = crate::LocalFs;
 
 impl RemoteStorage for UnreliableWrapper {
+    async fn list_prefixes(
+        &self,
+        prefix: Option<&RemotePath>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        self.attempt(RemoteOp::ListPrefixes(prefix.cloned()))
+            .map_err(DownloadError::Other)?;
+        self.inner.list_prefixes(prefix, cancel).await
+    }
+
+    async fn list_files(
+        &self,
+        folder: Option<&RemotePath>,
+        max_keys: Option<NonZeroU32>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        self.attempt(RemoteOp::ListPrefixes(folder.cloned()))
+            .map_err(DownloadError::Other)?;
+        self.inner.list_files(folder, max_keys, cancel).await
+    }
+
     async fn list(
         &self,
         prefix: Option<&RemotePath>,

libs/remote_storage/tests/common/tests.rs

@@ -1,6 +1,5 @@
 use anyhow::Context;
 use camino::Utf8Path;
-use remote_storage::ListingMode;
 use remote_storage::RemotePath;
 use std::sync::Arc;
 use std::{collections::HashSet, num::NonZeroU32};
@@ -55,9 +54,9 @@ async fn pagination_should_work(ctx: &mut MaybeEnabledStorageWithTestBlobs) -> a
     let base_prefix = RemotePath::new(Utf8Path::new(ctx.enabled.base_prefix))
         .context("common_prefix construction")?;
     let root_remote_prefixes = test_client
-        .list(None, ListingMode::WithDelimiter, None, &cancel)
-        .await?
-        .prefixes
+        .list_prefixes(None, &cancel)
+        .await
+        .context("client list root prefixes failure")?
         .into_iter()
         .collect::<HashSet<_>>();
     assert_eq!(
@@ -66,14 +65,9 @@ async fn pagination_should_work(ctx: &mut MaybeEnabledStorageWithTestBlobs) -> a
     );
 
     let nested_remote_prefixes = test_client
-        .list(
-            Some(&base_prefix.add_trailing_slash()),
-            ListingMode::WithDelimiter,
-            None,
-            &cancel,
-        )
-        .await?
-        .prefixes
+        .list_prefixes(Some(&base_prefix), &cancel)
+        .await
+        .context("client list nested prefixes failure")?
         .into_iter()
         .collect::<HashSet<_>>();
     let remote_only_prefixes = nested_remote_prefixes
@@ -96,13 +90,11 @@ async fn pagination_should_work(ctx: &mut MaybeEnabledStorageWithTestBlobs) -> a
 ///
 /// First, create a set of S3 objects with keys `random_prefix/folder{j}/blob_{i}.txt` in [`upload_remote_data`]
 /// Then performs the following queries:
-///    1. `list(None)`. This should return all files `random_prefix/folder{j}/blob_{i}.txt`
-///    2. `list("folder1")`.  This  should return all files `random_prefix/folder1/blob_{i}.txt`
+///    1. `list_files(None)`. This should return all files `random_prefix/folder{j}/blob_{i}.txt`
+///    2. `list_files("folder1")`.  This  should return all files `random_prefix/folder1/blob_{i}.txt`
 #[test_context(MaybeEnabledStorageWithSimpleTestBlobs)]
 #[tokio::test]
-async fn list_no_delimiter_works(
-    ctx: &mut MaybeEnabledStorageWithSimpleTestBlobs,
-) -> anyhow::Result<()> {
+async fn list_files_works(ctx: &mut MaybeEnabledStorageWithSimpleTestBlobs) -> anyhow::Result<()> {
     let ctx = match ctx {
         MaybeEnabledStorageWithSimpleTestBlobs::Enabled(ctx) => ctx,
         MaybeEnabledStorageWithSimpleTestBlobs::Disabled => return Ok(()),
@@ -115,36 +107,29 @@ async fn list_no_delimiter_works(
     let base_prefix =
         RemotePath::new(Utf8Path::new("folder1")).context("common_prefix construction")?;
     let root_files = test_client
-        .list(None, ListingMode::NoDelimiter, None, &cancel)
+        .list_files(None, None, &cancel)
         .await
         .context("client list root files failure")?
-        .keys
         .into_iter()
         .collect::<HashSet<_>>();
     assert_eq!(
         root_files,
         ctx.remote_blobs.clone(),
-        "remote storage list on root mismatches with the uploads."
+        "remote storage list_files on root mismatches with the uploads."
     );
 
     // Test that max_keys limit works. In total there are about 21 files (see
     // upload_simple_remote_data call in test_real_s3.rs).
     let limited_root_files = test_client
-        .list(
-            None,
-            ListingMode::NoDelimiter,
-            Some(NonZeroU32::new(2).unwrap()),
-            &cancel,
-        )
+        .list_files(None, Some(NonZeroU32::new(2).unwrap()), &cancel)
         .await
         .context("client list root files failure")?;
-    assert_eq!(limited_root_files.keys.len(), 2);
+    assert_eq!(limited_root_files.len(), 2);
 
     let nested_remote_files = test_client
-        .list(Some(&base_prefix), ListingMode::NoDelimiter, None, &cancel)
+        .list_files(Some(&base_prefix), None, &cancel)
         .await
         .context("client list nested files failure")?
-        .keys
         .into_iter()
         .collect::<HashSet<_>>();
     let trim_remote_blobs: HashSet<_> = ctx
@@ -156,7 +141,7 @@ async fn list_no_delimiter_works(
         .collect();
     assert_eq!(
         nested_remote_files, trim_remote_blobs,
-        "remote storage list on subdirrectory mismatches with the uploads."
+        "remote storage list_files on subdirrectory mismatches with the uploads."
     );
     Ok(())
 }
@@ -214,11 +199,7 @@ async fn delete_objects_works(ctx: &mut MaybeEnabledStorage) -> anyhow::Result<(
 
     ctx.client.delete_objects(&[path1, path2], &cancel).await?;
 
-    let prefixes = ctx
-        .client
-        .list(None, ListingMode::WithDelimiter, None, &cancel)
-        .await?
-        .prefixes;
+    let prefixes = ctx.client.list_prefixes(None, &cancel).await?;
 
     assert_eq!(prefixes.len(), 1);
 

libs/remote_storage/tests/test_real_azure.rs

@@ -132,6 +132,10 @@ impl AsyncTestContext for MaybeEnabledStorageWithTestBlobs {
     }
 }
 
+// NOTE: the setups for the list_prefixes test and the list_files test are very similar
+// However, they are not idential. The list_prefixes function is concerned with listing prefixes,
+// whereas the list_files function is concerned with listing files.
+// See `RemoteStorage::list_files` documentation for more details
 enum MaybeEnabledStorageWithSimpleTestBlobs {
     Enabled(AzureWithSimpleTestBlobs),
     Disabled,

libs/remote_storage/tests/test_real_s3.rs

@@ -12,8 +12,8 @@ use anyhow::Context;
 use camino::Utf8Path;
 use futures_util::StreamExt;
 use remote_storage::{
-    DownloadError, GenericRemoteStorage, ListingMode, RemotePath, RemoteStorageConfig,
-    RemoteStorageKind, S3Config,
+    DownloadError, GenericRemoteStorage, RemotePath, RemoteStorageConfig, RemoteStorageKind,
+    S3Config,
 };
 use test_context::test_context;
 use test_context::AsyncTestContext;
@@ -75,14 +75,11 @@ async fn s3_time_travel_recovery_works(ctx: &mut MaybeEnabledStorage) -> anyhow:
         client: &Arc<GenericRemoteStorage>,
         cancel: &CancellationToken,
     ) -> anyhow::Result<HashSet<RemotePath>> {
-        Ok(
-            retry(|| client.list(None, ListingMode::NoDelimiter, None, cancel))
-                .await
-                .context("list root files failure")?
-                .keys
-                .into_iter()
-                .collect::<HashSet<_>>(),
-        )
+        Ok(retry(|| client.list_files(None, None, cancel))
+            .await
+            .context("list root files failure")?
+            .into_iter()
+            .collect::<HashSet<_>>())
     }
 
     let cancel = CancellationToken::new();
@@ -297,6 +294,10 @@ impl AsyncTestContext for MaybeEnabledStorageWithTestBlobs {
     }
 }
 
+// NOTE: the setups for the list_prefixes test and the list_files test are very similar
+// However, they are not idential. The list_prefixes function is concerned with listing prefixes,
+// whereas the list_files function is concerned with listing files.
+// See `RemoteStorage::list_files` documentation for more details
 enum MaybeEnabledStorageWithSimpleTestBlobs {
     Enabled(S3WithSimpleTestBlobs),
     Disabled,
@@ -380,7 +381,6 @@ fn create_s3_client(
             endpoint: None,
             concurrency_limit: NonZeroUsize::new(100).unwrap(),
             max_keys_per_list_response,
-            upload_storage_class: None,
         }),
         timeout: RemoteStorageConfig::DEFAULT_TIMEOUT,
     };

libs/utils/src/generation.rs

@@ -34,8 +34,6 @@ pub enum Generation {
 /// scenarios where pageservers might otherwise issue conflicting writes to
 /// remote storage
 impl Generation {
-    pub const MAX: Self = Self::Valid(u32::MAX);
-
     /// Create a new Generation that represents a legacy key format with
     /// no generation suffix
     pub fn none() -> Self {

libs/utils/src/sync/heavier_once_cell.rs

@@ -192,14 +192,6 @@ impl<T> OnceCell<T> {
         }
     }
 
-    /// Like [`Guard::take_and_deinit`], but will return `None` if this OnceCell was never
-    /// initialized.
-    pub fn take_and_deinit(&mut self) -> Option<(T, InitPermit)> {
-        let inner = self.inner.get_mut().unwrap();
-
-        inner.take_and_deinit()
-    }
-
     /// Return the number of [`Self::get_or_init`] calls waiting for initialization to complete.
     pub fn initializer_count(&self) -> usize {
         self.initializers.load(Ordering::Relaxed)
@@ -254,23 +246,15 @@ impl<'a, T> Guard<'a, T> {
     /// The permit will be on a semaphore part of the new internal value, and any following
     /// [`OnceCell::get_or_init`] will wait on it to complete.
     pub fn take_and_deinit(mut self) -> (T, InitPermit) {
-        self.0
-            .take_and_deinit()
-            .expect("guard is not created unless value has been initialized")
-    }
-}
-
-impl<T> Inner<T> {
-    pub fn take_and_deinit(&mut self) -> Option<(T, InitPermit)> {
-        let value = self.value.take()?;
-
         let mut swapped = Inner::default();
         let sem = swapped.init_semaphore.clone();
         // acquire and forget right away, moving the control over to InitPermit
         sem.try_acquire().expect("we just created this").forget();
-        let permit = InitPermit(sem);
-        std::mem::swap(self, &mut swapped);
-        Some((value, permit))
+        std::mem::swap(&mut *self.0, &mut swapped);
+        swapped
+            .value
+            .map(|v| (v, InitPermit(sem)))
+            .expect("guard is not created unless value has been initialized")
     }
 }
 
@@ -279,13 +263,6 @@ impl<T> Inner<T> {
 /// On drop, this type will return the permit.
 pub struct InitPermit(Arc<tokio::sync::Semaphore>);
 
-impl std::fmt::Debug for InitPermit {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        let ptr = Arc::as_ptr(&self.0) as *const ();
-        f.debug_tuple("InitPermit").field(&ptr).finish()
-    }
-}
-
 impl Drop for InitPermit {
     fn drop(&mut self) {
         assert_eq!(
@@ -582,22 +559,4 @@ mod tests {
 
         assert_eq!(*target.get().unwrap(), 11);
     }
-
-    #[tokio::test]
-    async fn take_and_deinit_on_mut() {
-        use std::convert::Infallible;
-
-        let mut target = OnceCell::<u32>::default();
-        assert!(target.take_and_deinit().is_none());
-
-        target
-            .get_or_init(|permit| async move { Ok::<_, Infallible>((42, permit)) })
-            .await
-            .unwrap();
-
-        let again = target.take_and_deinit();
-        assert!(matches!(again, Some((42, _))), "{again:?}");
-
-        assert!(target.take_and_deinit().is_none());
-    }
 }

pageserver/Cargo.toml

@@ -70,7 +70,6 @@ tokio-stream.workspace = true
 tokio-util.workspace = true
 toml_edit = { workspace = true, features = [ "serde" ] }
 tracing.workspace = true
-twox-hash.workspace = true
 url.workspace = true
 walkdir.workspace = true
 metrics.workspace = true

pageserver/client/src/mgmt_api.rs

@@ -243,19 +243,6 @@ impl Client {
         Ok(())
     }
 
-    pub async fn tenant_scan_remote_storage(
-        &self,
-        tenant_id: TenantId,
-    ) -> Result<TenantScanRemoteStorageResponse> {
-        let uri = format!(
-            "{}/v1/tenant/{tenant_id}/scan_remote_storage",
-            self.mgmt_api_endpoint
-        );
-        let response = self.request(Method::GET, &uri, ()).await?;
-        let body = response.json().await.map_err(Error::ReceiveBody)?;
-        Ok(body)
-    }
-
     pub async fn tenant_config(&self, req: &TenantConfigRequest) -> Result<()> {
         let uri = format!("{}/v1/tenant/config", self.mgmt_api_endpoint);
         self.request(Method::PUT, &uri, req).await?;
@@ -292,7 +279,7 @@ impl Client {
         lazy: bool,
     ) -> Result<()> {
         let req_body = TenantLocationConfigRequest {
-            tenant_id: None,
+            tenant_id: Some(tenant_shard_id),
             config,
         };
 

pageserver/client/src/page_service.rs

@@ -60,7 +60,7 @@ impl Client {
     ) -> anyhow::Result<PagestreamClient> {
         let copy_both: tokio_postgres::CopyBothDuplex<bytes::Bytes> = self
             .client
-            .copy_both_simple(&format!("pagestream_v2 {tenant_id} {timeline_id}"))
+            .copy_both_simple(&format!("pagestream {tenant_id} {timeline_id}"))
             .await?;
         let Client {
             cancel_on_client_drop,

pageserver/ctl/src/draw_timeline_dir.rs

@@ -9,45 +9,18 @@
 //! Coordinates in both axis are compressed for better readability.
 //! (see <https://medium.com/algorithms-digest/coordinate-compression-2fff95326fb>)
 //!
-//! The plain text API was chosen so that we can easily work with filenames from various
-//! sources; see the Usage section below for examples.
-//!
-//! # Usage
-//!
-//! ## Producing the SVG
-//!
+//! Example use:
 //! ```bash
-//!
-//! # local timeline dir
-//! ls test_output/test_pgbench\[neon-45-684\]/repo/tenants/$TENANT/timelines/$TIMELINE | \
-//!     grep "__" | cargo run --release --bin pagectl draw-timeline-dir > out.svg
-//!
-//! # Layer map dump from `/v1/tenant/$TENANT/timeline/$TIMELINE/layer`
-//! (jq -r '.historic_layers[] | .layer_file_name' | cargo  run -p pagectl draw-timeline) < layer-map.json > out.svg
-//!
-//! # From an `index_part.json` in S3
-//! (jq -r '.layer_metadata | keys[]' | cargo  run -p pagectl draw-timeline ) < index_part.json-00000016 > out.svg
-//!
+//! $ ls test_output/test_pgbench\[neon-45-684\]/repo/tenants/$TENANT/timelines/$TIMELINE | \
+//! $   grep "__" | cargo run --release --bin pagectl draw-timeline-dir > out.svg
+//! $ firefox out.svg
 //! ```
 //!
-//! ## Viewing
+//! This API was chosen so that we can easily work with filenames extracted from ssh,
+//! or from pageserver log files.
 //!
-//! **Inkscape** is better than the built-in viewers in browsers.
-//!
-//! After selecting a layer file rectangle, use "Open XML Editor" (Ctrl|Cmd + Shift + X)
-//! to see the layer file name in the comment field.
-//!
-//! ```bash
-//!
-//! # Linux
-//! inkscape out.svg
-//!
-//! # macOS
-//! /Applications/Inkscape.app/Contents/MacOS/inkscape out.svg
-//!
-//! ```
-//!
-
+//! TODO Consider shipping this as a grafana panel plugin:
+//!      <https://grafana.com/tutorials/build-a-panel-plugin/>
 use anyhow::Result;
 use pageserver::repository::Key;
 use pageserver::METADATA_FILE_NAME;
@@ -92,12 +65,7 @@ fn parse_filename(name: &str) -> (Range<Key>, Range<Lsn>) {
 
 pub fn main() -> Result<()> {
     // Parse layer filenames from stdin
-    struct Layer {
-        filename: String,
-        key_range: Range<Key>,
-        lsn_range: Range<Lsn>,
-    }
-    let mut files: Vec<Layer> = vec![];
+    let mut ranges: Vec<(Range<Key>, Range<Lsn>)> = vec![];
     let stdin = io::stdin();
     for line in stdin.lock().lines() {
         let line = line.unwrap();
@@ -108,23 +76,14 @@ pub fn main() -> Result<()> {
             // Don't try and parse "metadata" like a key-lsn range
             continue;
         }
-        let (key_range, lsn_range) = parse_filename(filename);
-        files.push(Layer {
-            filename: filename.to_owned(),
-            key_range,
-            lsn_range,
-        });
+        let range = parse_filename(filename);
+        ranges.push(range);
     }
 
     // Collect all coordinates
     let mut keys: Vec<Key> = vec![];
     let mut lsns: Vec<Lsn> = vec![];
-    for Layer {
-        key_range: keyr,
-        lsn_range: lsnr,
-        ..
-    } in &files
-    {
+    for (keyr, lsnr) in &ranges {
         keys.push(keyr.start);
         keys.push(keyr.end);
         lsns.push(lsnr.start);
@@ -148,12 +107,7 @@ pub fn main() -> Result<()> {
             h: stretch * lsn_map.len() as f32
         }
     );
-    for Layer {
-        filename,
-        key_range: keyr,
-        lsn_range: lsnr,
-    } in &files
-    {
+    for (keyr, lsnr) in &ranges {
         let key_start = *key_map.get(&keyr.start).unwrap();
         let key_end = *key_map.get(&keyr.end).unwrap();
         let key_diff = key_end - key_start;
@@ -197,7 +151,6 @@ pub fn main() -> Result<()> {
             .fill(fill)
             .stroke(Stroke::Color(rgb(0, 0, 0), 0.1))
             .border_radius(0.4)
-            .comment(filename)
         );
     }
     println!("{}", EndSvg);

pageserver/pagebench/src/cmd/getpage_latest_lsn.rs

@@ -312,12 +312,8 @@ async fn main_impl(
                     let (rel_tag, block_no) =
                         key_to_rel_block(key).expect("we filter non-rel-block keys out above");
                     PagestreamGetPageRequest {
-                        request_lsn: if rng.gen_bool(args.req_latest_probability) {
-                            Lsn::MAX
-                        } else {
-                            r.timeline_lsn
-                        },
-                        not_modified_since: r.timeline_lsn,
+                        latest: rng.gen_bool(args.req_latest_probability),
+                        lsn: r.timeline_lsn,
                         rel: rel_tag,
                         blkno: block_no,
                     }

pageserver/src/aux_file.rs

@@ -1,112 +0,0 @@
-use pageserver_api::key::{Key, AUX_KEY_PREFIX, METADATA_KEY_SIZE};
-use tracing::warn;
-
-/// Create a metadata key from a hash, encoded as [AUX_KEY_PREFIX, 2B directory prefix, first 13B of 128b xxhash].
-fn aux_hash_to_metadata_key(dir_level1: u8, dir_level2: u8, data: &[u8]) -> Key {
-    let mut key = [0; METADATA_KEY_SIZE];
-    let hash = twox_hash::xxh3::hash128(data).to_be_bytes();
-    key[0] = AUX_KEY_PREFIX;
-    key[1] = dir_level1;
-    key[2] = dir_level2;
-    key[3..16].copy_from_slice(&hash[0..13]);
-    Key::from_metadata_key_fixed_size(&key)
-}
-
-const AUX_DIR_PG_LOGICAL: u8 = 0x01;
-const AUX_DIR_PG_REPLSLOT: u8 = 0x02;
-const AUX_DIR_PG_UNKNOWN: u8 = 0xFF;
-
-/// Encode the aux file into a fixed-size key.
-///
-/// The first byte is the AUX key prefix. We use the next 2 bytes of the key for the directory / aux file type.
-/// We have one-to-one mapping for each of the aux file that we support. We hash the remaining part of the path
-/// (usually a single file name, or several components) into 13-byte hash. The way we determine the 2-byte prefix
-/// is roughly based on the first two components of the path, one unique number for one component.
-///
-/// * pg_logical/mappings -> 0x0101
-/// * pg_logical/snapshots -> 0x0102
-/// * pg_logical/replorigin_checkpoint -> 0x0103
-/// * pg_logical/others -> 0x01FF
-/// * pg_replslot/ -> 0x0201
-/// * others -> 0xFFFF
-///
-/// If you add new AUX files to this function, please also add a test case to `test_encoding_portable`.
-/// The new file type must have never been written to the storage before. Otherwise, there could be data
-/// corruptions as the new file belongs to a new prefix but it might have been stored under the `others` prefix.
-pub fn encode_aux_file_key(path: &str) -> Key {
-    if let Some(fname) = path.strip_prefix("pg_logical/mappings/") {
-        aux_hash_to_metadata_key(AUX_DIR_PG_LOGICAL, 0x01, fname.as_bytes())
-    } else if let Some(fname) = path.strip_prefix("pg_logical/snapshots/") {
-        aux_hash_to_metadata_key(AUX_DIR_PG_LOGICAL, 0x02, fname.as_bytes())
-    } else if path == "pg_logical/replorigin_checkpoint" {
-        aux_hash_to_metadata_key(AUX_DIR_PG_LOGICAL, 0x03, b"")
-    } else if let Some(fname) = path.strip_prefix("pg_logical/") {
-        if cfg!(debug_assertions) {
-            warn!(
-                "unsupported pg_logical aux file type: {}, putting to 0x01FF, would affect path scanning",
-                path
-            );
-        }
-        aux_hash_to_metadata_key(AUX_DIR_PG_LOGICAL, 0xFF, fname.as_bytes())
-    } else if let Some(fname) = path.strip_prefix("pg_replslot/") {
-        aux_hash_to_metadata_key(AUX_DIR_PG_REPLSLOT, 0x01, fname.as_bytes())
-    } else {
-        if cfg!(debug_assertions) {
-            warn!(
-                "unsupported aux file type: {}, putting to 0xFFFF, would affect path scanning",
-                path
-            );
-        }
-        aux_hash_to_metadata_key(AUX_DIR_PG_UNKNOWN, 0xFF, path.as_bytes())
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_hash_portable() {
-        // AUX file encoding requires the hash to be portable across all platforms. This test case checks
-        // if the algorithm produces the same hash across different environments.
-        assert_eq!(
-            305317690835051308206966631765527126151,
-            twox_hash::xxh3::hash128("test1".as_bytes())
-        );
-        assert_eq!(
-            85104974691013376326742244813280798847,
-            twox_hash::xxh3::hash128("test/test2".as_bytes())
-        );
-        assert_eq!(0, twox_hash::xxh3::hash128("".as_bytes()));
-    }
-
-    #[test]
-    fn test_encoding_portable() {
-        // To correct retrieve AUX files, the generated keys for the same file must be the same for all versions
-        // of the page server.
-        assert_eq!(
-            "6200000101E5B20C5F8DD5AA3289D6D9EAFA",
-            encode_aux_file_key("pg_logical/mappings/test1").to_string()
-        );
-        assert_eq!(
-            "620000010239AAC544893139B26F501B97E6",
-            encode_aux_file_key("pg_logical/snapshots/test2").to_string()
-        );
-        assert_eq!(
-            "620000010300000000000000000000000000",
-            encode_aux_file_key("pg_logical/replorigin_checkpoint").to_string()
-        );
-        assert_eq!(
-            "62000001FF8635AF2134B7266EC5B4189FD6",
-            encode_aux_file_key("pg_logical/unsupported").to_string()
-        );
-        assert_eq!(
-            "6200000201772D0E5D71DE14DA86142A1619",
-            encode_aux_file_key("pg_replslot/test3").to_string()
-        );
-        assert_eq!(
-            "620000FFFF1866EBEB53B807B26A2416F317",
-            encode_aux_file_key("other_file_not_supported").to_string()
-        );
-    }
-}

pageserver/src/basebackup.rs

@@ -13,7 +13,7 @@
 use anyhow::{anyhow, bail, ensure, Context};
 use bytes::{BufMut, Bytes, BytesMut};
 use fail::fail_point;
-use pageserver_api::key::{key_to_slru_block, rel_block_to_key, Key};
+use pageserver_api::key::{key_to_slru_block, Key};
 use postgres_ffi::pg_constants;
 use std::fmt::Write as FmtWrite;
 use std::time::SystemTime;
@@ -297,20 +297,7 @@ where
                 if rel.forknum == INIT_FORKNUM {
                     // I doubt we need _init fork itself, but having it at least
                     // serves as a marker relation is unlogged.
-                    if let Err(_e) = self.add_rel(rel, rel).await {
-                        if self
-                            .timeline
-                            .get_shard_identity()
-                            .is_key_buggy_forknum(&rel_block_to_key(rel, 0x0))
-                        {
-                            // Workaround https://github.com/neondatabase/neon/issues/7451 -- if we have an unlogged relation
-                            // whose INIT_FORKNUM is not correctly on shard zero, then omit it in the basebackup.  This allows
-                            // postgres to start up.  The relation won't work, but it will be possible to DROP TABLE on it and
-                            // recreate.
-                            tracing::warn!("Omitting relation {rel} for issue #7451: drop and recreate this unlogged relation");
-                            continue;
-                        }
-                    };
+                    self.add_rel(rel, rel).await?;
                     self.add_rel(rel, rel.with_forknum(MAIN_FORKNUM)).await?;
                     continue;
                 }
@@ -376,7 +363,7 @@ where
     async fn add_rel(&mut self, src: RelTag, dst: RelTag) -> anyhow::Result<()> {
         let nblocks = self
             .timeline
-            .get_rel_size(src, Version::Lsn(self.lsn), self.ctx)
+            .get_rel_size(src, Version::Lsn(self.lsn), false, self.ctx)
             .await?;
 
         // If the relation is empty, create an empty file
@@ -397,7 +384,7 @@ where
             for blknum in startblk..endblk {
                 let img = self
                     .timeline
-                    .get_rel_page_at_lsn(src, blknum, Version::Lsn(self.lsn), self.ctx)
+                    .get_rel_page_at_lsn(src, blknum, Version::Lsn(self.lsn), false, self.ctx)
                     .await?;
                 segment_data.extend_from_slice(&img[..]);
             }

pageserver/src/bin/pageserver.rs

@@ -121,10 +121,8 @@ fn main() -> anyhow::Result<()> {
         &[("node_id", &conf.id.to_string())],
     );
 
-    // after setting up logging, log the effective IO engine choice and read path implementations
+    // after setting up logging, log the effective IO engine choice
     info!(?conf.virtual_file_io_engine, "starting with virtual_file IO engine");
-    info!(?conf.get_impl, "starting with get page implementation");
-    info!(?conf.get_vectored_impl, "starting with vectored get page implementation");
 
     let tenants_path = conf.tenants_path();
     if !tenants_path.exists() {

pageserver/src/config.rs

@@ -30,9 +30,9 @@ use utils::{
     logging::LogFormat,
 };
 
+use crate::tenant::config::TenantConfOpt;
 use crate::tenant::timeline::GetVectoredImpl;
 use crate::tenant::vectored_blob_io::MaxVectoredReadBytes;
-use crate::tenant::{config::TenantConfOpt, timeline::GetImpl};
 use crate::tenant::{
     TENANTS_SEGMENT_NAME, TENANT_DELETED_MARKER_FILE_NAME, TIMELINES_SEGMENT_NAME,
 };
@@ -91,8 +91,6 @@ pub mod defaults {
 
     pub const DEFAULT_GET_VECTORED_IMPL: &str = "sequential";
 
-    pub const DEFAULT_GET_IMPL: &str = "legacy";
-
     pub const DEFAULT_MAX_VECTORED_READ_BYTES: usize = 128 * 1024; // 128 KiB
 
     pub const DEFAULT_VALIDATE_VECTORED_GET: bool = true;
@@ -140,8 +138,6 @@ pub mod defaults {
 
 #get_vectored_impl = '{DEFAULT_GET_VECTORED_IMPL}'
 
-#get_impl = '{DEFAULT_GET_IMPL}'
-
 #max_vectored_read_bytes = '{DEFAULT_MAX_VECTORED_READ_BYTES}'
 
 #validate_vectored_get = '{DEFAULT_VALIDATE_VECTORED_GET}'
@@ -288,8 +284,6 @@ pub struct PageServerConf {
 
     pub get_vectored_impl: GetVectoredImpl,
 
-    pub get_impl: GetImpl,
-
     pub max_vectored_read_bytes: MaxVectoredReadBytes,
 
     pub validate_vectored_get: bool,
@@ -420,8 +414,6 @@ struct PageServerConfigBuilder {
 
     get_vectored_impl: BuilderValue<GetVectoredImpl>,
 
-    get_impl: BuilderValue<GetImpl>,
-
     max_vectored_read_bytes: BuilderValue<MaxVectoredReadBytes>,
 
     validate_vectored_get: BuilderValue<bool>,
@@ -511,7 +503,6 @@ impl PageServerConfigBuilder {
             virtual_file_io_engine: Set(DEFAULT_VIRTUAL_FILE_IO_ENGINE.parse().unwrap()),
 
             get_vectored_impl: Set(DEFAULT_GET_VECTORED_IMPL.parse().unwrap()),
-            get_impl: Set(DEFAULT_GET_IMPL.parse().unwrap()),
             max_vectored_read_bytes: Set(MaxVectoredReadBytes(
                 NonZeroUsize::new(DEFAULT_MAX_VECTORED_READ_BYTES).unwrap(),
             )),
@@ -690,10 +681,6 @@ impl PageServerConfigBuilder {
         self.get_vectored_impl = BuilderValue::Set(value);
     }
 
-    pub fn get_impl(&mut self, value: GetImpl) {
-        self.get_impl = BuilderValue::Set(value);
-    }
-
     pub fn get_max_vectored_read_bytes(&mut self, value: MaxVectoredReadBytes) {
         self.max_vectored_read_bytes = BuilderValue::Set(value);
     }
@@ -763,7 +750,6 @@ impl PageServerConfigBuilder {
                 secondary_download_concurrency,
                 ingest_batch_size,
                 get_vectored_impl,
-                get_impl,
                 max_vectored_read_bytes,
                 validate_vectored_get,
                 ephemeral_bytes_per_memory_kb,
@@ -1049,9 +1035,6 @@ impl PageServerConf {
                 "get_vectored_impl" => {
                     builder.get_vectored_impl(parse_toml_from_str("get_vectored_impl", item)?)
                 }
-                "get_impl" => {
-                    builder.get_impl(parse_toml_from_str("get_impl", item)?)
-                }
                 "max_vectored_read_bytes" => {
                     let bytes = parse_toml_u64("max_vectored_read_bytes", item)? as usize;
                     builder.get_max_vectored_read_bytes(
@@ -1143,7 +1126,6 @@ impl PageServerConf {
             ingest_batch_size: defaults::DEFAULT_INGEST_BATCH_SIZE,
             virtual_file_io_engine: DEFAULT_VIRTUAL_FILE_IO_ENGINE.parse().unwrap(),
             get_vectored_impl: defaults::DEFAULT_GET_VECTORED_IMPL.parse().unwrap(),
-            get_impl: defaults::DEFAULT_GET_IMPL.parse().unwrap(),
             max_vectored_read_bytes: MaxVectoredReadBytes(
                 NonZeroUsize::new(defaults::DEFAULT_MAX_VECTORED_READ_BYTES)
                     .expect("Invalid default constant"),
@@ -1383,7 +1365,6 @@ background_task_maximum_delay = '334 s'
                 ingest_batch_size: defaults::DEFAULT_INGEST_BATCH_SIZE,
                 virtual_file_io_engine: DEFAULT_VIRTUAL_FILE_IO_ENGINE.parse().unwrap(),
                 get_vectored_impl: defaults::DEFAULT_GET_VECTORED_IMPL.parse().unwrap(),
-                get_impl: defaults::DEFAULT_GET_IMPL.parse().unwrap(),
                 max_vectored_read_bytes: MaxVectoredReadBytes(
                     NonZeroUsize::new(defaults::DEFAULT_MAX_VECTORED_READ_BYTES)
                         .expect("Invalid default constant")
@@ -1457,7 +1438,6 @@ background_task_maximum_delay = '334 s'
                 ingest_batch_size: 100,
                 virtual_file_io_engine: DEFAULT_VIRTUAL_FILE_IO_ENGINE.parse().unwrap(),
                 get_vectored_impl: defaults::DEFAULT_GET_VECTORED_IMPL.parse().unwrap(),
-                get_impl: defaults::DEFAULT_GET_IMPL.parse().unwrap(),
                 max_vectored_read_bytes: MaxVectoredReadBytes(
                     NonZeroUsize::new(defaults::DEFAULT_MAX_VECTORED_READ_BYTES)
                         .expect("Invalid default constant")
@@ -1577,7 +1557,6 @@ broker_endpoint = '{broker_endpoint}'
                         endpoint: Some(endpoint.clone()),
                         concurrency_limit: s3_concurrency_limit,
                         max_keys_per_list_response: None,
-                        upload_storage_class: None,
                     }),
                     timeout: RemoteStorageConfig::DEFAULT_TIMEOUT,
                 },

pageserver/src/http/routes.rs

@@ -19,8 +19,6 @@ use pageserver_api::models::LocationConfigListResponse;
 use pageserver_api::models::ShardParameters;
 use pageserver_api::models::TenantDetails;
 use pageserver_api::models::TenantLocationConfigResponse;
-use pageserver_api::models::TenantScanRemoteStorageResponse;
-use pageserver_api::models::TenantScanRemoteStorageShard;
 use pageserver_api::models::TenantShardLocation;
 use pageserver_api::models::TenantShardSplitRequest;
 use pageserver_api::models::TenantShardSplitResponse;
@@ -31,7 +29,6 @@ use pageserver_api::models::{
 };
 use pageserver_api::shard::ShardCount;
 use pageserver_api::shard::TenantShardId;
-use remote_storage::DownloadError;
 use remote_storage::GenericRemoteStorage;
 use remote_storage::TimeTravelError;
 use tenant_size_model::{SizeResult, StorageModel};
@@ -57,9 +54,6 @@ use crate::tenant::mgr::{
 };
 use crate::tenant::mgr::{TenantSlot, UpsertLocationError};
 use crate::tenant::remote_timeline_client;
-use crate::tenant::remote_timeline_client::download_index_part;
-use crate::tenant::remote_timeline_client::list_remote_tenant_shards;
-use crate::tenant::remote_timeline_client::list_remote_timelines;
 use crate::tenant::secondary::SecondaryController;
 use crate::tenant::size::ModelInputs;
 use crate::tenant::storage_layer::LayerAccessStatsReset;
@@ -166,9 +160,6 @@ impl From<PageReconstructError> for ApiError {
     fn from(pre: PageReconstructError) -> ApiError {
         match pre {
             PageReconstructError::Other(pre) => ApiError::InternalServerError(pre),
-            PageReconstructError::MissingKey(e) => {
-                ApiError::InternalServerError(anyhow::anyhow!("{e}"))
-            }
             PageReconstructError::Cancelled => {
                 ApiError::InternalServerError(anyhow::anyhow!("request was cancelled"))
             }
@@ -2041,79 +2032,6 @@ async fn secondary_upload_handler(
     json_response(StatusCode::OK, ())
 }
 
-async fn tenant_scan_remote_handler(
-    request: Request<Body>,
-    cancel: CancellationToken,
-) -> Result<Response<Body>, ApiError> {
-    let state = get_state(&request);
-    let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
-
-    let Some(remote_storage) = state.remote_storage.as_ref() else {
-        return Err(ApiError::BadRequest(anyhow::anyhow!(
-            "Remote storage not configured"
-        )));
-    };
-
-    let mut response = TenantScanRemoteStorageResponse::default();
-
-    let (shards, _other_keys) =
-        list_remote_tenant_shards(remote_storage, tenant_id, cancel.clone())
-            .await
-            .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
-
-    for tenant_shard_id in shards {
-        let (timeline_ids, _other_keys) =
-            list_remote_timelines(remote_storage, tenant_shard_id, cancel.clone())
-                .await
-                .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
-
-        let mut generation = Generation::none();
-        for timeline_id in timeline_ids {
-            match download_index_part(
-                remote_storage,
-                &tenant_shard_id,
-                &timeline_id,
-                Generation::MAX,
-                &cancel,
-            )
-            .instrument(info_span!("download_index_part",
-                         tenant_id=%tenant_shard_id.tenant_id,
-                         shard_id=%tenant_shard_id.shard_slug(),
-                         %timeline_id))
-            .await
-            {
-                Ok((index_part, index_generation)) => {
-                    tracing::info!("Found timeline {tenant_shard_id}/{timeline_id} metadata (gen {index_generation:?}, {} layers, {} consistent LSN)",
-                        index_part.layer_metadata.len(), index_part.get_disk_consistent_lsn());
-                    generation = std::cmp::max(generation, index_generation);
-                }
-                Err(DownloadError::NotFound) => {
-                    // This is normal for tenants that were created with multiple shards: they have an unsharded path
-                    // containing the timeline's initdb tarball but no index.  Otherwise it is a bit strange.
-                    tracing::info!("Timeline path {tenant_shard_id}/{timeline_id} exists in remote storage but has no index, skipping");
-                    continue;
-                }
-                Err(e) => {
-                    return Err(ApiError::InternalServerError(anyhow::anyhow!(e)));
-                }
-            };
-        }
-
-        response.shards.push(TenantScanRemoteStorageShard {
-            tenant_shard_id,
-            generation: generation.into(),
-        });
-    }
-
-    if response.shards.is_empty() {
-        return Err(ApiError::NotFound(
-            anyhow::anyhow!("No shards found for tenant ID {tenant_id}").into(),
-        ));
-    }
-
-    json_response(StatusCode::OK, response)
-}
-
 async fn secondary_download_handler(
     request: Request<Body>,
     _cancel: CancellationToken,
@@ -2510,9 +2428,6 @@ pub fn make_router(
         .post("/v1/tenant/:tenant_shard_id/heatmap_upload", |r| {
             api_handler(r, secondary_upload_handler)
         })
-        .get("/v1/tenant/:tenant_id/scan_remote_storage", |r| {
-            api_handler(r, tenant_scan_remote_handler)
-        })
         .put("/v1/disk_usage_eviction/run", |r| {
             api_handler(r, disk_usage_eviction_run)
         })

pageserver/src/lib.rs

@@ -12,7 +12,6 @@ pub mod disk_usage_eviction_task;
 pub mod http;
 pub mod import_datadir;
 pub use pageserver_api::keyspace;
-pub mod aux_file;
 pub mod metrics;
 pub mod page_cache;
 pub mod page_service;

pageserver/src/metrics.rs

@@ -51,9 +51,6 @@ pub(crate) enum StorageTimeOperation {
     #[strum(serialize = "gc")]
     Gc,
 
-    #[strum(serialize = "update gc info")]
-    UpdateGcInfo,
-
     #[strum(serialize = "create tenant")]
     CreateTenant,
 }
@@ -89,58 +86,41 @@ pub(crate) static STORAGE_TIME_GLOBAL: Lazy<HistogramVec> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
-pub(crate) static READ_NUM_LAYERS_VISITED: Lazy<Histogram> = Lazy::new(|| {
+pub(crate) static READ_NUM_FS_LAYERS: Lazy<Histogram> = Lazy::new(|| {
     register_histogram!(
-        "pageserver_layers_visited_per_read_global",
-        "Number of layers visited to reconstruct one key",
-        vec![1.0, 4.0, 8.0, 16.0, 32.0, 64.0, 128.0, 256.0, 512.0, 1024.0],
-    )
-    .expect("failed to define a metric")
-});
-
-pub(crate) static VEC_READ_NUM_LAYERS_VISITED: Lazy<Histogram> = Lazy::new(|| {
-    register_histogram!(
-        "pageserver_layers_visited_per_vectored_read_global",
-        "Average number of layers visited to reconstruct one key",
-        vec![1.0, 4.0, 8.0, 16.0, 32.0, 64.0, 128.0, 256.0, 512.0, 1024.0],
+        "pageserver_read_num_fs_layers",
+        "Number of persistent layers accessed for processing a read request, including those in the cache",
+        vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 10.0, 20.0, 50.0, 100.0],
     )
     .expect("failed to define a metric")
 });
 
 // Metrics collected on operations on the storage repository.
-#[derive(
-    Clone, Copy, enum_map::Enum, strum_macros::EnumString, strum_macros::Display, IntoStaticStr,
-)]
-pub(crate) enum GetKind {
-    Singular,
-    Vectored,
-}
 
 pub(crate) struct ReconstructTimeMetrics {
-    singular: Histogram,
-    vectored: Histogram,
+    ok: Histogram,
+    err: Histogram,
 }
 
 pub(crate) static RECONSTRUCT_TIME: Lazy<ReconstructTimeMetrics> = Lazy::new(|| {
     let inner = register_histogram_vec!(
         "pageserver_getpage_reconstruct_seconds",
         "Time spent in reconstruct_value (reconstruct a page from deltas)",
-        &["get_kind"],
+        &["result"],
         CRITICAL_OP_BUCKETS.into(),
     )
     .expect("failed to define a metric");
-
     ReconstructTimeMetrics {
-        singular: inner.with_label_values(&[GetKind::Singular.into()]),
-        vectored: inner.with_label_values(&[GetKind::Vectored.into()]),
+        ok: inner.get_metric_with_label_values(&["ok"]).unwrap(),
+        err: inner.get_metric_with_label_values(&["err"]).unwrap(),
     }
 });
 
 impl ReconstructTimeMetrics {
-    pub(crate) fn for_get_kind(&self, get_kind: GetKind) -> &Histogram {
-        match get_kind {
-            GetKind::Singular => &self.singular,
-            GetKind::Vectored => &self.vectored,
+    pub(crate) fn for_result<T, E>(&self, result: &Result<T, E>) -> &Histogram {
+        match result {
+            Ok(_) => &self.ok,
+            Err(_) => &self.err,
         }
     }
 }
@@ -153,33 +133,13 @@ pub(crate) static MATERIALIZED_PAGE_CACHE_HIT_DIRECT: Lazy<IntCounter> = Lazy::n
     .expect("failed to define a metric")
 });
 
-pub(crate) struct ReconstructDataTimeMetrics {
-    singular: Histogram,
-    vectored: Histogram,
-}
-
-impl ReconstructDataTimeMetrics {
-    pub(crate) fn for_get_kind(&self, get_kind: GetKind) -> &Histogram {
-        match get_kind {
-            GetKind::Singular => &self.singular,
-            GetKind::Vectored => &self.vectored,
-        }
-    }
-}
-
-pub(crate) static GET_RECONSTRUCT_DATA_TIME: Lazy<ReconstructDataTimeMetrics> = Lazy::new(|| {
-    let inner = register_histogram_vec!(
+pub(crate) static GET_RECONSTRUCT_DATA_TIME: Lazy<Histogram> = Lazy::new(|| {
+    register_histogram!(
         "pageserver_getpage_get_reconstruct_data_seconds",
         "Time spent in get_reconstruct_value_data",
-        &["get_kind"],
         CRITICAL_OP_BUCKETS.into(),
     )
-    .expect("failed to define a metric");
-
-    ReconstructDataTimeMetrics {
-        singular: inner.with_label_values(&[GetKind::Singular.into()]),
-        vectored: inner.with_label_values(&[GetKind::Vectored.into()]),
-    }
+    .expect("failed to define a metric")
 });
 
 pub(crate) static MATERIALIZED_PAGE_CACHE_HIT: Lazy<IntCounter> = Lazy::new(|| {
@@ -1522,6 +1482,35 @@ pub(crate) static DELETION_QUEUE: Lazy<DeletionQueueMetrics> = Lazy::new(|| {
 }
 });
 
+pub(crate) struct WalIngestMetrics {
+    pub(crate) bytes_received: IntCounter,
+    pub(crate) records_received: IntCounter,
+    pub(crate) records_committed: IntCounter,
+    pub(crate) records_filtered: IntCounter,
+}
+
+pub(crate) static WAL_INGEST: Lazy<WalIngestMetrics> = Lazy::new(|| WalIngestMetrics {
+    bytes_received: register_int_counter!(
+        "pageserver_wal_ingest_bytes_received",
+        "Bytes of WAL ingested from safekeepers",
+    )
+    .unwrap(),
+    records_received: register_int_counter!(
+        "pageserver_wal_ingest_records_received",
+        "Number of WAL records received from safekeepers"
+    )
+    .expect("failed to define a metric"),
+    records_committed: register_int_counter!(
+        "pageserver_wal_ingest_records_committed",
+        "Number of WAL records which resulted in writes to pageserver storage"
+    )
+    .expect("failed to define a metric"),
+    records_filtered: register_int_counter!(
+        "pageserver_wal_ingest_records_filtered",
+        "Number of WAL records filtered out due to sharding"
+    )
+    .expect("failed to define a metric"),
+});
 pub(crate) struct SecondaryModeMetrics {
     pub(crate) upload_heatmap: IntCounter,
     pub(crate) upload_heatmap_errors: IntCounter,
@@ -1529,8 +1518,7 @@ pub(crate) struct SecondaryModeMetrics {
     pub(crate) download_heatmap: IntCounter,
     pub(crate) download_layer: IntCounter,
 }
-pub(crate) static SECONDARY_MODE: Lazy<SecondaryModeMetrics> = Lazy::new(|| {
-    SecondaryModeMetrics {
+pub(crate) static SECONDARY_MODE: Lazy<SecondaryModeMetrics> = Lazy::new(|| SecondaryModeMetrics {
     upload_heatmap: register_int_counter!(
         "pageserver_secondary_upload_heatmap",
         "Number of heatmaps written to remote storage by attached tenants"
@@ -1548,7 +1536,7 @@ pub(crate) static SECONDARY_MODE: Lazy<SecondaryModeMetrics> = Lazy::new(|| {
     .expect("failed to define a metric"),
     download_heatmap: register_int_counter!(
         "pageserver_secondary_download_heatmap",
-        "Number of downloads of heatmaps by secondary mode locations, including when it hasn't changed"
+        "Number of downloads of heatmaps by secondary mode locations"
     )
     .expect("failed to define a metric"),
     download_layer: register_int_counter!(
@@ -1556,7 +1544,6 @@ pub(crate) static SECONDARY_MODE: Lazy<SecondaryModeMetrics> = Lazy::new(|| {
         "Number of downloads of layers by secondary mode locations"
     )
     .expect("failed to define a metric"),
-}
 });
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
@@ -1723,43 +1710,6 @@ macro_rules! redo_bytes_histogram_count_buckets {
     };
 }
 
-pub(crate) struct WalIngestMetrics {
-    pub(crate) bytes_received: IntCounter,
-    pub(crate) records_received: IntCounter,
-    pub(crate) records_committed: IntCounter,
-    pub(crate) records_filtered: IntCounter,
-    pub(crate) time_spent_on_ingest: Histogram,
-}
-
-pub(crate) static WAL_INGEST: Lazy<WalIngestMetrics> = Lazy::new(|| WalIngestMetrics {
-    bytes_received: register_int_counter!(
-        "pageserver_wal_ingest_bytes_received",
-        "Bytes of WAL ingested from safekeepers",
-    )
-    .unwrap(),
-    records_received: register_int_counter!(
-        "pageserver_wal_ingest_records_received",
-        "Number of WAL records received from safekeepers"
-    )
-    .expect("failed to define a metric"),
-    records_committed: register_int_counter!(
-        "pageserver_wal_ingest_records_committed",
-        "Number of WAL records which resulted in writes to pageserver storage"
-    )
-    .expect("failed to define a metric"),
-    records_filtered: register_int_counter!(
-        "pageserver_wal_ingest_records_filtered",
-        "Number of WAL records filtered out due to sharding"
-    )
-    .expect("failed to define a metric"),
-    time_spent_on_ingest: register_histogram!(
-        "pageserver_wal_ingest_put_value_seconds",
-        "Actual time spent on ingesting a record",
-        redo_histogram_time_buckets!(),
-    )
-    .expect("failed to define a metric"),
-});
-
 pub(crate) static WAL_REDO_TIME: Lazy<Histogram> = Lazy::new(|| {
     register_histogram!(
         "pageserver_wal_redo_seconds",
@@ -1913,22 +1863,6 @@ impl StorageTimeMetricsTimer {
         self.metrics.timeline_count.inc();
         self.metrics.global_histogram.observe(duration);
     }
-
-    /// Turns this timer into a timer, which will always record -- usually this means recording
-    /// regardless an early `?` path was taken in a function.
-    pub(crate) fn record_on_drop(self) -> AlwaysRecordingStorageTimeMetricsTimer {
-        AlwaysRecordingStorageTimeMetricsTimer(Some(self))
-    }
-}
-
-pub(crate) struct AlwaysRecordingStorageTimeMetricsTimer(Option<StorageTimeMetricsTimer>);
-
-impl Drop for AlwaysRecordingStorageTimeMetricsTimer {
-    fn drop(&mut self) {
-        if let Some(inner) = self.0.take() {
-            inner.stop_and_record();
-        }
-    }
 }
 
 /// Timing facilities for an globally histogrammed metric, which is supported by per tenant and
@@ -1989,7 +1923,6 @@ pub(crate) struct TimelineMetrics {
     pub imitate_logical_size_histo: StorageTimeMetrics,
     pub load_layer_map_histo: StorageTimeMetrics,
     pub garbage_collect_histo: StorageTimeMetrics,
-    pub update_gc_info_histo: StorageTimeMetrics,
     pub last_record_gauge: IntGauge,
     resident_physical_size_gauge: UIntGauge,
     /// copy of LayeredTimeline.current_logical_size
@@ -2050,12 +1983,6 @@ impl TimelineMetrics {
             &shard_id,
             &timeline_id,
         );
-        let update_gc_info_histo = StorageTimeMetrics::new(
-            StorageTimeOperation::UpdateGcInfo,
-            &tenant_id,
-            &shard_id,
-            &timeline_id,
-        );
         let last_record_gauge = LAST_RECORD_LSN
             .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
             .unwrap();
@@ -2098,7 +2025,6 @@ impl TimelineMetrics {
             logical_size_histo,
             imitate_logical_size_histo,
             garbage_collect_histo,
-            update_gc_info_histo,
             load_layer_map_histo,
             last_record_gauge,
             resident_physical_size_gauge,
@@ -2843,8 +2769,7 @@ pub fn preinitialize_metrics() {
 
     // histograms
     [
-        &READ_NUM_LAYERS_VISITED,
-        &VEC_READ_NUM_LAYERS_VISITED,
+        &READ_NUM_FS_LAYERS,
         &WAIT_LSN_TIME,
         &WAL_REDO_TIME,
         &WAL_REDO_RECORDS_HISTOGRAM,

pageserver/src/page_service.rs

@@ -1,5 +1,13 @@
+//
 //! The Page Service listens for client connections and serves their GetPage@LSN
 //! requests.
+//
+//   It is possible to connect here using usual psql/pgbench/libpq. Following
+// commands are supported now:
+//     *status* -- show actual info about this pageserver,
+//     *pagestream* -- enter mode where smgr and pageserver talk with their
+//  custom protocol.
+//
 
 use anyhow::Context;
 use async_compression::tokio::write::GzipEncoder;
@@ -15,7 +23,7 @@ use pageserver_api::models::{
     PagestreamErrorResponse, PagestreamExistsRequest, PagestreamExistsResponse,
     PagestreamFeMessage, PagestreamGetPageRequest, PagestreamGetPageResponse,
     PagestreamGetSlruSegmentRequest, PagestreamGetSlruSegmentResponse, PagestreamNblocksRequest,
-    PagestreamNblocksResponse, PagestreamProtocolVersion,
+    PagestreamNblocksResponse,
 };
 use pageserver_api::shard::ShardIndex;
 use pageserver_api::shard::ShardNumber;
@@ -543,7 +551,6 @@ impl PageServerHandler {
         pgb: &mut PostgresBackend<IO>,
         tenant_id: TenantId,
         timeline_id: TimelineId,
-        protocol_version: PagestreamProtocolVersion,
         ctx: RequestContext,
     ) -> Result<(), QueryError>
     where
@@ -606,15 +613,14 @@ impl PageServerHandler {
                 t.trace(&copy_data_bytes)
             }
 
-            let neon_fe_msg =
-                PagestreamFeMessage::parse(&mut copy_data_bytes.reader(), protocol_version)?;
+            let neon_fe_msg = PagestreamFeMessage::parse(&mut copy_data_bytes.reader())?;
 
             // TODO: We could create a new per-request context here, with unique ID.
             // Currently we use the same per-timeline context for all requests
 
             let (response, span) = match neon_fe_msg {
                 PagestreamFeMessage::Exists(req) => {
-                    let span = tracing::info_span!("handle_get_rel_exists_request", rel = %req.rel, req_lsn = %req.request_lsn);
+                    let span = tracing::info_span!("handle_get_rel_exists_request", rel = %req.rel, req_lsn = %req.lsn);
                     (
                         self.handle_get_rel_exists_request(tenant_id, timeline_id, &req, &ctx)
                             .instrument(span.clone())
@@ -623,7 +629,7 @@ impl PageServerHandler {
                     )
                 }
                 PagestreamFeMessage::Nblocks(req) => {
-                    let span = tracing::info_span!("handle_get_nblocks_request", rel = %req.rel, req_lsn = %req.request_lsn);
+                    let span = tracing::info_span!("handle_get_nblocks_request", rel = %req.rel, req_lsn = %req.lsn);
                     (
                         self.handle_get_nblocks_request(tenant_id, timeline_id, &req, &ctx)
                             .instrument(span.clone())
@@ -633,7 +639,7 @@ impl PageServerHandler {
                 }
                 PagestreamFeMessage::GetPage(req) => {
                     // shard_id is filled in by the handler
-                    let span = tracing::info_span!("handle_get_page_at_lsn_request", rel = %req.rel, blkno = %req.blkno, req_lsn = %req.request_lsn);
+                    let span = tracing::info_span!("handle_get_page_at_lsn_request", rel = %req.rel, blkno = %req.blkno, req_lsn = %req.lsn);
                     (
                         self.handle_get_page_at_lsn_request(tenant_id, timeline_id, &req, &ctx)
                             .instrument(span.clone())
@@ -642,7 +648,7 @@ impl PageServerHandler {
                     )
                 }
                 PagestreamFeMessage::DbSize(req) => {
-                    let span = tracing::info_span!("handle_db_size_request", dbnode = %req.dbnode, req_lsn = %req.request_lsn);
+                    let span = tracing::info_span!("handle_db_size_request", dbnode = %req.dbnode, req_lsn = %req.lsn);
                     (
                         self.handle_db_size_request(tenant_id, timeline_id, &req, &ctx)
                             .instrument(span.clone())
@@ -651,7 +657,7 @@ impl PageServerHandler {
                     )
                 }
                 PagestreamFeMessage::GetSlruSegment(req) => {
-                    let span = tracing::info_span!("handle_get_slru_segment_request", kind = %req.kind, segno = %req.segno, req_lsn = %req.request_lsn);
+                    let span = tracing::info_span!("handle_get_slru_segment_request", kind = %req.kind, segno = %req.segno, req_lsn = %req.lsn);
                     (
                         self.handle_get_slru_segment_request(tenant_id, timeline_id, &req, &ctx)
                             .instrument(span.clone())
@@ -832,80 +838,78 @@ impl PageServerHandler {
     /// Helper function to handle the LSN from client request.
     ///
     /// Each GetPage (and Exists and Nblocks) request includes information about
-    /// which version of the page is being requested. The primary compute node
-    /// will always request the latest page version, by setting 'request_lsn' to
-    /// the last inserted or flushed WAL position, while a standby will request
-    /// a version at the LSN that it's currently caught up to.
+    /// which version of the page is being requested. The client can request the
+    /// latest version of the page, or the version that's valid at a particular
+    /// LSN. The primary compute node will always request the latest page
+    /// version, while a standby will request a version at the LSN that it's
+    /// currently caught up to.
     ///
     /// In either case, if the page server hasn't received the WAL up to the
     /// requested LSN yet, we will wait for it to arrive. The return value is
     /// the LSN that should be used to look up the page versions.
-    ///
-    /// In addition to the request LSN, each request carries another LSN,
-    /// 'not_modified_since', which is a hint to the pageserver that the client
-    /// knows that the page has not been modified between 'not_modified_since'
-    /// and the request LSN. This allows skipping the wait, as long as the WAL
-    /// up to 'not_modified_since' has arrived. If the client doesn't have any
-    /// information about when the page was modified, it will use
-    /// not_modified_since == lsn. If the client lies and sends a too low
-    /// not_modified_hint such that there are in fact later page versions, the
-    /// behavior is undefined: the pageserver may return any of the page versions
-    /// or an error.
     async fn wait_or_get_last_lsn(
         timeline: &Timeline,
-        request_lsn: Lsn,
-        not_modified_since: Lsn,
+        mut lsn: Lsn,
+        latest: bool,
         latest_gc_cutoff_lsn: &RcuReadGuard<Lsn>,
         ctx: &RequestContext,
     ) -> Result<Lsn, PageStreamError> {
-        let last_record_lsn = timeline.get_last_record_lsn();
+        if latest {
+            // Latest page version was requested. If LSN is given, it is a hint
+            // to the page server that there have been no modifications to the
+            // page after that LSN. If we haven't received WAL up to that point,
+            // wait until it arrives.
+            let last_record_lsn = timeline.get_last_record_lsn();
 
-        // Sanity check the request
-        if request_lsn < not_modified_since {
-            return Err(PageStreamError::BadRequest(
-                format!(
-                    "invalid request with request LSN {} and not_modified_since {}",
-                    request_lsn, not_modified_since,
-                )
-                .into(),
-            ));
-        }
-
-        if request_lsn < **latest_gc_cutoff_lsn {
-            // Check explicitly for INVALID just to get a less scary error message if the
-            // request is obviously bogus
-            return Err(if request_lsn == Lsn::INVALID {
-                PageStreamError::BadRequest("invalid LSN(0) in request".into())
+            // Note: this covers the special case that lsn == Lsn(0). That
+            // special case means "return the latest version whatever it is",
+            // and it's used for bootstrapping purposes, when the page server is
+            // connected directly to the compute node. That is needed because
+            // when you connect to the compute node, to receive the WAL, the
+            // walsender process will do a look up in the pg_authid catalog
+            // table for authentication. That poses a deadlock problem: the
+            // catalog table lookup will send a GetPage request, but the GetPage
+            // request will block in the page server because the recent WAL
+            // hasn't been received yet, and it cannot be received until the
+            // walsender completes the authentication and starts streaming the
+            // WAL.
+            if lsn <= last_record_lsn {
+                lsn = last_record_lsn;
             } else {
-                PageStreamError::BadRequest(format!(
-                        "tried to request a page version that was garbage collected. requested at {} gc cutoff {}",
-                        request_lsn, **latest_gc_cutoff_lsn
-                    ).into())
-            });
-        }
-
-        // Wait for WAL up to 'not_modified_since' to arrive, if necessary
-        if not_modified_since > last_record_lsn {
+                timeline
+                    .wait_lsn(
+                        lsn,
+                        crate::tenant::timeline::WaitLsnWaiter::PageService,
+                        ctx,
+                    )
+                    .await?;
+                // Since we waited for 'lsn' to arrive, that is now the last
+                // record LSN. (Or close enough for our purposes; the
+                // last-record LSN can advance immediately after we return
+                // anyway)
+            }
+        } else {
+            if lsn == Lsn(0) {
+                return Err(PageStreamError::BadRequest(
+                    "invalid LSN(0) in request".into(),
+                ));
+            }
             timeline
                 .wait_lsn(
-                    not_modified_since,
+                    lsn,
                     crate::tenant::timeline::WaitLsnWaiter::PageService,
                     ctx,
                 )
                 .await?;
-            // Since we waited for 'not_modified_since' to arrive, that is now the last
-            // record LSN. (Or close enough for our purposes; the last-record LSN can
-            // advance immediately after we return anyway)
-            Ok(not_modified_since)
-        } else {
-            // It might be better to use max(not_modified_since, latest_gc_cutoff_lsn)
-            // here instead. That would give the same result, since we know that there
-            // haven't been any modifications since 'not_modified_since'. Using an older
-            // LSN might be faster, because that could allow skipping recent layers when
-            // finding the page. However, we have historically used 'last_record_lsn', so
-            // stick to that for now.
-            Ok(std::cmp::min(last_record_lsn, request_lsn))
         }
+
+        if lsn < **latest_gc_cutoff_lsn {
+            return Err(PageStreamError::BadRequest(format!(
+                "tried to request a page version that was garbage collected. requested at {} gc cutoff {}",
+                lsn, **latest_gc_cutoff_lsn
+            ).into()));
+        }
+        Ok(lsn)
     }
 
     #[instrument(skip_all, fields(shard_id))]
@@ -922,17 +926,12 @@ impl PageServerHandler {
             .start_timer(metrics::SmgrQueryType::GetRelExists, ctx);
 
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
-        let lsn = Self::wait_or_get_last_lsn(
-            timeline,
-            req.request_lsn,
-            req.not_modified_since,
-            &latest_gc_cutoff_lsn,
-            ctx,
-        )
-        .await?;
+        let lsn =
+            Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
+                .await?;
 
         let exists = timeline
-            .get_rel_exists(req.rel, Version::Lsn(lsn), ctx)
+            .get_rel_exists(req.rel, Version::Lsn(lsn), req.latest, ctx)
             .await?;
 
         Ok(PagestreamBeMessage::Exists(PagestreamExistsResponse {
@@ -955,17 +954,12 @@ impl PageServerHandler {
             .start_timer(metrics::SmgrQueryType::GetRelSize, ctx);
 
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
-        let lsn = Self::wait_or_get_last_lsn(
-            timeline,
-            req.request_lsn,
-            req.not_modified_since,
-            &latest_gc_cutoff_lsn,
-            ctx,
-        )
-        .await?;
+        let lsn =
+            Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
+                .await?;
 
         let n_blocks = timeline
-            .get_rel_size(req.rel, Version::Lsn(lsn), ctx)
+            .get_rel_size(req.rel, Version::Lsn(lsn), req.latest, ctx)
             .await?;
 
         Ok(PagestreamBeMessage::Nblocks(PagestreamNblocksResponse {
@@ -988,17 +982,18 @@ impl PageServerHandler {
             .start_timer(metrics::SmgrQueryType::GetDbSize, ctx);
 
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
-        let lsn = Self::wait_or_get_last_lsn(
-            timeline,
-            req.request_lsn,
-            req.not_modified_since,
-            &latest_gc_cutoff_lsn,
-            ctx,
-        )
-        .await?;
+        let lsn =
+            Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
+                .await?;
 
         let total_blocks = timeline
-            .get_db_size(DEFAULTTABLESPACE_OID, req.dbnode, Version::Lsn(lsn), ctx)
+            .get_db_size(
+                DEFAULTTABLESPACE_OID,
+                req.dbnode,
+                Version::Lsn(lsn),
+                req.latest,
+                ctx,
+            )
             .await?;
         let db_size = total_blocks as i64 * BLCKSZ as i64;
 
@@ -1165,17 +1160,12 @@ impl PageServerHandler {
             .start_timer(metrics::SmgrQueryType::GetPageAtLsn, ctx);
 
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
-        let lsn = Self::wait_or_get_last_lsn(
-            timeline,
-            req.request_lsn,
-            req.not_modified_since,
-            &latest_gc_cutoff_lsn,
-            ctx,
-        )
-        .await?;
+        let lsn =
+            Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
+                .await?;
 
         let page = timeline
-            .get_rel_page_at_lsn(req.rel, req.blkno, Version::Lsn(lsn), ctx)
+            .get_rel_page_at_lsn(req.rel, req.blkno, Version::Lsn(lsn), req.latest, ctx)
             .await?;
 
         Ok(PagestreamBeMessage::GetPage(PagestreamGetPageResponse {
@@ -1198,14 +1188,9 @@ impl PageServerHandler {
             .start_timer(metrics::SmgrQueryType::GetSlruSegment, ctx);
 
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
-        let lsn = Self::wait_or_get_last_lsn(
-            timeline,
-            req.request_lsn,
-            req.not_modified_since,
-            &latest_gc_cutoff_lsn,
-            ctx,
-        )
-        .await?;
+        let lsn =
+            Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
+                .await?;
 
         let kind = SlruKind::from_repr(req.kind)
             .ok_or(PageStreamError::BadRequest("invalid SLRU kind".into()))?;
@@ -1216,10 +1201,6 @@ impl PageServerHandler {
         ))
     }
 
-    /// Note on "fullbackup":
-    /// Full basebackups should only be used for debugging purposes.
-    /// Originally, it was introduced to enable breaking storage format changes,
-    /// but that is not applicable anymore.
     #[allow(clippy::too_many_arguments)]
     #[instrument(skip_all, fields(shard_id, ?lsn, ?prev_lsn, %full_backup))]
     async fn handle_basebackup_request<IO>(
@@ -1423,34 +1404,7 @@ where
 
         let ctx = self.connection_ctx.attached_child();
         debug!("process query {query_string:?}");
-        if query_string.starts_with("pagestream_v2 ") {
-            let (_, params_raw) = query_string.split_at("pagestream_v2 ".len());
-            let params = params_raw.split(' ').collect::<Vec<_>>();
-            if params.len() != 2 {
-                return Err(QueryError::Other(anyhow::anyhow!(
-                    "invalid param number for pagestream command"
-                )));
-            }
-            let tenant_id = TenantId::from_str(params[0])
-                .with_context(|| format!("Failed to parse tenant id from {}", params[0]))?;
-            let timeline_id = TimelineId::from_str(params[1])
-                .with_context(|| format!("Failed to parse timeline id from {}", params[1]))?;
-
-            tracing::Span::current()
-                .record("tenant_id", field::display(tenant_id))
-                .record("timeline_id", field::display(timeline_id));
-
-            self.check_permission(Some(tenant_id))?;
-
-            self.handle_pagerequests(
-                pgb,
-                tenant_id,
-                timeline_id,
-                PagestreamProtocolVersion::V2,
-                ctx,
-            )
-            .await?;
-        } else if query_string.starts_with("pagestream ") {
+        if query_string.starts_with("pagestream ") {
             let (_, params_raw) = query_string.split_at("pagestream ".len());
             let params = params_raw.split(' ').collect::<Vec<_>>();
             if params.len() != 2 {
@@ -1469,14 +1423,8 @@ where
 
             self.check_permission(Some(tenant_id))?;
 
-            self.handle_pagerequests(
-                pgb,
-                tenant_id,
-                timeline_id,
-                PagestreamProtocolVersion::V1,
-                ctx,
-            )
-            .await?;
+            self.handle_pagerequests(pgb, tenant_id, timeline_id, ctx)
+                .await?;
         } else if query_string.starts_with("basebackup ") {
             let (_, params_raw) = query_string.split_at("basebackup ".len());
             let params = params_raw.split_whitespace().collect::<Vec<_>>();

pageserver/src/pgdatadir_mapping.rs

@@ -9,7 +9,6 @@
 use super::tenant::{PageReconstructError, Timeline};
 use crate::context::RequestContext;
 use crate::keyspace::{KeySpace, KeySpaceAccum};
-use crate::metrics::WAL_INGEST;
 use crate::repository::*;
 use crate::span::debug_assert_current_span_has_tenant_and_timeline_id_no_shard_id;
 use crate::walrecord::NeonWalRecord;
@@ -176,6 +175,7 @@ impl Timeline {
         tag: RelTag,
         blknum: BlockNumber,
         version: Version<'_>,
+        latest: bool,
         ctx: &RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         if tag.relnode == 0 {
@@ -184,7 +184,7 @@ impl Timeline {
             ));
         }
 
-        let nblocks = self.get_rel_size(tag, version, ctx).await?;
+        let nblocks = self.get_rel_size(tag, version, latest, ctx).await?;
         if blknum >= nblocks {
             debug!(
                 "read beyond EOF at {} blk {} at {}, size is {}: returning all-zeros page",
@@ -206,6 +206,7 @@ impl Timeline {
         spcnode: Oid,
         dbnode: Oid,
         version: Version<'_>,
+        latest: bool,
         ctx: &RequestContext,
     ) -> Result<usize, PageReconstructError> {
         let mut total_blocks = 0;
@@ -213,7 +214,7 @@ impl Timeline {
         let rels = self.list_rels(spcnode, dbnode, version, ctx).await?;
 
         for rel in rels {
-            let n_blocks = self.get_rel_size(rel, version, ctx).await?;
+            let n_blocks = self.get_rel_size(rel, version, latest, ctx).await?;
             total_blocks += n_blocks as usize;
         }
         Ok(total_blocks)
@@ -224,6 +225,7 @@ impl Timeline {
         &self,
         tag: RelTag,
         version: Version<'_>,
+        latest: bool,
         ctx: &RequestContext,
     ) -> Result<BlockNumber, PageReconstructError> {
         if tag.relnode == 0 {
@@ -237,7 +239,7 @@ impl Timeline {
         }
 
         if (tag.forknum == FSM_FORKNUM || tag.forknum == VISIBILITYMAP_FORKNUM)
-            && !self.get_rel_exists(tag, version, ctx).await?
+            && !self.get_rel_exists(tag, version, latest, ctx).await?
         {
             // FIXME: Postgres sometimes calls smgrcreate() to create
             // FSM, and smgrnblocks() on it immediately afterwards,
@@ -250,8 +252,16 @@ impl Timeline {
         let mut buf = version.get(self, key, ctx).await?;
         let nblocks = buf.get_u32_le();
 
-        self.update_cached_rel_size(tag, version.get_lsn(), nblocks);
-
+        if latest {
+            // Update relation size cache only if "latest" flag is set.
+            // This flag is set by compute when it is working with most recent version of relation.
+            // Typically master compute node always set latest=true.
+            // Please notice, that even if compute node "by mistake" specifies old LSN but set
+            // latest=true, then it can not cause cache corruption, because with latest=true
+            // pageserver choose max(request_lsn, last_written_lsn) and so cached value will be
+            // associated with most recent value of LSN.
+            self.update_cached_rel_size(tag, version.get_lsn(), nblocks);
+        }
         Ok(nblocks)
     }
 
@@ -260,6 +270,7 @@ impl Timeline {
         &self,
         tag: RelTag,
         version: Version<'_>,
+        _latest: bool,
         ctx: &RequestContext,
     ) -> Result<bool, PageReconstructError> {
         if tag.relnode == 0 {
@@ -455,12 +466,6 @@ impl Timeline {
                 // Didn't find any commit timestamps smaller than the request
                 Ok(LsnForTimestamp::Past(min_lsn))
             }
-            (true, _) if commit_lsn < min_lsn => {
-                // the search above did set found_smaller to true but it never increased the lsn.
-                // Then, low is still the old min_lsn, and the subtraction above gave a value
-                // below the min_lsn. We should never do that.
-                Ok(LsnForTimestamp::Past(min_lsn))
-            }
             (true, false) => {
                 // Only found commits with timestamps smaller than the request.
                 // It's still a valid case for branch creation, return it.
@@ -812,7 +817,7 @@ impl Timeline {
     /// Get cached size of relation if it not updated after specified LSN
     pub fn get_cached_rel_size(&self, tag: &RelTag, lsn: Lsn) -> Option<BlockNumber> {
         let rel_size_cache = self.rel_size_cache.read().unwrap();
-        if let Some((cached_lsn, nblocks)) = rel_size_cache.map.get(tag) {
+        if let Some((cached_lsn, nblocks)) = rel_size_cache.get(tag) {
             if lsn >= *cached_lsn {
                 return Some(*nblocks);
             }
@@ -823,16 +828,7 @@ impl Timeline {
     /// Update cached relation size if there is no more recent update
     pub fn update_cached_rel_size(&self, tag: RelTag, lsn: Lsn, nblocks: BlockNumber) {
         let mut rel_size_cache = self.rel_size_cache.write().unwrap();
-
-        if lsn < rel_size_cache.complete_as_of {
-            // Do not cache old values. It's safe to cache the size on read, as long as
-            // the read was at an LSN since we started the WAL ingestion. Reasoning: we
-            // never evict values from the cache, so if the relation size changed after
-            // 'lsn', the new value is already in the cache.
-            return;
-        }
-
-        match rel_size_cache.map.entry(tag) {
+        match rel_size_cache.entry(tag) {
             hash_map::Entry::Occupied(mut entry) => {
                 let cached_lsn = entry.get_mut();
                 if lsn >= cached_lsn.0 {
@@ -848,13 +844,13 @@ impl Timeline {
     /// Store cached relation size
     pub fn set_cached_rel_size(&self, tag: RelTag, lsn: Lsn, nblocks: BlockNumber) {
         let mut rel_size_cache = self.rel_size_cache.write().unwrap();
-        rel_size_cache.map.insert(tag, (lsn, nblocks));
+        rel_size_cache.insert(tag, (lsn, nblocks));
     }
 
     /// Remove cached relation size
     pub fn remove_cached_rel_size(&self, tag: &RelTag) {
         let mut rel_size_cache = self.rel_size_cache.write().unwrap();
-        rel_size_cache.map.remove(tag);
+        rel_size_cache.remove(tag);
     }
 }
 
@@ -1092,7 +1088,7 @@ impl<'a> DatadirModification<'a> {
     ) -> anyhow::Result<()> {
         let total_blocks = self
             .tline
-            .get_db_size(spcnode, dbnode, Version::Modified(self), ctx)
+            .get_db_size(spcnode, dbnode, Version::Modified(self), true, ctx)
             .await?;
 
         // Remove entry from dbdir
@@ -1191,7 +1187,7 @@ impl<'a> DatadirModification<'a> {
         anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
         if self
             .tline
-            .get_rel_exists(rel, Version::Modified(self), ctx)
+            .get_rel_exists(rel, Version::Modified(self), true, ctx)
             .await?
         {
             let size_key = rel_size_to_key(rel);
@@ -1405,7 +1401,7 @@ impl<'a> DatadirModification<'a> {
         let n_files;
         let mut aux_files = self.tline.aux_files.lock().await;
         if let Some(mut dir) = aux_files.dir.take() {
-            // We already updated aux files in `self`: emit a delta and update our latest value.
+            // We already updated aux files in `self`: emit a delta and update our latest value
             dir.upsert(file_path.clone(), content.clone());
             n_files = dir.files.len();
             if aux_files.n_deltas == MAX_AUX_FILE_DELTAS {
@@ -1450,14 +1446,10 @@ impl<'a> DatadirModification<'a> {
                     // reset the map.
                     return Err(e.into());
                 }
-                // Note: we added missing key error variant in https://github.com/neondatabase/neon/pull/7393 but
-                // the original code assumes all other errors are missing keys. Therefore, we keep the code path
-                // the same for now, though in theory, we should only match the `MissingKey` variant.
-                Err(
-                    PageReconstructError::Other(_)
-                    | PageReconstructError::WalRedo(_)
-                    | PageReconstructError::MissingKey { .. },
-                ) => {
+                // FIXME: PageReconstructError doesn't have an explicit variant for key-not-found, so
+                // we are assuming that all _other_ possible errors represents a missing key.  If some
+                // other error occurs, we may incorrectly reset the map of aux files.
+                Err(PageReconstructError::Other(_) | PageReconstructError::WalRedo(_)) => {
                     // Key is missing, we must insert an image as the basis for subsequent deltas.
 
                     let mut dir = AuxFilesDirectory {
@@ -1549,8 +1541,6 @@ impl<'a> DatadirModification<'a> {
     pub async fn commit(&mut self, ctx: &RequestContext) -> anyhow::Result<()> {
         let mut writer = self.tline.writer().await;
 
-        let timer = WAL_INGEST.time_spent_on_ingest.start_timer();
-
         let pending_nblocks = self.pending_nblocks;
         self.pending_nblocks = 0;
 
@@ -1590,8 +1580,6 @@ impl<'a> DatadirModification<'a> {
             writer.update_directory_entries_count(kind, count as u64);
         }
 
-        timer.observe_duration();
-
         Ok(())
     }
 

pageserver/src/repository.rs

@@ -33,52 +33,6 @@ impl Value {
     }
 }
 
-#[cfg(test)]
-#[derive(Debug, PartialEq)]
-pub(crate) enum InvalidInput {
-    TooShortValue,
-    TooShortPostgresRecord,
-}
-
-/// We could have a ValueRef where everything is `serde(borrow)`. Before implementing that, lets
-/// use this type for querying if a slice looks some particular way.
-#[cfg(test)]
-pub(crate) struct ValueBytes;
-
-#[cfg(test)]
-impl ValueBytes {
-    pub(crate) fn will_init(raw: &[u8]) -> Result<bool, InvalidInput> {
-        if raw.len() < 12 {
-            return Err(InvalidInput::TooShortValue);
-        }
-
-        let value_discriminator = &raw[0..4];
-
-        if value_discriminator == [0, 0, 0, 0] {
-            // Value::Image always initializes
-            return Ok(true);
-        }
-
-        if value_discriminator != [0, 0, 0, 1] {
-            // not a Value::WalRecord(..)
-            return Ok(false);
-        }
-
-        let walrecord_discriminator = &raw[4..8];
-
-        if walrecord_discriminator != [0, 0, 0, 0] {
-            // only NeonWalRecord::Postgres can have will_init
-            return Ok(false);
-        }
-
-        if raw.len() < 17 {
-            return Err(InvalidInput::TooShortPostgresRecord);
-        }
-
-        Ok(raw[8] == 1)
-    }
-}
-
 #[cfg(test)]
 mod test {
     use super::*;
@@ -116,8 +70,6 @@ mod test {
         ];
 
         roundtrip!(image, expected);
-
-        assert!(ValueBytes::will_init(&expected).unwrap());
     }
 
     #[test]
@@ -141,96 +93,6 @@ mod test {
         ];
 
         roundtrip!(rec, expected);
-
-        assert!(ValueBytes::will_init(&expected).unwrap());
-    }
-
-    #[test]
-    fn bytes_inspection_too_short_image() {
-        let rec = Value::Image(Bytes::from_static(b""));
-
-        #[rustfmt::skip]
-        let expected = [
-            // top level discriminator of 4 bytes
-            0x00, 0x00, 0x00, 0x00,
-            // 8 byte length
-            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-        ];
-
-        roundtrip!(rec, expected);
-
-        assert!(ValueBytes::will_init(&expected).unwrap());
-        assert_eq!(expected.len(), 12);
-        for len in 0..12 {
-            assert_eq!(
-                ValueBytes::will_init(&expected[..len]).unwrap_err(),
-                InvalidInput::TooShortValue
-            );
-        }
-    }
-
-    #[test]
-    fn bytes_inspection_too_short_postgres_record() {
-        let rec = NeonWalRecord::Postgres {
-            will_init: false,
-            rec: Bytes::from_static(b""),
-        };
-        let rec = Value::WalRecord(rec);
-
-        #[rustfmt::skip]
-        let expected = [
-            // flattened discriminator of total 8 bytes
-            0x00, 0x00, 0x00, 0x01,
-            0x00, 0x00, 0x00, 0x00,
-            // will_init
-            0x00,
-            // 8 byte length
-            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-        ];
-
-        roundtrip!(rec, expected);
-
-        assert!(!ValueBytes::will_init(&expected).unwrap());
-        assert_eq!(expected.len(), 17);
-        for len in 12..17 {
-            assert_eq!(
-                ValueBytes::will_init(&expected[..len]).unwrap_err(),
-                InvalidInput::TooShortPostgresRecord
-            )
-        }
-        for len in 0..12 {
-            assert_eq!(
-                ValueBytes::will_init(&expected[..len]).unwrap_err(),
-                InvalidInput::TooShortValue
-            )
-        }
-    }
-
-    #[test]
-    fn clear_visibility_map_flags_example() {
-        let rec = NeonWalRecord::ClearVisibilityMapFlags {
-            new_heap_blkno: Some(0x11),
-            old_heap_blkno: None,
-            flags: 0x03,
-        };
-        let rec = Value::WalRecord(rec);
-
-        #[rustfmt::skip]
-        let expected = [
-            // discriminators
-            0x00, 0x00, 0x00, 0x01,
-            0x00, 0x00, 0x00, 0x01,
-            // Some == 1 followed by 4 bytes
-            0x01, 0x00, 0x00, 0x00, 0x11,
-            // None == 0
-            0x00,
-            // flags
-            0x03
-        ];
-
-        roundtrip!(rec, expected);
-
-        assert!(!ValueBytes::will_init(&expected).unwrap());
     }
 }
 

pageserver/src/task_mgr.rs

@@ -361,8 +361,6 @@ pub enum TaskKind {
 
     DebugTool,
 
-    EphemeralFilePreWarmPageCache,
-
     #[cfg(test)]
     UnitTest,
 }

pageserver/src/tenant.rs

@@ -559,10 +559,9 @@ impl Tenant {
             // By doing what we do here, the index part upload is retried.
             // If control plane retries timeline creation in the meantime, the mgmt API handler
             // for timeline creation will coalesce on the upload we queue here.
-            // FIXME: this branch should be dead code as we no longer write local metadata.
             let rtc = timeline.remote_client.as_ref().unwrap();
             rtc.init_upload_queue_for_empty_remote(&metadata)?;
-            rtc.schedule_index_upload_for_full_metadata_update(&metadata)?;
+            rtc.schedule_index_upload_for_metadata_update(&metadata)?;
         }
 
         timeline
@@ -888,7 +887,7 @@ impl Tenant {
 
     #[instrument(skip_all)]
     pub(crate) async fn preload(
-        self: &Arc<Self>,
+        self: &Arc<Tenant>,
         remote_storage: &GenericRemoteStorage,
         cancel: CancellationToken,
     ) -> anyhow::Result<TenantPreload> {
@@ -918,13 +917,9 @@ impl Tenant {
 
         Ok(TenantPreload {
             deleting,
-            timelines: Self::load_timeline_metadata(
-                self,
-                remote_timeline_ids,
-                remote_storage,
-                cancel,
-            )
-            .await?,
+            timelines: self
+                .load_timeline_metadata(remote_timeline_ids, remote_storage, cancel)
+                .await?,
         })
     }
 
@@ -2874,23 +2869,20 @@ impl Tenant {
                 }
             }
 
-            let cutoff = timeline
-                .get_last_record_lsn()
-                .checked_sub(horizon)
-                .unwrap_or(Lsn(0));
+            if let Some(cutoff) = timeline.get_last_record_lsn().checked_sub(horizon) {
+                let branchpoints: Vec<Lsn> = all_branchpoints
+                    .range((
+                        Included((timeline_id, Lsn(0))),
+                        Included((timeline_id, Lsn(u64::MAX))),
+                    ))
+                    .map(|&x| x.1)
+                    .collect();
+                timeline
+                    .update_gc_info(branchpoints, cutoff, pitr, cancel, ctx)
+                    .await?;
 
-            let branchpoints: Vec<Lsn> = all_branchpoints
-                .range((
-                    Included((timeline_id, Lsn(0))),
-                    Included((timeline_id, Lsn(u64::MAX))),
-                ))
-                .map(|&x| x.1)
-                .collect();
-            timeline
-                .update_gc_info(branchpoints, cutoff, pitr, cancel, ctx)
-                .await?;
-
-            gc_timelines.push(timeline);
+                gc_timelines.push(timeline);
+            }
         }
         drop(gc_cs);
         Ok(gc_timelines)
@@ -3035,7 +3027,7 @@ impl Tenant {
         // See also https://github.com/neondatabase/neon/issues/3865
         if let Some(remote_client) = new_timeline.remote_client.as_ref() {
             remote_client
-                .schedule_index_upload_for_full_metadata_update(&metadata)
+                .schedule_index_upload_for_metadata_update(&metadata)
                 .context("branch initial metadata upload")?;
         }
 
@@ -3406,11 +3398,7 @@ impl Tenant {
         // is in progress (which is not a common case).
         //
         // See more for on the issue #2748 condenced out of the initial PR review.
-        let mut shared_cache = tokio::select! {
-            locked = self.cached_logical_sizes.lock() => locked,
-            _ = cancel.cancelled() => anyhow::bail!("cancelled"),
-            _ = self.cancel.cancelled() => anyhow::bail!("tenant is shutting down"),
-        };
+        let mut shared_cache = self.cached_logical_sizes.lock().await;
 
         size::gather_inputs(
             self,
@@ -3672,7 +3660,6 @@ pub(crate) mod harness {
                 image_layer_creation_check_threshold: Some(
                     tenant_conf.image_layer_creation_check_threshold,
                 ),
-                switch_to_aux_file_v2: Some(tenant_conf.switch_to_aux_file_v2),
             }
         }
     }
@@ -3861,8 +3848,6 @@ pub(crate) mod harness {
 
 #[cfg(test)]
 mod tests {
-    use std::collections::BTreeMap;
-
     use super::*;
     use crate::keyspace::KeySpaceAccum;
     use crate::repository::{Key, Value};
@@ -3871,11 +3856,9 @@ mod tests {
     use crate::DEFAULT_PG_VERSION;
     use bytes::BytesMut;
     use hex_literal::hex;
-    use pageserver_api::key::NON_INHERITED_RANGE;
     use pageserver_api::keyspace::KeySpace;
     use rand::{thread_rng, Rng};
-    use tests::storage_layer::ValuesReconstructState;
-    use tests::timeline::{GetVectoredError, ShutdownMode};
+    use tests::timeline::ShutdownMode;
 
     static TEST_KEY: Lazy<Key> =
         Lazy::new(|| Key::from_slice(&hex!("010000000033333333444444445500000001")));
@@ -4663,9 +4646,7 @@ mod tests {
         for read in reads {
             info!("Doing vectored read on {:?}", read);
 
-            let vectored_res = tline
-                .get_vectored_impl(read.clone(), reads_lsn, ValuesReconstructState::new(), &ctx)
-                .await;
+            let vectored_res = tline.get_vectored_impl(read.clone(), reads_lsn, &ctx).await;
             tline
                 .validate_get_vectored_impl(&vectored_res, read, reads_lsn, &ctx)
                 .await;
@@ -4674,67 +4655,6 @@ mod tests {
         Ok(())
     }
 
-    #[tokio::test]
-    async fn test_get_vectored_aux_files() -> anyhow::Result<()> {
-        let harness = TenantHarness::create("test_get_vectored_aux_files")?;
-
-        let (tenant, ctx) = harness.load().await;
-        let tline = tenant
-            .create_empty_timeline(TIMELINE_ID, Lsn(0), DEFAULT_PG_VERSION, &ctx)
-            .await?;
-        let tline = tline.raw_timeline().unwrap();
-
-        let mut modification = tline.begin_modification(Lsn(0x1000));
-        modification.put_file("foo/bar1", b"content1", &ctx).await?;
-        modification.set_lsn(Lsn(0x1008))?;
-        modification.put_file("foo/bar2", b"content2", &ctx).await?;
-        modification.commit(&ctx).await?;
-
-        let child_timeline_id = TimelineId::generate();
-        tenant
-            .branch_timeline_test(
-                tline,
-                child_timeline_id,
-                Some(tline.get_last_record_lsn()),
-                &ctx,
-            )
-            .await?;
-
-        let child_timeline = tenant
-            .get_timeline(child_timeline_id, true)
-            .expect("Should have the branched timeline");
-
-        let aux_keyspace = KeySpace {
-            ranges: vec![NON_INHERITED_RANGE],
-        };
-        let read_lsn = child_timeline.get_last_record_lsn();
-
-        let vectored_res = child_timeline
-            .get_vectored_impl(
-                aux_keyspace.clone(),
-                read_lsn,
-                ValuesReconstructState::new(),
-                &ctx,
-            )
-            .await;
-
-        child_timeline
-            .validate_get_vectored_impl(&vectored_res, aux_keyspace, read_lsn, &ctx)
-            .await;
-
-        let images = vectored_res?;
-        let mut key = NON_INHERITED_RANGE.start;
-        while key < NON_INHERITED_RANGE.end {
-            assert!(matches!(
-                images[&key],
-                Err(PageReconstructError::MissingKey(_))
-            ));
-            key = key.next();
-        }
-
-        Ok(())
-    }
-
     // Test that vectored get handles layer gaps correctly
     // by advancing into the next ancestor timeline if required.
     //
@@ -4863,12 +4783,7 @@ mod tests {
             ranges: vec![key_near_gap..gap_at_key.next(), key_near_end..current_key],
         };
         let results = child_timeline
-            .get_vectored_impl(
-                read.clone(),
-                current_lsn,
-                ValuesReconstructState::new(),
-                &ctx,
-            )
+            .get_vectored_impl(read.clone(), current_lsn, &ctx)
             .await?;
 
         for (key, img_res) in results {
@@ -4879,167 +4794,6 @@ mod tests {
         Ok(())
     }
 
-    // Test that vectored get descends into ancestor timelines correctly and
-    // does not return an image that's newer than requested.
-    //
-    // The diagram below ilustrates an interesting case. We have a parent timeline
-    // (top of the Lsn range) and a child timeline. The request key cannot be reconstructed
-    // from the child timeline, so the parent timeline must be visited. When advacing into
-    // the child timeline, the read path needs to remember what the requested Lsn was in
-    // order to avoid returning an image that's too new. The test below constructs such
-    // a timeline setup and does a few queries around the Lsn of each page image.
-    // ```
-    //    LSN
-    //     ^
-    //     |
-    //     |
-    // 500 | --------------------------------------> branch point
-    // 400 |        X
-    // 300 |        X
-    // 200 | --------------------------------------> requested lsn
-    // 100 |        X
-    //     |---------------------------------------> Key
-    //              |
-    //              ------> requested key
-    //
-    // Legend:
-    // * X - page images
-    // ```
-    #[tokio::test]
-    async fn test_get_vectored_ancestor_descent() -> anyhow::Result<()> {
-        let harness = TenantHarness::create("test_get_vectored_on_lsn_axis")?;
-        let (tenant, ctx) = harness.load().await;
-
-        let start_key = Key::from_hex("010000000033333333444444445500000000").unwrap();
-        let end_key = start_key.add(1000);
-        let child_gap_at_key = start_key.add(500);
-        let mut parent_gap_lsns: BTreeMap<Lsn, String> = BTreeMap::new();
-
-        let mut current_lsn = Lsn(0x10);
-
-        let timeline_id = TimelineId::generate();
-        let parent_timeline = tenant
-            .create_test_timeline(timeline_id, current_lsn, DEFAULT_PG_VERSION, &ctx)
-            .await?;
-
-        current_lsn += 0x100;
-
-        for _ in 0..3 {
-            let mut key = start_key;
-            while key < end_key {
-                current_lsn += 0x10;
-
-                let image_value = format!("{} at {}", child_gap_at_key, current_lsn);
-
-                let mut writer = parent_timeline.writer().await;
-                writer
-                    .put(
-                        key,
-                        current_lsn,
-                        &Value::Image(test_img(&image_value)),
-                        &ctx,
-                    )
-                    .await?;
-                writer.finish_write(current_lsn);
-
-                if key == child_gap_at_key {
-                    parent_gap_lsns.insert(current_lsn, image_value);
-                }
-
-                key = key.next();
-            }
-
-            parent_timeline.freeze_and_flush().await?;
-        }
-
-        let child_timeline_id = TimelineId::generate();
-
-        let child_timeline = tenant
-            .branch_timeline_test(&parent_timeline, child_timeline_id, Some(current_lsn), &ctx)
-            .await?;
-
-        let mut key = start_key;
-        while key < end_key {
-            if key == child_gap_at_key {
-                key = key.next();
-                continue;
-            }
-
-            current_lsn += 0x10;
-
-            let mut writer = child_timeline.writer().await;
-            writer
-                .put(
-                    key,
-                    current_lsn,
-                    &Value::Image(test_img(&format!("{} at {}", key, current_lsn))),
-                    &ctx,
-                )
-                .await?;
-            writer.finish_write(current_lsn);
-
-            key = key.next();
-        }
-
-        child_timeline.freeze_and_flush().await?;
-
-        let lsn_offsets: [i64; 5] = [-10, -1, 0, 1, 10];
-        let mut query_lsns = Vec::new();
-        for image_lsn in parent_gap_lsns.keys().rev() {
-            for offset in lsn_offsets {
-                query_lsns.push(Lsn(image_lsn
-                    .0
-                    .checked_add_signed(offset)
-                    .expect("Shouldn't overflow")));
-            }
-        }
-
-        for query_lsn in query_lsns {
-            let results = child_timeline
-                .get_vectored_impl(
-                    KeySpace {
-                        ranges: vec![child_gap_at_key..child_gap_at_key.next()],
-                    },
-                    query_lsn,
-                    ValuesReconstructState::new(),
-                    &ctx,
-                )
-                .await;
-
-            let expected_item = parent_gap_lsns
-                .iter()
-                .rev()
-                .find(|(lsn, _)| **lsn <= query_lsn);
-
-            info!(
-                "Doing vectored read at LSN {}. Expecting image to be: {:?}",
-                query_lsn, expected_item
-            );
-
-            match expected_item {
-                Some((_, img_value)) => {
-                    let key_results = results.expect("No vectored get error expected");
-                    let key_result = &key_results[&child_gap_at_key];
-                    let returned_img = key_result
-                        .as_ref()
-                        .expect("No page reconstruct error expected");
-
-                    info!(
-                        "Vectored read at LSN {} returned image {}",
-                        query_lsn,
-                        std::str::from_utf8(returned_img)?
-                    );
-                    assert_eq!(*returned_img, test_img(img_value));
-                }
-                None => {
-                    assert!(matches!(results, Err(GetVectoredError::MissingKey(_))));
-                }
-            }
-        }
-
-        Ok(())
-    }
-
     #[tokio::test]
     async fn test_random_updates() -> anyhow::Result<()> {
         let harness = TenantHarness::create("test_random_updates")?;

pageserver/src/tenant/config.rs

@@ -369,10 +369,6 @@ pub struct TenantConf {
     // How much WAL must be ingested before checking again whether a new image layer is required.
     // Expresed in multiples of checkpoint distance.
     pub image_layer_creation_check_threshold: u8,
-
-    /// Switch to aux file v2. Switching this flag requires the user has not written any aux file into
-    /// the storage before, and this flag cannot be switched back. Otherwise there will be data corruptions.
-    pub switch_to_aux_file_v2: bool,
 }
 
 /// Same as TenantConf, but this struct preserves the information about
@@ -468,10 +464,6 @@ pub struct TenantConfOpt {
 
     #[serde(skip_serializing_if = "Option::is_none")]
     pub image_layer_creation_check_threshold: Option<u8>,
-
-    #[serde(skip_serializing_if = "Option::is_none")]
-    #[serde(default)]
-    pub switch_to_aux_file_v2: Option<bool>,
 }
 
 impl TenantConfOpt {
@@ -529,9 +521,6 @@ impl TenantConfOpt {
             image_layer_creation_check_threshold: self
                 .image_layer_creation_check_threshold
                 .unwrap_or(global_conf.image_layer_creation_check_threshold),
-            switch_to_aux_file_v2: self
-                .switch_to_aux_file_v2
-                .unwrap_or(global_conf.switch_to_aux_file_v2),
         }
     }
 }
@@ -573,7 +562,6 @@ impl Default for TenantConf {
             lazy_slru_download: false,
             timeline_get_throttle: crate::tenant::throttle::Config::disabled(),
             image_layer_creation_check_threshold: DEFAULT_IMAGE_LAYER_CREATION_CHECK_THRESHOLD,
-            switch_to_aux_file_v2: false,
         }
     }
 }
@@ -648,7 +636,6 @@ impl From<TenantConfOpt> for models::TenantConfig {
             lazy_slru_download: value.lazy_slru_download,
             timeline_get_throttle: value.timeline_get_throttle.map(ThrottleConfig::from),
             image_layer_creation_check_threshold: value.image_layer_creation_check_threshold,
-            switch_to_aux_file_v2: value.switch_to_aux_file_v2,
         }
     }
 }

pageserver/src/tenant/ephemeral_file.rs

@@ -21,6 +21,7 @@ pub struct EphemeralFile {
 }
 
 mod page_caching;
+mod zero_padded_buffer;
 mod zero_padded_read_write;
 
 impl EphemeralFile {

pageserver/src/tenant/ephemeral_file/page_caching.rs

@@ -6,9 +6,7 @@ use crate::page_cache::{self, PAGE_SZ};
 use crate::tenant::block_io::BlockLease;
 use crate::virtual_file::VirtualFile;
 
-use once_cell::sync::Lazy;
-use std::io::{self, ErrorKind};
-use tokio_epoll_uring::BoundedBuf;
+use std::io;
 use tracing::*;
 
 use super::zero_padded_read_write;
@@ -16,18 +14,14 @@ use super::zero_padded_read_write;
 /// See module-level comment.
 pub struct RW {
     page_cache_file_id: page_cache::FileId,
-    rw: super::zero_padded_read_write::RW<PreWarmingWriter>,
+    rw: super::zero_padded_read_write::RW,
 }
 
 impl RW {
     pub fn new(file: VirtualFile) -> Self {
-        let page_cache_file_id = page_cache::next_file_id();
         Self {
-            page_cache_file_id,
-            rw: super::zero_padded_read_write::RW::new(PreWarmingWriter::new(
-                page_cache_file_id,
-                file,
-            )),
+            page_cache_file_id: page_cache::next_file_id(),
+            rw: super::zero_padded_read_write::RW::new(file),
         }
     }
 
@@ -51,7 +45,7 @@ impl RW {
         ctx: &RequestContext,
     ) -> Result<BlockLease, io::Error> {
         match self.rw.read_blk(blknum).await? {
-            zero_padded_read_write::ReadResult::NeedsReadFromWriter { writer } => {
+            zero_padded_read_write::ReadResult::NeedsReadFromVirtualFile { virtual_file } => {
                 let cache = page_cache::get();
                 match cache
                     .read_immutable_buf(self.page_cache_file_id, blknum, ctx)
@@ -63,7 +57,7 @@ impl RW {
                             format!(
                                 "ephemeral file: read immutable page #{}: {}: {:#}",
                                 blknum,
-                                self.rw.as_writer().file.path,
+                                self.rw.as_inner_virtual_file().path,
                                 e,
                             ),
                         )
@@ -72,8 +66,7 @@ impl RW {
                         return Ok(BlockLease::PageReadGuard(guard))
                     }
                     page_cache::ReadBufResult::NotFound(write_guard) => {
-                        let write_guard = writer
-                            .file
+                        let write_guard = virtual_file
                             .read_exact_at_page(write_guard, blknum as u64 * PAGE_SZ as u64)
                             .await?;
                         let read_guard = write_guard.mark_valid();
@@ -94,7 +87,7 @@ impl Drop for RW {
         // We leave them there, [`crate::page_cache::PageCache::find_victim`] will evict them when needed.
 
         // unlink the file
-        let res = std::fs::remove_file(&self.rw.as_writer().file.path);
+        let res = std::fs::remove_file(&self.rw.as_inner_virtual_file().path);
         if let Err(e) = res {
             if e.kind() != std::io::ErrorKind::NotFound {
                 // just never log the not found errors, we cannot do anything for them; on detach
@@ -103,116 +96,10 @@ impl Drop for RW {
                 // not found files might also be related to https://github.com/neondatabase/neon/issues/2442
                 error!(
                     "could not remove ephemeral file '{}': {}",
-                    self.rw.as_writer().file.path,
+                    self.rw.as_inner_virtual_file().path,
                     e
                 );
             }
         }
     }
 }
-
-struct PreWarmingWriter {
-    nwritten_blocks: u32,
-    page_cache_file_id: page_cache::FileId,
-    file: VirtualFile,
-}
-
-impl PreWarmingWriter {
-    fn new(page_cache_file_id: page_cache::FileId, file: VirtualFile) -> Self {
-        Self {
-            nwritten_blocks: 0,
-            page_cache_file_id,
-            file,
-        }
-    }
-}
-
-impl crate::virtual_file::owned_buffers_io::write::OwnedAsyncWriter for PreWarmingWriter {
-    async fn write_all<
-        B: tokio_epoll_uring::BoundedBuf<Buf = Buf>,
-        Buf: tokio_epoll_uring::IoBuf + Send,
-    >(
-        &mut self,
-        buf: B,
-    ) -> std::io::Result<(usize, B::Buf)> {
-        let buf = buf.slice(..);
-        let saved_bounds = buf.bounds(); // save for reconstructing the Slice from iobuf after the IO is done
-        let check_bounds_stuff_works = if cfg!(test) && cfg!(debug_assertions) {
-            Some(buf.to_vec())
-        } else {
-            None
-        };
-        let buflen = buf.len();
-        assert_eq!(
-            buflen % PAGE_SZ,
-            0,
-            "{buflen} ; we know TAIL_SZ is a PAGE_SZ multiple, and write_buffered_borrowed is used"
-        );
-
-        // Do the IO.
-        let iobuf = match self.file.write_all(buf).await {
-            (iobuf, Ok(nwritten)) => {
-                assert_eq!(nwritten, buflen);
-                iobuf
-            }
-            (_, Err(e)) => {
-                return Err(std::io::Error::new(
-                    ErrorKind::Other,
-                    // order error before path because path is long and error is short
-                    format!(
-                        "ephemeral_file: write_blob: write-back tail self.nwritten_blocks={}, buflen={}, {:#}: {}",
-                        self.nwritten_blocks, buflen, e, self.file.path,
-                    ),
-                ));
-            }
-        };
-
-        // Reconstruct the Slice (the write path consumed the Slice and returned us the underlying IoBuf)
-        let buf = tokio_epoll_uring::Slice::from_buf_bounds(iobuf, saved_bounds);
-        if let Some(check_bounds_stuff_works) = check_bounds_stuff_works {
-            assert_eq!(&check_bounds_stuff_works, &*buf);
-        }
-
-        // Pre-warm page cache with the contents.
-        // At least in isolated bulk ingest benchmarks (test_bulk_insert.py), the pre-warming
-        // benefits the code that writes InMemoryLayer=>L0 layers.
-        let nblocks = buflen / PAGE_SZ;
-        let nblocks32 = u32::try_from(nblocks).unwrap();
-        let cache = page_cache::get();
-        static CTX: Lazy<RequestContext> = Lazy::new(|| {
-            RequestContext::new(
-                crate::task_mgr::TaskKind::EphemeralFilePreWarmPageCache,
-                crate::context::DownloadBehavior::Error,
-            )
-        });
-        for blknum_in_buffer in 0..nblocks {
-            let blk_in_buffer = &buf[blknum_in_buffer * PAGE_SZ..(blknum_in_buffer + 1) * PAGE_SZ];
-            let blknum = self
-                .nwritten_blocks
-                .checked_add(blknum_in_buffer as u32)
-                .unwrap();
-            match cache
-                .read_immutable_buf(self.page_cache_file_id, blknum, &CTX)
-                .await
-            {
-                Err(e) => {
-                    error!("ephemeral_file write_blob failed to get immutable buf to pre-warm page cache: {e:?}");
-                    // fail gracefully, it's not the end of the world if we can't pre-warm the cache here
-                }
-                Ok(v) => match v {
-                    page_cache::ReadBufResult::Found(_guard) => {
-                        // This function takes &mut self, so, it shouldn't be possible to reach this point.
-                        unreachable!("we just wrote block {blknum} to the VirtualFile, which is owned by Self, \
-                                      and this function takes &mut self, so, no concurrent read_blk is possible");
-                    }
-                    page_cache::ReadBufResult::NotFound(mut write_guard) => {
-                        write_guard.copy_from_slice(blk_in_buffer);
-                        let _ = write_guard.mark_valid();
-                    }
-                },
-            }
-        }
-        self.nwritten_blocks = self.nwritten_blocks.checked_add(nblocks32).unwrap();
-        Ok((buflen, buf.into_inner()))
-    }
-}

pageserver/src/tenant/ephemeral_file/zero_padded_read_write.rs

@@ -3,60 +3,56 @@
 //! # Writes
 //!
 //! [`super::EphemeralFile`] writes small, borrowed buffers using [`RW::write_all_borrowed`].
-//! The [`RW`] batches these into [`TAIL_SZ`] bigger writes, using [`owned_buffers_io::write::BufferedWriter`].
+//! The [`RW`] batches these into [`RW::TAIL_SZ`] bigger writes, using [`owned_buffers_io::write::BufferedWriter`].
 //!
 //! # Reads
 //!
 //! [`super::EphemeralFile`] always reads full [`PAGE_SZ`]ed blocks using [`RW::read_blk`].
 //!
 //! The [`RW`] serves these reads either from the buffered writer's in-memory buffer
-//! or redirects the caller to read from the underlying [`OwnedAsyncWriter`]
-//! if the read is for the prefix that has already been flushed.
+//! or redirects the caller to read from the underlying [`VirtualFile`]` if they have already
+//! been flushed.
 //!
-//! # Current Usage
-//!
-//! The current user of this module is [`super::page_caching::RW`].
+//! The current caller is [`super::page_caching::RW`]. In case it gets redirected to read from
+//! [`VirtualFile`], it consults the [`crate::page_cache`] first.
 
-mod zero_padded;
+mod zero_padded_buffer;
 
 use crate::{
     page_cache::PAGE_SZ,
-    virtual_file::owned_buffers_io::{
-        self,
-        write::{Buffer, OwnedAsyncWriter},
+    tenant::ephemeral_file::zero_padded_buffer,
+    virtual_file::{
+        owned_buffers_io::{self, write::Buffer},
+        VirtualFile,
     },
 };
 
-const TAIL_SZ: usize = 64 * 1024;
-
 /// See module-level comment.
-pub struct RW<W: OwnedAsyncWriter> {
+pub struct RW {
     buffered_writer: owned_buffers_io::write::BufferedWriter<
-        zero_padded::Buffer<TAIL_SZ>,
-        owned_buffers_io::util::size_tracking_writer::Writer<W>,
+        zero_padded_buffer::Buf<{ Self::TAIL_SZ }>,
+        owned_buffers_io::util::size_tracking_writer::Writer<VirtualFile>,
     >,
 }
 
-pub enum ReadResult<'a, W> {
-    NeedsReadFromWriter { writer: &'a W },
+pub enum ReadResult<'a> {
+    NeedsReadFromVirtualFile { virtual_file: &'a VirtualFile },
     ServedFromZeroPaddedMutableTail { buffer: &'a [u8; PAGE_SZ] },
 }
 
-impl<W> RW<W>
-where
-    W: OwnedAsyncWriter,
-{
-    pub fn new(writer: W) -> Self {
-        let bytes_flushed_tracker =
-            owned_buffers_io::util::size_tracking_writer::Writer::new(writer);
+impl RW {
+    const TAIL_SZ: usize = 64 * 1024;
+
+    pub fn new(file: VirtualFile) -> Self {
+        let bytes_flushed_tracker = owned_buffers_io::util::size_tracking_writer::Writer::new(file);
         let buffered_writer = owned_buffers_io::write::BufferedWriter::new(
             bytes_flushed_tracker,
-            zero_padded::Buffer::default(),
+            zero_padded_buffer::Buf::default(),
         );
         Self { buffered_writer }
     }
 
-    pub(crate) fn as_writer(&self) -> &W {
+    pub(crate) fn as_inner_virtual_file(&self) -> &VirtualFile {
         self.buffered_writer.as_inner().as_inner()
     }
 
@@ -66,35 +62,20 @@ where
 
     pub fn bytes_written(&self) -> u64 {
         let flushed_offset = self.buffered_writer.as_inner().bytes_written();
-        let buffer: &zero_padded::Buffer<TAIL_SZ> = self.buffered_writer.inspect_buffer();
+        let buffer: &zero_padded_buffer::Buf<{ Self::TAIL_SZ }> =
+            self.buffered_writer.inspect_buffer();
         flushed_offset + u64::try_from(buffer.pending()).unwrap()
     }
 
-    pub(crate) async fn read_blk(&self, blknum: u32) -> Result<ReadResult<'_, W>, std::io::Error> {
+    pub(crate) async fn read_blk(&self, blknum: u32) -> Result<ReadResult, std::io::Error> {
         let flushed_offset = self.buffered_writer.as_inner().bytes_written();
-        let buffer: &zero_padded::Buffer<TAIL_SZ> = self.buffered_writer.inspect_buffer();
+        let buffer: &zero_padded_buffer::Buf<{ Self::TAIL_SZ }> =
+            self.buffered_writer.inspect_buffer();
         let buffered_offset = flushed_offset + u64::try_from(buffer.pending()).unwrap();
         let read_offset = (blknum as u64) * (PAGE_SZ as u64);
 
-        // The trailing page ("block") might only be partially filled,
-        // yet the blob_io code relies on us to return a full PAGE_SZed slice anyway.
-        // Moreover, it has to be zero-padded, because when we still had
-        // a write-back page cache, it provided pre-zeroed pages, and blob_io came to rely on it.
-        // DeltaLayer probably has the same issue, not sure why it needs no special treatment.
-        // => check here that the read doesn't go beyond this potentially trailing
-        // => the zero-padding is done in the `else` branch below
-        let blocks_written = if buffered_offset % (PAGE_SZ as u64) == 0 {
-            buffered_offset / (PAGE_SZ as u64)
-        } else {
-            (buffered_offset / (PAGE_SZ as u64)) + 1
-        };
-        if (blknum as u64) >= blocks_written {
-            return Err(std::io::Error::new(std::io::ErrorKind::Other, anyhow::anyhow!("read past end of ephemeral_file: read=0x{read_offset:x} buffered=0x{buffered_offset:x} flushed=0x{flushed_offset}")));
-        }
-
-        // assertions for the `if-else` below
         assert_eq!(
-            flushed_offset % (TAIL_SZ as u64), 0,
+            flushed_offset % (Self::TAIL_SZ as u64), 0,
             "we only use write_buffered_borrowed to write to the buffered writer, so it's guaranteed that flushes happen buffer.cap()-sized chunks"
         );
         assert_eq!(
@@ -104,11 +85,29 @@ where
         );
 
         if read_offset < flushed_offset {
-            assert!(read_offset + (PAGE_SZ as u64) <= flushed_offset);
-            Ok(ReadResult::NeedsReadFromWriter {
-                writer: self.as_writer(),
+            assert!(
+                read_offset + (PAGE_SZ as u64) <= flushed_offset,
+                "this impl can't deal with pages spread across flushed & buffered part"
+            );
+            Ok(ReadResult::NeedsReadFromVirtualFile {
+                virtual_file: self.as_inner_virtual_file(),
             })
         } else {
+            let read_until_offset = read_offset + (PAGE_SZ as u64);
+            if !(0..buffered_offset).contains(&read_until_offset) {
+                // The blob_io code relies on the reader allowing reads past
+                // the end of what was written, up to end of the current PAGE_SZ chunk.
+                // This is a relict of the past where we would get a pre-zeroed page from the page cache.
+                //
+                // DeltaLayer probably has the same issue, not sure why it needs no special treatment.
+                let nbytes_past_end = read_until_offset.checked_sub(buffered_offset).unwrap();
+                if nbytes_past_end >= (PAGE_SZ as u64) {
+                    // TODO: treat this as error. Pre-existing issue before this patch.
+                    panic!(
+                        "return IO error: read past end of file: read=0x{read_offset:x} buffered=0x{buffered_offset:x} flushed=0x{flushed_offset}"
+                    )
+                }
+            }
             let read_offset_in_buffer = read_offset
                 .checked_sub(flushed_offset)
                 .expect("would have taken `if` branch instead of this one");

pageserver/src/tenant/ephemeral_file/zero_padded_read_write/zero_padded_buffer.rs

@@ -1,17 +1,11 @@
-//! A [`crate::virtual_file::owned_buffers_io::write::Buffer`] whose
-//! unwritten range is guaranteed to be zero-initialized.
-//! This is used by [`crate::tenant::ephemeral_file::zero_padded_read_write::RW::read_blk`]
-//! to serve page-sized reads of the trailing page when the trailing page has only been partially filled.
-
 use std::mem::MaybeUninit;
 
-/// See module-level comment.
-pub struct Buffer<const N: usize> {
+pub struct Buf<const N: usize> {
     allocation: Box<[u8; N]>,
     written: usize,
 }
 
-impl<const N: usize> Default for Buffer<N> {
+impl<const N: usize> Default for Buf<N> {
     fn default() -> Self {
         Self {
             allocation: Box::new(
@@ -23,7 +17,7 @@ impl<const N: usize> Default for Buffer<N> {
     }
 }
 
-impl<const N: usize> Buffer<N> {
+impl<const N: usize> Buf<N> {
     #[inline(always)]
     fn invariants(&self) {
         // don't check by default, unoptimized is too expensive even for debug mode
@@ -38,7 +32,7 @@ impl<const N: usize> Buffer<N> {
     }
 }
 
-impl<const N: usize> crate::virtual_file::owned_buffers_io::write::Buffer for Buffer<N> {
+impl<const N: usize> crate::virtual_file::owned_buffers_io::write::Buffer for Buf<N> {
     type IoBuf = Self;
 
     fn cap(&self) -> usize {
@@ -47,7 +41,7 @@ impl<const N: usize> crate::virtual_file::owned_buffers_io::write::Buffer for Bu
 
     fn extend_from_slice(&mut self, other: &[u8]) {
         self.invariants();
-        let remaining = self.allocation.len() - self.written;
+        let remaining = self.cap() - other.len();
         if other.len() > remaining {
             panic!("calling extend_from_slice() with insufficient remaining capacity");
         }
@@ -81,24 +75,20 @@ impl<const N: usize> crate::virtual_file::owned_buffers_io::write::Buffer for Bu
     }
 }
 
-/// We have this trait impl so that the `flush` method in the `Buffer` impl above can produce a
-/// [`tokio_epoll_uring::BoundedBuf::slice`] of the [`Self::written`] range of the data.
+/// We have this implementation so that [`Buf<N>`] can be used with [`crate::virtual_file::owned_buffers_io::write::BufferedWriter`].
 ///
-/// Remember that bytes_init is generally _not_ a tracker of the amount
-/// of valid data in the io buffer; we use `Slice` for that.
-/// The `IoBuf` is _only_ for keeping track of uninitialized memory, a bit like MaybeUninit.
 ///
 /// SAFETY:
 ///
 /// The [`Self::allocation`] is stable becauses boxes are stable.
 /// The memory is zero-initialized, so, bytes_init is always N.
-unsafe impl<const N: usize> tokio_epoll_uring::IoBuf for Buffer<N> {
+///
+unsafe impl<const N: usize> tokio_epoll_uring::IoBuf for Buf<N> {
     fn stable_ptr(&self) -> *const u8 {
         self.allocation.as_ptr()
     }
 
     fn bytes_init(&self) -> usize {
-        // Yes, N, not self.written; Read the full comment of this impl block!
         N
     }
 

pageserver/src/tenant/metadata.rs

@@ -235,12 +235,6 @@ impl TimelineMetadata {
         let bytes = instance.to_bytes().unwrap();
         Self::from_bytes(&bytes).unwrap()
     }
-
-    pub(crate) fn apply(&mut self, update: &MetadataUpdate) {
-        self.body.disk_consistent_lsn = update.disk_consistent_lsn;
-        self.body.prev_record_lsn = update.prev_record_lsn;
-        self.body.latest_gc_cutoff_lsn = update.latest_gc_cutoff_lsn;
-    }
 }
 
 impl<'de> Deserialize<'de> for TimelineMetadata {
@@ -265,27 +259,6 @@ impl Serialize for TimelineMetadata {
     }
 }
 
-/// Parts of the metadata which are regularly modified.
-pub(crate) struct MetadataUpdate {
-    disk_consistent_lsn: Lsn,
-    prev_record_lsn: Option<Lsn>,
-    latest_gc_cutoff_lsn: Lsn,
-}
-
-impl MetadataUpdate {
-    pub(crate) fn new(
-        disk_consistent_lsn: Lsn,
-        prev_record_lsn: Option<Lsn>,
-        latest_gc_cutoff_lsn: Lsn,
-    ) -> Self {
-        Self {
-            disk_consistent_lsn,
-            prev_record_lsn,
-            latest_gc_cutoff_lsn,
-        }
-    }
-}
-
 #[cfg(test)]
 mod tests {
     use super::*;

pageserver/src/tenant/mgr.rs

@@ -678,19 +678,12 @@ pub async fn init_tenant_mgr(
                     }
                 }
             }
-            LocationMode::Secondary(secondary_conf) => {
-                info!(
-                    tenant_id = %tenant_shard_id.tenant_id,
-                    shard_id = %tenant_shard_id.shard_slug(),
-                    "Starting secondary tenant"
-                );
-                TenantSlot::Secondary(SecondaryTenant::new(
-                    tenant_shard_id,
-                    shard_identity,
-                    location_conf.tenant_conf,
-                    &secondary_conf,
-                ))
-            }
+            LocationMode::Secondary(secondary_conf) => TenantSlot::Secondary(SecondaryTenant::new(
+                tenant_shard_id,
+                shard_identity,
+                location_conf.tenant_conf,
+                &secondary_conf,
+            )),
         };
 
         tenants.insert(tenant_shard_id, slot);

pageserver/src/tenant/remote_timeline_client.rs

@@ -202,9 +202,7 @@ use std::sync::atomic::{AtomicU32, Ordering};
 use std::sync::{Arc, Mutex};
 use std::time::Duration;
 
-use remote_storage::{
-    DownloadError, GenericRemoteStorage, ListingMode, RemotePath, TimeoutOrCancel,
-};
+use remote_storage::{DownloadError, GenericRemoteStorage, RemotePath, TimeoutOrCancel};
 use std::ops::DerefMut;
 use tracing::{debug, error, info, instrument, warn};
 use tracing::{info_span, Instrument};
@@ -238,14 +236,11 @@ use utils::id::{TenantId, TimelineId};
 
 use self::index::IndexPart;
 
-use super::metadata::MetadataUpdate;
 use super::storage_layer::{Layer, LayerFileName, ResidentLayer};
 use super::upload_queue::SetDeletedFlagProgress;
 use super::Generation;
 
-pub(crate) use download::{
-    download_index_part, is_temp_download_file, list_remote_tenant_shards, list_remote_timelines,
-};
+pub(crate) use download::{is_temp_download_file, list_remote_timelines};
 pub(crate) use index::LayerFileMetadata;
 
 // Occasional network issues and such can cause remote operations to fail, and
@@ -474,7 +469,7 @@ impl RemoteTimelineClient {
             },
         );
 
-        let (index_part, _index_generation) = download::download_index_part(
+        let index_part = download::download_index_part(
             &self.storage_impl,
             &self.tenant_shard_id,
             &self.timeline_id,
@@ -541,10 +536,9 @@ impl RemoteTimelineClient {
     // Upload operations.
     //
 
-    /// Launch an index-file upload operation in the background, with
-    /// fully updated metadata.
     ///
-    /// This should only be used to upload initial metadata to remote storage.
+    /// Launch an index-file upload operation in the background, with
+    /// updated metadata.
     ///
     /// The upload will be added to the queue immediately, but it
     /// won't be performed until all previously scheduled layer file
@@ -556,7 +550,7 @@ impl RemoteTimelineClient {
     /// If there were any changes to the list of files, i.e. if any
     /// layer file uploads were scheduled, since the last index file
     /// upload, those will be included too.
-    pub fn schedule_index_upload_for_full_metadata_update(
+    pub fn schedule_index_upload_for_metadata_update(
         self: &Arc<Self>,
         metadata: &TimelineMetadata,
     ) -> anyhow::Result<()> {
@@ -572,27 +566,6 @@ impl RemoteTimelineClient {
         Ok(())
     }
 
-    /// Launch an index-file upload operation in the background, with only parts of the metadata
-    /// updated.
-    ///
-    /// This is the regular way of updating metadata on layer flushes or Gc.
-    ///
-    /// Using this lighter update mechanism allows for reparenting and detaching without changes to
-    /// `index_part.json`, while being more clear on what values update regularly.
-    pub(crate) fn schedule_index_upload_for_metadata_update(
-        self: &Arc<Self>,
-        update: &MetadataUpdate,
-    ) -> anyhow::Result<()> {
-        let mut guard = self.upload_queue.lock().unwrap();
-        let upload_queue = guard.initialized_mut()?;
-
-        upload_queue.latest_metadata.apply(update);
-
-        self.schedule_index_upload(upload_queue, upload_queue.latest_metadata.clone());
-
-        Ok(())
-    }
-
     ///
     /// Launch an index-file upload operation in the background, if necessary.
     ///
@@ -1149,7 +1122,7 @@ impl RemoteTimelineClient {
         // and retry will arrive to different pageserver there wont be any traces of it on remote storage
         let timeline_storage_path = remote_timeline_path(&self.tenant_shard_id, &self.timeline_id);
 
-        // Execute all pending deletions, so that when we proceed to do a listing below, we aren't
+        // Execute all pending deletions, so that when we proceed to do a list_prefixes below, we aren't
         // taking the burden of listing all the layers that we already know we should delete.
         self.flush_deletion_queue().await?;
 
@@ -1158,20 +1131,14 @@ impl RemoteTimelineClient {
         let remaining = download_retry(
             || async {
                 self.storage_impl
-                    .list(
-                        Some(&timeline_storage_path),
-                        ListingMode::NoDelimiter,
-                        None,
-                        &cancel,
-                    )
+                    .list_files(Some(&timeline_storage_path), None, &cancel)
                     .await
             },
             "list remaining files",
             &cancel,
         )
         .await
-        .context("list files remaining files")?
-        .keys;
+        .context("list files remaining files")?;
 
         // We will delete the current index_part object last, since it acts as a deletion
         // marker via its deleted_at attribute
@@ -1718,11 +1685,6 @@ impl RemoteTimelineClient {
     }
 }
 
-pub fn remote_tenant_path(tenant_shard_id: &TenantShardId) -> RemotePath {
-    let path = format!("tenants/{tenant_shard_id}");
-    RemotePath::from_string(&path).expect("Failed to construct path")
-}
-
 pub fn remote_timelines_path(tenant_shard_id: &TenantShardId) -> RemotePath {
     let path = format!("tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}");
     RemotePath::from_string(&path).expect("Failed to construct path")
@@ -2062,7 +2024,7 @@ mod tests {
         // Schedule upload of index. Check that it is queued
         let metadata = dummy_metadata(Lsn(0x20));
         client
-            .schedule_index_upload_for_full_metadata_update(&metadata)
+            .schedule_index_upload_for_metadata_update(&metadata)
             .unwrap();
         {
             let mut guard = client.upload_queue.lock().unwrap();

pageserver/src/tenant/remote_timeline_client/download.rs

@@ -5,9 +5,9 @@
 
 use std::collections::HashSet;
 use std::future::Future;
-use std::str::FromStr;
 
 use anyhow::{anyhow, Context};
+use bytes::BytesMut;
 use camino::{Utf8Path, Utf8PathBuf};
 use pageserver_api::shard::TenantShardId;
 use tokio::fs::{self, File, OpenOptions};
@@ -26,13 +26,13 @@ use crate::virtual_file::{on_fatal_io_error, MaybeFatalIo, VirtualFile};
 use crate::TEMP_FILE_SUFFIX;
 use remote_storage::{DownloadError, GenericRemoteStorage, ListingMode, RemotePath};
 use utils::crashsafe::path_with_suffix_extension;
-use utils::id::{TenantId, TimelineId};
+use utils::id::TimelineId;
 
 use super::index::{IndexPart, LayerFileMetadata};
 use super::{
     parse_remote_index_path, remote_index_path, remote_initdb_archive_path,
-    remote_initdb_preserved_archive_path, remote_tenant_path, FAILED_DOWNLOAD_WARN_THRESHOLD,
-    FAILED_REMOTE_OP_RETRIES, INITDB_PATH,
+    remote_initdb_preserved_archive_path, FAILED_DOWNLOAD_WARN_THRESHOLD, FAILED_REMOTE_OP_RETRIES,
+    INITDB_PATH,
 };
 
 ///
@@ -183,7 +183,6 @@ async fn download_object<'a>(
         #[cfg(target_os = "linux")]
         crate::virtual_file::io_engine::IoEngine::TokioEpollUring => {
             use crate::virtual_file::owned_buffers_io::{self, util::size_tracking_writer};
-            use bytes::BytesMut;
             async {
                 let destination_file = VirtualFile::create(dst_path)
                     .await
@@ -254,31 +253,42 @@ pub(crate) fn is_temp_download_file(path: &Utf8Path) -> bool {
     }
 }
 
-async fn list_identifiers<T>(
+/// List timelines of given tenant in remote storage
+pub async fn list_remote_timelines(
     storage: &GenericRemoteStorage,
-    prefix: RemotePath,
+    tenant_shard_id: TenantShardId,
     cancel: CancellationToken,
-) -> anyhow::Result<(HashSet<T>, HashSet<String>)>
-where
-    T: FromStr + Eq + std::hash::Hash,
-{
+) -> anyhow::Result<(HashSet<TimelineId>, HashSet<String>)> {
+    let remote_path = remote_timelines_path(&tenant_shard_id);
+
+    fail::fail_point!("storage-sync-list-remote-timelines", |_| {
+        anyhow::bail!("storage-sync-list-remote-timelines");
+    });
+
     let listing = download_retry_forever(
-        || storage.list(Some(&prefix), ListingMode::WithDelimiter, None, &cancel),
-        &format!("list identifiers in prefix {prefix}"),
+        || {
+            storage.list(
+                Some(&remote_path),
+                ListingMode::WithDelimiter,
+                None,
+                &cancel,
+            )
+        },
+        &format!("list timelines for {tenant_shard_id}"),
         &cancel,
     )
     .await?;
 
-    let mut parsed_ids = HashSet::new();
+    let mut timeline_ids = HashSet::new();
     let mut other_prefixes = HashSet::new();
 
-    for id_remote_storage_key in listing.prefixes {
-        let object_name = id_remote_storage_key.object_name().ok_or_else(|| {
-            anyhow::anyhow!("failed to get object name for key {id_remote_storage_key}")
+    for timeline_remote_storage_key in listing.prefixes {
+        let object_name = timeline_remote_storage_key.object_name().ok_or_else(|| {
+            anyhow::anyhow!("failed to get timeline id for remote tenant {tenant_shard_id}")
         })?;
 
-        match object_name.parse::<T>() {
-            Ok(t) => parsed_ids.insert(t),
+        match object_name.parse::<TimelineId>() {
+            Ok(t) => timeline_ids.insert(t),
             Err(_) => other_prefixes.insert(object_name.to_string()),
         };
     }
@@ -290,31 +300,7 @@ where
         other_prefixes.insert(object_name.to_string());
     }
 
-    Ok((parsed_ids, other_prefixes))
-}
-
-/// List shards of given tenant in remote storage
-pub(crate) async fn list_remote_tenant_shards(
-    storage: &GenericRemoteStorage,
-    tenant_id: TenantId,
-    cancel: CancellationToken,
-) -> anyhow::Result<(HashSet<TenantShardId>, HashSet<String>)> {
-    let remote_path = remote_tenant_path(&TenantShardId::unsharded(tenant_id));
-    list_identifiers::<TenantShardId>(storage, remote_path, cancel).await
-}
-
-/// List timelines of given tenant shard in remote storage
-pub async fn list_remote_timelines(
-    storage: &GenericRemoteStorage,
-    tenant_shard_id: TenantShardId,
-    cancel: CancellationToken,
-) -> anyhow::Result<(HashSet<TimelineId>, HashSet<String>)> {
-    fail::fail_point!("storage-sync-list-remote-timelines", |_| {
-        anyhow::bail!("storage-sync-list-remote-timelines");
-    });
-
-    let remote_path = remote_timelines_path(&tenant_shard_id).add_trailing_slash();
-    list_identifiers::<TimelineId>(storage, remote_path, cancel).await
+    Ok((timeline_ids, other_prefixes))
 }
 
 async fn do_download_index_part(
@@ -323,7 +309,7 @@ async fn do_download_index_part(
     timeline_id: &TimelineId,
     index_generation: Generation,
     cancel: &CancellationToken,
-) -> Result<(IndexPart, Generation), DownloadError> {
+) -> Result<IndexPart, DownloadError> {
     let remote_path = remote_index_path(tenant_shard_id, timeline_id, index_generation);
 
     let index_part_bytes = download_retry_forever(
@@ -348,7 +334,7 @@ async fn do_download_index_part(
         .with_context(|| format!("deserialize index part file at {remote_path:?}"))
         .map_err(DownloadError::Other)?;
 
-    Ok((index_part, index_generation))
+    Ok(index_part)
 }
 
 /// index_part.json objects are suffixed with a generation number, so we cannot
@@ -357,13 +343,13 @@ async fn do_download_index_part(
 /// In this function we probe for the most recent index in a generation <= our current generation.
 /// See "Finding the remote indices for timelines" in docs/rfcs/025-generation-numbers.md
 #[tracing::instrument(skip_all, fields(generation=?my_generation))]
-pub(crate) async fn download_index_part(
+pub(super) async fn download_index_part(
     storage: &GenericRemoteStorage,
     tenant_shard_id: &TenantShardId,
     timeline_id: &TimelineId,
     my_generation: Generation,
     cancel: &CancellationToken,
-) -> Result<(IndexPart, Generation), DownloadError> {
+) -> Result<IndexPart, DownloadError> {
     debug_assert_current_span_has_tenant_and_timeline_id();
 
     if my_generation.is_none() {
@@ -432,16 +418,11 @@ pub(crate) async fn download_index_part(
     let index_prefix = remote_index_path(tenant_shard_id, timeline_id, Generation::none());
 
     let indices = download_retry(
-        || async {
-            storage
-                .list(Some(&index_prefix), ListingMode::NoDelimiter, None, cancel)
-                .await
-        },
+        || async { storage.list_files(Some(&index_prefix), None, cancel).await },
         "list index_part files",
         cancel,
     )
-    .await?
-    .keys;
+    .await?;
 
     // General case logic for which index to use: the latest index whose generation
     // is <= our own.  See "Finding the remote indices for timelines" in docs/rfcs/025-generation-numbers.md

pageserver/src/tenant/secondary/downloader.rs

@@ -312,7 +312,7 @@ impl JobGenerator<PendingDownload, RunningDownload, CompleteDownload, DownloadCo
                     (detail.last_download, detail.next_download.unwrap())
                 };
 
-                if now > next_download {
+                if now < next_download {
                     Some(PendingDownload {
                         secondary_state: secondary_tenant,
                         last_download,
@@ -647,12 +647,6 @@ impl<'a> TenantDownloader<'a> {
                 progress.bytes_downloaded += layer_byte_count;
                 progress.layers_downloaded += layer_count;
             }
-
-            for delete_timeline in &delete_timelines {
-                // We haven't removed from disk yet, but optimistically remove from in-memory state: if removal
-                // from disk fails that will be a fatal error.
-                detail.timelines.remove(delete_timeline);
-            }
         }
 
         // Execute accumulated deletions
@@ -716,14 +710,13 @@ impl<'a> TenantDownloader<'a> {
                     .await
                     .map_err(UpdateError::from)?;
 
-                SECONDARY_MODE.download_heatmap.inc();
-
                 if Some(&download.etag) == prev_etag {
                     Ok(HeatMapDownload::Unmodified)
                 } else {
                     let mut heatmap_bytes = Vec::new();
                     let mut body = tokio_util::io::StreamReader::new(download.download_stream);
                     let _size = tokio::io::copy_buf(&mut body, &mut heatmap_bytes).await?;
+                    SECONDARY_MODE.download_heatmap.inc();
                     Ok(HeatMapDownload::Modified(HeatMapModified {
                         etag: download.etag,
                         last_modified: download.last_modified,

pageserver/src/tenant/size.rs

@@ -118,9 +118,6 @@ pub(super) async fn gather_inputs(
     ctx: &RequestContext,
 ) -> anyhow::Result<ModelInputs> {
     // refresh is needed to update gc related pitr_cutoff and horizon_cutoff
-    //
-    // FIXME: if a single timeline is deleted while refresh gc info is ongoing, we will fail the
-    // whole computation. It does not make sense from the billing perspective.
     tenant
         .refresh_gc_info(cancel, ctx)
         .await

pageserver/src/tenant/storage_layer.rs

@@ -118,7 +118,6 @@ pub(crate) struct ValuesReconstructState {
     pub(crate) keys: HashMap<Key, Result<VectoredValueReconstructState, PageReconstructError>>,
 
     keys_done: KeySpaceRandomAccum,
-    layers_visited: u32,
 }
 
 impl ValuesReconstructState {
@@ -126,7 +125,6 @@ impl ValuesReconstructState {
         Self {
             keys: HashMap::new(),
             keys_done: KeySpaceRandomAccum::new(),
-            layers_visited: 0,
         }
     }
 
@@ -140,37 +138,6 @@ impl ValuesReconstructState {
         }
     }
 
-    pub(crate) fn on_layer_visited(&mut self) {
-        self.layers_visited += 1;
-    }
-
-    pub(crate) fn get_layers_visited(&self) -> u32 {
-        self.layers_visited
-    }
-
-    /// This function is called after reading a keyspace from a layer.
-    /// It checks if the read path has now moved past the cached Lsn for any keys.
-    ///
-    /// Implementation note: We intentionally iterate over the keys for which we've
-    /// already collected some reconstruct data. This avoids scaling complexity with
-    /// the size of the search space.
-    pub(crate) fn on_lsn_advanced(&mut self, keyspace: &KeySpace, advanced_to: Lsn) {
-        for (key, value) in self.keys.iter_mut() {
-            if !keyspace.contains(key) {
-                continue;
-            }
-
-            if let Ok(state) = value {
-                if state.situation != ValueReconstructSituation::Complete
-                    && state.get_cached_lsn() >= Some(advanced_to)
-                {
-                    state.situation = ValueReconstructSituation::Complete;
-                    self.keys_done.add_key(*key);
-                }
-            }
-        }
-    }
-
     /// Update the state collected for a given key.
     /// Returns true if this was the last value needed for the key and false otherwise.
     ///
@@ -195,18 +162,11 @@ impl ValuesReconstructState {
                         true
                     }
                     Value::WalRecord(rec) => {
-                        debug_assert!(
-                            Some(lsn) > state.get_cached_lsn(),
-                            "Attempt to collect a record below cached LSN for walredo: {} < {}",
-                            lsn,
-                            state
-                                .get_cached_lsn()
-                                .expect("Assertion can only fire if a cached lsn is present")
-                        );
-
+                        let reached_cache =
+                            state.get_cached_lsn().map(|clsn| clsn + 1) == Some(lsn);
                         let will_init = rec.will_init();
                         state.records.push((lsn, rec));
-                        will_init
+                        will_init || reached_cache
                     }
                 },
             };

pageserver/src/tenant/storage_layer/delta_layer.rs

@@ -20,8 +20,8 @@
 //!    000000067F000032BE0000400000000020B6-000000067F000032BE0000400000000030B6__000000578C6B29-0000000057A50051
 //! ```
 //!
-//! Every delta file consists of three parts: "summary", "values", and
-//! "index". The summary is a fixed size header at the beginning of the file,
+//! Every delta file consists of three parts: "summary", "index", and
+//! "values". The summary is a fixed size header at the beginning of the file,
 //! and it contains basic information about the layer, and offsets to the other
 //! parts. The "index" is a B-tree, mapping from Key and LSN to an offset in the
 //! "values" part.  The actual page images and WAL records are stored in the
@@ -217,7 +217,6 @@ pub struct DeltaLayerInner {
     // values copied from summary
     index_start_blk: u32,
     index_root_blk: u32,
-    lsn_range: Range<Lsn>,
 
     file: VirtualFile,
     file_id: FileId,
@@ -729,9 +728,6 @@ impl DeltaLayerInner {
             // production code path
             expected_summary.index_start_blk = actual_summary.index_start_blk;
             expected_summary.index_root_blk = actual_summary.index_root_blk;
-            // mask out the timeline_id, but still require the layers to be from the same tenant
-            expected_summary.timeline_id = actual_summary.timeline_id;
-
             if actual_summary != expected_summary {
                 bail!(
                     "in-file summary does not match expected summary. actual = {:?} expected = {:?}",
@@ -746,7 +742,6 @@ impl DeltaLayerInner {
             file_id,
             index_start_blk: actual_summary.index_start_blk,
             index_root_blk: actual_summary.index_root_blk,
-            lsn_range: actual_summary.lsn_range,
             max_vectored_read_bytes,
         }))
     }
@@ -868,10 +863,10 @@ impl DeltaLayerInner {
                 .into(),
         );
 
-        let data_end_offset = self.index_start_offset();
+        let data_end_offset = self.index_start_blk as u64 * PAGE_SZ as u64;
 
         let reads = Self::plan_reads(
-            &keyspace,
+            keyspace,
             lsn_range,
             data_end_offset,
             index_reader,
@@ -885,13 +880,11 @@ impl DeltaLayerInner {
         self.do_reads_and_update_state(reads, reconstruct_state)
             .await;
 
-        reconstruct_state.on_lsn_advanced(&keyspace, self.lsn_range.start);
-
         Ok(())
     }
 
     async fn plan_reads<Reader>(
-        keyspace: &KeySpace,
+        keyspace: KeySpace,
         lsn_range: Range<Lsn>,
         data_end_offset: u64,
         index_reader: DiskBtreeReader<Reader, DELTA_KEY_SIZE>,
@@ -1110,195 +1103,11 @@ impl DeltaLayerInner {
         if let Some(last) = all_keys.last_mut() {
             // Last key occupies all space till end of value storage,
             // which corresponds to beginning of the index
-            last.size = self.index_start_offset() - last.size;
+            last.size = self.index_start_blk as u64 * PAGE_SZ as u64 - last.size;
         }
         Ok(all_keys)
     }
 
-    /// Using the given writer, write out a truncated version, where LSNs higher than the
-    /// truncate_at are missing.
-    #[cfg(test)]
-    pub(super) async fn copy_prefix(
-        &self,
-        writer: &mut DeltaLayerWriter,
-        truncate_at: Lsn,
-        ctx: &RequestContext,
-    ) -> anyhow::Result<()> {
-        use crate::tenant::vectored_blob_io::{
-            BlobMeta, VectoredReadBuilder, VectoredReadExtended,
-        };
-        use futures::stream::TryStreamExt;
-
-        #[derive(Debug)]
-        enum Item {
-            Actual(Key, Lsn, BlobRef),
-            Sentinel,
-        }
-
-        impl From<Item> for Option<(Key, Lsn, BlobRef)> {
-            fn from(value: Item) -> Self {
-                match value {
-                    Item::Actual(key, lsn, blob) => Some((key, lsn, blob)),
-                    Item::Sentinel => None,
-                }
-            }
-        }
-
-        impl Item {
-            fn offset(&self) -> Option<BlobRef> {
-                match self {
-                    Item::Actual(_, _, blob) => Some(*blob),
-                    Item::Sentinel => None,
-                }
-            }
-
-            fn is_last(&self) -> bool {
-                matches!(self, Item::Sentinel)
-            }
-        }
-
-        let block_reader = FileBlockReader::new(&self.file, self.file_id);
-        let tree_reader = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
-            self.index_start_blk,
-            self.index_root_blk,
-            block_reader,
-        );
-
-        let stream = self.stream_index_forwards(&tree_reader, &[0u8; DELTA_KEY_SIZE], ctx);
-        let stream = stream.map_ok(|(key, lsn, pos)| Item::Actual(key, lsn, pos));
-        // put in a sentinel value for getting the end offset for last item, and not having to
-        // repeat the whole read part
-        let stream = stream.chain(futures::stream::once(futures::future::ready(Ok(
-            Item::Sentinel,
-        ))));
-        let mut stream = std::pin::pin!(stream);
-
-        let mut prev: Option<(Key, Lsn, BlobRef)> = None;
-
-        let mut read_builder: Option<VectoredReadBuilder> = None;
-
-        let max_read_size = self
-            .max_vectored_read_bytes
-            .map(|x| x.0.get())
-            .unwrap_or(8192);
-
-        let mut buffer = Some(BytesMut::with_capacity(max_read_size));
-
-        // FIXME: buffering of DeltaLayerWriter
-        let mut per_blob_copy = Vec::new();
-
-        while let Some(item) = stream.try_next().await? {
-            tracing::debug!(?item, "popped");
-            let offset = item
-                .offset()
-                .unwrap_or(BlobRef::new(self.index_start_offset(), false));
-
-            let actionable = if let Some((key, lsn, start_offset)) = prev.take() {
-                let end_offset = offset;
-
-                Some((BlobMeta { key, lsn }, start_offset..end_offset))
-            } else {
-                None
-            };
-
-            let is_last = item.is_last();
-
-            prev = Option::from(item);
-
-            let actionable = actionable.filter(|x| x.0.lsn < truncate_at);
-
-            let builder = if let Some((meta, offsets)) = actionable {
-                // extend or create a new builder
-                if read_builder
-                    .as_mut()
-                    .map(|x| x.extend(offsets.start.pos(), offsets.end.pos(), meta))
-                    .unwrap_or(VectoredReadExtended::No)
-                    == VectoredReadExtended::Yes
-                {
-                    None
-                } else {
-                    read_builder.replace(VectoredReadBuilder::new(
-                        offsets.start.pos(),
-                        offsets.end.pos(),
-                        meta,
-                        max_read_size,
-                    ))
-                }
-            } else {
-                // nothing to do, except perhaps flush any existing for the last element
-                None
-            };
-
-            // flush the possible older builder and also the new one if the item was the last one
-            let builders = builder.into_iter();
-            let builders = if is_last {
-                builders.chain(read_builder.take())
-            } else {
-                builders.chain(None)
-            };
-
-            for builder in builders {
-                let read = builder.build();
-
-                let reader = VectoredBlobReader::new(&self.file);
-
-                let mut buf = buffer.take().unwrap();
-
-                buf.clear();
-                buf.reserve(read.size());
-                let res = reader.read_blobs(&read, buf).await?;
-
-                for blob in res.blobs {
-                    let key = blob.meta.key;
-                    let lsn = blob.meta.lsn;
-                    let data = &res.buf[blob.start..blob.end];
-
-                    #[cfg(debug_assertions)]
-                    Value::des(data)
-                        .with_context(|| {
-                            format!(
-                                "blob failed to deserialize for {}@{}, {}..{}: {:?}",
-                                blob.meta.key,
-                                blob.meta.lsn,
-                                blob.start,
-                                blob.end,
-                                utils::Hex(data)
-                            )
-                        })
-                        .unwrap();
-
-                    // is it an image or will_init walrecord?
-                    // FIXME: this could be handled by threading the BlobRef to the
-                    // VectoredReadBuilder
-                    let will_init = crate::repository::ValueBytes::will_init(data)
-                        .inspect_err(|_e| {
-                            #[cfg(feature = "testing")]
-                            tracing::error!(data=?utils::Hex(data), err=?_e, "failed to parse will_init out of serialized value");
-                        })
-                        .unwrap_or(false);
-
-                    per_blob_copy.clear();
-                    per_blob_copy.extend_from_slice(data);
-
-                    let (tmp, res) = writer
-                        .put_value_bytes(key, lsn, std::mem::take(&mut per_blob_copy), will_init)
-                        .await;
-                    per_blob_copy = tmp;
-                    res?;
-                }
-
-                buffer = Some(res.buf);
-            }
-        }
-
-        assert!(
-            read_builder.is_none(),
-            "with the sentinel above loop should had handled all"
-        );
-
-        Ok(())
-    }
-
     pub(super) async fn dump(&self, ctx: &RequestContext) -> anyhow::Result<()> {
         println!(
             "index_start_blk: {}, root {}",
@@ -1368,44 +1177,6 @@ impl DeltaLayerInner {
 
         Ok(())
     }
-
-    #[cfg(test)]
-    fn stream_index_forwards<'a, R>(
-        &'a self,
-        reader: &'a DiskBtreeReader<R, DELTA_KEY_SIZE>,
-        start: &'a [u8; DELTA_KEY_SIZE],
-        ctx: &'a RequestContext,
-    ) -> impl futures::stream::Stream<
-        Item = Result<(Key, Lsn, BlobRef), crate::tenant::disk_btree::DiskBtreeError>,
-    > + 'a
-    where
-        R: BlockReader,
-    {
-        use futures::stream::TryStreamExt;
-        let stream = reader.get_stream_from(start, ctx);
-        stream.map_ok(|(key, value)| {
-            let key = DeltaKey::from_slice(&key);
-            let (key, lsn) = (key.key(), key.lsn());
-            let offset = BlobRef(value);
-
-            (key, lsn, offset)
-        })
-    }
-
-    /// The file offset to the first block of index.
-    ///
-    /// The file structure is summary, values, and index. We often need this for the size of last blob.
-    fn index_start_offset(&self) -> u64 {
-        let offset = self.index_start_blk as u64 * PAGE_SZ as u64;
-        let bref = BlobRef(offset);
-        tracing::debug!(
-            index_start_blk = self.index_start_blk,
-            offset,
-            pos = bref.pos(),
-            "index_start_offset"
-        );
-        offset
-    }
 }
 
 /// A set of data associated with a delta layer key and its value
@@ -1539,7 +1310,7 @@ mod test {
 
         // Plan and validate
         let vectored_reads = DeltaLayerInner::plan_reads(
-            &keyspace,
+            keyspace.clone(),
             lsn_range.clone(),
             disk_offset,
             reader,
@@ -1767,7 +1538,7 @@ mod test {
 
         let resident = writer.finish(entries_meta.key_range.end, &timeline).await?;
 
-        let inner = resident.as_delta(&ctx).await?;
+        let inner = resident.get_inner_delta(&ctx).await?;
 
         let file_size = inner.file.metadata().await?.len();
         tracing::info!(
@@ -1791,7 +1562,7 @@ mod test {
             let data_end_offset = inner.index_start_blk as u64 * PAGE_SZ as u64;
 
             let vectored_reads = DeltaLayerInner::plan_reads(
-                &keyspace,
+                keyspace.clone(),
                 entries_meta.lsn_range.clone(),
                 data_end_offset,
                 index_reader,
@@ -1823,217 +1594,4 @@ mod test {
 
         Ok(())
     }
-
-    #[tokio::test]
-    async fn copy_delta_prefix_smoke() {
-        use crate::walrecord::NeonWalRecord;
-        use bytes::Bytes;
-
-        let h = crate::tenant::harness::TenantHarness::create("truncate_delta_smoke").unwrap();
-        let (tenant, ctx) = h.load().await;
-        let ctx = &ctx;
-        let timeline = tenant
-            .create_test_timeline(TimelineId::generate(), Lsn(0x10), 14, ctx)
-            .await
-            .unwrap();
-
-        let initdb_layer = timeline
-            .layers
-            .read()
-            .await
-            .likely_resident_layers()
-            .next()
-            .unwrap();
-
-        {
-            let mut writer = timeline.writer().await;
-
-            let data = [
-                (0x20, 12, Value::Image(Bytes::from_static(b"foobar"))),
-                (
-                    0x30,
-                    12,
-                    Value::WalRecord(NeonWalRecord::Postgres {
-                        will_init: false,
-                        rec: Bytes::from_static(b"1"),
-                    }),
-                ),
-                (
-                    0x40,
-                    12,
-                    Value::WalRecord(NeonWalRecord::Postgres {
-                        will_init: true,
-                        rec: Bytes::from_static(b"2"),
-                    }),
-                ),
-                // build an oversized value so we cannot extend and existing read over
-                // this
-                (
-                    0x50,
-                    12,
-                    Value::WalRecord(NeonWalRecord::Postgres {
-                        will_init: true,
-                        rec: {
-                            let mut buf =
-                                vec![0u8; tenant.conf.max_vectored_read_bytes.0.get() + 1024];
-                            buf.iter_mut()
-                                .enumerate()
-                                .for_each(|(i, slot)| *slot = (i % 256) as u8);
-                            Bytes::from(buf)
-                        },
-                    }),
-                ),
-                // because the oversized read cannot be extended further, we are sure to exercise the
-                // builder created on the last round with this:
-                (
-                    0x60,
-                    12,
-                    Value::WalRecord(NeonWalRecord::Postgres {
-                        will_init: true,
-                        rec: Bytes::from_static(b"3"),
-                    }),
-                ),
-                (
-                    0x60,
-                    9,
-                    Value::Image(Bytes::from_static(b"something for a different key")),
-                ),
-            ];
-
-            let mut last_lsn = None;
-
-            for (lsn, key, value) in data {
-                let key = Key::from_i128(key);
-                writer.put(key, Lsn(lsn), &value, ctx).await.unwrap();
-                last_lsn = Some(lsn);
-            }
-
-            writer.finish_write(Lsn(last_lsn.unwrap()));
-        }
-        timeline.freeze_and_flush().await.unwrap();
-
-        let new_layer = timeline
-            .layers
-            .read()
-            .await
-            .likely_resident_layers()
-            .find(|x| x != &initdb_layer)
-            .unwrap();
-
-        // create a copy for the timeline, so we don't overwrite the file
-        let branch = tenant
-            .branch_timeline_test(&timeline, TimelineId::generate(), None, ctx)
-            .await
-            .unwrap();
-
-        assert_eq!(branch.get_ancestor_lsn(), Lsn(0x60));
-
-        // truncating at 0x61 gives us a full copy, otherwise just go backwards until there's just
-        // a single key
-
-        for truncate_at in [0x61, 0x51, 0x41, 0x31, 0x21] {
-            let truncate_at = Lsn(truncate_at);
-
-            let mut writer = DeltaLayerWriter::new(
-                tenant.conf,
-                branch.timeline_id,
-                tenant.tenant_shard_id,
-                Key::MIN,
-                Lsn(0x11)..truncate_at,
-            )
-            .await
-            .unwrap();
-
-            let new_layer = new_layer.download_and_keep_resident().await.unwrap();
-
-            new_layer
-                .copy_delta_prefix(&mut writer, truncate_at, ctx)
-                .await
-                .unwrap();
-
-            let copied_layer = writer.finish(Key::MAX, &branch).await.unwrap();
-
-            copied_layer.as_delta(ctx).await.unwrap();
-
-            assert_keys_and_values_eq(
-                new_layer.as_delta(ctx).await.unwrap(),
-                copied_layer.as_delta(ctx).await.unwrap(),
-                truncate_at,
-                ctx,
-            )
-            .await;
-        }
-    }
-
-    async fn assert_keys_and_values_eq(
-        source: &DeltaLayerInner,
-        truncated: &DeltaLayerInner,
-        truncated_at: Lsn,
-        ctx: &RequestContext,
-    ) {
-        use futures::future::ready;
-        use futures::stream::TryStreamExt;
-
-        let start_key = [0u8; DELTA_KEY_SIZE];
-
-        let source_reader = FileBlockReader::new(&source.file, source.file_id);
-        let source_tree = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
-            source.index_start_blk,
-            source.index_root_blk,
-            &source_reader,
-        );
-        let source_stream = source.stream_index_forwards(&source_tree, &start_key, ctx);
-        let source_stream = source_stream.filter(|res| match res {
-            Ok((_, lsn, _)) => ready(lsn < &truncated_at),
-            _ => ready(true),
-        });
-        let mut source_stream = std::pin::pin!(source_stream);
-
-        let truncated_reader = FileBlockReader::new(&truncated.file, truncated.file_id);
-        let truncated_tree = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
-            truncated.index_start_blk,
-            truncated.index_root_blk,
-            &truncated_reader,
-        );
-        let truncated_stream = truncated.stream_index_forwards(&truncated_tree, &start_key, ctx);
-        let mut truncated_stream = std::pin::pin!(truncated_stream);
-
-        let mut scratch_left = Vec::new();
-        let mut scratch_right = Vec::new();
-
-        loop {
-            let (src, truncated) = (source_stream.try_next(), truncated_stream.try_next());
-            let (src, truncated) = tokio::try_join!(src, truncated).unwrap();
-
-            if src.is_none() {
-                assert!(truncated.is_none());
-                break;
-            }
-
-            let (src, truncated) = (src.unwrap(), truncated.unwrap());
-
-            // because we've filtered the source with Lsn, we should always have the same keys from both.
-            assert_eq!(src.0, truncated.0);
-            assert_eq!(src.1, truncated.1);
-
-            // if this is needed for something else, just drop this assert.
-            assert!(
-                src.2.pos() >= truncated.2.pos(),
-                "value position should not go backwards {} vs. {}",
-                src.2.pos(),
-                truncated.2.pos()
-            );
-
-            scratch_left.clear();
-            let src_cursor = source_reader.block_cursor();
-            let left = src_cursor.read_blob_into_buf(src.2.pos(), &mut scratch_left, ctx);
-            scratch_right.clear();
-            let trunc_cursor = truncated_reader.block_cursor();
-            let right = trunc_cursor.read_blob_into_buf(truncated.2.pos(), &mut scratch_right, ctx);
-
-            tokio::try_join!(left, right).unwrap();
-
-            assert_eq!(utils::Hex(&scratch_left), utils::Hex(&scratch_right));
-        }
-    }
 }

pageserver/src/tenant/storage_layer/image_layer.rs

@@ -396,8 +396,6 @@ impl ImageLayerInner {
             // production code path
             expected_summary.index_start_blk = actual_summary.index_start_blk;
             expected_summary.index_root_blk = actual_summary.index_root_blk;
-            // mask out the timeline_id, but still require the layers to be from the same tenant
-            expected_summary.timeline_id = actual_summary.timeline_id;
 
             if actual_summary != expected_summary {
                 bail!(

pageserver/src/tenant/storage_layer/inmemory_layer.rs

@@ -26,7 +26,7 @@ use utils::{bin_ser::BeSer, id::TimelineId, lsn::Lsn, vec_map::VecMap};
 // while being able to use std::fmt::Write's methods
 use crate::metrics::TIMELINE_EPHEMERAL_BYTES;
 use std::cmp::Ordering;
-use std::fmt::Write;
+use std::fmt::Write as _;
 use std::ops::Range;
 use std::sync::atomic::Ordering as AtomicOrdering;
 use std::sync::atomic::{AtomicU64, AtomicUsize};
@@ -54,12 +54,6 @@ pub struct InMemoryLayer {
     /// Writes are only allowed when this is `None`.
     end_lsn: OnceLock<Lsn>,
 
-    /// Used for traversal path. Cached representation of the in-memory layer before frozen.
-    local_path_str: Arc<str>,
-
-    /// Used for traversal path. Cached representation of the in-memory layer after frozen.
-    frozen_local_path_str: OnceLock<Arc<str>>,
-
     opened_at: Instant,
 
     /// The above fields never change, except for `end_lsn`, which is only set once.
@@ -247,12 +241,6 @@ impl InMemoryLayer {
         self.start_lsn..self.end_lsn_or_max()
     }
 
-    pub(crate) fn local_path_str(&self) -> &Arc<str> {
-        self.frozen_local_path_str
-            .get()
-            .unwrap_or(&self.local_path_str)
-    }
-
     /// debugging function to print out the contents of the layer
     ///
     /// this is likely completly unused
@@ -438,30 +426,14 @@ impl InMemoryLayer {
             }
         }
 
-        reconstruct_state.on_lsn_advanced(&keyspace, self.start_lsn);
-
         Ok(())
     }
 }
 
-fn inmem_layer_display(mut f: impl Write, start_lsn: Lsn, end_lsn: Lsn) -> std::fmt::Result {
-    write!(f, "inmem-{:016X}-{:016X}", start_lsn.0, end_lsn.0)
-}
-
-fn inmem_layer_log_display(
-    mut f: impl Write,
-    timeline: TimelineId,
-    start_lsn: Lsn,
-    end_lsn: Lsn,
-) -> std::fmt::Result {
-    write!(f, "timeline {} in-memory ", timeline)?;
-    inmem_layer_display(f, start_lsn, end_lsn)
-}
-
 impl std::fmt::Display for InMemoryLayer {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
         let end_lsn = self.end_lsn_or_max();
-        inmem_layer_display(f, self.start_lsn, end_lsn)
+        write!(f, "inmem-{:016X}-{:016X}", self.start_lsn.0, end_lsn.0)
     }
 }
 
@@ -486,12 +458,6 @@ impl InMemoryLayer {
 
         Ok(InMemoryLayer {
             file_id: key,
-            local_path_str: {
-                let mut buf = String::new();
-                inmem_layer_log_display(&mut buf, timeline_id, start_lsn, Lsn::MAX).unwrap();
-                buf.into()
-            },
-            frozen_local_path_str: OnceLock::new(),
             conf,
             timeline_id,
             tenant_shard_id,
@@ -586,15 +552,6 @@ impl InMemoryLayer {
         );
         self.end_lsn.set(end_lsn).expect("end_lsn set only once");
 
-        self.frozen_local_path_str
-            .set({
-                let mut buf = String::new();
-                inmem_layer_log_display(&mut buf, self.get_timeline_id(), self.start_lsn, end_lsn)
-                    .unwrap();
-                buf.into()
-            })
-            .expect("frozen_local_path_str set only once");
-
         for vec_map in inner.index.values() {
             for (lsn, _pos) in vec_map.as_slice() {
                 assert!(*lsn < end_lsn);

pageserver/src/tenant/storage_layer/layer.rs

@@ -116,12 +116,6 @@ impl AsLayerDesc for Layer {
     }
 }
 
-impl PartialEq for Layer {
-    fn eq(&self, other: &Self) -> bool {
-        Arc::as_ptr(&self.0) == Arc::as_ptr(&other.0)
-    }
-}
-
 impl Layer {
     /// Creates a layer value for a file we know to not be resident.
     pub(crate) fn for_evicted(
@@ -336,12 +330,6 @@ impl Layer {
             .get_values_reconstruct_data(keyspace, lsn_range, reconstruct_data, &self.0, ctx)
             .instrument(tracing::debug_span!("get_values_reconstruct_data", layer=%self))
             .await
-            .map_err(|err| match err {
-                GetVectoredError::Other(err) => GetVectoredError::Other(
-                    err.context(format!("get_values_reconstruct_data for layer {self}")),
-                ),
-                err => err,
-            })
     }
 
     /// Download the layer if evicted.
@@ -401,10 +389,6 @@ impl Layer {
         &self.0.path
     }
 
-    pub(crate) fn local_path_str(&self) -> &Arc<str> {
-        &self.0.path_str
-    }
-
     pub(crate) fn metadata(&self) -> LayerFileMetadata {
         self.0.metadata()
     }
@@ -527,9 +511,6 @@ struct LayerInner {
     /// Full path to the file; unclear if this should exist anymore.
     path: Utf8PathBuf,
 
-    /// String representation of the full path, used for traversal id.
-    path_str: Arc<str>,
-
     desc: PersistentLayerDesc,
 
     /// Timeline access is needed for remote timeline client and metrics.
@@ -623,17 +604,9 @@ enum Status {
 
 impl Drop for LayerInner {
     fn drop(&mut self) {
-        // if there was a pending eviction, mark it cancelled here to balance metrics
-        if let Some((ResidentOrWantedEvicted::WantedEvicted(..), _)) = self.inner.take_and_deinit()
-        {
-            // eviction has already been started
-            LAYER_IMPL_METRICS.inc_eviction_cancelled(EvictionCancelled::LayerGone);
-
-            // eviction request is intentionally not honored as no one is present to wait for it
-            // and we could be delaying shutdown for nothing.
-        }
-
         if !*self.wanted_deleted.get_mut() {
+            // should we try to evict if the last wish was for eviction? seems more like a hazard
+            // than a clear win.
             return;
         }
 
@@ -735,7 +708,6 @@ impl LayerInner {
 
         LayerInner {
             conf,
-            path_str: path.to_string().into(),
             path,
             desc,
             timeline: Arc::downgrade(timeline),
@@ -1580,8 +1552,8 @@ impl Drop for DownloadedLayer {
         if let Some(owner) = self.owner.upgrade() {
             owner.on_downloaded_layer_drop(self.version);
         } else {
-            // Layer::drop will handle cancelling the eviction; because of drop order and
-            // `DownloadedLayer` never leaking, we cannot know here if eviction was requested.
+            // no need to do anything, we are shutting down
+            LAYER_IMPL_METRICS.inc_eviction_cancelled(EvictionCancelled::LayerGone);
         }
     }
 }
@@ -1780,28 +1752,6 @@ impl ResidentLayer {
         }
     }
 
-    /// FIXME: truncate is bad name because we are not truncating anything, but copying the
-    /// filtered parts.
-    #[cfg(test)]
-    pub(super) async fn copy_delta_prefix(
-        &self,
-        writer: &mut super::delta_layer::DeltaLayerWriter,
-        truncate_at: Lsn,
-        ctx: &RequestContext,
-    ) -> anyhow::Result<()> {
-        use LayerKind::*;
-
-        let owner = &self.owner.0;
-
-        match self.downloaded.get(owner, ctx).await? {
-            Delta(ref d) => d
-                .copy_prefix(writer, truncate_at, ctx)
-                .await
-                .with_context(|| format!("truncate {self}")),
-            Image(_) => anyhow::bail!(format!("cannot truncate image layer {self}")),
-        }
-    }
-
     pub(crate) fn local_path(&self) -> &Utf8Path {
         &self.owner.0.path
     }
@@ -1811,14 +1761,14 @@ impl ResidentLayer {
     }
 
     #[cfg(test)]
-    pub(crate) async fn as_delta(
-        &self,
+    pub(crate) async fn get_inner_delta<'a>(
+        &'a self,
         ctx: &RequestContext,
-    ) -> anyhow::Result<&delta_layer::DeltaLayerInner> {
-        use LayerKind::*;
-        match self.downloaded.get(&self.owner.0, ctx).await? {
-            Delta(ref d) => Ok(d),
-            Image(_) => Err(anyhow::anyhow!("image layer")),
+    ) -> anyhow::Result<&'a delta_layer::DeltaLayerInner> {
+        let owner = &self.owner.0;
+        match self.downloaded.get(owner, ctx).await? {
+            LayerKind::Delta(d) => Ok(d),
+            LayerKind::Image(_) => Err(anyhow::anyhow!("Expected a delta layer")),
         }
     }
 }

pageserver/src/tenant/storage_layer/layer/tests.rs

@@ -721,110 +721,11 @@ async fn evict_and_wait_does_not_wait_for_download() {
     layer.evict_and_wait(FOREVER).await.unwrap();
 }
 
-/// Asserts that there is no miscalculation when Layer is dropped while it is being kept resident,
-/// which is the last value.
-///
-/// Also checks that the same does not happen on a non-evicted layer (regression test).
-#[tokio::test(start_paused = true)]
-async fn eviction_cancellation_on_drop() {
-    use crate::repository::Value;
-    use bytes::Bytes;
-
-    // this is the runtime on which Layer spawns the blocking tasks on
-    let handle = tokio::runtime::Handle::current();
-
-    let h = TenantHarness::create("eviction_cancellation_on_drop").unwrap();
-    utils::logging::replace_panic_hook_with_tracing_panic_hook().forget();
-    let (tenant, ctx) = h.load().await;
-
-    let timeline = tenant
-        .create_test_timeline(TimelineId::generate(), Lsn(0x10), 14, &ctx)
-        .await
-        .unwrap();
-
-    {
-        // create_test_timeline wrote us one layer, write another
-        let mut writer = timeline.writer().await;
-        writer
-            .put(
-                Key::from_i128(5),
-                Lsn(0x20),
-                &Value::Image(Bytes::from_static(b"this does not matter either")),
-                &ctx,
-            )
-            .await
-            .unwrap();
-
-        writer.finish_write(Lsn(0x20));
-    }
-
-    timeline.freeze_and_flush().await.unwrap();
-
-    // wait for the upload to complete so our Arc::strong_count assertion holds
-    timeline
-        .remote_client
-        .as_ref()
-        .unwrap()
-        .wait_completion()
-        .await
-        .unwrap();
-
-    let (evicted_layer, not_evicted) = {
-        let mut layers = {
-            let mut guard = timeline.layers.write().await;
-            let layers = guard.likely_resident_layers().collect::<Vec<_>>();
-            // remove the layers from layermap
-            guard.finish_gc_timeline(&layers);
-
-            layers
-        };
-
-        assert_eq!(layers.len(), 2);
-
-        (layers.pop().unwrap(), layers.pop().unwrap())
-    };
-
-    let victims = [(evicted_layer, true), (not_evicted, false)];
-
-    for (victim, evict) in victims {
-        let resident = victim.keep_resident().await.unwrap();
-        drop(victim);
-
-        assert_eq!(Arc::strong_count(&resident.owner.0), 1);
-
-        if evict {
-            let evict_and_wait = resident.owner.evict_and_wait(FOREVER);
-
-            // drive the future to await on the status channel, and then drop it
-            tokio::time::timeout(ADVANCE, evict_and_wait)
-                .await
-                .expect_err("should had been a timeout since we are holding the layer resident");
-        }
-
-        // 1 == we only evict one of the layers
-        assert_eq!(1, LAYER_IMPL_METRICS.started_evictions.get());
-
-        drop(resident);
-
-        // run any spawned
-        tokio::time::sleep(ADVANCE).await;
-
-        SpawnBlockingPoolHelper::consume_and_release_all_of_spawn_blocking_threads(&handle).await;
-
-        assert_eq!(
-            1,
-            LAYER_IMPL_METRICS.cancelled_evictions[EvictionCancelled::LayerGone].get()
-        );
-    }
-}
-
-/// A test case to remind you the cost of these structures. You can bump the size limit
-/// below if it is really necessary to add more fields to the structures.
 #[test]
 fn layer_size() {
     assert_eq!(std::mem::size_of::<LayerAccessStats>(), 2040);
     assert_eq!(std::mem::size_of::<PersistentLayerDesc>(), 104);
-    assert_eq!(std::mem::size_of::<LayerInner>(), 2344);
+    assert_eq!(std::mem::size_of::<LayerInner>(), 2328);
     // it also has the utf8 path
 }
 

pageserver/src/tenant/tasks.rs

@@ -62,7 +62,7 @@ impl BackgroundLoopKind {
 pub(crate) async fn concurrent_background_tasks_rate_limit_permit(
     loop_kind: BackgroundLoopKind,
     _ctx: &RequestContext,
-) -> tokio::sync::SemaphorePermit<'static> {
+) -> impl Drop {
     let _guard = crate::metrics::BACKGROUND_LOOP_SEMAPHORE_WAIT_GAUGE
         .with_label_values(&[loop_kind.as_static_str()])
         .guard();

pageserver/src/tenant/timeline.rs

@@ -16,14 +16,14 @@ use enumset::EnumSet;
 use fail::fail_point;
 use once_cell::sync::Lazy;
 use pageserver_api::{
-    key::{AUX_FILES_KEY, NON_INHERITED_RANGE},
+    key::AUX_FILES_KEY,
     keyspace::KeySpaceAccum,
     models::{
         CompactionAlgorithm, DownloadRemoteLayersTaskInfo, DownloadRemoteLayersTaskSpawnRequest,
         EvictionPolicy, InMemoryLayerInfo, LayerMapInfo, TimelineState,
     },
     reltag::BlockNumber,
-    shard::{ShardIdentity, ShardNumber, TenantShardId},
+    shard::{ShardIdentity, TenantShardId},
 };
 use rand::Rng;
 use serde_with::serde_as;
@@ -86,7 +86,7 @@ use crate::{
 use crate::config::PageServerConf;
 use crate::keyspace::{KeyPartitioning, KeySpace};
 use crate::metrics::{
-    GetKind, TimelineMetrics, MATERIALIZED_PAGE_CACHE_HIT, MATERIALIZED_PAGE_CACHE_HIT_DIRECT,
+    TimelineMetrics, MATERIALIZED_PAGE_CACHE_HIT, MATERIALIZED_PAGE_CACHE_HIT_DIRECT,
 };
 use crate::pgdatadir_mapping::CalculateLogicalSizeError;
 use crate::tenant::config::TenantConfOpt;
@@ -119,8 +119,8 @@ use self::layer_manager::LayerManager;
 use self::logical_size::LogicalSize;
 use self::walreceiver::{WalReceiver, WalReceiverConf};
 
+use super::config::TenantConf;
 use super::secondary::heatmap::{HeatMapLayer, HeatMapTimeline};
-use super::{config::TenantConf, storage_layer::VectoredValueReconstructState};
 use super::{debug_assert_current_span_has_tenant_and_timeline_id, AttachedTenantConf};
 use super::{remote_timeline_client::index::IndexPart, storage_layer::LayerFringe};
 use super::{remote_timeline_client::RemoteTimelineClient, storage_layer::ReadableLayer};
@@ -182,16 +182,6 @@ pub(crate) struct AuxFilesState {
     pub(crate) n_deltas: usize,
 }
 
-/// The relation size cache caches relation sizes at the end of the timeline. It speeds up WAL
-/// ingestion considerably, because WAL ingestion needs to check on most records if the record
-/// implicitly extends the relation.  At startup, `complete_as_of` is initialized to the current end
-/// of the timeline (disk_consistent_lsn).  It's used on reads of relation sizes to check if the
-/// value can be used to also update the cache, see [`Timeline::update_cached_rel_size`].
-pub(crate) struct RelSizeCache {
-    pub(crate) complete_as_of: Lsn,
-    pub(crate) map: HashMap<RelTag, (Lsn, BlockNumber)>,
-}
-
 pub struct Timeline {
     conf: &'static PageServerConf,
     tenant_conf: Arc<ArcSwap<AttachedTenantConf>>,
@@ -334,7 +324,7 @@ pub struct Timeline {
     pub walreceiver: Mutex<Option<WalReceiver>>,
 
     /// Relation size cache
-    pub(crate) rel_size_cache: RwLock<RelSizeCache>,
+    pub rel_size_cache: RwLock<HashMap<RelTag, (Lsn, BlockNumber)>>,
 
     download_all_remote_layers_task_info: RwLock<Option<DownloadRemoteLayersTaskInfo>>,
 
@@ -438,62 +428,6 @@ pub(crate) enum PageReconstructError {
     /// An error happened replaying WAL records
     #[error(transparent)]
     WalRedo(anyhow::Error),
-
-    #[error("{0}")]
-    MissingKey(MissingKeyError),
-}
-
-#[derive(Debug)]
-pub struct MissingKeyError {
-    stuck_at_lsn: bool,
-    key: Key,
-    shard: ShardNumber,
-    cont_lsn: Lsn,
-    request_lsn: Lsn,
-    ancestor_lsn: Option<Lsn>,
-    traversal_path: Vec<TraversalPathItem>,
-    backtrace: Option<std::backtrace::Backtrace>,
-}
-
-impl std::fmt::Display for MissingKeyError {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        if self.stuck_at_lsn {
-            // Records are found in this timeline but no image layer or initial delta record was found.
-            write!(
-                f,
-                "could not find layer with more data for key {} (shard {:?}) at LSN {}, request LSN {}",
-                self.key, self.shard, self.cont_lsn, self.request_lsn
-            )?;
-            if let Some(ref ancestor_lsn) = self.ancestor_lsn {
-                write!(f, ", ancestor {}", ancestor_lsn)?;
-            }
-        } else {
-            // No records in this timeline.
-            write!(
-                f,
-                "could not find data for key {} (shard {:?}) at LSN {}, for request at LSN {}",
-                self.key, self.shard, self.cont_lsn, self.request_lsn
-            )?;
-        }
-
-        if !self.traversal_path.is_empty() {
-            writeln!(f)?;
-        }
-
-        for (r, c, l) in &self.traversal_path {
-            writeln!(
-                f,
-                "layer traversal: result {:?}, cont_lsn {}, layer: {}",
-                r, c, l,
-            )?;
-        }
-
-        if let Some(ref backtrace) = self.backtrace {
-            write!(f, "\n{}", backtrace)?;
-        }
-
-        Ok(())
-    }
 }
 
 impl PageReconstructError {
@@ -505,7 +439,6 @@ impl PageReconstructError {
             AncestorLsnTimeout(_) => false,
             Cancelled | AncestorStopping(_) => true,
             WalRedo(_) => false,
-            MissingKey { .. } => false,
         }
     }
 }
@@ -653,19 +586,6 @@ impl From<GetVectoredError> for CreateImageLayersError {
     }
 }
 
-impl From<GetVectoredError> for PageReconstructError {
-    fn from(e: GetVectoredError) -> Self {
-        match e {
-            GetVectoredError::Cancelled => PageReconstructError::Cancelled,
-            GetVectoredError::InvalidLsn(_) => PageReconstructError::Other(anyhow!("Invalid LSN")),
-            err @ GetVectoredError::Oversized(_) => PageReconstructError::Other(err.into()),
-            err @ GetVectoredError::MissingKey(_) => PageReconstructError::Other(err.into()),
-            GetVectoredError::GetReadyAncestorError(err) => PageReconstructError::from(err),
-            GetVectoredError::Other(err) => PageReconstructError::Other(err),
-        }
-    }
-}
-
 impl From<GetReadyAncestorError> for PageReconstructError {
     fn from(e: GetReadyAncestorError) -> Self {
         use GetReadyAncestorError::*;
@@ -695,23 +615,6 @@ pub enum GetVectoredImpl {
     Vectored,
 }
 
-#[derive(
-    Eq,
-    PartialEq,
-    Debug,
-    Copy,
-    Clone,
-    strum_macros::EnumString,
-    strum_macros::Display,
-    serde_with::DeserializeFromStr,
-    serde_with::SerializeDisplay,
-)]
-#[strum(serialize_all = "kebab-case")]
-pub enum GetImpl {
-    Legacy,
-    Vectored,
-}
-
 pub(crate) enum WaitLsnWaiter<'a> {
     Timeline(&'a Timeline),
     Tenant,
@@ -773,6 +676,16 @@ impl Timeline {
         key: Key,
         lsn: Lsn,
         ctx: &RequestContext,
+    ) -> Result<Bytes, PageReconstructError> {
+        self.timeline_get_throttle.throttle(ctx, 1).await;
+        self.get_impl(key, lsn, ctx).await
+    }
+    /// Not subject to [`Self::timeline_get_throttle`].
+    async fn get_impl(
+        &self,
+        key: Key,
+        lsn: Lsn,
+        ctx: &RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         if !lsn.is_valid() {
             return Err(PageReconstructError::Other(anyhow::anyhow!("Invalid LSN")));
@@ -783,7 +696,13 @@ impl Timeline {
         // page_service.
         debug_assert!(!self.shard_identity.is_key_disposable(&key));
 
-        self.timeline_get_throttle.throttle(ctx, 1).await;
+        // XXX: structured stats collection for layer eviction here.
+        trace!(
+            "get page request for {}@{} from task kind {:?}",
+            key,
+            lsn,
+            ctx.task_kind()
+        );
 
         // Check the page cache. We will get back the most recent page with lsn <= `lsn`.
         // The cached image can be returned directly if there is no WAL between the cached image
@@ -806,85 +725,12 @@ impl Timeline {
             None => None,
         };
 
-        match self.conf.get_impl {
-            GetImpl::Legacy => {
-                let reconstruct_state = ValueReconstructState {
-                    records: Vec::new(),
-                    img: cached_page_img,
-                };
+        let mut reconstruct_state = ValueReconstructState {
+            records: Vec::new(),
+            img: cached_page_img,
+        };
 
-                self.get_impl(key, lsn, reconstruct_state, ctx).await
-            }
-            GetImpl::Vectored => {
-                let keyspace = KeySpace {
-                    ranges: vec![key..key.next()],
-                };
-
-                // Initialise the reconstruct state for the key with the cache
-                // entry returned above.
-                let mut reconstruct_state = ValuesReconstructState::new();
-                let mut key_state = VectoredValueReconstructState::default();
-                key_state.img = cached_page_img;
-                reconstruct_state.keys.insert(key, Ok(key_state));
-
-                let vectored_res = self
-                    .get_vectored_impl(keyspace.clone(), lsn, reconstruct_state, ctx)
-                    .await;
-
-                if self.conf.validate_vectored_get {
-                    self.validate_get_vectored_impl(&vectored_res, keyspace, lsn, ctx)
-                        .await;
-                }
-
-                let key_value = vectored_res?.pop_first();
-                match key_value {
-                    Some((got_key, value)) => {
-                        if got_key != key {
-                            error!(
-                                "Expected {}, but singular vectored get returned {}",
-                                key, got_key
-                            );
-                            Err(PageReconstructError::Other(anyhow!(
-                                "Singular vectored get returned wrong key"
-                            )))
-                        } else {
-                            value
-                        }
-                    }
-                    None => {
-                        error!(
-                            "Expected {}, but singular vectored get returned nothing",
-                            key
-                        );
-                        Err(PageReconstructError::Other(anyhow!(
-                            "Singular vectored get did not return a value for {}",
-                            key
-                        )))
-                    }
-                }
-            }
-        }
-    }
-
-    /// Not subject to [`Self::timeline_get_throttle`].
-    async fn get_impl(
-        &self,
-        key: Key,
-        lsn: Lsn,
-        mut reconstruct_state: ValueReconstructState,
-        ctx: &RequestContext,
-    ) -> Result<Bytes, PageReconstructError> {
-        // XXX: structured stats collection for layer eviction here.
-        trace!(
-            "get page request for {}@{} from task kind {:?}",
-            key,
-            lsn,
-            ctx.task_kind()
-        );
-
-        let timer = crate::metrics::GET_RECONSTRUCT_DATA_TIME
-            .for_get_kind(GetKind::Singular)
-            .start_timer();
+        let timer = crate::metrics::GET_RECONSTRUCT_DATA_TIME.start_timer();
         let path = self
             .get_reconstruct_data(key, lsn, &mut reconstruct_state, ctx)
             .await?;
@@ -894,7 +740,7 @@ impl Timeline {
         let res = self.reconstruct_value(key, lsn, reconstruct_state).await;
         let elapsed = start.elapsed();
         crate::metrics::RECONSTRUCT_TIME
-            .for_get_kind(GetKind::Singular)
+            .for_result(&res)
             .observe(elapsed.as_secs_f64());
 
         if cfg!(feature = "testing") && res.is_err() {
@@ -907,7 +753,7 @@ impl Timeline {
                 writeln!(
                     msg,
                     "- layer traversal: result {res:?}, cont_lsn {cont_lsn}, layer: {}",
-                    layer,
+                    layer(),
                 )
                 .expect("string grows")
             });
@@ -973,9 +819,7 @@ impl Timeline {
                 self.get_vectored_sequential_impl(keyspace, lsn, ctx).await
             }
             GetVectoredImpl::Vectored => {
-                let vectored_res = self
-                    .get_vectored_impl(keyspace.clone(), lsn, ValuesReconstructState::new(), ctx)
-                    .await;
+                let vectored_res = self.get_vectored_impl(keyspace.clone(), lsn, ctx).await;
 
                 if self.conf.validate_vectored_get {
                     self.validate_get_vectored_impl(&vectored_res, keyspace, lsn, ctx)
@@ -1021,42 +865,15 @@ impl Timeline {
         for range in keyspace.ranges {
             let mut key = range.start;
             while key != range.end {
-                let block = self
-                    .get_impl(key, lsn, ValueReconstructState::default(), ctx)
-                    .await;
+                let block = self.get_impl(key, lsn, ctx).await;
 
                 use PageReconstructError::*;
                 match block {
                     Err(Cancelled | AncestorStopping(_)) => {
                         return Err(GetVectoredError::Cancelled)
                     }
-                    // we only capture stuck_at_lsn=false now until we figure out https://github.com/neondatabase/neon/issues/7380
-                    Err(MissingKey(MissingKeyError {
-                        stuck_at_lsn: false,
-                        ..
-                    })) if !NON_INHERITED_RANGE.contains(&key) => {
-                        // The vectored read path handles non inherited keys specially.
-                        // If such a a key cannot be reconstructed from the current timeline,
-                        // the vectored read path returns a key level error as opposed to a top
-                        // level error.
-                        return Err(GetVectoredError::MissingKey(key));
-                    }
-                    Err(Other(err))
-                        if err
-                            .to_string()
-                            .contains("downloading evicted layer file failed") =>
-                    {
-                        return Err(GetVectoredError::Other(err))
-                    }
-                    Err(Other(err))
-                        if err
-                            .chain()
-                            .any(|cause| cause.to_string().contains("layer loading failed")) =>
-                    {
-                        // The intent here is to achieve error parity with the vectored read path.
-                        // When vectored read fails to load a layer it fails the whole read, hence
-                        // we mimic this behaviour here to keep the validation happy.
-                        return Err(GetVectoredError::Other(err));
+                    Err(Other(err)) if err.to_string().contains("could not find data for key") => {
+                        return Err(GetVectoredError::MissingKey(key))
                     }
                     _ => {
                         values.insert(key, block);
@@ -1073,27 +890,14 @@ impl Timeline {
         &self,
         keyspace: KeySpace,
         lsn: Lsn,
-        mut reconstruct_state: ValuesReconstructState,
         ctx: &RequestContext,
     ) -> Result<BTreeMap<Key, Result<Bytes, PageReconstructError>>, GetVectoredError> {
-        let get_kind = if keyspace.total_size() == 1 {
-            GetKind::Singular
-        } else {
-            GetKind::Vectored
-        };
+        let mut reconstruct_state = ValuesReconstructState::new();
 
-        let get_data_timer = crate::metrics::GET_RECONSTRUCT_DATA_TIME
-            .for_get_kind(get_kind)
-            .start_timer();
         self.get_vectored_reconstruct_data(keyspace, lsn, &mut reconstruct_state, ctx)
             .await?;
-        get_data_timer.stop_and_record();
 
-        let reconstruct_timer = crate::metrics::RECONSTRUCT_TIME
-            .for_get_kind(get_kind)
-            .start_timer();
         let mut results: BTreeMap<Key, Result<Bytes, PageReconstructError>> = BTreeMap::new();
-        let layers_visited = reconstruct_state.get_layers_visited();
         for (key, res) in reconstruct_state.keys {
             match res {
                 Err(err) => {
@@ -1107,13 +911,6 @@ impl Timeline {
                 }
             }
         }
-        reconstruct_timer.stop_and_record();
-
-        // Note that this is an approximation. Tracking the exact number of layers visited
-        // per key requires virtually unbounded memory usage and is inefficient
-        // (i.e. segment tree tracking each range queried from a layer)
-        crate::metrics::VEC_READ_NUM_LAYERS_VISITED
-            .observe(layers_visited as f64 / results.len() as f64);
 
         Ok(results)
     }
@@ -1871,15 +1668,6 @@ const REPARTITION_FREQ_IN_CHECKPOINT_DISTANCE: u64 = 10;
 
 // Private functions
 impl Timeline {
-    #[allow(dead_code)]
-    pub(crate) fn get_switch_to_aux_file_v2(&self) -> bool {
-        let tenant_conf = self.tenant_conf.load();
-        tenant_conf
-            .tenant_conf
-            .switch_to_aux_file_v2
-            .unwrap_or(self.conf.default_tenant_conf.switch_to_aux_file_v2)
-    }
-
     pub(crate) fn get_lazy_slru_download(&self) -> bool {
         let tenant_conf = self.tenant_conf.load();
         tenant_conf
@@ -2104,10 +1892,7 @@ impl Timeline {
                 last_image_layer_creation_check_at: AtomicLsn::new(0),
 
                 last_received_wal: Mutex::new(None),
-                rel_size_cache: RwLock::new(RelSizeCache {
-                    complete_as_of: disk_consistent_lsn,
-                    map: HashMap::new(),
-                }),
+                rel_size_cache: RwLock::new(HashMap::new()),
 
                 download_all_remote_layers_task_info: RwLock::new(None),
 
@@ -2907,7 +2692,7 @@ impl Timeline {
     }
 }
 
-type TraversalId = Arc<str>;
+type TraversalId = String;
 
 trait TraversalLayerExt {
     fn traversal_id(&self) -> TraversalId;
@@ -2915,13 +2700,13 @@ trait TraversalLayerExt {
 
 impl TraversalLayerExt for Layer {
     fn traversal_id(&self) -> TraversalId {
-        Arc::clone(self.local_path_str())
+        self.local_path().to_string()
     }
 }
 
 impl TraversalLayerExt for Arc<InMemoryLayer> {
     fn traversal_id(&self) -> TraversalId {
-        Arc::clone(self.local_path_str())
+        format!("timeline {} in-memory {self}", self.get_timeline_id())
     }
 }
 
@@ -2950,7 +2735,7 @@ impl Timeline {
         let mut timeline = self;
 
         let mut read_count = scopeguard::guard(0, |cnt| {
-            crate::metrics::READ_NUM_LAYERS_VISITED.observe(cnt as f64)
+            crate::metrics::READ_NUM_FS_LAYERS.observe(cnt as f64)
         });
 
         // For debugging purposes, collect the path of layers that we traversed
@@ -2990,35 +2775,32 @@ impl Timeline {
                         if prev <= cont_lsn {
                             // Didn't make any progress in last iteration. Error out to avoid
                             // getting stuck in the loop.
-                            return Err(PageReconstructError::MissingKey(MissingKeyError {
-                                stuck_at_lsn: true,
+                            return Err(layer_traversal_error(format!(
+                                "could not find layer with more data for key {} at LSN {}, request LSN {}, ancestor {}",
                                 key,
-                                shard: self.shard_identity.get_shard_number(&key),
-                                cont_lsn: Lsn(cont_lsn.0 - 1),
+                                Lsn(cont_lsn.0 - 1),
                                 request_lsn,
-                                ancestor_lsn: Some(timeline.ancestor_lsn),
-                                traversal_path,
-                                backtrace: None,
-                            }));
+                                timeline.ancestor_lsn
+                            ), traversal_path));
                         }
                     }
                     prev_lsn = Some(cont_lsn);
                 }
                 ValueReconstructResult::Missing => {
-                    return Err(PageReconstructError::MissingKey(MissingKeyError {
-                        stuck_at_lsn: false,
-                        key,
-                        shard: self.shard_identity.get_shard_number(&key),
-                        cont_lsn,
-                        request_lsn,
-                        ancestor_lsn: None,
-                        traversal_path,
-                        backtrace: if cfg!(test) {
-                            Some(std::backtrace::Backtrace::force_capture())
+                    return Err(layer_traversal_error(
+                        if cfg!(test) {
+                            format!(
+                                "could not find data for key {} (shard {:?}) at LSN {}, for request at LSN {}\n{}",
+                                key, self.shard_identity.get_shard_number(&key), cont_lsn, request_lsn, std::backtrace::Backtrace::force_capture(),
+                            )
                         } else {
-                            None
+                            format!(
+                                "could not find data for key {} (shard {:?}) at LSN {}, for request at LSN {}",
+                                key, self.shard_identity.get_shard_number(&key), cont_lsn, request_lsn
+                            )
                         },
-                    }));
+                        traversal_path,
+                    ));
                 }
             }
 
@@ -3065,8 +2847,12 @@ impl Timeline {
                         Err(e) => return Err(PageReconstructError::from(e)),
                     };
                     cont_lsn = lsn_floor;
-                    *read_count += 1;
-                    traversal_path.push((result, cont_lsn, open_layer.traversal_id()));
+                    // metrics: open_layer does not count as fs access, so we are not updating `read_count`
+                    traversal_path.push((
+                        result,
+                        cont_lsn,
+                        Box::new(move || open_layer.traversal_id()),
+                    ));
                     continue 'outer;
                 }
             }
@@ -3092,8 +2878,12 @@ impl Timeline {
                         Err(e) => return Err(PageReconstructError::from(e)),
                     };
                     cont_lsn = lsn_floor;
-                    *read_count += 1;
-                    traversal_path.push((result, cont_lsn, frozen_layer.traversal_id()));
+                    // metrics: open_layer does not count as fs access, so we are not updating `read_count`
+                    traversal_path.push((
+                        result,
+                        cont_lsn,
+                        Box::new(move || frozen_layer.traversal_id()),
+                    ));
                     continue 'outer;
                 }
             }
@@ -3114,7 +2904,14 @@ impl Timeline {
                 };
                 cont_lsn = lsn_floor;
                 *read_count += 1;
-                traversal_path.push((result, cont_lsn, layer.traversal_id()));
+                traversal_path.push((
+                    result,
+                    cont_lsn,
+                    Box::new({
+                        let layer = layer.to_owned();
+                        move || layer.traversal_id()
+                    }),
+                ));
                 continue 'outer;
             } else if timeline.ancestor_timeline.is_some() {
                 // Nothing on this timeline. Traverse to parent
@@ -3167,47 +2964,11 @@ impl Timeline {
             .await?;
 
             keyspace.remove_overlapping_with(&completed);
-
-            // Do not descend into the ancestor timeline for aux files.
-            // We don't return a blanket [`GetVectoredError::MissingKey`] to avoid
-            // stalling compaction.
-            // TODO(chi): this will need to be updated for aux files v2 storage
-            if keyspace.overlaps(&NON_INHERITED_RANGE) {
-                let removed = keyspace.remove_overlapping_with(&KeySpace {
-                    ranges: vec![NON_INHERITED_RANGE],
-                });
-
-                for range in removed.ranges {
-                    let mut key = range.start;
-                    while key < range.end {
-                        reconstruct_state.on_key_error(
-                            key,
-                            PageReconstructError::MissingKey(MissingKeyError {
-                                stuck_at_lsn: false,
-                                key,
-                                shard: self.shard_identity.get_shard_number(&key),
-                                cont_lsn,
-                                request_lsn,
-                                ancestor_lsn: None,
-                                traversal_path: Vec::default(),
-                                backtrace: if cfg!(test) {
-                                    Some(std::backtrace::Backtrace::force_capture())
-                                } else {
-                                    None
-                                },
-                            }),
-                        );
-                        key = key.next();
-                    }
-                }
-            }
-
             if keyspace.total_size() == 0 || timeline.ancestor_timeline.is_none() {
                 break;
             }
 
-            // Take the min to avoid reconstructing a page with data newer than request Lsn.
-            cont_lsn = std::cmp::min(Lsn(request_lsn.0 + 1), Lsn(timeline.ancestor_lsn.0 + 1));
+            cont_lsn = Lsn(timeline.ancestor_lsn.0 + 1);
             timeline_owned = timeline
                 .get_ready_ancestor_timeline(ctx)
                 .await
@@ -3257,62 +3018,56 @@ impl Timeline {
             unmapped_keyspace.remove_overlapping_with(&keys_done_last_step);
             completed_keyspace.merge(&keys_done_last_step);
 
-            // Do not descent any further if the last layer we visited
-            // completed all keys in the keyspace it inspected. This is not
-            // required for correctness, but avoids visiting extra layers
-            // which turns out to be a perf bottleneck in some cases.
-            if !unmapped_keyspace.is_empty() {
-                let guard = timeline.layers.read().await;
-                let layers = guard.layer_map();
+            let guard = timeline.layers.read().await;
+            let layers = guard.layer_map();
 
-                let in_memory_layer = layers.find_in_memory_layer(|l| {
-                    let start_lsn = l.get_lsn_range().start;
-                    cont_lsn > start_lsn
-                });
+            let in_memory_layer = layers.find_in_memory_layer(|l| {
+                let start_lsn = l.get_lsn_range().start;
+                cont_lsn > start_lsn
+            });
 
-                match in_memory_layer {
-                    Some(l) => {
-                        let lsn_range = l.get_lsn_range().start..cont_lsn;
-                        fringe.update(
-                            ReadableLayer::InMemoryLayer(l),
-                            unmapped_keyspace.clone(),
-                            lsn_range,
-                        );
-                    }
-                    None => {
-                        for range in unmapped_keyspace.ranges.iter() {
-                            let results = layers.range_search(range.clone(), cont_lsn);
+            match in_memory_layer {
+                Some(l) => {
+                    let lsn_range = l.get_lsn_range().start..cont_lsn;
+                    fringe.update(
+                        ReadableLayer::InMemoryLayer(l),
+                        unmapped_keyspace.clone(),
+                        lsn_range,
+                    );
+                }
+                None => {
+                    for range in unmapped_keyspace.ranges.iter() {
+                        let results = layers.range_search(range.clone(), cont_lsn);
 
-                            results
-                                .found
-                                .into_iter()
-                                .map(|(SearchResult { layer, lsn_floor }, keyspace_accum)| {
-                                    (
-                                        ReadableLayer::PersistentLayer(guard.get_from_desc(&layer)),
-                                        keyspace_accum.to_keyspace(),
-                                        lsn_floor..cont_lsn,
-                                    )
-                                })
-                                .for_each(|(layer, keyspace, lsn_range)| {
-                                    fringe.update(layer, keyspace, lsn_range)
-                                });
-                        }
+                        results
+                            .found
+                            .into_iter()
+                            .map(|(SearchResult { layer, lsn_floor }, keyspace_accum)| {
+                                (
+                                    ReadableLayer::PersistentLayer(guard.get_from_desc(&layer)),
+                                    keyspace_accum.to_keyspace(),
+                                    lsn_floor..cont_lsn,
+                                )
+                            })
+                            .for_each(|(layer, keyspace, lsn_range)| {
+                                fringe.update(layer, keyspace, lsn_range)
+                            });
                     }
                 }
-
-                // It's safe to drop the layer map lock after planning the next round of reads.
-                // The fringe keeps readable handles for the layers which are safe to read even
-                // if layers were compacted or flushed.
-                //
-                // The more interesting consideration is: "Why is the read algorithm still correct
-                // if the layer map changes while it is operating?". Doing a vectored read on a
-                // timeline boils down to pushing an imaginary lsn boundary downwards for each range
-                // covered by the read. The layer map tells us how to move the lsn downwards for a
-                // range at *a particular point in time*. It is fine for the answer to be different
-                // at two different time points.
-                drop(guard);
             }
 
+            // It's safe to drop the layer map lock after planning the next round of reads.
+            // The fringe keeps readable handles for the layers which are safe to read even
+            // if layers were compacted or flushed.
+            //
+            // The more interesting consideration is: "Why is the read algorithm still correct
+            // if the layer map changes while it is operating?". Doing a vectored read on a
+            // timeline boils down to pushing an imaginary lsn boundary downwards for each range
+            // covered by the read. The layer map tells us how to move the lsn downwards for a
+            // range at *a particular point in time*. It is fine for the answer to be different
+            // at two different time points.
+            drop(guard);
+
             if let Some((layer_to_read, keyspace_to_read, lsn_range)) = fringe.next_layer() {
                 let next_cont_lsn = lsn_range.start;
                 layer_to_read
@@ -3326,8 +3081,6 @@ impl Timeline {
 
                 unmapped_keyspace = keyspace_to_read;
                 cont_lsn = next_cont_lsn;
-
-                reconstruct_state.on_layer_visited();
             } else {
                 break;
             }
@@ -3771,7 +3524,7 @@ impl Timeline {
         &self,
         disk_consistent_lsn: Lsn,
         layers_to_upload: impl IntoIterator<Item = ResidentLayer>,
-    ) -> anyhow::Result<()> {
+    ) -> anyhow::Result<TimelineMetadata> {
         // We can only save a valid 'prev_record_lsn' value on disk if we
         // flushed *all* in-memory changes to disk. We only track
         // 'prev_record_lsn' in memory for the latest processed record, so we
@@ -3788,10 +3541,19 @@ impl Timeline {
             None
         };
 
-        let update = crate::tenant::metadata::MetadataUpdate::new(
+        let ancestor_timeline_id = self
+            .ancestor_timeline
+            .as_ref()
+            .map(|ancestor| ancestor.timeline_id);
+
+        let metadata = TimelineMetadata::new(
             disk_consistent_lsn,
             ondisk_prev_record_lsn,
+            ancestor_timeline_id,
+            self.ancestor_lsn,
             *self.latest_gc_cutoff_lsn.read(),
+            self.initdb_lsn,
+            self.pg_version,
         );
 
         fail_point!("checkpoint-before-saving-metadata", |x| bail!(
@@ -3803,10 +3565,10 @@ impl Timeline {
             for layer in layers_to_upload {
                 remote_client.schedule_layer_file_upload(layer)?;
             }
-            remote_client.schedule_index_upload_for_metadata_update(&update)?;
+            remote_client.schedule_index_upload_for_metadata_update(&metadata)?;
         }
 
-        Ok(())
+        Ok(metadata)
     }
 
     pub(crate) async fn preserve_initdb_archive(&self) -> anyhow::Result<()> {
@@ -3915,6 +3677,24 @@ impl Timeline {
 
     // Is it time to create a new image layer for the given partition?
     async fn time_for_new_image_layer(&self, partition: &KeySpace, lsn: Lsn) -> bool {
+        let last = self.last_image_layer_creation_check_at.load();
+        if lsn != Lsn(0) {
+            let distance = lsn
+                .checked_sub(last)
+                .expect("Attempt to compact with LSN going backwards");
+
+            let min_distance = self.get_image_layer_creation_check_threshold() as u64
+                * self.get_checkpoint_distance();
+
+            // Skip the expensive delta layer counting below if we've not ingested
+            // sufficient WAL since the last check.
+            if distance.0 < min_distance {
+                return false;
+            }
+        }
+
+        self.last_image_layer_creation_check_at.store(lsn);
+
         let threshold = self.get_image_creation_threshold();
 
         let guard = self.layers.read().await;
@@ -3986,37 +3766,9 @@ impl Timeline {
         // image layers  <100000000..100000099> and <200000000..200000199> are not completely covering it.
         let mut start = Key::MIN;
 
-        let check_for_image_layers = {
-            let last_checks_at = self.last_image_layer_creation_check_at.load();
-            let distance = lsn
-                .checked_sub(last_checks_at)
-                .expect("Attempt to compact with LSN going backwards");
-            let min_distance = self.get_image_layer_creation_check_threshold() as u64
-                * self.get_checkpoint_distance();
-
-            // Skip the expensive delta layer counting if this timeline has not ingested sufficient
-            // WAL since the last check.
-            distance.0 >= min_distance
-        };
-
-        if check_for_image_layers {
-            self.last_image_layer_creation_check_at.store(lsn);
-        }
-
         for partition in partitioning.parts.iter() {
             let img_range = start..partition.ranges.last().unwrap().end;
-
-            let do_it = if force {
-                true
-            } else if check_for_image_layers {
-                // [`Self::time_for_new_image_layer`] is CPU expensive,
-                // so skip if we've not collected enough WAL since the last time
-                self.time_for_new_image_layer(partition, lsn).await
-            } else {
-                false
-            };
-
-            if !do_it {
+            if !force && !self.time_for_new_image_layer(partition, lsn).await {
                 start = img_range.end;
                 continue;
             }
@@ -4345,12 +4097,6 @@ impl Timeline {
         cancel: &CancellationToken,
         ctx: &RequestContext,
     ) -> anyhow::Result<()> {
-        let _timer = self
-            .metrics
-            .update_gc_info_histo
-            .start_timer()
-            .record_on_drop();
-
         // First, calculate pitr_cutoff_timestamp and then convert it to LSN.
         //
         // Some unit tests depend on garbage-collection working even when
@@ -4396,8 +4142,9 @@ impl Timeline {
                 *self.get_latest_gc_cutoff_lsn()
             }
         } else {
-            // No time-based retention was configured. Interpret this as "keep no history".
-            self.get_last_record_lsn()
+            // No time-based retention was configured. Set time-based cutoff to
+            // same as LSN based.
+            cutoff_horizon
         };
 
         // Grab the lock and update the values
@@ -4917,7 +4664,35 @@ impl Timeline {
     }
 }
 
-type TraversalPathItem = (ValueReconstructResult, Lsn, TraversalId);
+type TraversalPathItem = (
+    ValueReconstructResult,
+    Lsn,
+    Box<dyn Send + FnOnce() -> TraversalId>,
+);
+
+/// Helper function for get_reconstruct_data() to add the path of layers traversed
+/// to an error, as anyhow context information.
+fn layer_traversal_error(msg: String, path: Vec<TraversalPathItem>) -> PageReconstructError {
+    // We want the original 'msg' to be the outermost context. The outermost context
+    // is the most high-level information, which also gets propagated to the client.
+    let mut msg_iter = path
+        .into_iter()
+        .map(|(r, c, l)| {
+            format!(
+                "layer traversal: result {:?}, cont_lsn {}, layer: {}",
+                r,
+                c,
+                l(),
+            )
+        })
+        .chain(std::iter::once(msg));
+    // Construct initial message from the first traversed layer
+    let err = anyhow!(msg_iter.next().unwrap());
+
+    // Append all subsequent traversals, and the error message 'msg', as contexts.
+    let msg = msg_iter.fold(err, |err, msg| err.context(msg));
+    PageReconstructError::from(msg)
+}
 
 struct TimelineWriterState {
     open_layer: Arc<InMemoryLayer>,

pageserver/src/tenant/timeline/eviction_task.rs

@@ -188,10 +188,24 @@ impl Timeline {
     ) -> ControlFlow<()> {
         let now = SystemTime::now();
 
-        let permit = self.acquire_imitation_permit(cancel, ctx).await?;
+        let acquire_permit = crate::tenant::tasks::concurrent_background_tasks_rate_limit_permit(
+            BackgroundLoopKind::Eviction,
+            ctx,
+        );
 
-        self.imitate_layer_accesses(tenant, p, cancel, gate, permit, ctx)
-            .await?;
+        let _permit = tokio::select! {
+            permit = acquire_permit => permit,
+            _ = cancel.cancelled() => return ControlFlow::Break(()),
+            _ = self.cancel.cancelled() => return ControlFlow::Break(()),
+        };
+
+        match self
+            .imitate_layer_accesses(tenant, p, cancel, gate, ctx)
+            .await
+        {
+            ControlFlow::Break(()) => return ControlFlow::Break(()),
+            ControlFlow::Continue(()) => (),
+        }
 
         #[derive(Debug, Default)]
         struct EvictionStats {
@@ -316,27 +330,19 @@ impl Timeline {
         gate: &GateGuard,
         ctx: &RequestContext,
     ) -> ControlFlow<()> {
-        let permit = self.acquire_imitation_permit(cancel, ctx).await?;
-
-        self.imitate_layer_accesses(tenant, p, cancel, gate, permit, ctx)
-            .await
-    }
-
-    async fn acquire_imitation_permit(
-        &self,
-        cancel: &CancellationToken,
-        ctx: &RequestContext,
-    ) -> ControlFlow<(), tokio::sync::SemaphorePermit<'static>> {
         let acquire_permit = crate::tenant::tasks::concurrent_background_tasks_rate_limit_permit(
             BackgroundLoopKind::Eviction,
             ctx,
         );
 
-        tokio::select! {
-            permit = acquire_permit => ControlFlow::Continue(permit),
-            _ = cancel.cancelled() => ControlFlow::Break(()),
-            _ = self.cancel.cancelled() => ControlFlow::Break(()),
-        }
+        let _permit = tokio::select! {
+            permit = acquire_permit => permit,
+            _ = cancel.cancelled() => return ControlFlow::Break(()),
+            _ = self.cancel.cancelled() => return ControlFlow::Break(()),
+        };
+
+        self.imitate_layer_accesses(tenant, p, cancel, gate, ctx)
+            .await
     }
 
     /// If we evict layers but keep cached values derived from those layers, then
@@ -370,7 +376,6 @@ impl Timeline {
         p: &EvictionPolicyLayerAccessThreshold,
         cancel: &CancellationToken,
         gate: &GateGuard,
-        permit: tokio::sync::SemaphorePermit<'static>,
         ctx: &RequestContext,
     ) -> ControlFlow<()> {
         if !self.tenant_shard_id.is_shard_zero() {
@@ -403,28 +408,7 @@ impl Timeline {
         // Make one of the tenant's timelines draw the short straw and run the calculation.
         // The others wait until the calculation is done so that they take into account the
         // imitated accesses that the winner made.
-        let (mut state, _permit) = {
-            if let Ok(locked) = tenant.eviction_task_tenant_state.try_lock() {
-                (locked, permit)
-            } else {
-                // we might need to wait for a long time here in case of pathological synthetic
-                // size calculation performance
-                drop(permit);
-                let locked = tokio::select! {
-                    locked = tenant.eviction_task_tenant_state.lock() => locked,
-                    _ = self.cancel.cancelled() => {
-                        return ControlFlow::Break(())
-                    },
-                    _ = cancel.cancelled() => {
-                        return ControlFlow::Break(())
-                    }
-                };
-                // then reacquire -- this will be bad if there is a lot of traffic, but because we
-                // released the permit, the overall latency will be much better.
-                let permit = self.acquire_imitation_permit(cancel, ctx).await?;
-                (locked, permit)
-            }
-        };
+        let mut state = tenant.eviction_task_tenant_state.lock().await;
         match state.last_layer_access_imitation {
             Some(ts) if ts.elapsed() < inter_imitate_period => { /* no need to run */ }
             _ => {

pageserver/src/tenant/vectored_blob_io.rs

@@ -61,18 +61,18 @@ pub struct VectoredRead {
 }
 
 impl VectoredRead {
-    pub(crate) fn size(&self) -> usize {
+    pub fn size(&self) -> usize {
         (self.end - self.start) as usize
     }
 }
 
 #[derive(Eq, PartialEq)]
-pub(crate) enum VectoredReadExtended {
+enum VectoredReadExtended {
     Yes,
     No,
 }
 
-pub(crate) struct VectoredReadBuilder {
+struct VectoredReadBuilder {
     start: u64,
     end: u64,
     blobs_at: VecMap<u64, BlobMeta>,
@@ -80,17 +80,7 @@ pub(crate) struct VectoredReadBuilder {
 }
 
 impl VectoredReadBuilder {
-    /// Start building a new vectored read.
-    ///
-    /// Note that by design, this does not check against reading more than `max_read_size` to
-    /// support reading larger blobs than the configuration value. The builder will be single use
-    /// however after that.
-    pub(crate) fn new(
-        start_offset: u64,
-        end_offset: u64,
-        meta: BlobMeta,
-        max_read_size: usize,
-    ) -> Self {
+    fn new(start_offset: u64, end_offset: u64, meta: BlobMeta, max_read_size: usize) -> Self {
         let mut blobs_at = VecMap::default();
         blobs_at
             .append(start_offset, meta)
@@ -107,8 +97,7 @@ impl VectoredReadBuilder {
     /// Attempt to extend the current read with a new blob if the start
     /// offset matches with the current end of the vectored read
     /// and the resuting size is below the max read size
-    pub(crate) fn extend(&mut self, start: u64, end: u64, meta: BlobMeta) -> VectoredReadExtended {
-        tracing::trace!(start, end, "trying to extend");
+    fn extend(&mut self, start: u64, end: u64, meta: BlobMeta) -> VectoredReadExtended {
         let size = (end - start) as usize;
         if self.end == start && self.size() + size <= self.max_read_size {
             self.end = end;
@@ -122,11 +111,11 @@ impl VectoredReadBuilder {
         VectoredReadExtended::No
     }
 
-    pub(crate) fn size(&self) -> usize {
+    fn size(&self) -> usize {
         (self.end - self.start) as usize
     }
 
-    pub(crate) fn build(self) -> VectoredRead {
+    fn build(self) -> VectoredRead {
         VectoredRead {
             start: self.start,
             end: self.end,

pageserver/src/virtual_file.rs

@@ -32,11 +32,12 @@ pub use io_engine::feature_test as io_engine_feature_test;
 pub use io_engine::FeatureTestResult as IoEngineFeatureTestResult;
 mod metadata;
 mod open_options;
-use self::owned_buffers_io::write::OwnedAsyncWriter;
 pub(crate) use io_engine::IoEngineKind;
 pub(crate) use metadata::Metadata;
 pub(crate) use open_options::*;
 
+use self::owned_buffers_io::write::OwnedAsyncWriter;
+
 pub(crate) mod owned_buffers_io {
     //! Abstractions for IO with owned buffers.
     //!

pageserver/src/virtual_file/owned_buffers_io/util/size_tracking_writer.rs

@@ -13,18 +13,14 @@ impl<W> Writer<W> {
             bytes_amount: 0,
         }
     }
-
     pub fn bytes_written(&self) -> u64 {
         self.bytes_amount
     }
-
     pub fn as_inner(&self) -> &W {
         &self.dst
     }
-
     /// Returns the wrapped `VirtualFile` object as well as the number
     /// of bytes that were written to it through this object.
-    #[cfg_attr(target_os = "macos", allow(dead_code))]
     pub fn into_inner(self) -> (u64, W) {
         (self.bytes_amount, self.dst)
     }

pageserver/src/virtual_file/owned_buffers_io/write.rs

@@ -56,7 +56,6 @@ where
         self.buf()
     }
 
-    #[cfg_attr(target_os = "macos", allow(dead_code))]
     pub async fn flush_and_into_inner(mut self) -> std::io::Result<W> {
         self.flush().await?;
         let Self { buf, writer } = self;
@@ -71,7 +70,6 @@ where
             .expect("must not use after we returned an error")
     }
 
-    #[cfg_attr(target_os = "macos", allow(dead_code))]
     pub async fn write_buffered<S: IoBuf>(&mut self, chunk: Slice<S>) -> std::io::Result<(usize, S)>
     where
         S: IoBuf + Send,

pageserver/src/walingest.rs

@@ -1034,7 +1034,7 @@ impl WalIngest {
 
             let nblocks = modification
                 .tline
-                .get_rel_size(src_rel, Version::Modified(modification), ctx)
+                .get_rel_size(src_rel, Version::Modified(modification), true, ctx)
                 .await?;
             let dst_rel = RelTag {
                 spcnode: tablespace_id,
@@ -1068,7 +1068,13 @@ impl WalIngest {
 
                 let content = modification
                     .tline
-                    .get_rel_page_at_lsn(src_rel, blknum, Version::Modified(modification), ctx)
+                    .get_rel_page_at_lsn(
+                        src_rel,
+                        blknum,
+                        Version::Modified(modification),
+                        true,
+                        ctx,
+                    )
                     .await?;
                 modification.put_rel_page_image(dst_rel, blknum, content)?;
                 num_blocks_copied += 1;
@@ -1236,7 +1242,7 @@ impl WalIngest {
                 };
                 if modification
                     .tline
-                    .get_rel_exists(rel, Version::Modified(modification), ctx)
+                    .get_rel_exists(rel, Version::Modified(modification), true, ctx)
                     .await?
                 {
                     self.put_rel_drop(modification, rel, ctx).await?;
@@ -1535,7 +1541,7 @@ impl WalIngest {
             nblocks
         } else if !modification
             .tline
-            .get_rel_exists(rel, Version::Modified(modification), ctx)
+            .get_rel_exists(rel, Version::Modified(modification), true, ctx)
             .await?
         {
             // create it with 0 size initially, the logic below will extend it
@@ -1547,7 +1553,7 @@ impl WalIngest {
         } else {
             modification
                 .tline
-                .get_rel_size(rel, Version::Modified(modification), ctx)
+                .get_rel_size(rel, Version::Modified(modification), true, ctx)
                 .await?
         };
 
@@ -1644,14 +1650,14 @@ async fn get_relsize(
 ) -> anyhow::Result<BlockNumber> {
     let nblocks = if !modification
         .tline
-        .get_rel_exists(rel, Version::Modified(modification), ctx)
+        .get_rel_exists(rel, Version::Modified(modification), true, ctx)
         .await?
     {
         0
     } else {
         modification
             .tline
-            .get_rel_size(rel, Version::Modified(modification), ctx)
+            .get_rel_size(rel, Version::Modified(modification), true, ctx)
             .await?
     };
     Ok(nblocks)
@@ -1726,29 +1732,29 @@ mod tests {
         // The relation was created at LSN 2, not visible at LSN 1 yet.
         assert_eq!(
             tline
-                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x10)), &ctx)
+                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x10)), false, &ctx)
                 .await?,
             false
         );
         assert!(tline
-            .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x10)), &ctx)
+            .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x10)), false, &ctx)
             .await
             .is_err());
         assert_eq!(
             tline
-                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x20)), &ctx)
+                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x20)), false, &ctx)
                 .await?,
             true
         );
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x20)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x20)), false, &ctx)
                 .await?,
             1
         );
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x50)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x50)), false, &ctx)
                 .await?,
             3
         );
@@ -1756,46 +1762,46 @@ mod tests {
         // Check page contents at each LSN
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x20)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x20)), false, &ctx)
                 .await?,
             test_img("foo blk 0 at 2")
         );
 
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x30)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x30)), false, &ctx)
                 .await?,
             test_img("foo blk 0 at 3")
         );
 
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x40)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x40)), false, &ctx)
                 .await?,
             test_img("foo blk 0 at 3")
         );
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 1, Version::Lsn(Lsn(0x40)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 1, Version::Lsn(Lsn(0x40)), false, &ctx)
                 .await?,
             test_img("foo blk 1 at 4")
         );
 
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x50)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x50)), false, &ctx)
                 .await?,
             test_img("foo blk 0 at 3")
         );
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 1, Version::Lsn(Lsn(0x50)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 1, Version::Lsn(Lsn(0x50)), false, &ctx)
                 .await?,
             test_img("foo blk 1 at 4")
         );
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 2, Version::Lsn(Lsn(0x50)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 2, Version::Lsn(Lsn(0x50)), false, &ctx)
                 .await?,
             test_img("foo blk 2 at 5")
         );
@@ -1811,19 +1817,19 @@ mod tests {
         // Check reported size and contents after truncation
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x60)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x60)), false, &ctx)
                 .await?,
             2
         );
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x60)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x60)), false, &ctx)
                 .await?,
             test_img("foo blk 0 at 3")
         );
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 1, Version::Lsn(Lsn(0x60)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 1, Version::Lsn(Lsn(0x60)), false, &ctx)
                 .await?,
             test_img("foo blk 1 at 4")
         );
@@ -1831,13 +1837,13 @@ mod tests {
         // should still see the truncated block with older LSN
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x50)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x50)), false, &ctx)
                 .await?,
             3
         );
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 2, Version::Lsn(Lsn(0x50)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 2, Version::Lsn(Lsn(0x50)), false, &ctx)
                 .await?,
             test_img("foo blk 2 at 5")
         );
@@ -1850,7 +1856,7 @@ mod tests {
         m.commit(&ctx).await?;
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x68)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x68)), false, &ctx)
                 .await?,
             0
         );
@@ -1863,19 +1869,19 @@ mod tests {
         m.commit(&ctx).await?;
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x70)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x70)), false, &ctx)
                 .await?,
             2
         );
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x70)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x70)), false, &ctx)
                 .await?,
             ZERO_PAGE
         );
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 1, Version::Lsn(Lsn(0x70)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 1, Version::Lsn(Lsn(0x70)), false, &ctx)
                 .await?,
             test_img("foo blk 1")
         );
@@ -1888,21 +1894,21 @@ mod tests {
         m.commit(&ctx).await?;
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x80)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x80)), false, &ctx)
                 .await?,
             1501
         );
         for blk in 2..1500 {
             assert_eq!(
                 tline
-                    .get_rel_page_at_lsn(TESTREL_A, blk, Version::Lsn(Lsn(0x80)), &ctx)
+                    .get_rel_page_at_lsn(TESTREL_A, blk, Version::Lsn(Lsn(0x80)), false, &ctx)
                     .await?,
                 ZERO_PAGE
             );
         }
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 1500, Version::Lsn(Lsn(0x80)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 1500, Version::Lsn(Lsn(0x80)), false, &ctx)
                 .await?,
             test_img("foo blk 1500")
         );
@@ -1929,13 +1935,13 @@ mod tests {
         // Check that rel exists and size is correct
         assert_eq!(
             tline
-                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x20)), &ctx)
+                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x20)), false, &ctx)
                 .await?,
             true
         );
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x20)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x20)), false, &ctx)
                 .await?,
             1
         );
@@ -1948,7 +1954,7 @@ mod tests {
         // Check that rel is not visible anymore
         assert_eq!(
             tline
-                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x30)), &ctx)
+                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x30)), false, &ctx)
                 .await?,
             false
         );
@@ -1966,13 +1972,13 @@ mod tests {
         // Check that rel exists and size is correct
         assert_eq!(
             tline
-                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x40)), &ctx)
+                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x40)), false, &ctx)
                 .await?,
             true
         );
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x40)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x40)), false, &ctx)
                 .await?,
             1
         );
@@ -2005,24 +2011,24 @@ mod tests {
         // The relation was created at LSN 20, not visible at LSN 1 yet.
         assert_eq!(
             tline
-                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x10)), &ctx)
+                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x10)), false, &ctx)
                 .await?,
             false
         );
         assert!(tline
-            .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x10)), &ctx)
+            .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x10)), false, &ctx)
             .await
             .is_err());
 
         assert_eq!(
             tline
-                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x20)), &ctx)
+                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x20)), false, &ctx)
                 .await?,
             true
         );
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x20)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x20)), false, &ctx)
                 .await?,
             relsize
         );
@@ -2033,7 +2039,7 @@ mod tests {
             let data = format!("foo blk {} at {}", blkno, lsn);
             assert_eq!(
                 tline
-                    .get_rel_page_at_lsn(TESTREL_A, blkno, Version::Lsn(lsn), &ctx)
+                    .get_rel_page_at_lsn(TESTREL_A, blkno, Version::Lsn(lsn), false, &ctx)
                     .await?,
                 test_img(&data)
             );
@@ -2050,7 +2056,7 @@ mod tests {
         // Check reported size and contents after truncation
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x60)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x60)), false, &ctx)
                 .await?,
             1
         );
@@ -2060,7 +2066,7 @@ mod tests {
             let data = format!("foo blk {} at {}", blkno, lsn);
             assert_eq!(
                 tline
-                    .get_rel_page_at_lsn(TESTREL_A, blkno, Version::Lsn(Lsn(0x60)), &ctx)
+                    .get_rel_page_at_lsn(TESTREL_A, blkno, Version::Lsn(Lsn(0x60)), false, &ctx)
                     .await?,
                 test_img(&data)
             );
@@ -2069,7 +2075,7 @@ mod tests {
         // should still see all blocks with older LSN
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x50)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x50)), false, &ctx)
                 .await?,
             relsize
         );
@@ -2078,7 +2084,7 @@ mod tests {
             let data = format!("foo blk {} at {}", blkno, lsn);
             assert_eq!(
                 tline
-                    .get_rel_page_at_lsn(TESTREL_A, blkno, Version::Lsn(Lsn(0x50)), &ctx)
+                    .get_rel_page_at_lsn(TESTREL_A, blkno, Version::Lsn(Lsn(0x50)), false, &ctx)
                     .await?,
                 test_img(&data)
             );
@@ -2098,13 +2104,13 @@ mod tests {
 
         assert_eq!(
             tline
-                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x80)), &ctx)
+                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x80)), false, &ctx)
                 .await?,
             true
         );
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x80)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x80)), false, &ctx)
                 .await?,
             relsize
         );
@@ -2114,7 +2120,7 @@ mod tests {
             let data = format!("foo blk {} at {}", blkno, lsn);
             assert_eq!(
                 tline
-                    .get_rel_page_at_lsn(TESTREL_A, blkno, Version::Lsn(Lsn(0x80)), &ctx)
+                    .get_rel_page_at_lsn(TESTREL_A, blkno, Version::Lsn(Lsn(0x80)), false, &ctx)
                     .await?,
                 test_img(&data)
             );
@@ -2148,7 +2154,7 @@ mod tests {
 
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(lsn)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(lsn)), false, &ctx)
                 .await?,
             RELSEG_SIZE + 1
         );
@@ -2162,7 +2168,7 @@ mod tests {
         m.commit(&ctx).await?;
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(lsn)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(lsn)), false, &ctx)
                 .await?,
             RELSEG_SIZE
         );
@@ -2177,7 +2183,7 @@ mod tests {
         m.commit(&ctx).await?;
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(lsn)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(lsn)), false, &ctx)
                 .await?,
             RELSEG_SIZE - 1
         );
@@ -2195,7 +2201,7 @@ mod tests {
             m.commit(&ctx).await?;
             assert_eq!(
                 tline
-                    .get_rel_size(TESTREL_A, Version::Lsn(Lsn(lsn)), &ctx)
+                    .get_rel_size(TESTREL_A, Version::Lsn(Lsn(lsn)), false, &ctx)
                     .await?,
                 size as BlockNumber
             );

pageserver/src/walrecord.rs

@@ -55,7 +55,6 @@ impl NeonWalRecord {
     /// Does replaying this WAL record initialize the page from scratch, or does
     /// it need to be applied over the previous image of the page?
     pub fn will_init(&self) -> bool {
-        // If you change this function, you'll also need to change ValueBytes::will_init
         match self {
             NeonWalRecord::Postgres { will_init, rec: _ } => *will_init,
 

pgxn/neon/libpagestore.c

@@ -49,8 +49,6 @@ char	   *neon_auth_token;
 int			readahead_buffer_size = 128;
 int			flush_every_n_requests = 8;
 
-int         neon_protocol_version = 1;
-
 static int	n_reconnect_attempts = 0;
 static int	max_reconnect_attempts = 60;
 static int	stripe_size;
@@ -381,17 +379,7 @@ pageserver_connect(shardno_t shard_no, int elevel)
 		pfree(msg);
 		return false;
 	}
-	switch (neon_protocol_version)
-	{
-		case 2:
-			query = psprintf("pagestream_v2 %s %s", neon_tenant, neon_timeline);
-			break;
-		case 1:
-			query = psprintf("pagestream %s %s", neon_tenant, neon_timeline);
-			break;
-		default:
-			elog(ERROR, "unexpected neon_protocol_version %d", neon_protocol_version);
-	}
+	query = psprintf("pagestream %s %s", neon_tenant, neon_timeline);
 	ret = PQsendQuery(conn, query);
 	pfree(query);
 	if (ret != 1)
@@ -452,7 +440,7 @@ pageserver_connect(shardno_t shard_no, int elevel)
 		return false;
 	}
 
-	neon_shard_log(shard_no, LOG, "libpagestore: connected to '%s' with protocol version %d", connstr, neon_protocol_version);
+	neon_shard_log(shard_no, LOG, "libpagestore: connected to '%s'", connstr);
 	page_servers[shard_no].conn = conn;
 	page_servers[shard_no].wes = wes;
 
@@ -856,16 +844,6 @@ pg_init_libpagestore(void)
 							PGC_USERSET,
 							0,	/* no flags required */
 							NULL, (GucIntAssignHook) &readahead_buffer_resize, NULL);
-	DefineCustomIntVariable("neon.protocol_version",
-							"Version of compute<->page server protocol",
-							NULL,
-							&neon_protocol_version,
-							1, /* default to old protocol for now */
-							1, /* min */
-							2, /* max */
-							PGC_SU_BACKEND,
-							0,	/* no flags required */
-							NULL, NULL, NULL);
 
 	relsize_hash_init();
 

pgxn/neon/pagestore_client.h

@@ -69,33 +69,18 @@ typedef enum {
 	SLRU_MULTIXACT_OFFSETS
 } SlruKind;
 
-/*--
- * supertype of all the Neon*Request structs below.
+/*
+ * supertype of all the Neon*Request structs below
  *
- * All requests contain two LSNs:
- *
- * lsn:                request page (or relation size, etc) at this LSN
- * not_modified_since: Hint that the page hasn't been modified between
- *                     this LSN and the request LSN (`lsn`).
- *
- * To request the latest version of a page, you can use MAX_LSN as the request
- * LSN.
- *
- * If you don't know any better, you can always set 'not_modified_since' equal
- * to 'lsn', but providing a lower value can speed up processing the request
- * in the pageserver, as it doesn't need to wait for the WAL to arrive, and it
- * can skip traversing through recent layers which we know to not contain any
- * versions for the requested page.
- *
- * These structs describe the V2 of these requests. The old V1 protocol contained
- * just one LSN and a boolean 'latest' flag. If the neon_protocol_version GUC is
- * set to 1, we will convert these to the V1 requests before sending.
+ * If 'latest' is true, we are requesting the latest page version, and 'lsn'
+ * is just a hint to the server that we know there are no versions of the page
+ * (or relation size, for exists/nblocks requests) later than the 'lsn'.
  */
 typedef struct
 {
 	NeonMessageTag tag;
-	XLogRecPtr	lsn;
-	XLogRecPtr	not_modified_since;
+	bool		latest;			/* if true, request latest page version */
+	XLogRecPtr	lsn;			/* request page version @ this LSN */
 } NeonRequest;
 
 typedef struct
@@ -208,7 +193,6 @@ extern int	readahead_buffer_size;
 extern char *neon_timeline;
 extern char *neon_tenant;
 extern int32 max_cluster_size;
-extern int  neon_protocol_version;
 
 extern shardno_t get_shard_number(BufferTag* tag);
 
@@ -241,14 +225,14 @@ extern bool neon_prefetch(SMgrRelation reln, ForkNumber forknum,
 extern void neon_read(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
 					  char *buffer);
 extern PGDLLEXPORT void neon_read_at_lsn(NRelFileInfo rnode, ForkNumber forkNum, BlockNumber blkno,
-										 XLogRecPtr request_lsn, XLogRecPtr not_modified_since, char *buffer);
+										 XLogRecPtr request_lsn, bool request_latest, char *buffer);
 extern void neon_write(SMgrRelation reln, ForkNumber forknum,
 					   BlockNumber blocknum, char *buffer, bool skipFsync);
 #else
 extern void neon_read(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
 					  void *buffer);
 extern PGDLLEXPORT void neon_read_at_lsn(NRelFileInfo rnode, ForkNumber forkNum, BlockNumber blkno,
-										 XLogRecPtr request_lsn, XLogRecPtr not_modified_since, void *buffer);
+										 XLogRecPtr request_lsn, bool request_latest, void *buffer);
 extern void neon_write(SMgrRelation reln, ForkNumber forknum,
 					   BlockNumber blocknum, const void *buffer, bool skipFsync);
 #endif

pgxn/neon/pagestore_smgr.c

@@ -168,8 +168,8 @@ typedef enum PrefetchStatus
 typedef struct PrefetchRequest
 {
 	BufferTag	buftag;			/* must be first entry in the struct */
-	XLogRecPtr	request_lsn;
-	XLogRecPtr	not_modified_since;
+	XLogRecPtr	effective_request_lsn;
+	XLogRecPtr	actual_request_lsn;
 	NeonResponse *response;		/* may be null */
 	PrefetchStatus status;
 	shardno_t   shard_no;
@@ -269,19 +269,19 @@ static PrefetchState *MyPState;
 	) \
 )
 
+static XLogRecPtr prefetch_lsn = 0;
+
 static bool compact_prefetch_buffers(void);
 static void consume_prefetch_responses(void);
-static uint64 prefetch_register_buffer(BufferTag tag, XLogRecPtr *force_request_lsn, XLogRecPtr *force_not_modified_since);
+static uint64 prefetch_register_buffer(BufferTag tag, bool *force_latest, XLogRecPtr *force_lsn);
 static bool prefetch_read(PrefetchRequest *slot);
-static void prefetch_do_request(PrefetchRequest *slot, XLogRecPtr *force_request_lsn, XLogRecPtr *force_not_modified_since);
+static void prefetch_do_request(PrefetchRequest *slot, bool *force_latest, XLogRecPtr *force_lsn);
 static bool prefetch_wait_for(uint64 ring_index);
 static void prefetch_cleanup_trailing_unused(void);
 static inline void prefetch_set_unused(uint64 ring_index);
 
-static void neon_get_request_lsn(NRelFileInfo rinfo, ForkNumber forknum, BlockNumber blkno,
-								 XLogRecPtr *request_lsn, XLogRecPtr *not_modified_since);
-static bool neon_prefetch_response_usable(XLogRecPtr request_lsn, XLogRecPtr not_modified_since,
-										  PrefetchRequest *slot);
+static XLogRecPtr neon_get_request_lsn(bool *latest, NRelFileInfo rinfo,
+									   ForkNumber forknum, BlockNumber blkno);
 
 static bool
 compact_prefetch_buffers(void)
@@ -338,8 +338,8 @@ compact_prefetch_buffers(void)
 		target_slot->shard_no = source_slot->shard_no;
 		target_slot->status = source_slot->status;
 		target_slot->response = source_slot->response;
-		target_slot->request_lsn = source_slot->request_lsn;
-		target_slot->not_modified_since = source_slot->not_modified_since;
+		target_slot->effective_request_lsn = source_slot->effective_request_lsn;
+		target_slot->actual_request_lsn = source_slot->actual_request_lsn;
 		target_slot->my_ring_index = empty_ring_index;
 
 		prfh_delete(MyPState->prf_hash, source_slot);
@@ -358,8 +358,7 @@ compact_prefetch_buffers(void)
 		};
 		source_slot->response = NULL;
 		source_slot->my_ring_index = 0;
-		source_slot->request_lsn = InvalidXLogRecPtr;
-		source_slot->not_modified_since = InvalidXLogRecPtr;
+		source_slot->effective_request_lsn = 0;
 
 		/* update bookkeeping */
 		n_moved++;
@@ -684,39 +683,56 @@ prefetch_set_unused(uint64 ring_index)
 		compact_prefetch_buffers();
 }
 
-/*
- * Send one prefetch request to the pageserver. To wait for the response, call
- * prefetch_wait_for().
- */
 static void
-prefetch_do_request(PrefetchRequest *slot, XLogRecPtr *force_request_lsn, XLogRecPtr *force_not_modified_since)
+prefetch_do_request(PrefetchRequest *slot, bool *force_latest, XLogRecPtr *force_lsn)
 {
 	bool		found;
 	NeonGetPageRequest request = {
 		.req.tag = T_NeonGetPageRequest,
-		/* lsn and not_modified_since are filled in below */
+		.req.latest = false,
+		.req.lsn = 0,
 		.rinfo = BufTagGetNRelFileInfo(slot->buftag),
 		.forknum = slot->buftag.forkNum,
 		.blkno = slot->buftag.blockNum,
 	};
 
-	Assert(((force_request_lsn != NULL) == (force_not_modified_since != NULL)));
-
-	if (force_request_lsn)
+	if (force_lsn && force_latest)
 	{
-		request.req.lsn = *force_request_lsn;
-		request.req.not_modified_since = *force_not_modified_since;
+		request.req.lsn = *force_lsn;
+		request.req.latest = *force_latest;
+		slot->actual_request_lsn = slot->effective_request_lsn = *force_lsn;
 	}
 	else
 	{
-		neon_get_request_lsn(BufTagGetNRelFileInfo(slot->buftag),
-							 slot->buftag.forkNum,
-							 slot->buftag.blockNum,
-							 &request.req.lsn,
-							 &request.req.not_modified_since);
+		XLogRecPtr	lsn = neon_get_request_lsn(
+											   &request.req.latest,
+											   BufTagGetNRelFileInfo(slot->buftag),
+											   slot->buftag.forkNum,
+											   slot->buftag.blockNum
+			);
+
+		/*
+		 * Note: effective_request_lsn is potentially higher than the
+		 * requested LSN, but still correct:
+		 *
+		 * We know there are no changes between the actual requested LSN and
+		 * the value of effective_request_lsn: If there were, the page would
+		 * have been in cache and evicted between those LSN values, which then
+		 * would have had to result in a larger request LSN for this page.
+		 *
+		 * It is possible that a concurrent backend loads the page, modifies
+		 * it and then evicts it again, but the LSN of that eviction cannot be
+		 * smaller than the current WAL insert/redo pointer, which is already
+		 * larger than this prefetch_lsn. So in any case, that would
+		 * invalidate this cache.
+		 *
+		 * The best LSN to use for effective_request_lsn would be
+		 * XLogCtl->Insert.RedoRecPtr, but that's expensive to access.
+		 */
+		slot->actual_request_lsn = request.req.lsn = lsn;
+		prefetch_lsn = Max(prefetch_lsn, lsn);
+		slot->effective_request_lsn = prefetch_lsn;
 	}
-	slot->request_lsn = request.req.lsn;
-	slot->not_modified_since = request.req.not_modified_since;
 
 	Assert(slot->response == NULL);
 	Assert(slot->my_ring_index == MyPState->ring_unused);
@@ -733,6 +749,7 @@ prefetch_do_request(PrefetchRequest *slot, XLogRecPtr *force_request_lsn, XLogRe
 	/* update slot state */
 	slot->status = PRFS_REQUESTED;
 
+
 	prfh_insert(MyPState->prf_hash, slot, &found);
 	Assert(!found);
 }
@@ -742,25 +759,22 @@ prefetch_do_request(PrefetchRequest *slot, XLogRecPtr *force_request_lsn, XLogRe
  *
  * Register that we may want the contents of BufferTag in the near future.
  *
- * If force_request_lsn and force_not_modified_since are not NULL, those
- * values are sent to the pageserver. If they are NULL, we utilize the
- * lastWrittenLsn -infrastructure to fill them in.
+ * If force_latest and force_lsn are not NULL, those values are sent to the
+ * pageserver. If they are NULL, we utilize the lastWrittenLsn -infrastructure
+ * to fill in these values manually.
  *
  * NOTE: this function may indirectly update MyPState->pfs_hash; which
  * invalidates any active pointers into the hash table.
  */
 
 static uint64
-prefetch_register_buffer(BufferTag tag, XLogRecPtr *force_request_lsn,
-						 XLogRecPtr *force_not_modified_since)
+prefetch_register_buffer(BufferTag tag, bool *force_latest, XLogRecPtr *force_lsn)
 {
 	uint64		ring_index;
 	PrefetchRequest req;
 	PrefetchRequest *slot;
 	PrfHashEntry *entry;
 
-	Assert(((force_request_lsn != NULL) == (force_not_modified_since != NULL)));
-
 	/* use an intermediate PrefetchRequest struct to ensure correct alignment */
 	req.buftag = tag;
 Retry:
@@ -778,19 +792,40 @@ Retry:
 		Assert(BUFFERTAGS_EQUAL(slot->buftag, tag));
 
 		/*
-		 * If the caller specified a request LSN to use, only accept prefetch
-		 * responses that satisfy that request.
+		 * If we want a specific lsn, we do not accept requests that were made
+		 * with a potentially different LSN.
 		 */
-		if (force_request_lsn)
+		if (force_latest && force_lsn)
 		{
-			if (!neon_prefetch_response_usable(*force_request_lsn,
-											   *force_not_modified_since, slot))
+			/*
+			 * if we want the latest version, any effective_request_lsn <
+			 * request lsn is OK
+			 */
+			if (*force_latest)
 			{
-				/* Wait for the old request to finish and discard it */
-				if (!prefetch_wait_for(ring_index))
-					goto Retry;
-				prefetch_set_unused(ring_index);
-				entry = NULL;
+				if (*force_lsn > slot->effective_request_lsn)
+				{
+					if (!prefetch_wait_for(ring_index))
+						goto Retry;
+					prefetch_set_unused(ring_index);
+					entry = NULL;
+				}
+
+			}
+
+			/*
+			 * if we don't want the latest version, only accept requests with
+			 * the exact same LSN
+			 */
+			else
+			{
+				if (*force_lsn != slot->effective_request_lsn)
+				{
+					if (!prefetch_wait_for(ring_index))
+						goto Retry;
+					prefetch_set_unused(ring_index);
+					entry = NULL;
+				}
 			}
 		}
 
@@ -886,7 +921,7 @@ Retry:
 	slot->shard_no = get_shard_number(&tag);
 	slot->my_ring_index = ring_index;
 
-	prefetch_do_request(slot, force_request_lsn, force_not_modified_since);
+	prefetch_do_request(slot, force_latest, force_lsn);
 	Assert(slot->status == PRFS_REQUESTED);
 	Assert(MyPState->ring_last <= ring_index &&
 		   ring_index < MyPState->ring_unused);
@@ -915,7 +950,7 @@ page_server_request(void const *req)
 	BufferTag tag = {0};
 	shardno_t shard_no;
 
-	switch (messageTag(req))
+	switch (((NeonRequest *) req)->tag)
 	{
 		case T_NeonExistsRequest:
 			CopyNRelFileInfoToBufTag(tag, ((NeonExistsRequest *) req)->rinfo);
@@ -931,10 +966,11 @@ page_server_request(void const *req)
 			tag.blockNum = ((NeonGetPageRequest *) req)->blkno;
 			break;
 		default:
-			neon_log(ERROR, "Unexpected request tag: %d", messageTag(req));
+			neon_log(ERROR, "Unexpected request tag: %d", ((NeonRequest *) req)->tag);
 	}
 	shard_no = get_shard_number(&tag);
 
+
 	/*
 	 * Current sharding model assumes that all metadata is present only at shard 0.
 	 * We still need to call get_shard_no() to check if shard map is up-to-date.
@@ -961,52 +997,8 @@ nm_pack_request(NeonRequest *msg)
 	StringInfoData s;
 
 	initStringInfo(&s);
+	pq_sendbyte(&s, msg->tag);
 
-	if (neon_protocol_version >= 2)
-	{
-		pq_sendbyte(&s, msg->tag);
-		pq_sendint64(&s, msg->lsn);
-		pq_sendint64(&s, msg->not_modified_since);
-	}
-	else
-	{
-		bool		latest;
-		XLogRecPtr	lsn;
-
-		/*
-		 * In primary, we always request the latest page version.
-		 */
-		if (!RecoveryInProgress())
-		{
-			latest = true;
-			lsn = msg->not_modified_since;
-		}
-		else
-		{
-			/*
-			 * In the protocol V1, we cannot represent that we want to read
-			 * page at LSN X, and we know that it hasn't been modified since
-			 * Y. We can either use 'not_modified_lsn' as the request LSN, and
-			 * risk getting an error if that LSN is too old and has already
-			 * fallen out of the pageserver's GC horizon, or we can send
-			 * 'request_lsn', causing the pageserver to possibly wait for the
-			 * recent WAL to arrive unnecessarily. Or something in between. We
-			 * choose to use the old LSN and risk GC errors, because that's
-			 * what we've done historically.
-			 */
-			latest = false;
-			lsn = msg->not_modified_since;
-		}
-
-		pq_sendbyte(&s, msg->tag);
-		pq_sendbyte(&s, latest);
-		pq_sendint64(&s, lsn);
-	}
-
-	/*
-	 * The rest of the request messages are the same between protocol V1 and
-	 * V2
-	 */
 	switch (messageTag(msg))
 	{
 			/* pagestore_client -> pagestore */
@@ -1014,6 +1006,8 @@ nm_pack_request(NeonRequest *msg)
 			{
 				NeonExistsRequest *msg_req = (NeonExistsRequest *) msg;
 
+				pq_sendbyte(&s, msg_req->req.latest);
+				pq_sendint64(&s, msg_req->req.lsn);
 				pq_sendint32(&s, NInfoGetSpcOid(msg_req->rinfo));
 				pq_sendint32(&s, NInfoGetDbOid(msg_req->rinfo));
 				pq_sendint32(&s, NInfoGetRelNumber(msg_req->rinfo));
@@ -1025,6 +1019,8 @@ nm_pack_request(NeonRequest *msg)
 			{
 				NeonNblocksRequest *msg_req = (NeonNblocksRequest *) msg;
 
+				pq_sendbyte(&s, msg_req->req.latest);
+				pq_sendint64(&s, msg_req->req.lsn);
 				pq_sendint32(&s, NInfoGetSpcOid(msg_req->rinfo));
 				pq_sendint32(&s, NInfoGetDbOid(msg_req->rinfo));
 				pq_sendint32(&s, NInfoGetRelNumber(msg_req->rinfo));
@@ -1036,6 +1032,8 @@ nm_pack_request(NeonRequest *msg)
 			{
 				NeonDbSizeRequest *msg_req = (NeonDbSizeRequest *) msg;
 
+				pq_sendbyte(&s, msg_req->req.latest);
+				pq_sendint64(&s, msg_req->req.lsn);
 				pq_sendint32(&s, msg_req->dbNode);
 
 				break;
@@ -1044,6 +1042,8 @@ nm_pack_request(NeonRequest *msg)
 			{
 				NeonGetPageRequest *msg_req = (NeonGetPageRequest *) msg;
 
+				pq_sendbyte(&s, msg_req->req.latest);
+				pq_sendint64(&s, msg_req->req.lsn);
 				pq_sendint32(&s, NInfoGetSpcOid(msg_req->rinfo));
 				pq_sendint32(&s, NInfoGetDbOid(msg_req->rinfo));
 				pq_sendint32(&s, NInfoGetRelNumber(msg_req->rinfo));
@@ -1057,6 +1057,8 @@ nm_pack_request(NeonRequest *msg)
 			{
 				NeonGetSlruSegmentRequest *msg_req = (NeonGetSlruSegmentRequest *) msg;
 
+				pq_sendbyte(&s, msg_req->req.latest);
+				pq_sendint64(&s, msg_req->req.lsn);
 				pq_sendbyte(&s, msg_req->kind);
 				pq_sendint32(&s, msg_req->segno);
 
@@ -1207,7 +1209,7 @@ nm_to_string(NeonMessage *msg)
 				appendStringInfo(&s, ", \"rinfo\": \"%u/%u/%u\"", RelFileInfoFmt(msg_req->rinfo));
 				appendStringInfo(&s, ", \"forknum\": %d", msg_req->forknum);
 				appendStringInfo(&s, ", \"lsn\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.lsn));
-				appendStringInfo(&s, ", \"not_modified_since\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.not_modified_since));
+				appendStringInfo(&s, ", \"latest\": %d", msg_req->req.latest);
 				appendStringInfoChar(&s, '}');
 				break;
 			}
@@ -1220,7 +1222,7 @@ nm_to_string(NeonMessage *msg)
 				appendStringInfo(&s, ", \"rinfo\": \"%u/%u/%u\"", RelFileInfoFmt(msg_req->rinfo));
 				appendStringInfo(&s, ", \"forknum\": %d", msg_req->forknum);
 				appendStringInfo(&s, ", \"lsn\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.lsn));
-				appendStringInfo(&s, ", \"not_modified_since\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.not_modified_since));
+				appendStringInfo(&s, ", \"latest\": %d", msg_req->req.latest);
 				appendStringInfoChar(&s, '}');
 				break;
 			}
@@ -1234,7 +1236,7 @@ nm_to_string(NeonMessage *msg)
 				appendStringInfo(&s, ", \"forknum\": %d", msg_req->forknum);
 				appendStringInfo(&s, ", \"blkno\": %u", msg_req->blkno);
 				appendStringInfo(&s, ", \"lsn\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.lsn));
-				appendStringInfo(&s, ", \"not_modified_since\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.not_modified_since));
+				appendStringInfo(&s, ", \"latest\": %d", msg_req->req.latest);
 				appendStringInfoChar(&s, '}');
 				break;
 			}
@@ -1245,7 +1247,7 @@ nm_to_string(NeonMessage *msg)
 				appendStringInfoString(&s, "{\"type\": \"NeonDbSizeRequest\"");
 				appendStringInfo(&s, ", \"dbnode\": \"%u\"", msg_req->dbNode);
 				appendStringInfo(&s, ", \"lsn\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.lsn));
-				appendStringInfo(&s, ", \"not_modified_since\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.not_modified_since));
+				appendStringInfo(&s, ", \"latest\": %d", msg_req->req.latest);
 				appendStringInfoChar(&s, '}');
 				break;
 			}
@@ -1257,7 +1259,7 @@ nm_to_string(NeonMessage *msg)
 				appendStringInfo(&s, ", \"kind\": %u", msg_req->kind);
 				appendStringInfo(&s, ", \"segno\": %u", msg_req->segno);
 				appendStringInfo(&s, ", \"lsn\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.lsn));
-				appendStringInfo(&s, ", \"not_modified_since\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.not_modified_since));
+				appendStringInfo(&s, ", \"latest\": %d", msg_req->req.latest);
 				appendStringInfoChar(&s, '}');
 				break;
 			}
@@ -1529,38 +1531,44 @@ nm_adjust_lsn(XLogRecPtr lsn)
 /*
  * Return LSN for requesting pages and number of blocks from page server
  */
-static void
-neon_get_request_lsn(NRelFileInfo rinfo, ForkNumber forknum, BlockNumber blkno,
-					 XLogRecPtr *request_lsn, XLogRecPtr *not_modified_since)
+static XLogRecPtr
+neon_get_request_lsn(bool *latest, NRelFileInfo rinfo, ForkNumber forknum, BlockNumber blkno)
 {
-	XLogRecPtr	last_written_lsn;
-
-	last_written_lsn = GetLastWrittenLSN(rinfo, forknum, blkno);
-	last_written_lsn = nm_adjust_lsn(last_written_lsn);
-	Assert(last_written_lsn != InvalidXLogRecPtr);
+	XLogRecPtr	lsn;
 
 	if (RecoveryInProgress())
 	{
-		/* Request the page at the last replayed LSN. */
-		*request_lsn = GetXLogReplayRecPtr(NULL);
-		*not_modified_since = last_written_lsn;
-		Assert(last_written_lsn <= *request_lsn);
+		/*
+		 * We don't know if WAL has been generated but not yet replayed, so
+		 * we're conservative in our estimates about latest pages.
+		 */
+		*latest = false;
 
-		neon_log(DEBUG1, "neon_get_request_lsn request lsn %X/%X, not_modified_since %X/%X",
-				 LSN_FORMAT_ARGS(*request_lsn), LSN_FORMAT_ARGS(*not_modified_since));
+		/*
+		 * Get the last written LSN of this page.
+		 */
+		lsn = GetLastWrittenLSN(rinfo, forknum, blkno);
+		lsn = nm_adjust_lsn(lsn);
+
+		neon_log(DEBUG1, "neon_get_request_lsn GetXLogReplayRecPtr %X/%X request lsn 0 ",
+			 (uint32) ((lsn) >> 32), (uint32) (lsn));
 	}
 	else
 	{
 		XLogRecPtr	flushlsn;
 
 		/*
-		 * Use the latest LSN that was evicted from the buffer cache as the
-		 * 'not_modified_since' hint. Any pages modified by later WAL records
-		 * must still in the buffer cache, so our request cannot concern
-		 * those.
+		 * Use the latest LSN that was evicted from the buffer cache. Any
+		 * pages modified by later WAL records must still in the buffer cache,
+		 * so our request cannot concern those.
 		 */
+		*latest = true;
+		lsn = GetLastWrittenLSN(rinfo, forknum, blkno);
+		Assert(lsn != InvalidXLogRecPtr);
 		neon_log(DEBUG1, "neon_get_request_lsn GetLastWrittenLSN lsn %X/%X ",
-				 LSN_FORMAT_ARGS(last_written_lsn));
+			 (uint32) ((lsn) >> 32), (uint32) (lsn));
+
+		lsn = nm_adjust_lsn(lsn);
 
 		/*
 		 * Is it possible that the last-written LSN is ahead of last flush
@@ -1575,109 +1583,16 @@ neon_get_request_lsn(NRelFileInfo rinfo, ForkNumber forknum, BlockNumber blkno,
 #else
 		flushlsn = GetFlushRecPtr();
 #endif
-		if (last_written_lsn > flushlsn)
+		if (lsn > flushlsn)
 		{
 			neon_log(DEBUG5, "last-written LSN %X/%X is ahead of last flushed LSN %X/%X",
-					 LSN_FORMAT_ARGS(last_written_lsn),
-					 LSN_FORMAT_ARGS(flushlsn));
-			XLogFlush(last_written_lsn);
-			flushlsn = last_written_lsn;
+				 (uint32) (lsn >> 32), (uint32) lsn,
+				 (uint32) (flushlsn >> 32), (uint32) flushlsn);
+			XLogFlush(lsn);
 		}
-
-		/*
-		 * Request the latest version of the page. The most up-to-date request
-		 * LSN we could use would be the current insert LSN, but to avoid the
-		 * overhead of looking it up, use 'flushlsn' instead. This relies on
-		 * the assumption that if the page was modified since the last WAL
-		 * flush, it should still be in the buffer cache, and we wouldn't be
-		 * requesting it.
-		 */
-		*request_lsn = flushlsn;
-		*not_modified_since = last_written_lsn;
-	}
-}
-
-/*
- *  neon_prefetch_response_usable -- Can a new request be satisfied by old one?
- *
- * This is used to check if the response to a prefetch request can be used to
- * satisfy a page read now.
- */
-static bool
-neon_prefetch_response_usable(XLogRecPtr request_lsn, XLogRecPtr not_modified_since,
-							  PrefetchRequest *slot)
-{
-	/* sanity check the LSN's on the old and the new request */
-	Assert(request_lsn >= not_modified_since);
-	Assert(slot->request_lsn >= slot->not_modified_since);
-	Assert(slot->status != PRFS_UNUSED);
-
-	/*
-	 * The new request's LSN should never be older than the old one.  This
-	 * could be an Assert, except that for testing purposes, we do provide an
-	 * interface in neon_test_utils to fetch pages at arbitary LSNs, which
-	 * violates this.
-	 *
-	 * Similarly, the not_modified_since value calculated for a page should
-	 * never move backwards. This assumption is a bit fragile; if we updated
-	 * the last-written cache when we read in a page, for example, then it
-	 * might. But as the code stands, it should not.
-	 *
-	 * (If two backends issue a request at the same time, they might race and
-	 * calculate LSNs "out of order" with each other, but the prefetch queue
-	 * is backend-private at the moment.)
-	 */
-	if (request_lsn < slot->request_lsn || not_modified_since < slot->not_modified_since)
-	{
-		ereport(LOG,
-				(errcode(ERRCODE_IO_ERROR),
-				 errmsg(NEON_TAG "request with unexpected LSN after prefetch"),
-				 errdetail("Request %X/%X not_modified_since %X/%X, prefetch %X/%X not_modified_since %X/%X)",
-						   LSN_FORMAT_ARGS(request_lsn), LSN_FORMAT_ARGS(not_modified_since),
-						   LSN_FORMAT_ARGS(slot->request_lsn), LSN_FORMAT_ARGS(slot->not_modified_since))));
-		return false;
 	}
 
-	/*---
-	 * Each request to the pageserver carries two LSN values:
-	 * `not_modified_since` and `request_lsn`. The (not_modified_since,
-	 * request_lsn] range of each request is effectively a claim that the page
-	 * has not been modified between those LSNs.  If the range of the old
-	 * request in the queue overlaps with the new request, we know that the
-	 * page hasn't been modified in the union of the ranges. We can use the
-	 * response to old request to satisfy the new request in that case. For
-	 * example:
-	 *
-	 *              100      500
-	 * Old request:  +--------+
-	 *
-	 *                     400      800
-	 * New request:         +--------+
-	 *
-	 * The old request claims that the page was not modified between LSNs 100
-	 * and 500, and the second claims that it was not modified between 400 and
-	 * 800. Together they mean that the page was not modified between 100 and
-	 * 800. Therefore the response to the old request is also valid for the
-	 * new request.
-	 *
-	 * This logic also holds at the boundary case that the old request's LSN
-	 * matches the new request's not_modified_since LSN exactly:
-	 *
-	 *              100      500
-	 * Old request:  +--------+
-	 *
-	 *                       500      900
-	 * New request:           +--------+
-	 *
-	 * The response to the old request is the page as it was at LSN 500, and
-	 * the page hasn't been changed in the range (500, 900], therefore the
-	 * response is valid also for the new request.
-	 */
-
-	/* this follows from the checks above */
-	Assert(request_lsn >= slot->not_modified_since);
-
-	return not_modified_since <= slot->request_lsn;
+	return lsn;
 }
 
 /*
@@ -1689,8 +1604,8 @@ neon_exists(SMgrRelation reln, ForkNumber forkNum)
 	bool		exists;
 	NeonResponse *resp;
 	BlockNumber n_blocks;
+	bool		latest;
 	XLogRecPtr	request_lsn;
-	XLogRecPtr	not_modified_since;
 
 	switch (reln->smgr_relpersistence)
 	{
@@ -1745,13 +1660,12 @@ neon_exists(SMgrRelation reln, ForkNumber forkNum)
 		return false;
 	}
 
-	neon_get_request_lsn(InfoFromSMgrRel(reln), forkNum, REL_METADATA_PSEUDO_BLOCKNO,
-						 &request_lsn, &not_modified_since);
+	request_lsn = neon_get_request_lsn(&latest, InfoFromSMgrRel(reln), forkNum, REL_METADATA_PSEUDO_BLOCKNO);
 	{
 		NeonExistsRequest request = {
 			.req.tag = T_NeonExistsRequest,
+			.req.latest = latest,
 			.req.lsn = request_lsn,
-			.req.not_modified_since = not_modified_since,
 			.rinfo = InfoFromSMgrRel(reln),
 		.forknum = forkNum};
 
@@ -2188,10 +2102,10 @@ neon_writeback(SMgrRelation reln, ForkNumber forknum,
 void
 #if PG_MAJORVERSION_NUM < 16
 neon_read_at_lsn(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
-				 XLogRecPtr request_lsn, XLogRecPtr not_modified_since, char *buffer)
+				 XLogRecPtr request_lsn, bool request_latest, char *buffer)
 #else
 neon_read_at_lsn(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
-				 XLogRecPtr request_lsn, XLogRecPtr not_modified_since, void *buffer)
+				 XLogRecPtr request_lsn, bool request_latest, void *buffer)
 #endif
 {
 	NeonResponse *resp;
@@ -2234,16 +2148,15 @@ neon_read_at_lsn(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
 	if (entry != NULL)
 	{
 		slot = entry->slot;
-		if (neon_prefetch_response_usable(request_lsn, not_modified_since, slot))
+		if (slot->effective_request_lsn >= request_lsn)
 		{
 			ring_index = slot->my_ring_index;
 			pgBufferUsage.prefetch.hits += 1;
 		}
-		else
+		else					/* the current prefetch LSN is not large
+								 * enough, so drop the prefetch */
 		{
 			/*
-			 * Cannot use this prefetch, discard it
-			 *
 			 * We can't drop cache for not-yet-received requested items. It is
 			 * unlikely this happens, but it can happen if prefetch distance
 			 * is large enough and a backend didn't consume all prefetch
@@ -2268,8 +2181,8 @@ neon_read_at_lsn(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
 		{
 			pgBufferUsage.prefetch.misses += 1;
 
-			ring_index = prefetch_register_buffer(buftag, &request_lsn,
-												  &not_modified_since);
+			ring_index = prefetch_register_buffer(buftag, &request_latest,
+												  &request_lsn);
 			slot = GetPrfSlot(ring_index);
 		}
 		else
@@ -2333,8 +2246,8 @@ neon_read(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno, char *buffer
 neon_read(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno, void *buffer)
 #endif
 {
+	bool		latest;
 	XLogRecPtr	request_lsn;
-	XLogRecPtr	not_modified_since;
 
 	switch (reln->smgr_relpersistence)
 	{
@@ -2359,9 +2272,8 @@ neon_read(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno, void *buffer
 		return;
 	}
 
-	neon_get_request_lsn(InfoFromSMgrRel(reln), forkNum, blkno,
-						 &request_lsn, &not_modified_since);
-	neon_read_at_lsn(InfoFromSMgrRel(reln), forkNum, blkno, request_lsn, not_modified_since, buffer);
+	request_lsn = neon_get_request_lsn(&latest, InfoFromSMgrRel(reln), forkNum, blkno);
+	neon_read_at_lsn(InfoFromSMgrRel(reln), forkNum, blkno, request_lsn, latest, buffer);
 
 #ifdef DEBUG_COMPARE_LOCAL
 	if (forkNum == MAIN_FORKNUM && IS_LOCAL_REL(reln))
@@ -2530,8 +2442,8 @@ neon_nblocks(SMgrRelation reln, ForkNumber forknum)
 {
 	NeonResponse *resp;
 	BlockNumber n_blocks;
+	bool		latest;
 	XLogRecPtr	request_lsn;
-	XLogRecPtr	not_modified_since;
 
 	switch (reln->smgr_relpersistence)
 	{
@@ -2558,13 +2470,12 @@ neon_nblocks(SMgrRelation reln, ForkNumber forknum)
 		return n_blocks;
 	}
 
-	neon_get_request_lsn(InfoFromSMgrRel(reln), forknum, REL_METADATA_PSEUDO_BLOCKNO,
-						 &request_lsn, &not_modified_since);
+	request_lsn = neon_get_request_lsn(&latest, InfoFromSMgrRel(reln), forknum, REL_METADATA_PSEUDO_BLOCKNO);
 	{
 		NeonNblocksRequest request = {
 			.req.tag = T_NeonNblocksRequest,
+			.req.latest = latest,
 			.req.lsn = request_lsn,
-			.req.not_modified_since = not_modified_since,
 			.rinfo = InfoFromSMgrRel(reln),
 			.forknum = forknum,
 		};
@@ -2612,17 +2523,16 @@ neon_dbsize(Oid dbNode)
 {
 	NeonResponse *resp;
 	int64		db_size;
-	XLogRecPtr	request_lsn,
-				not_modified_since;
+	XLogRecPtr	request_lsn;
+	bool		latest;
 	NRelFileInfo dummy_node = {0};
 
-	neon_get_request_lsn(dummy_node, MAIN_FORKNUM, REL_METADATA_PSEUDO_BLOCKNO,
-						 &request_lsn, &not_modified_since);
+	request_lsn = neon_get_request_lsn(&latest, dummy_node, MAIN_FORKNUM, REL_METADATA_PSEUDO_BLOCKNO);
 	{
 		NeonDbSizeRequest request = {
 			.req.tag = T_NeonDbSizeRequest,
+			.req.latest = latest,
 			.req.lsn = request_lsn,
-			.req.not_modified_since = not_modified_since,
 			.dbNode = dbNode,
 		};
 
@@ -2695,6 +2605,7 @@ neon_truncate(SMgrRelation reln, ForkNumber forknum, BlockNumber nblocks)
 	 * the most recently inserted WAL record's LSN.
 	 */
 	lsn = GetXLogInsertRecPtr();
+
 	lsn = nm_adjust_lsn(lsn);
 
 	/*
@@ -2894,33 +2805,14 @@ neon_end_unlogged_build(SMgrRelation reln)
 static int
 neon_read_slru_segment(SMgrRelation reln, const char* path, int segno, void* buffer)
 {
-	XLogRecPtr request_lsn,
-		not_modified_since;
-
-	if (RecoveryInProgress())
-	{
-		request_lsn = GetXLogReplayRecPtr(NULL);
-		if (request_lsn == InvalidXLogRecPtr)
-		{
-			/*
-			 * This happens in neon startup, we start up without replaying any
-			 * records.
-			 */
-			request_lsn = GetRedoStartLsn();
-		}
-	}
-	else
-		request_lsn = GetXLogInsertRecPtr();
-	request_lsn = nm_adjust_lsn(request_lsn);
-
+	XLogRecPtr request_lsn;
 	/*
-	 * GetRedoStartLsn() returns LSN of basebackup. We know that the SLRU
-	 * segment has not changed since the basebackup, because in order to
-	 * modify it, we would have had to download it already. And once
-	 * downloaded, we never evict SLRU segments from local disk.
+	 * GetRedoStartLsn() returns LSN of basebackup.
+	 * We need to download SLRU segments only once after node startup,
+	 * then SLRUs are maintained locally.
 	 */
-	not_modified_since = GetRedoStartLsn();
-
+	request_lsn = GetRedoStartLsn();
+	request_lsn = nm_adjust_lsn(request_lsn);
 	SlruKind kind;
 
     if (STRPREFIX(path, "pg_xact"))
@@ -2935,8 +2827,8 @@ neon_read_slru_segment(SMgrRelation reln, const char* path, int segno, void* buf
 	NeonResponse *resp;
 	NeonGetSlruSegmentRequest request = {
 		.req.tag = T_NeonGetSlruSegmentRequest,
+		.req.latest = false,
 		.req.lsn = request_lsn,
-		.req.not_modified_since = not_modified_since,
 
 		.kind = kind,
 		.segno = segno
@@ -3064,9 +2956,6 @@ neon_extend_rel_size(NRelFileInfo rinfo, ForkNumber forknum, BlockNumber blkno,
 {
 	BlockNumber relsize;
 
-	/* This is only used in WAL replay */
-	Assert(RecoveryInProgress());
-
 	/* Extend the relation if we know its size */
 	if (get_cached_relsize(rinfo, forknum, &relsize))
 	{
@@ -3085,13 +2974,14 @@ neon_extend_rel_size(NRelFileInfo rinfo, ForkNumber forknum, BlockNumber blkno,
 		 * This length is later reused when we open the smgr to read the
 		 * block, which is fine and expected.
 		 */
+
 		NeonResponse *response;
 		NeonNblocksResponse *nbresponse;
 		NeonNblocksRequest request = {
 			.req = (NeonRequest) {
-				.tag = T_NeonNblocksRequest,
 				.lsn = end_recptr,
-				.not_modified_since = end_recptr,
+				.latest = false,
+				.tag = T_NeonNblocksRequest,
 			},
 			.rinfo = rinfo,
 			.forknum = forknum,

pgxn/neon_test_utils/Makefile

@@ -7,7 +7,7 @@ OBJS = \
 	neontest.o
 
 EXTENSION = neon_test_utils
-DATA = neon_test_utils--1.1.sql
+DATA = neon_test_utils--1.0.sql
 PGFILEDESC = "neon_test_utils - helpers for neon testing and debugging"
 
 PG_CONFIG = pg_config

pgxn/neon_test_utils/neon_test_utils--1.0.sql

@@ -31,12 +31,12 @@ AS 'MODULE_PATHNAME', 'clear_buffer_cache'
 LANGUAGE C STRICT
 PARALLEL UNSAFE;
 
-CREATE FUNCTION get_raw_page_at_lsn(relname text, forkname text, blocknum int8, request_lsn pg_lsn, not_modified_since pg_lsn)
+CREATE FUNCTION get_raw_page_at_lsn(relname text, forkname text, blocknum int8, lsn pg_lsn)
 RETURNS bytea
 AS 'MODULE_PATHNAME', 'get_raw_page_at_lsn'
 LANGUAGE C PARALLEL UNSAFE;
 
-CREATE FUNCTION get_raw_page_at_lsn(tbspc oid, db oid, relfilenode oid, forknum int8, blocknum int8, request_lsn pg_lsn, not_modified_since pg_lsn)
+CREATE FUNCTION get_raw_page_at_lsn(tbspc oid, db oid, relfilenode oid, forknum int8, blocknum int8, lsn pg_lsn)
 RETURNS bytea
 AS 'MODULE_PATHNAME', 'get_raw_page_at_lsn_ex'
 LANGUAGE C PARALLEL UNSAFE;

pgxn/neon_test_utils/neon_test_utils.control

@@ -1,6 +1,6 @@
 # neon_test_utils extension
 comment = 'helpers for neon testing and debugging'
-default_version = '1.1'
+default_version = '1.0'
 module_pathname = '$libdir/neon_test_utils'
 relocatable = true
 trusted = true

pgxn/neon_test_utils/neontest.c

@@ -48,10 +48,10 @@ PG_FUNCTION_INFO_V1(neon_xlogflush);
  */
 #if PG_MAJORVERSION_NUM < 16
 typedef void (*neon_read_at_lsn_type) (NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
-									   XLogRecPtr request_lsn, XLogRecPtr not_modified_since, char *buffer);
+									   XLogRecPtr request_lsn, bool request_latest, char *buffer);
 #else
 typedef void (*neon_read_at_lsn_type) (NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
-									   XLogRecPtr request_lsn, XLogRecPtr not_modified_since, void *buffer);
+									   XLogRecPtr request_lsn, bool request_latest, void *buffer);
 #endif
 
 static neon_read_at_lsn_type neon_read_at_lsn_ptr;
@@ -299,11 +299,8 @@ get_raw_page_at_lsn(PG_FUNCTION_ARGS)
 	text	   *forkname;
 	uint32		blkno;
 
-	XLogRecPtr	request_lsn;
-	XLogRecPtr	not_modified_since;
-
-	if (PG_NARGS() != 5)
-		elog(ERROR, "unexpected number of arguments in SQL function signature");
+	bool		request_latest = PG_ARGISNULL(3);
+	uint64		read_lsn = request_latest ? GetXLogInsertRecPtr() : PG_GETARG_INT64(3);
 
 	if (PG_ARGISNULL(0) || PG_ARGISNULL(1) || PG_ARGISNULL(2))
 		PG_RETURN_NULL();
@@ -312,9 +309,6 @@ get_raw_page_at_lsn(PG_FUNCTION_ARGS)
 	forkname = PG_GETARG_TEXT_PP(1);
 	blkno = PG_GETARG_UINT32(2);
 
-	request_lsn = PG_ARGISNULL(3) ? GetXLogInsertRecPtr() : PG_GETARG_LSN(3);
-	not_modified_since = PG_ARGISNULL(4) ? request_lsn : PG_GETARG_LSN(4);
-
 	if (!superuser())
 		ereport(ERROR,
 				(errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
@@ -367,7 +361,7 @@ get_raw_page_at_lsn(PG_FUNCTION_ARGS)
 	SET_VARSIZE(raw_page, BLCKSZ + VARHDRSZ);
 	raw_page_data = VARDATA(raw_page);
 
-	neon_read_at_lsn(InfoFromRelation(rel), forknum, blkno, request_lsn, not_modified_since, raw_page_data);
+	neon_read_at_lsn(InfoFromRelation(rel), forknum, blkno, read_lsn, request_latest, raw_page_data);
 
 	relation_close(rel, AccessShareLock);
 
@@ -386,9 +380,6 @@ get_raw_page_at_lsn_ex(PG_FUNCTION_ARGS)
 {
 	char	   *raw_page_data;
 
-	if (PG_NARGS() != 7)
-		elog(ERROR, "unexpected number of arguments in SQL function signature");
-
 	if (!superuser())
 		ereport(ERROR,
 				(errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
@@ -412,20 +403,18 @@ get_raw_page_at_lsn_ex(PG_FUNCTION_ARGS)
 		};
 
 		ForkNumber	forknum = PG_GETARG_UINT32(3);
+
 		uint32		blkno = PG_GETARG_UINT32(4);
-		XLogRecPtr	request_lsn;
-		XLogRecPtr	not_modified_since;
+		bool		request_latest = PG_ARGISNULL(5);
+		uint64		read_lsn = request_latest ? GetXLogInsertRecPtr() : PG_GETARG_INT64(5);
 
 		/* Initialize buffer to copy to */
 		bytea	   *raw_page = (bytea *) palloc(BLCKSZ + VARHDRSZ);
 
-		request_lsn = PG_ARGISNULL(5) ? GetXLogInsertRecPtr() : PG_GETARG_LSN(5);
-		not_modified_since = PG_ARGISNULL(6) ? request_lsn : PG_GETARG_LSN(6);
-
 		SET_VARSIZE(raw_page, BLCKSZ + VARHDRSZ);
 		raw_page_data = VARDATA(raw_page);
 
-		neon_read_at_lsn(rinfo, forknum, blkno, request_lsn, not_modified_since, raw_page_data);
+		neon_read_at_lsn(rinfo, forknum, blkno, read_lsn, request_latest, raw_page_data);
 		PG_RETURN_BYTEA_P(raw_page);
 	}
 }

poetry.lock

@@ -2,87 +2,87 @@
 
 [[package]]
 name = "aiohttp"
-version = "3.9.4"
+version = "3.9.2"
 description = "Async http client/server framework (asyncio)"
 optional = false
 python-versions = ">=3.8"
 files = [
-    {file = "aiohttp-3.9.4-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:76d32588ef7e4a3f3adff1956a0ba96faabbdee58f2407c122dd45aa6e34f372"},
-    {file = "aiohttp-3.9.4-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:56181093c10dbc6ceb8a29dfeea1e815e1dfdc020169203d87fd8d37616f73f9"},
-    {file = "aiohttp-3.9.4-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:c7a5b676d3c65e88b3aca41816bf72831898fcd73f0cbb2680e9d88e819d1e4d"},
-    {file = "aiohttp-3.9.4-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:d1df528a85fb404899d4207a8d9934cfd6be626e30e5d3a5544a83dbae6d8a7e"},
-    {file = "aiohttp-3.9.4-cp310-cp310-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:f595db1bceabd71c82e92df212dd9525a8a2c6947d39e3c994c4f27d2fe15b11"},
-    {file = "aiohttp-3.9.4-cp310-cp310-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:9c0b09d76e5a4caac3d27752027fbd43dc987b95f3748fad2b924a03fe8632ad"},
-    {file = "aiohttp-3.9.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:689eb4356649ec9535b3686200b231876fb4cab4aca54e3bece71d37f50c1d13"},
-    {file = "aiohttp-3.9.4-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:a3666cf4182efdb44d73602379a66f5fdfd5da0db5e4520f0ac0dcca644a3497"},
-    {file = "aiohttp-3.9.4-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:b65b0f8747b013570eea2f75726046fa54fa8e0c5db60f3b98dd5d161052004a"},
-    {file = "aiohttp-3.9.4-cp310-cp310-musllinux_1_1_i686.whl", hash = "sha256:a1885d2470955f70dfdd33a02e1749613c5a9c5ab855f6db38e0b9389453dce7"},
-    {file = "aiohttp-3.9.4-cp310-cp310-musllinux_1_1_ppc64le.whl", hash = "sha256:0593822dcdb9483d41f12041ff7c90d4d1033ec0e880bcfaf102919b715f47f1"},
-    {file = "aiohttp-3.9.4-cp310-cp310-musllinux_1_1_s390x.whl", hash = "sha256:47f6eb74e1ecb5e19a78f4a4228aa24df7fbab3b62d4a625d3f41194a08bd54f"},
-    {file = "aiohttp-3.9.4-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:c8b04a3dbd54de6ccb7604242fe3ad67f2f3ca558f2d33fe19d4b08d90701a89"},
-    {file = "aiohttp-3.9.4-cp310-cp310-win32.whl", hash = "sha256:8a78dfb198a328bfb38e4308ca8167028920fb747ddcf086ce706fbdd23b2926"},
-    {file = "aiohttp-3.9.4-cp310-cp310-win_amd64.whl", hash = "sha256:e78da6b55275987cbc89141a1d8e75f5070e577c482dd48bd9123a76a96f0bbb"},
-    {file = "aiohttp-3.9.4-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:c111b3c69060d2bafc446917534150fd049e7aedd6cbf21ba526a5a97b4402a5"},
-    {file = "aiohttp-3.9.4-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:efbdd51872cf170093998c87ccdf3cb5993add3559341a8e5708bcb311934c94"},
-    {file = "aiohttp-3.9.4-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:7bfdb41dc6e85d8535b00d73947548a748e9534e8e4fddd2638109ff3fb081df"},
-    {file = "aiohttp-3.9.4-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:2bd9d334412961125e9f68d5b73c1d0ab9ea3f74a58a475e6b119f5293eee7ba"},
-    {file = "aiohttp-3.9.4-cp311-cp311-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:35d78076736f4a668d57ade00c65d30a8ce28719d8a42471b2a06ccd1a2e3063"},
-    {file = "aiohttp-3.9.4-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:824dff4f9f4d0f59d0fa3577932ee9a20e09edec8a2f813e1d6b9f89ced8293f"},
-    {file = "aiohttp-3.9.4-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:52b8b4e06fc15519019e128abedaeb56412b106ab88b3c452188ca47a25c4093"},
-    {file = "aiohttp-3.9.4-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:eae569fb1e7559d4f3919965617bb39f9e753967fae55ce13454bec2d1c54f09"},
-    {file = "aiohttp-3.9.4-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:69b97aa5792428f321f72aeb2f118e56893371f27e0b7d05750bcad06fc42ca1"},
-    {file = "aiohttp-3.9.4-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:4d79aad0ad4b980663316f26d9a492e8fab2af77c69c0f33780a56843ad2f89e"},
-    {file = "aiohttp-3.9.4-cp311-cp311-musllinux_1_1_ppc64le.whl", hash = "sha256:d6577140cd7db19e430661e4b2653680194ea8c22c994bc65b7a19d8ec834403"},
-    {file = "aiohttp-3.9.4-cp311-cp311-musllinux_1_1_s390x.whl", hash = "sha256:9860d455847cd98eb67897f5957b7cd69fbcb436dd3f06099230f16a66e66f79"},
-    {file = "aiohttp-3.9.4-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:69ff36d3f8f5652994e08bd22f093e11cfd0444cea310f92e01b45a4e46b624e"},
-    {file = "aiohttp-3.9.4-cp311-cp311-win32.whl", hash = "sha256:e27d3b5ed2c2013bce66ad67ee57cbf614288bda8cdf426c8d8fe548316f1b5f"},
-    {file = "aiohttp-3.9.4-cp311-cp311-win_amd64.whl", hash = "sha256:d6a67e26daa686a6fbdb600a9af8619c80a332556245fa8e86c747d226ab1a1e"},
-    {file = "aiohttp-3.9.4-cp312-cp312-macosx_10_9_universal2.whl", hash = "sha256:c5ff8ff44825736a4065d8544b43b43ee4c6dd1530f3a08e6c0578a813b0aa35"},
-    {file = "aiohttp-3.9.4-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:d12a244627eba4e9dc52cbf924edef905ddd6cafc6513849b4876076a6f38b0e"},
-    {file = "aiohttp-3.9.4-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:dcad56c8d8348e7e468899d2fb3b309b9bc59d94e6db08710555f7436156097f"},
-    {file = "aiohttp-3.9.4-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:4f7e69a7fd4b5ce419238388e55abd220336bd32212c673ceabc57ccf3d05b55"},
-    {file = "aiohttp-3.9.4-cp312-cp312-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:c4870cb049f10d7680c239b55428916d84158798eb8f353e74fa2c98980dcc0b"},
-    {file = "aiohttp-3.9.4-cp312-cp312-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:3b2feaf1b7031ede1bc0880cec4b0776fd347259a723d625357bb4b82f62687b"},
-    {file = "aiohttp-3.9.4-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:939393e8c3f0a5bcd33ef7ace67680c318dc2ae406f15e381c0054dd658397de"},
-    {file = "aiohttp-3.9.4-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:7d2334e387b2adcc944680bebcf412743f2caf4eeebd550f67249c1c3696be04"},
-    {file = "aiohttp-3.9.4-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:e0198ea897680e480845ec0ffc5a14e8b694e25b3f104f63676d55bf76a82f1a"},
-    {file = "aiohttp-3.9.4-cp312-cp312-musllinux_1_1_i686.whl", hash = "sha256:e40d2cd22914d67c84824045861a5bb0fb46586b15dfe4f046c7495bf08306b2"},
-    {file = "aiohttp-3.9.4-cp312-cp312-musllinux_1_1_ppc64le.whl", hash = "sha256:aba80e77c227f4234aa34a5ff2b6ff30c5d6a827a91d22ff6b999de9175d71bd"},
-    {file = "aiohttp-3.9.4-cp312-cp312-musllinux_1_1_s390x.whl", hash = "sha256:fb68dc73bc8ac322d2e392a59a9e396c4f35cb6fdbdd749e139d1d6c985f2527"},
-    {file = "aiohttp-3.9.4-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:f3460a92638dce7e47062cf088d6e7663adb135e936cb117be88d5e6c48c9d53"},
-    {file = "aiohttp-3.9.4-cp312-cp312-win32.whl", hash = "sha256:32dc814ddbb254f6170bca198fe307920f6c1308a5492f049f7f63554b88ef36"},
-    {file = "aiohttp-3.9.4-cp312-cp312-win_amd64.whl", hash = "sha256:63f41a909d182d2b78fe3abef557fcc14da50c7852f70ae3be60e83ff64edba5"},
-    {file = "aiohttp-3.9.4-cp38-cp38-macosx_10_9_universal2.whl", hash = "sha256:c3770365675f6be220032f6609a8fbad994d6dcf3ef7dbcf295c7ee70884c9af"},
-    {file = "aiohttp-3.9.4-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:305edae1dea368ce09bcb858cf5a63a064f3bff4767dec6fa60a0cc0e805a1d3"},
-    {file = "aiohttp-3.9.4-cp38-cp38-macosx_11_0_arm64.whl", hash = "sha256:6f121900131d116e4a93b55ab0d12ad72573f967b100e49086e496a9b24523ea"},
-    {file = "aiohttp-3.9.4-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:b71e614c1ae35c3d62a293b19eface83d5e4d194e3eb2fabb10059d33e6e8cbf"},
-    {file = "aiohttp-3.9.4-cp38-cp38-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:419f009fa4cfde4d16a7fc070d64f36d70a8d35a90d71aa27670bba2be4fd039"},
-    {file = "aiohttp-3.9.4-cp38-cp38-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:7b39476ee69cfe64061fd77a73bf692c40021f8547cda617a3466530ef63f947"},
-    {file = "aiohttp-3.9.4-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:b33f34c9c7decdb2ab99c74be6443942b730b56d9c5ee48fb7df2c86492f293c"},
-    {file = "aiohttp-3.9.4-cp38-cp38-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:c78700130ce2dcebb1a8103202ae795be2fa8c9351d0dd22338fe3dac74847d9"},
-    {file = "aiohttp-3.9.4-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:268ba22d917655d1259af2d5659072b7dc11b4e1dc2cb9662fdd867d75afc6a4"},
-    {file = "aiohttp-3.9.4-cp38-cp38-musllinux_1_1_i686.whl", hash = "sha256:17e7c051f53a0d2ebf33013a9cbf020bb4e098c4bc5bce6f7b0c962108d97eab"},
-    {file = "aiohttp-3.9.4-cp38-cp38-musllinux_1_1_ppc64le.whl", hash = "sha256:7be99f4abb008cb38e144f85f515598f4c2c8932bf11b65add0ff59c9c876d99"},
-    {file = "aiohttp-3.9.4-cp38-cp38-musllinux_1_1_s390x.whl", hash = "sha256:d58a54d6ff08d2547656356eea8572b224e6f9bbc0cf55fa9966bcaac4ddfb10"},
-    {file = "aiohttp-3.9.4-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:7673a76772bda15d0d10d1aa881b7911d0580c980dbd16e59d7ba1422b2d83cd"},
-    {file = "aiohttp-3.9.4-cp38-cp38-win32.whl", hash = "sha256:e4370dda04dc8951012f30e1ce7956a0a226ac0714a7b6c389fb2f43f22a250e"},
-    {file = "aiohttp-3.9.4-cp38-cp38-win_amd64.whl", hash = "sha256:eb30c4510a691bb87081192a394fb661860e75ca3896c01c6d186febe7c88530"},
-    {file = "aiohttp-3.9.4-cp39-cp39-macosx_10_9_universal2.whl", hash = "sha256:84e90494db7df3be5e056f91412f9fa9e611fbe8ce4aaef70647297f5943b276"},
-    {file = "aiohttp-3.9.4-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:7d4845f8501ab28ebfdbeab980a50a273b415cf69e96e4e674d43d86a464df9d"},
-    {file = "aiohttp-3.9.4-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:69046cd9a2a17245c4ce3c1f1a4ff8c70c7701ef222fce3d1d8435f09042bba1"},
-    {file = "aiohttp-3.9.4-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:8b73a06bafc8dcc508420db43b4dd5850e41e69de99009d0351c4f3007960019"},
-    {file = "aiohttp-3.9.4-cp39-cp39-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:418bb0038dfafeac923823c2e63226179976c76f981a2aaad0ad5d51f2229bca"},
-    {file = "aiohttp-3.9.4-cp39-cp39-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:71a8f241456b6c2668374d5d28398f8e8cdae4cce568aaea54e0f39359cd928d"},
-    {file = "aiohttp-3.9.4-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:935c369bf8acc2dc26f6eeb5222768aa7c62917c3554f7215f2ead7386b33748"},
-    {file = "aiohttp-3.9.4-cp39-cp39-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:74e4e48c8752d14ecfb36d2ebb3d76d614320570e14de0a3aa7a726ff150a03c"},
-    {file = "aiohttp-3.9.4-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:916b0417aeddf2c8c61291238ce25286f391a6acb6f28005dd9ce282bd6311b6"},
-    {file = "aiohttp-3.9.4-cp39-cp39-musllinux_1_1_i686.whl", hash = "sha256:9b6787b6d0b3518b2ee4cbeadd24a507756ee703adbac1ab6dc7c4434b8c572a"},
-    {file = "aiohttp-3.9.4-cp39-cp39-musllinux_1_1_ppc64le.whl", hash = "sha256:221204dbda5ef350e8db6287937621cf75e85778b296c9c52260b522231940ed"},
-    {file = "aiohttp-3.9.4-cp39-cp39-musllinux_1_1_s390x.whl", hash = "sha256:10afd99b8251022ddf81eaed1d90f5a988e349ee7d779eb429fb07b670751e8c"},
-    {file = "aiohttp-3.9.4-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:2506d9f7a9b91033201be9ffe7d89c6a54150b0578803cce5cb84a943d075bc3"},
-    {file = "aiohttp-3.9.4-cp39-cp39-win32.whl", hash = "sha256:e571fdd9efd65e86c6af2f332e0e95dad259bfe6beb5d15b3c3eca3a6eb5d87b"},
-    {file = "aiohttp-3.9.4-cp39-cp39-win_amd64.whl", hash = "sha256:7d29dd5319d20aa3b7749719ac9685fbd926f71ac8c77b2477272725f882072d"},
-    {file = "aiohttp-3.9.4.tar.gz", hash = "sha256:6ff71ede6d9a5a58cfb7b6fffc83ab5d4a63138276c771ac91ceaaddf5459644"},
+    {file = "aiohttp-3.9.2-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:772fbe371788e61c58d6d3d904268e48a594ba866804d08c995ad71b144f94cb"},
+    {file = "aiohttp-3.9.2-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:edd4f1af2253f227ae311ab3d403d0c506c9b4410c7fc8d9573dec6d9740369f"},
+    {file = "aiohttp-3.9.2-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:cfee9287778399fdef6f8a11c9e425e1cb13cc9920fd3a3df8f122500978292b"},
+    {file = "aiohttp-3.9.2-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:3cc158466f6a980a6095ee55174d1de5730ad7dec251be655d9a6a9dd7ea1ff9"},
+    {file = "aiohttp-3.9.2-cp310-cp310-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:54ec82f45d57c9a65a1ead3953b51c704f9587440e6682f689da97f3e8defa35"},
+    {file = "aiohttp-3.9.2-cp310-cp310-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:abeb813a18eb387f0d835ef51f88568540ad0325807a77a6e501fed4610f864e"},
+    {file = "aiohttp-3.9.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:cc91d07280d7d169f3a0f9179d8babd0ee05c79d4d891447629ff0d7d8089ec2"},
+    {file = "aiohttp-3.9.2-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:b65e861f4bebfb660f7f0f40fa3eb9f2ab9af10647d05dac824390e7af8f75b7"},
+    {file = "aiohttp-3.9.2-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:04fd8ffd2be73d42bcf55fd78cde7958eeee6d4d8f73c3846b7cba491ecdb570"},
+    {file = "aiohttp-3.9.2-cp310-cp310-musllinux_1_1_i686.whl", hash = "sha256:3d8d962b439a859b3ded9a1e111a4615357b01620a546bc601f25b0211f2da81"},
+    {file = "aiohttp-3.9.2-cp310-cp310-musllinux_1_1_ppc64le.whl", hash = "sha256:8ceb658afd12b27552597cf9a65d9807d58aef45adbb58616cdd5ad4c258c39e"},
+    {file = "aiohttp-3.9.2-cp310-cp310-musllinux_1_1_s390x.whl", hash = "sha256:0e4ee4df741670560b1bc393672035418bf9063718fee05e1796bf867e995fad"},
+    {file = "aiohttp-3.9.2-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:2dec87a556f300d3211decf018bfd263424f0690fcca00de94a837949fbcea02"},
+    {file = "aiohttp-3.9.2-cp310-cp310-win32.whl", hash = "sha256:3e1a800f988ce7c4917f34096f81585a73dbf65b5c39618b37926b1238cf9bc4"},
+    {file = "aiohttp-3.9.2-cp310-cp310-win_amd64.whl", hash = "sha256:ea510718a41b95c236c992b89fdfc3d04cc7ca60281f93aaada497c2b4e05c46"},
+    {file = "aiohttp-3.9.2-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:6aaa6f99256dd1b5756a50891a20f0d252bd7bdb0854c5d440edab4495c9f973"},
+    {file = "aiohttp-3.9.2-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:a27d8c70ad87bcfce2e97488652075a9bdd5b70093f50b10ae051dfe5e6baf37"},
+    {file = "aiohttp-3.9.2-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:54287bcb74d21715ac8382e9de146d9442b5f133d9babb7e5d9e453faadd005e"},
+    {file = "aiohttp-3.9.2-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:5bb3d05569aa83011fcb346b5266e00b04180105fcacc63743fc2e4a1862a891"},
+    {file = "aiohttp-3.9.2-cp311-cp311-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:c8534e7d69bb8e8d134fe2be9890d1b863518582f30c9874ed7ed12e48abe3c4"},
+    {file = "aiohttp-3.9.2-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:4bd9d5b989d57b41e4ff56ab250c5ddf259f32db17159cce630fd543376bd96b"},
+    {file = "aiohttp-3.9.2-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:fa6904088e6642609981f919ba775838ebf7df7fe64998b1a954fb411ffb4663"},
+    {file = "aiohttp-3.9.2-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:bda42eb410be91b349fb4ee3a23a30ee301c391e503996a638d05659d76ea4c2"},
+    {file = "aiohttp-3.9.2-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:193cc1ccd69d819562cc7f345c815a6fc51d223b2ef22f23c1a0f67a88de9a72"},
+    {file = "aiohttp-3.9.2-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:b9f1cb839b621f84a5b006848e336cf1496688059d2408e617af33e3470ba204"},
+    {file = "aiohttp-3.9.2-cp311-cp311-musllinux_1_1_ppc64le.whl", hash = "sha256:d22a0931848b8c7a023c695fa2057c6aaac19085f257d48baa24455e67df97ec"},
+    {file = "aiohttp-3.9.2-cp311-cp311-musllinux_1_1_s390x.whl", hash = "sha256:4112d8ba61fbd0abd5d43a9cb312214565b446d926e282a6d7da3f5a5aa71d36"},
+    {file = "aiohttp-3.9.2-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:c4ad4241b52bb2eb7a4d2bde060d31c2b255b8c6597dd8deac2f039168d14fd7"},
+    {file = "aiohttp-3.9.2-cp311-cp311-win32.whl", hash = "sha256:ee2661a3f5b529f4fc8a8ffee9f736ae054adfb353a0d2f78218be90617194b3"},
+    {file = "aiohttp-3.9.2-cp311-cp311-win_amd64.whl", hash = "sha256:4deae2c165a5db1ed97df2868ef31ca3cc999988812e82386d22937d9d6fed52"},
+    {file = "aiohttp-3.9.2-cp312-cp312-macosx_10_9_universal2.whl", hash = "sha256:6f4cdba12539215aaecf3c310ce9d067b0081a0795dd8a8805fdb67a65c0572a"},
+    {file = "aiohttp-3.9.2-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:84e843b33d5460a5c501c05539809ff3aee07436296ff9fbc4d327e32aa3a326"},
+    {file = "aiohttp-3.9.2-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:8008d0f451d66140a5aa1c17e3eedc9d56e14207568cd42072c9d6b92bf19b52"},
+    {file = "aiohttp-3.9.2-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:61c47ab8ef629793c086378b1df93d18438612d3ed60dca76c3422f4fbafa792"},
+    {file = "aiohttp-3.9.2-cp312-cp312-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:bc71f748e12284312f140eaa6599a520389273174b42c345d13c7e07792f4f57"},
+    {file = "aiohttp-3.9.2-cp312-cp312-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:a1c3a4d0ab2f75f22ec80bca62385db2e8810ee12efa8c9e92efea45c1849133"},
+    {file = "aiohttp-3.9.2-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:9a87aa0b13bbee025faa59fa58861303c2b064b9855d4c0e45ec70182bbeba1b"},
+    {file = "aiohttp-3.9.2-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:e2cc0d04688b9f4a7854c56c18aa7af9e5b0a87a28f934e2e596ba7e14783192"},
+    {file = "aiohttp-3.9.2-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:1956e3ac376b1711c1533266dec4efd485f821d84c13ce1217d53e42c9e65f08"},
+    {file = "aiohttp-3.9.2-cp312-cp312-musllinux_1_1_i686.whl", hash = "sha256:114da29f39eccd71b93a0fcacff178749a5c3559009b4a4498c2c173a6d74dff"},
+    {file = "aiohttp-3.9.2-cp312-cp312-musllinux_1_1_ppc64le.whl", hash = "sha256:3f17999ae3927d8a9a823a1283b201344a0627272f92d4f3e3a4efe276972fe8"},
+    {file = "aiohttp-3.9.2-cp312-cp312-musllinux_1_1_s390x.whl", hash = "sha256:f31df6a32217a34ae2f813b152a6f348154f948c83213b690e59d9e84020925c"},
+    {file = "aiohttp-3.9.2-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:7a75307ffe31329928a8d47eae0692192327c599113d41b278d4c12b54e1bd11"},
+    {file = "aiohttp-3.9.2-cp312-cp312-win32.whl", hash = "sha256:972b63d589ff8f305463593050a31b5ce91638918da38139b9d8deaba9e0fed7"},
+    {file = "aiohttp-3.9.2-cp312-cp312-win_amd64.whl", hash = "sha256:200dc0246f0cb5405c80d18ac905c8350179c063ea1587580e3335bfc243ba6a"},
+    {file = "aiohttp-3.9.2-cp38-cp38-macosx_10_9_universal2.whl", hash = "sha256:158564d0d1020e0d3fe919a81d97aadad35171e13e7b425b244ad4337fc6793a"},
+    {file = "aiohttp-3.9.2-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:da1346cd0ccb395f0ed16b113ebb626fa43b7b07fd7344fce33e7a4f04a8897a"},
+    {file = "aiohttp-3.9.2-cp38-cp38-macosx_11_0_arm64.whl", hash = "sha256:eaa9256de26ea0334ffa25f1913ae15a51e35c529a1ed9af8e6286dd44312554"},
+    {file = "aiohttp-3.9.2-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:1543e7fb00214fb4ccead42e6a7d86f3bb7c34751ec7c605cca7388e525fd0b4"},
+    {file = "aiohttp-3.9.2-cp38-cp38-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:186e94570433a004e05f31f632726ae0f2c9dee4762a9ce915769ce9c0a23d89"},
+    {file = "aiohttp-3.9.2-cp38-cp38-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:d52d20832ac1560f4510d68e7ba8befbc801a2b77df12bd0cd2bcf3b049e52a4"},
+    {file = "aiohttp-3.9.2-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:1c45e4e815ac6af3b72ca2bde9b608d2571737bb1e2d42299fc1ffdf60f6f9a1"},
+    {file = "aiohttp-3.9.2-cp38-cp38-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:aa906b9bdfd4a7972dd0628dbbd6413d2062df5b431194486a78f0d2ae87bd55"},
+    {file = "aiohttp-3.9.2-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:68bbee9e17d66f17bb0010aa15a22c6eb28583edcc8b3212e2b8e3f77f3ebe2a"},
+    {file = "aiohttp-3.9.2-cp38-cp38-musllinux_1_1_i686.whl", hash = "sha256:4c189b64bd6d9a403a1a3f86a3ab3acbc3dc41a68f73a268a4f683f89a4dec1f"},
+    {file = "aiohttp-3.9.2-cp38-cp38-musllinux_1_1_ppc64le.whl", hash = "sha256:8a7876f794523123bca6d44bfecd89c9fec9ec897a25f3dd202ee7fc5c6525b7"},
+    {file = "aiohttp-3.9.2-cp38-cp38-musllinux_1_1_s390x.whl", hash = "sha256:d23fba734e3dd7b1d679b9473129cd52e4ec0e65a4512b488981a56420e708db"},
+    {file = "aiohttp-3.9.2-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:b141753be581fab842a25cb319f79536d19c2a51995d7d8b29ee290169868eab"},
+    {file = "aiohttp-3.9.2-cp38-cp38-win32.whl", hash = "sha256:103daf41ff3b53ba6fa09ad410793e2e76c9d0269151812e5aba4b9dd674a7e8"},
+    {file = "aiohttp-3.9.2-cp38-cp38-win_amd64.whl", hash = "sha256:328918a6c2835861ff7afa8c6d2c70c35fdaf996205d5932351bdd952f33fa2f"},
+    {file = "aiohttp-3.9.2-cp39-cp39-macosx_10_9_universal2.whl", hash = "sha256:5264d7327c9464786f74e4ec9342afbbb6ee70dfbb2ec9e3dfce7a54c8043aa3"},
+    {file = "aiohttp-3.9.2-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:07205ae0015e05c78b3288c1517afa000823a678a41594b3fdc870878d645305"},
+    {file = "aiohttp-3.9.2-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:ae0a1e638cffc3ec4d4784b8b4fd1cf28968febc4bd2718ffa25b99b96a741bd"},
+    {file = "aiohttp-3.9.2-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:d43302a30ba1166325974858e6ef31727a23bdd12db40e725bec0f759abce505"},
+    {file = "aiohttp-3.9.2-cp39-cp39-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:16a967685907003765855999af11a79b24e70b34dc710f77a38d21cd9fc4f5fe"},
+    {file = "aiohttp-3.9.2-cp39-cp39-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:6fa3ee92cd441d5c2d07ca88d7a9cef50f7ec975f0117cd0c62018022a184308"},
+    {file = "aiohttp-3.9.2-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:0b500c5ad9c07639d48615a770f49618130e61be36608fc9bc2d9bae31732b8f"},
+    {file = "aiohttp-3.9.2-cp39-cp39-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:c07327b368745b1ce2393ae9e1aafed7073d9199e1dcba14e035cc646c7941bf"},
+    {file = "aiohttp-3.9.2-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:cc7d6502c23a0ec109687bf31909b3fb7b196faf198f8cff68c81b49eb316ea9"},
+    {file = "aiohttp-3.9.2-cp39-cp39-musllinux_1_1_i686.whl", hash = "sha256:07be2be7071723c3509ab5c08108d3a74f2181d4964e869f2504aaab68f8d3e8"},
+    {file = "aiohttp-3.9.2-cp39-cp39-musllinux_1_1_ppc64le.whl", hash = "sha256:122468f6fee5fcbe67cb07014a08c195b3d4c41ff71e7b5160a7bcc41d585a5f"},
+    {file = "aiohttp-3.9.2-cp39-cp39-musllinux_1_1_s390x.whl", hash = "sha256:00a9abcea793c81e7f8778ca195a1714a64f6d7436c4c0bb168ad2a212627000"},
+    {file = "aiohttp-3.9.2-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:7a9825fdd64ecac5c670234d80bb52bdcaa4139d1f839165f548208b3779c6c6"},
+    {file = "aiohttp-3.9.2-cp39-cp39-win32.whl", hash = "sha256:5422cd9a4a00f24c7244e1b15aa9b87935c85fb6a00c8ac9b2527b38627a9211"},
+    {file = "aiohttp-3.9.2-cp39-cp39-win_amd64.whl", hash = "sha256:7d579dcd5d82a86a46f725458418458fa43686f6a7b252f2966d359033ffc8ab"},
+    {file = "aiohttp-3.9.2.tar.gz", hash = "sha256:b0ad0a5e86ce73f5368a164c10ada10504bf91869c05ab75d982c6048217fbf7"},
 ]
 
 [package.dependencies]
@@ -2900,4 +2900,4 @@ cffi = ["cffi (>=1.11)"]
 [metadata]
 lock-version = "2.0"
 python-versions = "^3.9"
-content-hash = "b3452b50901123fd5f2c385ce8a0c1c492296393b8a7926a322b6df0ea3ac572"
+content-hash = "df7161da4fdc3cba0a445176fc9dda2a0e8a53e13a7aa8a864385ca259381b41"

proxy/src/bin/pg_sni_router.rs

@@ -279,7 +279,7 @@ async fn handle_client(
 
     // doesn't yet matter as pg-sni-router doesn't report analytics logs
     ctx.set_success();
-    ctx.log_connect();
+    ctx.log();
 
     // Starting from here we only proxy the client's traffic.
     info!("performing the proxy pass...");

proxy/src/bin/proxy.rs

@@ -200,12 +200,6 @@ struct ProxyCliArgs {
     /// Size of each event is no more than 400 bytes, so 2**22 is about 200MB before the compression.
     #[clap(long, default_value = "4194304")]
     metric_backup_collection_chunk_size: usize,
-    /// Whether to retry the connection to the compute node
-    #[clap(long, default_value = config::RetryConfig::CONNECT_TO_COMPUTE_DEFAULT_VALUES)]
-    connect_to_compute_retry: String,
-    /// Whether to retry the wake_compute request
-    #[clap(long, default_value = config::RetryConfig::WAKE_COMPUTE_DEFAULT_VALUES)]
-    wake_compute_retry: String,
 }
 
 #[derive(clap::Args, Clone, Copy, Debug)]
@@ -337,6 +331,7 @@ async fn main() -> anyhow::Result<()> {
     let proxy_listener = TcpListener::bind(proxy_address).await?;
     let cancellation_token = CancellationToken::new();
 
+    let endpoint_rate_limiter = Arc::new(EndpointRateLimiter::new(&config.endpoint_rps_limit));
     let cancel_map = CancelMap::default();
 
     let redis_publisher = match &regional_redis_client {
@@ -362,6 +357,7 @@ async fn main() -> anyhow::Result<()> {
         config,
         proxy_listener,
         cancellation_token.clone(),
+        endpoint_rate_limiter.clone(),
         cancellation_handler.clone(),
     ));
 
@@ -376,6 +372,7 @@ async fn main() -> anyhow::Result<()> {
             config,
             serverless_listener,
             cancellation_token.clone(),
+            endpoint_rate_limiter.clone(),
             cancellation_handler.clone(),
         ));
     }
@@ -403,43 +400,27 @@ async fn main() -> anyhow::Result<()> {
         maintenance_tasks.spawn(usage_metrics::task_main(metrics_config));
         client_tasks.spawn(usage_metrics::task_backup(
             &metrics_config.backup_metric_collection_config,
-            cancellation_token.clone(),
+            cancellation_token,
         ));
     }
 
     if let auth::BackendType::Console(api, _) = &config.auth_backend {
         if let proxy::console::provider::ConsoleBackend::Console(api) = &**api {
-            match (redis_notifications_client, regional_redis_client.clone()) {
-                (None, None) => {}
-                (client1, client2) => {
-                    let cache = api.caches.project_info.clone();
-                    if let Some(client) = client1 {
-                        maintenance_tasks.spawn(notifications::task_main(
-                            client,
-                            cache.clone(),
-                            cancel_map.clone(),
-                            args.region.clone(),
-                        ));
-                    }
-                    if let Some(client) = client2 {
-                        maintenance_tasks.spawn(notifications::task_main(
-                            client,
-                            cache.clone(),
-                            cancel_map.clone(),
-                            args.region.clone(),
-                        ));
-                    }
-                    maintenance_tasks.spawn(async move { cache.clone().gc_worker().await });
-                }
+            if let Some(redis_notifications_client) = redis_notifications_client {
+                let cache = api.caches.project_info.clone();
+                maintenance_tasks.spawn(notifications::task_main(
+                    redis_notifications_client,
+                    cache.clone(),
+                    cancel_map.clone(),
+                    args.region.clone(),
+                ));
+                maintenance_tasks.spawn(async move { cache.clone().gc_worker().await });
             }
             if let Some(regional_redis_client) = regional_redis_client {
                 let cache = api.caches.endpoints_cache.clone();
                 let con = regional_redis_client;
                 let span = tracing::info_span!("endpoints_cache");
-                maintenance_tasks.spawn(
-                    async move { cache.do_read(con, cancellation_token.clone()).await }
-                        .instrument(span),
-                );
+                maintenance_tasks.spawn(async move { cache.do_read(con).await }.instrument(span));
             }
         }
     }
@@ -552,11 +533,7 @@ fn build_config(args: &ProxyCliArgs) -> anyhow::Result<&'static ProxyConfig> {
             let url = args.auth_endpoint.parse()?;
             let endpoint = http::Endpoint::new(url, http::new_client());
 
-            let mut endpoint_rps_limit = args.endpoint_rps_limit.clone();
-            RateBucketInfo::validate(&mut endpoint_rps_limit)?;
-            let endpoint_rate_limiter = Arc::new(EndpointRateLimiter::new(endpoint_rps_limit));
-            let api =
-                console::provider::neon::Api::new(endpoint, caches, locks, endpoint_rate_limiter);
+            let api = console::provider::neon::Api::new(endpoint, caches, locks);
             let api = console::provider::ConsoleBackend::Console(api);
             auth::BackendType::Console(MaybeOwned::Owned(api), ())
         }
@@ -590,6 +567,8 @@ fn build_config(args: &ProxyCliArgs) -> anyhow::Result<&'static ProxyConfig> {
         rate_limit_ip_subnet: args.auth_rate_limit_ip_subnet,
     };
 
+    let mut endpoint_rps_limit = args.endpoint_rps_limit.clone();
+    RateBucketInfo::validate(&mut endpoint_rps_limit)?;
     let mut redis_rps_limit = args.redis_rps_limit.clone();
     RateBucketInfo::validate(&mut redis_rps_limit)?;
 
@@ -602,14 +581,11 @@ fn build_config(args: &ProxyCliArgs) -> anyhow::Result<&'static ProxyConfig> {
         authentication_config,
         require_client_ip: args.require_client_ip,
         disable_ip_check_for_http: args.disable_ip_check_for_http,
+        endpoint_rps_limit,
         redis_rps_limit,
         handshake_timeout: args.handshake_timeout,
         region: args.region.clone(),
         aws_region: args.aws_region.clone(),
-        wake_compute_retry_config: config::RetryConfig::parse(&args.wake_compute_retry)?,
-        connect_to_compute_retry_config: config::RetryConfig::parse(
-            &args.connect_to_compute_retry,
-        )?,
     }));
 
     Ok(config)

proxy/src/cache/endpoints.rs

@@ -4,7 +4,6 @@ use std::{
         atomic::{AtomicBool, Ordering},
         Arc,
     },
-    time::Duration,
 };
 
 use dashmap::DashSet;
@@ -14,7 +13,6 @@ use redis::{
 };
 use serde::Deserialize;
 use tokio::sync::Mutex;
-use tokio_util::sync::CancellationToken;
 use tracing::info;
 
 use crate::{
@@ -72,14 +70,20 @@ impl EndpointsCache {
         if !self.ready.load(Ordering::Acquire) {
             return true;
         }
-        let rejected = self.should_reject(endpoint);
-        ctx.set_rejected(rejected);
-        info!(?rejected, "check endpoint is valid, disabled cache");
-        // If cache is disabled, just collect the metrics and return or
-        // If the limiter allows, we don't need to check the cache.
-        if self.config.disable_cache || self.limiter.lock().await.check() {
+        // If cache is disabled, just collect the metrics and return.
+        if self.config.disable_cache {
+            let rejected = self.should_reject(endpoint);
+            ctx.set_rejected(rejected);
+            info!(?rejected, "check endpoint is valid, disabled cache");
             return true;
         }
+        // If the limiter allows, we don't need to check the cache.
+        if self.limiter.lock().await.check() {
+            return true;
+        }
+        let rejected = self.should_reject(endpoint);
+        info!(?rejected, "check endpoint is valid, enabled cache");
+        ctx.set_rejected(rejected);
         !rejected
     }
     fn should_reject(&self, endpoint: &EndpointId) -> bool {
@@ -113,22 +117,16 @@ impl EndpointsCache {
     pub async fn do_read(
         &self,
         mut con: ConnectionWithCredentialsProvider,
-        cancellation_token: CancellationToken,
     ) -> anyhow::Result<Infallible> {
         let mut last_id = "0-0".to_string();
         loop {
+            self.ready.store(false, Ordering::Release);
             if let Err(e) = con.connect().await {
                 tracing::error!("error connecting to redis: {:?}", e);
-                self.ready.store(false, Ordering::Release);
+                continue;
             }
             if let Err(e) = self.read_from_stream(&mut con, &mut last_id).await {
                 tracing::error!("error reading from redis: {:?}", e);
-                self.ready.store(false, Ordering::Release);
-            }
-            if cancellation_token.is_cancelled() {
-                info!("cancellation token is cancelled, exiting");
-                tokio::time::sleep(Duration::from_secs(60 * 60 * 24 * 7)).await;
-                // 1 week.
             }
             tokio::time::sleep(self.config.retry_interval).await;
         }

proxy/src/compute.rs

@@ -260,9 +260,7 @@ impl ConnCfg {
         aux: MetricsAuxInfo,
         timeout: Duration,
     ) -> Result<PostgresConnection, ConnectionError> {
-        let pause = ctx.latency_timer.pause(crate::metrics::Waiting::Compute);
         let (socket_addr, stream, host) = self.connect_raw(timeout).await?;
-        drop(pause);
 
         let tls_connector = native_tls::TlsConnector::builder()
             .danger_accept_invalid_certs(allow_self_signed_compute)
@@ -272,9 +270,7 @@ impl ConnCfg {
         let tls = MakeTlsConnect::<tokio::net::TcpStream>::make_tls_connect(&mut mk_tls, host)?;
 
         // connect_raw() will not use TLS if sslmode is "disable"
-        let pause = ctx.latency_timer.pause(crate::metrics::Waiting::Compute);
         let (client, connection) = self.0.connect_raw(stream, tls).await?;
-        drop(pause);
         tracing::Span::current().record("pid", &tracing::field::display(client.get_process_id()));
         let stream = connection.stream.into_inner();
 

proxy/src/config.rs

@@ -29,12 +29,11 @@ pub struct ProxyConfig {
     pub authentication_config: AuthenticationConfig,
     pub require_client_ip: bool,
     pub disable_ip_check_for_http: bool,
+    pub endpoint_rps_limit: Vec<RateBucketInfo>,
     pub redis_rps_limit: Vec<RateBucketInfo>,
     pub region: String,
     pub handshake_timeout: Duration,
     pub aws_region: String,
-    pub wake_compute_retry_config: RetryConfig,
-    pub connect_to_compute_retry_config: RetryConfig,
 }
 
 #[derive(Debug)]
@@ -519,59 +518,6 @@ impl FromStr for ProjectInfoCacheOptions {
     }
 }
 
-/// This is a config for connect to compute and wake compute.
-#[derive(Clone, Copy, Debug)]
-pub struct RetryConfig {
-    /// Number of times we should retry.
-    pub max_retries: u32,
-    /// Retry duration is base_delay * backoff_factor ^ n, where n starts at 0
-    pub base_delay: tokio::time::Duration,
-    /// Exponential base for retry wait duration
-    pub backoff_factor: f64,
-}
-
-impl RetryConfig {
-    /// Default options for RetryConfig.
-
-    /// Total delay for 8 retries with 100ms base delay and 1.6 backoff factor is about 7s.
-    pub const CONNECT_TO_COMPUTE_DEFAULT_VALUES: &'static str =
-        "num_retries=8,base_retry_wait_duration=100ms,retry_wait_exponent_base=1.6";
-    /// Total delay for 8 retries with 100ms base delay and 1.6 backoff factor is about 7s.
-    /// Cplane has timeout of 60s on each request. 8m7s in total.
-    pub const WAKE_COMPUTE_DEFAULT_VALUES: &'static str =
-        "num_retries=8,base_retry_wait_duration=100ms,retry_wait_exponent_base=1.6";
-
-    /// Parse retry options passed via cmdline.
-    /// Example: [`Self::CONNECT_TO_COMPUTE_DEFAULT_VALUES`].
-    pub fn parse(options: &str) -> anyhow::Result<Self> {
-        let mut num_retries = None;
-        let mut base_retry_wait_duration = None;
-        let mut retry_wait_exponent_base = None;
-
-        for option in options.split(',') {
-            let (key, value) = option
-                .split_once('=')
-                .with_context(|| format!("bad key-value pair: {option}"))?;
-
-            match key {
-                "num_retries" => num_retries = Some(value.parse()?),
-                "base_retry_wait_duration" => {
-                    base_retry_wait_duration = Some(humantime::parse_duration(value)?)
-                }
-                "retry_wait_exponent_base" => retry_wait_exponent_base = Some(value.parse()?),
-                unknown => bail!("unknown key: {unknown}"),
-            }
-        }
-
-        Ok(Self {
-            max_retries: num_retries.context("missing `num_retries`")?,
-            base_delay: base_retry_wait_duration.context("missing `base_retry_wait_duration`")?,
-            backoff_factor: retry_wait_exponent_base
-                .context("missing `retry_wait_exponent_base`")?,
-        })
-    }
-}
-
 /// Helper for cmdline cache options parsing.
 pub struct WakeComputeLockOptions {
     /// The number of shards the lock map should have

proxy/src/console/provider.rs

@@ -208,9 +208,6 @@ pub mod errors {
         #[error(transparent)]
         ApiError(ApiError),
 
-        #[error("Too many connections attempts")]
-        TooManyConnections,
-
         #[error("Timeout waiting to acquire wake compute lock")]
         TimeoutError,
     }
@@ -243,8 +240,6 @@ pub mod errors {
                 // However, API might return a meaningful error.
                 ApiError(e) => e.to_string_client(),
 
-                TooManyConnections => self.to_string(),
-
                 TimeoutError => "timeout while acquiring the compute resource lock".to_owned(),
             }
         }
@@ -255,7 +250,6 @@ pub mod errors {
             match self {
                 WakeComputeError::BadComputeAddress(_) => crate::error::ErrorKind::ControlPlane,
                 WakeComputeError::ApiError(e) => e.get_error_kind(),
-                WakeComputeError::TooManyConnections => crate::error::ErrorKind::RateLimit,
                 WakeComputeError::TimeoutError => crate::error::ErrorKind::ServiceRateLimit,
             }
         }

proxy/src/console/provider/neon.rs

@@ -12,7 +12,6 @@ use crate::{
     console::messages::ColdStartInfo,
     http,
     metrics::{CacheOutcome, Metrics},
-    rate_limiter::EndpointRateLimiter,
     scram, Normalize,
 };
 use crate::{cache::Cached, context::RequestMonitoring};
@@ -26,7 +25,6 @@ pub struct Api {
     endpoint: http::Endpoint,
     pub caches: &'static ApiCaches,
     pub locks: &'static ApiLocks,
-    pub endpoint_rate_limiter: Arc<EndpointRateLimiter>,
     jwt: String,
 }
 
@@ -36,7 +34,6 @@ impl Api {
         endpoint: http::Endpoint,
         caches: &'static ApiCaches,
         locks: &'static ApiLocks,
-        endpoint_rate_limiter: Arc<EndpointRateLimiter>,
     ) -> Self {
         let jwt: String = match std::env::var("NEON_PROXY_TO_CONTROLPLANE_TOKEN") {
             Ok(v) => v,
@@ -46,7 +43,6 @@ impl Api {
             endpoint,
             caches,
             locks,
-            endpoint_rate_limiter,
             jwt,
         }
     }
@@ -281,14 +277,6 @@ impl super::Api for Api {
             return Ok(cached);
         }
 
-        // check rate limit
-        if !self
-            .endpoint_rate_limiter
-            .check(user_info.endpoint.normalize().into(), 1)
-        {
-            return Err(WakeComputeError::TooManyConnections);
-        }
-
         let permit = self.locks.get_wake_compute_permit(&key).await?;
 
         // after getting back a permit - it's possible the cache was filled

proxy/src/context.rs

@@ -20,8 +20,7 @@ use self::parquet::RequestData;
 
 pub mod parquet;
 
-pub static LOG_CHAN: OnceCell<mpsc::WeakUnboundedSender<RequestData>> = OnceCell::new();
-pub static LOG_CHAN_DISCONNECT: OnceCell<mpsc::WeakUnboundedSender<RequestData>> = OnceCell::new();
+static LOG_CHAN: OnceCell<mpsc::WeakUnboundedSender<RequestData>> = OnceCell::new();
 
 /// Context data for a single request to connect to a database.
 ///
@@ -50,12 +49,9 @@ pub struct RequestMonitoring {
     // extra
     // This sender is here to keep the request monitoring channel open while requests are taking place.
     sender: Option<mpsc::UnboundedSender<RequestData>>,
-    // This sender is only used to log the length of session in case of success.
-    disconnect_sender: Option<mpsc::UnboundedSender<RequestData>>,
     pub latency_timer: LatencyTimer,
     // Whether proxy decided that it's not a valid endpoint end rejected it before going to cplane.
-    rejected: Option<bool>,
-    disconnect_timestamp: Option<chrono::DateTime<Utc>>,
+    rejected: bool,
 }
 
 #[derive(Clone, Debug)]
@@ -100,13 +96,11 @@ impl RequestMonitoring {
             error_kind: None,
             auth_method: None,
             success: false,
-            rejected: None,
+            rejected: false,
             cold_start_info: ColdStartInfo::Unknown,
 
             sender: LOG_CHAN.get().and_then(|tx| tx.upgrade()),
-            disconnect_sender: LOG_CHAN_DISCONNECT.get().and_then(|tx| tx.upgrade()),
             latency_timer: LatencyTimer::new(protocol),
-            disconnect_timestamp: None,
         }
     }
 
@@ -124,7 +118,7 @@ impl RequestMonitoring {
     }
 
     pub fn set_rejected(&mut self, rejected: bool) {
-        self.rejected = Some(rejected);
+        self.rejected = rejected;
     }
 
     pub fn set_cold_start_info(&mut self, info: ColdStartInfo) {
@@ -196,55 +190,39 @@ impl RequestMonitoring {
         self.success = true;
     }
 
-    pub fn log_connect(&mut self) {
+    pub fn log(self) {}
+}
+
+impl Drop for RequestMonitoring {
+    fn drop(&mut self) {
         let outcome = if self.success {
             ConnectOutcome::Success
         } else {
             ConnectOutcome::Failed
         };
-        if let Some(rejected) = self.rejected {
-            let ep = self
-                .endpoint_id
-                .as_ref()
-                .map(|x| x.as_str())
-                .unwrap_or_default();
-            // This makes sense only if cache is disabled
-            info!(
-                ?outcome,
-                ?rejected,
-                ?ep,
-                "check endpoint is valid with outcome"
-            );
-            Metrics::get()
-                .proxy
-                .invalid_endpoints_total
-                .inc(InvalidEndpointsGroup {
-                    protocol: self.protocol,
-                    rejected: rejected.into(),
-                    outcome,
-                });
-        }
+        let rejected = self.rejected;
+        let ep = self
+            .endpoint_id
+            .as_ref()
+            .map(|x| x.as_str())
+            .unwrap_or_default();
+        // This makes sense only if cache is disabled
+        info!(
+            ?ep,
+            ?outcome,
+            ?rejected,
+            "check endpoint is valid with outcome"
+        );
+        Metrics::get()
+            .proxy
+            .invalid_endpoints_total
+            .inc(InvalidEndpointsGroup {
+                protocol: self.protocol,
+                rejected: rejected.into(),
+                outcome,
+            });
         if let Some(tx) = self.sender.take() {
             let _: Result<(), _> = tx.send(RequestData::from(&*self));
         }
     }
-
-    fn log_disconnect(&mut self) {
-        // If we are here, it's guaranteed that the user successfully connected to the endpoint.
-        // Here we log the length of the session.
-        self.disconnect_timestamp = Some(Utc::now());
-        if let Some(tx) = self.disconnect_sender.take() {
-            let _: Result<(), _> = tx.send(RequestData::from(&*self));
-        }
-    }
-}
-
-impl Drop for RequestMonitoring {
-    fn drop(&mut self) {
-        if self.sender.is_some() {
-            self.log_connect();
-        } else {
-            self.log_disconnect();
-        }
-    }
 }

proxy/src/context/parquet.rs

@@ -19,10 +19,7 @@ use tokio_util::sync::CancellationToken;
 use tracing::{debug, info, Span};
 use utils::backoff;
 
-use crate::{
-    config::{remote_storage_from_toml, OptRemoteStorageConfig},
-    context::LOG_CHAN_DISCONNECT,
-};
+use crate::config::{remote_storage_from_toml, OptRemoteStorageConfig};
 
 use super::{RequestMonitoring, LOG_CHAN};
 
@@ -34,9 +31,6 @@ pub struct ParquetUploadArgs {
     #[clap(long, default_value = "{}", value_parser = remote_storage_from_toml)]
     parquet_upload_remote_storage: OptRemoteStorageConfig,
 
-    #[clap(long, default_value = "{}", value_parser = remote_storage_from_toml)]
-    parquet_upload_disconnect_events_remote_storage: OptRemoteStorageConfig,
-
     /// How many rows to include in a row group
     #[clap(long, default_value_t = 8192)]
     parquet_upload_row_group_size: usize,
@@ -97,8 +91,6 @@ pub struct RequestData {
     /// Tracks time from session start (HTTP request/libpq TCP handshake)
     /// Through to success/failure
     duration_us: u64,
-    /// If the session was successful after the disconnect, will be created one more event with filled `disconnect_timestamp`.
-    disconnect_timestamp: Option<chrono::NaiveDateTime>,
 }
 
 impl From<&RequestMonitoring> for RequestData {
@@ -128,7 +120,6 @@ impl From<&RequestMonitoring> for RequestData {
                 .elapsed()
                 .unwrap_or_default()
                 .as_micros() as u64, // 584 millenia... good enough
-            disconnect_timestamp: value.disconnect_timestamp.map(|x| x.naive_utc()),
         }
     }
 }
@@ -150,9 +141,8 @@ pub async fn worker(
     LOG_CHAN.set(tx.downgrade()).unwrap();
 
     // setup row stream that will close on cancellation
-    let cancellation_token2 = cancellation_token.clone();
     tokio::spawn(async move {
-        cancellation_token2.cancelled().await;
+        cancellation_token.cancelled().await;
         // dropping this sender will cause the channel to close only once
         // all the remaining inflight requests have been completed.
         drop(tx);
@@ -177,38 +167,9 @@ pub async fn worker(
         test_remote_failures: 0,
     };
 
-    // TODO(anna): consider moving this to a separate function.
-    if let Some(disconnect_events_storage_config) =
-        config.parquet_upload_disconnect_events_remote_storage
-    {
-        let (tx_disconnect, mut rx_disconnect) = mpsc::unbounded_channel();
-        LOG_CHAN_DISCONNECT.set(tx_disconnect.downgrade()).unwrap();
-
-        // setup row stream that will close on cancellation
-        tokio::spawn(async move {
-            cancellation_token.cancelled().await;
-            // dropping this sender will cause the channel to close only once
-            // all the remaining inflight requests have been completed.
-            drop(tx_disconnect);
-        });
-        let rx_disconnect = futures::stream::poll_fn(move |cx| rx_disconnect.poll_recv(cx));
-        let rx_disconnect = rx_disconnect.map(RequestData::from);
-
-        let storage_disconnect =
-            GenericRemoteStorage::from_config(&disconnect_events_storage_config)
-                .context("remote storage for disconnect events init")?;
-        let parquet_config_disconnect = parquet_config.clone();
-        tokio::try_join!(
-            worker_inner(storage, rx, parquet_config),
-            worker_inner(storage_disconnect, rx_disconnect, parquet_config_disconnect)
-        )
-        .map(|_| ())
-    } else {
-        worker_inner(storage, rx, parquet_config).await
-    }
+    worker_inner(storage, rx, parquet_config).await
 }
 
-#[derive(Clone, Debug)]
 struct ParquetConfig {
     propeties: WriterPropertiesPtr,
     rows_per_group: usize,
@@ -452,7 +413,6 @@ mod tests {
                     )
                     .unwrap(),
                     max_keys_per_list_response: DEFAULT_MAX_KEYS_PER_LIST_RESPONSE,
-                    upload_storage_class: None,
                 }),
                 timeout: RemoteStorageConfig::DEFAULT_TIMEOUT,
             })
@@ -491,7 +451,6 @@ mod tests {
             success: rng.gen(),
             cold_start_info: "no",
             duration_us: rng.gen_range(0..30_000_000),
-            disconnect_timestamp: None,
         }
     }
 
@@ -560,15 +519,15 @@ mod tests {
         assert_eq!(
             file_stats,
             [
-                (1315008, 3, 6000),
-                (1315001, 3, 6000),
-                (1315061, 3, 6000),
-                (1315018, 3, 6000),
-                (1315148, 3, 6000),
-                (1314990, 3, 6000),
-                (1314782, 3, 6000),
-                (1315018, 3, 6000),
-                (438575, 1, 2000)
+                (1314385, 3, 6000),
+                (1314378, 3, 6000),
+                (1314438, 3, 6000),
+                (1314395, 3, 6000),
+                (1314525, 3, 6000),
+                (1314367, 3, 6000),
+                (1314159, 3, 6000),
+                (1314395, 3, 6000),
+                (438352, 1, 2000)
             ]
         );
 
@@ -598,11 +557,11 @@ mod tests {
         assert_eq!(
             file_stats,
             [
-                (1221738, 5, 10000),
-                (1227888, 5, 10000),
-                (1229682, 5, 10000),
-                (1229044, 5, 10000),
-                (1220322, 5, 10000)
+                (1220633, 5, 10000),
+                (1226783, 5, 10000),
+                (1228577, 5, 10000),
+                (1227939, 5, 10000),
+                (1219217, 5, 10000)
             ]
         );
 
@@ -634,11 +593,11 @@ mod tests {
         assert_eq!(
             file_stats,
             [
-                (1207385, 5, 10000),
-                (1207116, 5, 10000),
-                (1207409, 5, 10000),
-                (1207397, 5, 10000),
-                (1207652, 5, 10000)
+                (1206280, 5, 10000),
+                (1206011, 5, 10000),
+                (1206304, 5, 10000),
+                (1206292, 5, 10000),
+                (1206547, 5, 10000)
             ]
         );
 
@@ -663,15 +622,15 @@ mod tests {
         assert_eq!(
             file_stats,
             [
-                (1315008, 3, 6000),
-                (1315001, 3, 6000),
-                (1315061, 3, 6000),
-                (1315018, 3, 6000),
-                (1315148, 3, 6000),
-                (1314990, 3, 6000),
-                (1314782, 3, 6000),
-                (1315018, 3, 6000),
-                (438575, 1, 2000)
+                (1314385, 3, 6000),
+                (1314378, 3, 6000),
+                (1314438, 3, 6000),
+                (1314395, 3, 6000),
+                (1314525, 3, 6000),
+                (1314367, 3, 6000),
+                (1314159, 3, 6000),
+                (1314395, 3, 6000),
+                (438352, 1, 2000)
             ]
         );
 
@@ -708,7 +667,7 @@ mod tests {
         // files are smaller than the size threshold, but they took too long to fill so were flushed early
         assert_eq!(
             file_stats,
-            [(659240, 2, 3001), (658954, 2, 3000), (658750, 2, 2999)]
+            [(658823, 2, 3001), (658537, 2, 3000), (658333, 2, 2999)]
         );
 
         tmpdir.close().unwrap();

proxy/src/metrics.rs

@@ -119,10 +119,6 @@ pub struct ProxyMetrics {
 
     /// Number of invalid endpoints (per protocol, per rejected).
     pub invalid_endpoints_total: CounterVec<InvalidEndpointsSet>,
-
-    /// Number of retries (per outcome, per retry_type).
-    #[metric(metadata = Thresholds::with_buckets([0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]))]
-    pub retries_metric: HistogramVec<RetriesMetricSet, 9>,
 }
 
 #[derive(MetricGroup)]
@@ -284,8 +280,6 @@ pub struct ComputeConnectionLatencyGroup {
 pub enum LatencyExclusions {
     Client,
     ClientAndCplane,
-    ClientCplaneCompute,
-    ClientCplaneComputeRetry,
 }
 
 #[derive(FixedCardinalityLabel, Copy, Clone)]
@@ -354,7 +348,6 @@ pub enum Waiting {
     Cplane,
     Client,
     Compute,
-    RetryTimeout,
 }
 
 #[derive(Default)]
@@ -362,7 +355,6 @@ struct Accumulated {
     cplane: time::Duration,
     client: time::Duration,
     compute: time::Duration,
-    retry: time::Duration,
 }
 
 pub struct LatencyTimer {
@@ -425,7 +417,6 @@ impl Drop for LatencyTimerPause<'_> {
             Waiting::Cplane => self.timer.accumulated.cplane += dur,
             Waiting::Client => self.timer.accumulated.client += dur,
             Waiting::Compute => self.timer.accumulated.compute += dur,
-            Waiting::RetryTimeout => self.timer.accumulated.retry += dur,
         }
     }
 }
@@ -469,34 +460,6 @@ impl Drop for LatencyTimer {
             },
             duration.saturating_sub(accumulated_total).as_secs_f64(),
         );
-
-        // Exclude client cplane, compue communication from the accumulated time.
-        let accumulated_total =
-            self.accumulated.client + self.accumulated.cplane + self.accumulated.compute;
-        metric.observe(
-            ComputeConnectionLatencyGroup {
-                protocol: self.protocol,
-                cold_start_info: self.cold_start_info,
-                outcome: self.outcome,
-                excluded: LatencyExclusions::ClientCplaneCompute,
-            },
-            duration.saturating_sub(accumulated_total).as_secs_f64(),
-        );
-
-        // Exclude client cplane, compue, retry communication from the accumulated time.
-        let accumulated_total = self.accumulated.client
-            + self.accumulated.cplane
-            + self.accumulated.compute
-            + self.accumulated.retry;
-        metric.observe(
-            ComputeConnectionLatencyGroup {
-                protocol: self.protocol,
-                cold_start_info: self.cold_start_info,
-                outcome: self.outcome,
-                excluded: LatencyExclusions::ClientCplaneComputeRetry,
-            },
-            duration.saturating_sub(accumulated_total).as_secs_f64(),
-        );
     }
 }
 
@@ -517,16 +480,3 @@ pub struct InvalidEndpointsGroup {
     pub rejected: Bool,
     pub outcome: ConnectOutcome,
 }
-
-#[derive(LabelGroup)]
-#[label(set = RetriesMetricSet)]
-pub struct RetriesMetricGroup {
-    pub outcome: ConnectOutcome,
-    pub retry_type: RetryType,
-}
-
-#[derive(FixedCardinalityLabel, Clone, Copy, Debug)]
-pub enum RetryType {
-    WakeCompute,
-    ConnectToCompute,
-}

proxy/src/protocol2.rs

@@ -1,26 +1,42 @@
 //! Proxy Protocol V2 implementation
 
 use std::{
+    future::{poll_fn, Future},
     io,
     net::SocketAddr,
-    pin::Pin,
-    task::{Context, Poll},
+    pin::{pin, Pin},
+    task::{ready, Context, Poll},
 };
 
-use bytes::BytesMut;
+use bytes::{Buf, BytesMut};
+use hyper::server::conn::AddrIncoming;
 use pin_project_lite::pin_project;
 use tokio::io::{AsyncRead, AsyncReadExt, AsyncWrite, ReadBuf};
 
+pub struct ProxyProtocolAccept {
+    pub incoming: AddrIncoming,
+    pub protocol: &'static str,
+}
+
 pin_project! {
-    /// A chained [`AsyncRead`] with [`AsyncWrite`] passthrough
-    pub struct ChainRW<T> {
+    pub struct WithClientIp<T> {
         #[pin]
         pub inner: T,
         buf: BytesMut,
+        tlv_bytes: u16,
+        state: ProxyParse,
     }
 }
 
-impl<T: AsyncWrite> AsyncWrite for ChainRW<T> {
+#[derive(Clone, PartialEq, Debug)]
+enum ProxyParse {
+    NotStarted,
+
+    Finished(SocketAddr),
+    None,
+}
+
+impl<T: AsyncWrite> AsyncWrite for WithClientIp<T> {
     #[inline]
     fn poll_write(
         self: Pin<&mut Self>,
@@ -55,174 +71,267 @@ impl<T: AsyncWrite> AsyncWrite for ChainRW<T> {
     }
 }
 
+impl<T> WithClientIp<T> {
+    pub fn new(inner: T) -> Self {
+        WithClientIp {
+            inner,
+            buf: BytesMut::with_capacity(128),
+            tlv_bytes: 0,
+            state: ProxyParse::NotStarted,
+        }
+    }
+
+    pub fn client_addr(&self) -> Option<SocketAddr> {
+        match self.state {
+            ProxyParse::Finished(socket) => Some(socket),
+            _ => None,
+        }
+    }
+}
+
+impl<T: AsyncRead + Unpin> WithClientIp<T> {
+    pub async fn wait_for_addr(&mut self) -> io::Result<Option<SocketAddr>> {
+        match self.state {
+            ProxyParse::NotStarted => {
+                let mut pin = Pin::new(&mut *self);
+                let addr = poll_fn(|cx| pin.as_mut().poll_client_ip(cx)).await?;
+                match addr {
+                    Some(addr) => self.state = ProxyParse::Finished(addr),
+                    None => self.state = ProxyParse::None,
+                }
+                Ok(addr)
+            }
+            ProxyParse::Finished(addr) => Ok(Some(addr)),
+            ProxyParse::None => Ok(None),
+        }
+    }
+}
+
 /// Proxy Protocol Version 2 Header
 const HEADER: [u8; 12] = [
     0x0D, 0x0A, 0x0D, 0x0A, 0x00, 0x0D, 0x0A, 0x51, 0x55, 0x49, 0x54, 0x0A,
 ];
 
-pub async fn read_proxy_protocol<T: AsyncRead + Unpin>(
-    mut read: T,
-) -> std::io::Result<(ChainRW<T>, Option<SocketAddr>)> {
-    let mut buf = BytesMut::with_capacity(128);
-    while buf.len() < 16 {
-        let bytes_read = read.read_buf(&mut buf).await?;
+impl<T: AsyncRead> WithClientIp<T> {
+    /// implementation of <https://www.haproxy.org/download/2.4/doc/proxy-protocol.txt>
+    /// Version 2 (Binary Format)
+    fn poll_client_ip(
+        mut self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+    ) -> Poll<io::Result<Option<SocketAddr>>> {
+        // The binary header format starts with a constant 12 bytes block containing the protocol signature :
+        //    \x0D \x0A \x0D \x0A \x00 \x0D \x0A \x51 \x55 \x49 \x54 \x0A
+        while self.buf.len() < 16 {
+            let mut this = self.as_mut().project();
+            let bytes_read = pin!(this.inner.read_buf(this.buf)).poll(cx)?;
 
-        // exit for bad header
-        let len = usize::min(buf.len(), HEADER.len());
-        if buf[..len] != HEADER[..len] {
-            return Ok((ChainRW { inner: read, buf }, None));
+            // exit for bad header
+            let len = usize::min(self.buf.len(), HEADER.len());
+            if self.buf[..len] != HEADER[..len] {
+                return Poll::Ready(Ok(None));
+            }
+
+            // if no more bytes available then exit
+            if ready!(bytes_read) == 0 {
+                return Poll::Ready(Ok(None));
+            };
         }
 
-        // if no more bytes available then exit
-        if bytes_read == 0 {
-            return Ok((ChainRW { inner: read, buf }, None));
-        };
-    }
-
-    let header = buf.split_to(16);
-
-    // The next byte (the 13th one) is the protocol version and command.
-    // The highest four bits contains the version. As of this specification, it must
-    // always be sent as \x2 and the receiver must only accept this value.
-    let vc = header[12];
-    let version = vc >> 4;
-    let command = vc & 0b1111;
-    if version != 2 {
-        return Err(io::Error::new(
-            io::ErrorKind::Other,
-            "invalid proxy protocol version. expected version 2",
-        ));
-    }
-    match command {
-        // the connection was established on purpose by the proxy
-        // without being relayed. The connection endpoints are the sender and the
-        // receiver. Such connections exist when the proxy sends health-checks to the
-        // server. The receiver must accept this connection as valid and must use the
-        // real connection endpoints and discard the protocol block including the
-        // family which is ignored.
-        0 => {}
-        // the connection was established on behalf of another node,
-        // and reflects the original connection endpoints. The receiver must then use
-        // the information provided in the protocol block to get original the address.
-        1 => {}
-        // other values are unassigned and must not be emitted by senders. Receivers
-        // must drop connections presenting unexpected values here.
-        _ => {
-            return Err(io::Error::new(
+        // The next byte (the 13th one) is the protocol version and command.
+        // The highest four bits contains the version. As of this specification, it must
+        // always be sent as \x2 and the receiver must only accept this value.
+        let vc = self.buf[12];
+        let version = vc >> 4;
+        let command = vc & 0b1111;
+        if version != 2 {
+            return Poll::Ready(Err(io::Error::new(
                 io::ErrorKind::Other,
-                "invalid proxy protocol command. expected local (0) or proxy (1)",
-            ))
+                "invalid proxy protocol version. expected version 2",
+            )));
         }
-    };
+        match command {
+            // the connection was established on purpose by the proxy
+            // without being relayed. The connection endpoints are the sender and the
+            // receiver. Such connections exist when the proxy sends health-checks to the
+            // server. The receiver must accept this connection as valid and must use the
+            // real connection endpoints and discard the protocol block including the
+            // family which is ignored.
+            0 => {}
+            // the connection was established on behalf of another node,
+            // and reflects the original connection endpoints. The receiver must then use
+            // the information provided in the protocol block to get original the address.
+            1 => {}
+            // other values are unassigned and must not be emitted by senders. Receivers
+            // must drop connections presenting unexpected values here.
+            _ => {
+                return Poll::Ready(Err(io::Error::new(
+                    io::ErrorKind::Other,
+                    "invalid proxy protocol command. expected local (0) or proxy (1)",
+                )))
+            }
+        };
 
-    // The 14th byte contains the transport protocol and address family. The highest 4
-    // bits contain the address family, the lowest 4 bits contain the protocol.
-    let ft = header[13];
-    let address_length = match ft {
-        // - \x11 : TCP over IPv4 : the forwarded connection uses TCP over the AF_INET
-        //   protocol family. Address length is 2*4 + 2*2 = 12 bytes.
-        // - \x12 : UDP over IPv4 : the forwarded connection uses UDP over the AF_INET
-        //   protocol family. Address length is 2*4 + 2*2 = 12 bytes.
-        0x11 | 0x12 => 12,
-        // - \x21 : TCP over IPv6 : the forwarded connection uses TCP over the AF_INET6
-        //   protocol family. Address length is 2*16 + 2*2 = 36 bytes.
-        // - \x22 : UDP over IPv6 : the forwarded connection uses UDP over the AF_INET6
-        //   protocol family. Address length is 2*16 + 2*2 = 36 bytes.
-        0x21 | 0x22 => 36,
-        // unspecified or unix stream. ignore the addresses
-        _ => 0,
-    };
+        // The 14th byte contains the transport protocol and address family. The highest 4
+        // bits contain the address family, the lowest 4 bits contain the protocol.
+        let ft = self.buf[13];
+        let address_length = match ft {
+            // - \x11 : TCP over IPv4 : the forwarded connection uses TCP over the AF_INET
+            //   protocol family. Address length is 2*4 + 2*2 = 12 bytes.
+            // - \x12 : UDP over IPv4 : the forwarded connection uses UDP over the AF_INET
+            //   protocol family. Address length is 2*4 + 2*2 = 12 bytes.
+            0x11 | 0x12 => 12,
+            // - \x21 : TCP over IPv6 : the forwarded connection uses TCP over the AF_INET6
+            //   protocol family. Address length is 2*16 + 2*2 = 36 bytes.
+            // - \x22 : UDP over IPv6 : the forwarded connection uses UDP over the AF_INET6
+            //   protocol family. Address length is 2*16 + 2*2 = 36 bytes.
+            0x21 | 0x22 => 36,
+            // unspecified or unix stream. ignore the addresses
+            _ => 0,
+        };
 
-    // The 15th and 16th bytes is the address length in bytes in network endian order.
-    // It is used so that the receiver knows how many address bytes to skip even when
-    // it does not implement the presented protocol. Thus the length of the protocol
-    // header in bytes is always exactly 16 + this value. When a sender presents a
-    // LOCAL connection, it should not present any address so it sets this field to
-    // zero. Receivers MUST always consider this field to skip the appropriate number
-    // of bytes and must not assume zero is presented for LOCAL connections. When a
-    // receiver accepts an incoming connection showing an UNSPEC address family or
-    // protocol, it may or may not decide to log the address information if present.
-    let remaining_length = u16::from_be_bytes(header[14..16].try_into().unwrap());
-    if remaining_length < address_length {
-        return Err(io::Error::new(
-            io::ErrorKind::Other,
-            "invalid proxy protocol length. not enough to fit requested IP addresses",
-        ));
-    }
-    drop(header);
-
-    while buf.len() < remaining_length as usize {
-        if read.read_buf(&mut buf).await? == 0 {
-            return Err(io::Error::new(
-                io::ErrorKind::UnexpectedEof,
-                "stream closed while waiting for proxy protocol addresses",
-            ));
+        // The 15th and 16th bytes is the address length in bytes in network endian order.
+        // It is used so that the receiver knows how many address bytes to skip even when
+        // it does not implement the presented protocol. Thus the length of the protocol
+        // header in bytes is always exactly 16 + this value. When a sender presents a
+        // LOCAL connection, it should not present any address so it sets this field to
+        // zero. Receivers MUST always consider this field to skip the appropriate number
+        // of bytes and must not assume zero is presented for LOCAL connections. When a
+        // receiver accepts an incoming connection showing an UNSPEC address family or
+        // protocol, it may or may not decide to log the address information if present.
+        let remaining_length = u16::from_be_bytes(self.buf[14..16].try_into().unwrap());
+        if remaining_length < address_length {
+            return Poll::Ready(Err(io::Error::new(
+                io::ErrorKind::Other,
+                "invalid proxy protocol length. not enough to fit requested IP addresses",
+            )));
         }
+
+        while self.buf.len() < 16 + address_length as usize {
+            let mut this = self.as_mut().project();
+            if ready!(pin!(this.inner.read_buf(this.buf)).poll(cx)?) == 0 {
+                return Poll::Ready(Err(io::Error::new(
+                    io::ErrorKind::UnexpectedEof,
+                    "stream closed while waiting for proxy protocol addresses",
+                )));
+            }
+        }
+
+        let this = self.as_mut().project();
+
+        // we are sure this is a proxy protocol v2 entry and we have read all the bytes we need
+        // discard the header we have parsed
+        this.buf.advance(16);
+
+        // Starting from the 17th byte, addresses are presented in network byte order.
+        // The address order is always the same :
+        //   - source layer 3 address in network byte order
+        //   - destination layer 3 address in network byte order
+        //   - source layer 4 address if any, in network byte order (port)
+        //   - destination layer 4 address if any, in network byte order (port)
+        let addresses = this.buf.split_to(address_length as usize);
+        let socket = match address_length {
+            12 => {
+                let src_addr: [u8; 4] = addresses[0..4].try_into().unwrap();
+                let src_port = u16::from_be_bytes(addresses[8..10].try_into().unwrap());
+                Some(SocketAddr::from((src_addr, src_port)))
+            }
+            36 => {
+                let src_addr: [u8; 16] = addresses[0..16].try_into().unwrap();
+                let src_port = u16::from_be_bytes(addresses[32..34].try_into().unwrap());
+                Some(SocketAddr::from((src_addr, src_port)))
+            }
+            _ => None,
+        };
+
+        *this.tlv_bytes = remaining_length - address_length;
+        self.as_mut().skip_tlv_inner();
+
+        Poll::Ready(Ok(socket))
     }
 
-    // Starting from the 17th byte, addresses are presented in network byte order.
-    // The address order is always the same :
-    //   - source layer 3 address in network byte order
-    //   - destination layer 3 address in network byte order
-    //   - source layer 4 address if any, in network byte order (port)
-    //   - destination layer 4 address if any, in network byte order (port)
-    let addresses = buf.split_to(remaining_length as usize);
-    let socket = match address_length {
-        12 => {
-            let src_addr: [u8; 4] = addresses[0..4].try_into().unwrap();
-            let src_port = u16::from_be_bytes(addresses[8..10].try_into().unwrap());
-            Some(SocketAddr::from((src_addr, src_port)))
+    #[cold]
+    fn read_ip(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>> {
+        let ip = ready!(self.as_mut().poll_client_ip(cx)?);
+        match ip {
+            Some(x) => *self.as_mut().project().state = ProxyParse::Finished(x),
+            None => *self.as_mut().project().state = ProxyParse::None,
         }
-        36 => {
-            let src_addr: [u8; 16] = addresses[0..16].try_into().unwrap();
-            let src_port = u16::from_be_bytes(addresses[32..34].try_into().unwrap());
-            Some(SocketAddr::from((src_addr, src_port)))
-        }
-        _ => None,
-    };
+        Poll::Ready(Ok(()))
+    }
 
-    Ok((ChainRW { inner: read, buf }, socket))
+    #[cold]
+    fn skip_tlv(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>> {
+        let mut this = self.as_mut().project();
+        // we know that this.buf is empty
+        debug_assert_eq!(this.buf.len(), 0);
+
+        this.buf.reserve((*this.tlv_bytes).clamp(0, 1024) as usize);
+        ready!(pin!(this.inner.read_buf(this.buf)).poll(cx)?);
+        self.skip_tlv_inner();
+
+        Poll::Ready(Ok(()))
+    }
+
+    fn skip_tlv_inner(self: Pin<&mut Self>) {
+        let tlv_bytes_read = match u16::try_from(self.buf.len()) {
+            // we read more than u16::MAX therefore we must have read the full tlv_bytes
+            Err(_) => self.tlv_bytes,
+            // we might not have read the full tlv bytes yet
+            Ok(n) => u16::min(n, self.tlv_bytes),
+        };
+        let this = self.project();
+        *this.tlv_bytes -= tlv_bytes_read;
+        this.buf.advance(tlv_bytes_read as usize);
+    }
 }
 
-impl<T: AsyncRead> AsyncRead for ChainRW<T> {
+impl<T: AsyncRead> AsyncRead for WithClientIp<T> {
     #[inline]
     fn poll_read(
-        self: Pin<&mut Self>,
+        mut self: Pin<&mut Self>,
         cx: &mut Context<'_>,
         buf: &mut ReadBuf<'_>,
     ) -> Poll<io::Result<()>> {
-        if self.buf.is_empty() {
-            self.project().inner.poll_read(cx, buf)
-        } else {
-            self.read_from_buf(buf)
-        }
-    }
-}
+        // I'm assuming these 3 comparisons will be easy to branch predict.
+        // especially with the cold attributes
+        // which should make this read wrapper almost invisible
+
+        if let ProxyParse::NotStarted = self.state {
+            ready!(self.as_mut().read_ip(cx)?);
+        }
+
+        while self.tlv_bytes > 0 {
+            ready!(self.as_mut().skip_tlv(cx)?)
+        }
 
-impl<T: AsyncRead> ChainRW<T> {
-    #[cold]
-    fn read_from_buf(self: Pin<&mut Self>, buf: &mut ReadBuf<'_>) -> Poll<io::Result<()>> {
-        debug_assert!(!self.buf.is_empty());
         let this = self.project();
-
-        let write = usize::min(this.buf.len(), buf.remaining());
-        let slice = this.buf.split_to(write).freeze();
-        buf.put_slice(&slice);
-
-        // reset the allocation so it can be freed
         if this.buf.is_empty() {
-            *this.buf = BytesMut::new();
-        }
+            this.inner.poll_read(cx, buf)
+        } else {
+            // we know that tlv_bytes is 0
+            debug_assert_eq!(*this.tlv_bytes, 0);
 
-        Poll::Ready(Ok(()))
+            let write = usize::min(this.buf.len(), buf.remaining());
+            let slice = this.buf.split_to(write).freeze();
+            buf.put_slice(&slice);
+
+            // reset the allocation so it can be freed
+            if this.buf.is_empty() {
+                *this.buf = BytesMut::new();
+            }
+
+            Poll::Ready(Ok(()))
+        }
     }
 }
 
 #[cfg(test)]
 mod tests {
+    use std::pin::pin;
+
     use tokio::io::AsyncReadExt;
 
-    use crate::protocol2::read_proxy_protocol;
+    use crate::protocol2::{ProxyParse, WithClientIp};
 
     #[tokio::test]
     async fn test_ipv4() {
@@ -244,15 +353,16 @@ mod tests {
 
         let extra_data = [0x55; 256];
 
-        let (mut read, addr) = read_proxy_protocol(header.chain(extra_data.as_slice()))
-            .await
-            .unwrap();
+        let mut read = pin!(WithClientIp::new(header.chain(extra_data.as_slice())));
 
         let mut bytes = vec![];
         read.read_to_end(&mut bytes).await.unwrap();
 
         assert_eq!(bytes, extra_data);
-        assert_eq!(addr, Some(([127, 0, 0, 1], 65535).into()));
+        assert_eq!(
+            read.state,
+            ProxyParse::Finished(([127, 0, 0, 1], 65535).into())
+        );
     }
 
     #[tokio::test]
@@ -275,17 +385,17 @@ mod tests {
 
         let extra_data = [0x55; 256];
 
-        let (mut read, addr) = read_proxy_protocol(header.chain(extra_data.as_slice()))
-            .await
-            .unwrap();
+        let mut read = pin!(WithClientIp::new(header.chain(extra_data.as_slice())));
 
         let mut bytes = vec![];
         read.read_to_end(&mut bytes).await.unwrap();
 
         assert_eq!(bytes, extra_data);
         assert_eq!(
-            addr,
-            Some(([15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0], 257).into())
+            read.state,
+            ProxyParse::Finished(
+                ([15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0], 257).into()
+            )
         );
     }
 
@@ -293,24 +403,24 @@ mod tests {
     async fn test_invalid() {
         let data = [0x55; 256];
 
-        let (mut read, addr) = read_proxy_protocol(data.as_slice()).await.unwrap();
+        let mut read = pin!(WithClientIp::new(data.as_slice()));
 
         let mut bytes = vec![];
         read.read_to_end(&mut bytes).await.unwrap();
         assert_eq!(bytes, data);
-        assert_eq!(addr, None);
+        assert_eq!(read.state, ProxyParse::None);
     }
 
     #[tokio::test]
     async fn test_short() {
         let data = [0x55; 10];
 
-        let (mut read, addr) = read_proxy_protocol(data.as_slice()).await.unwrap();
+        let mut read = pin!(WithClientIp::new(data.as_slice()));
 
         let mut bytes = vec![];
         read.read_to_end(&mut bytes).await.unwrap();
         assert_eq!(bytes, data);
-        assert_eq!(addr, None);
+        assert_eq!(read.state, ProxyParse::None);
     }
 
     #[tokio::test]
@@ -336,14 +446,15 @@ mod tests {
 
         let extra_data = [0xaa; 256];
 
-        let (mut read, addr) = read_proxy_protocol(header.chain(extra_data.as_slice()))
-            .await
-            .unwrap();
+        let mut read = pin!(WithClientIp::new(header.chain(extra_data.as_slice())));
 
         let mut bytes = vec![];
         read.read_to_end(&mut bytes).await.unwrap();
 
         assert_eq!(bytes, extra_data);
-        assert_eq!(addr, Some(([55, 56, 57, 58], 65535).into()));
+        assert_eq!(
+            read.state,
+            ProxyParse::Finished(([55, 56, 57, 58], 65535).into())
+        );
     }
 }

proxy/src/proxy.rs

@@ -17,10 +17,11 @@ use crate::{
     context::RequestMonitoring,
     error::ReportableError,
     metrics::{Metrics, NumClientConnectionsGuard},
-    protocol2::read_proxy_protocol,
+    protocol2::WithClientIp,
     proxy::handshake::{handshake, HandshakeData},
+    rate_limiter::EndpointRateLimiter,
     stream::{PqStream, Stream},
-    EndpointCacheKey,
+    EndpointCacheKey, Normalize,
 };
 use futures::TryFutureExt;
 use itertools::Itertools;
@@ -60,6 +61,7 @@ pub async fn task_main(
     config: &'static ProxyConfig,
     listener: tokio::net::TcpListener,
     cancellation_token: CancellationToken,
+    endpoint_rate_limiter: Arc<EndpointRateLimiter>,
     cancellation_handler: Arc<CancellationHandlerMain>,
 ) -> anyhow::Result<()> {
     scopeguard::defer! {
@@ -84,22 +86,25 @@ pub async fn task_main(
 
         let session_id = uuid::Uuid::new_v4();
         let cancellation_handler = Arc::clone(&cancellation_handler);
+        let endpoint_rate_limiter = endpoint_rate_limiter.clone();
 
         tracing::info!(protocol = "tcp", %session_id, "accepted new TCP connection");
 
         connections.spawn(async move {
-            let (socket, peer_addr) = match read_proxy_protocol(socket).await{
-                Ok((socket, Some(addr))) => (socket, addr.ip()),
+            let mut socket = WithClientIp::new(socket);
+            let mut peer_addr = peer_addr.ip();
+            match socket.wait_for_addr().await {
+                Ok(Some(addr)) => peer_addr = addr.ip(),
                 Err(e) => {
                     error!("per-client task finished with an error: {e:#}");
                     return;
                 }
-                Ok((_socket, None)) if config.require_client_ip => {
+                Ok(None) if config.require_client_ip => {
                     error!("missing required client IP");
                     return;
                 }
-                Ok((socket, None)) => (socket, peer_addr.ip())
-            };
+                Ok(None) => {}
+            }
 
             match socket.inner.set_nodelay(true) {
                 Ok(()) => {},
@@ -123,6 +128,7 @@ pub async fn task_main(
                 cancellation_handler,
                 socket,
                 ClientMode::Tcp,
+                endpoint_rate_limiter,
                 conn_gauge,
             )
             .instrument(span.clone())
@@ -132,14 +138,16 @@ pub async fn task_main(
                 Err(e) => {
                     // todo: log and push to ctx the error kind
                     ctx.set_error_kind(e.get_error_kind());
+                    ctx.log();
                     error!(parent: &span, "per-client task finished with an error: {e:#}");
                 }
                 Ok(None) => {
                     ctx.set_success();
+                    ctx.log();
                 }
                 Ok(Some(p)) => {
                     ctx.set_success();
-                    ctx.log_connect();
+                    ctx.log();
                     match p.proxy_pass().instrument(span.clone()).await {
                         Ok(()) => {}
                         Err(e) => {
@@ -234,6 +242,7 @@ pub async fn handle_client<S: AsyncRead + AsyncWrite + Unpin>(
     cancellation_handler: Arc<CancellationHandlerMain>,
     stream: S,
     mode: ClientMode,
+    endpoint_rate_limiter: Arc<EndpointRateLimiter>,
     conn_gauge: NumClientConnectionsGuard<'static>,
 ) -> Result<Option<ProxyPassthrough<CancellationHandlerMainInternal, S>>, ClientRequestError> {
     info!(
@@ -279,6 +288,15 @@ pub async fn handle_client<S: AsyncRead + AsyncWrite + Unpin>(
         Err(e) => stream.throw_error(e).await?,
     };
 
+    // check rate limit
+    if let Some(ep) = user_info.get_endpoint() {
+        if !endpoint_rate_limiter.check(ep.normalize(), 1) {
+            return stream
+                .throw_error(auth::AuthError::too_many_connections())
+                .await?;
+        }
+    }
+
     let user = user_info.get_user().to_owned();
     let user_info = match user_info
         .authenticate(
@@ -304,8 +322,6 @@ pub async fn handle_client<S: AsyncRead + AsyncWrite + Unpin>(
         &TcpMechanism { params: &params },
         &user_info,
         mode.allow_self_signed_compute(config),
-        config.wake_compute_retry_config,
-        config.connect_to_compute_retry_config,
     )
     .or_else(|e| stream.throw_error(e))
     .await?;

proxy/src/proxy/connect_compute.rs

@@ -1,11 +1,10 @@
 use crate::{
     auth::backend::ComputeCredentialKeys,
     compute::{self, PostgresConnection},
-    config::RetryConfig,
     console::{self, errors::WakeComputeError, CachedNodeInfo, NodeInfo},
     context::RequestMonitoring,
     error::ReportableError,
-    metrics::{ConnectOutcome, ConnectionFailureKind, Metrics, RetriesMetricGroup, RetryType},
+    metrics::{ConnectionFailureKind, Metrics},
     proxy::{
         retry::{retry_after, ShouldRetry},
         wake_compute::wake_compute,
@@ -94,23 +93,19 @@ pub async fn connect_to_compute<M: ConnectMechanism, B: ComputeConnectBackend>(
     mechanism: &M,
     user_info: &B,
     allow_self_signed_compute: bool,
-    wake_compute_retry_config: RetryConfig,
-    connect_to_compute_retry_config: RetryConfig,
 ) -> Result<M::Connection, M::Error>
 where
     M::ConnectError: ShouldRetry + std::fmt::Debug,
     M::Error: From<WakeComputeError>,
 {
     let mut num_retries = 0;
-    let mut node_info =
-        wake_compute(&mut num_retries, ctx, user_info, wake_compute_retry_config).await?;
+    let mut node_info = wake_compute(&mut num_retries, ctx, user_info).await?;
     if let Some(keys) = user_info.get_keys() {
         node_info.set_keys(keys);
     }
     node_info.allow_self_signed_compute = allow_self_signed_compute;
     // let mut node_info = credentials.get_node_info(ctx, user_info).await?;
     mechanism.update_connect_config(&mut node_info.config);
-    let retry_type = RetryType::ConnectToCompute;
 
     // try once
     let err = match mechanism
@@ -119,13 +114,6 @@ where
     {
         Ok(res) => {
             ctx.latency_timer.success();
-            Metrics::get().proxy.retries_metric.observe(
-                RetriesMetricGroup {
-                    outcome: ConnectOutcome::Success,
-                    retry_type,
-                },
-                num_retries.into(),
-            );
             return Ok(res);
         }
         Err(e) => e,
@@ -133,17 +121,10 @@ where
 
     error!(error = ?err, "could not connect to compute node");
 
-    let node_info = if !node_info.cached() || !err.should_retry_database_address() {
+    let node_info = if !node_info.cached() {
         // If we just recieved this from cplane and dodn't get it from cache, we shouldn't retry.
         // Do not need to retrieve a new node_info, just return the old one.
-        if !err.should_retry(num_retries, connect_to_compute_retry_config) {
-            Metrics::get().proxy.retries_metric.observe(
-                RetriesMetricGroup {
-                    outcome: ConnectOutcome::Failed,
-                    retry_type,
-                },
-                num_retries.into(),
-            );
+        if !err.should_retry(num_retries) {
             return Err(err.into());
         }
         node_info
@@ -151,8 +132,7 @@ where
         // if we failed to connect, it's likely that the compute node was suspended, wake a new compute node
         info!("compute node's state has likely changed; requesting a wake-up");
         let old_node_info = invalidate_cache(node_info);
-        let mut node_info =
-            wake_compute(&mut num_retries, ctx, user_info, wake_compute_retry_config).await?;
+        let mut node_info = wake_compute(&mut num_retries, ctx, user_info).await?;
         node_info.reuse_settings(old_node_info);
 
         mechanism.update_connect_config(&mut node_info.config);
@@ -171,40 +151,21 @@ where
         {
             Ok(res) => {
                 ctx.latency_timer.success();
-                Metrics::get().proxy.retries_metric.observe(
-                    RetriesMetricGroup {
-                        outcome: ConnectOutcome::Success,
-                        retry_type,
-                    },
-                    num_retries.into(),
-                );
-                info!(?num_retries, "connected to compute node after");
                 return Ok(res);
             }
             Err(e) => {
-                let retriable = e.should_retry(num_retries, connect_to_compute_retry_config);
+                let retriable = e.should_retry(num_retries);
                 if !retriable {
                     error!(error = ?e, num_retries, retriable, "couldn't connect to compute node");
-                    Metrics::get().proxy.retries_metric.observe(
-                        RetriesMetricGroup {
-                            outcome: ConnectOutcome::Failed,
-                            retry_type,
-                        },
-                        num_retries.into(),
-                    );
                     return Err(e.into());
                 }
                 warn!(error = ?e, num_retries, retriable, "couldn't connect to compute node");
             }
         }
 
-        let wait_duration = retry_after(num_retries, connect_to_compute_retry_config);
+        let wait_duration = retry_after(num_retries);
         num_retries += 1;
 
-        let pause = ctx
-            .latency_timer
-            .pause(crate::metrics::Waiting::RetryTimeout);
         time::sleep(wait_duration).await;
-        drop(pause);
     }
 }

proxy/src/proxy/retry.rs

@@ -1,18 +1,21 @@
-use crate::{compute, config::RetryConfig};
+use crate::compute;
 use std::{error::Error, io};
 use tokio::time;
 
+/// Number of times we should retry the `/proxy_wake_compute` http request.
+/// Retry duration is BASE_RETRY_WAIT_DURATION * RETRY_WAIT_EXPONENT_BASE ^ n, where n starts at 0
+pub const NUM_RETRIES_CONNECT: u32 = 16;
+const BASE_RETRY_WAIT_DURATION: time::Duration = time::Duration::from_millis(25);
+const RETRY_WAIT_EXPONENT_BASE: f64 = std::f64::consts::SQRT_2;
+
 pub trait ShouldRetry {
     fn could_retry(&self) -> bool;
-    fn should_retry(&self, num_retries: u32, config: RetryConfig) -> bool {
+    fn should_retry(&self, num_retries: u32) -> bool {
         match self {
-            _ if num_retries >= config.max_retries => false,
+            _ if num_retries >= NUM_RETRIES_CONNECT => false,
             err => err.could_retry(),
         }
     }
-    fn should_retry_database_address(&self) -> bool {
-        true
-    }
 }
 
 impl ShouldRetry for io::Error {
@@ -36,21 +39,6 @@ impl ShouldRetry for tokio_postgres::error::DbError {
                 | &SqlState::SQLCLIENT_UNABLE_TO_ESTABLISH_SQLCONNECTION,
         )
     }
-    fn should_retry_database_address(&self) -> bool {
-        use tokio_postgres::error::SqlState;
-        // Here are errors that happens after the user successfully authenticated to the database.
-        // TODO: there are pgbouncer errors that should be retried, but they are not listed here.
-        !matches!(
-            self.code(),
-            &SqlState::TOO_MANY_CONNECTIONS
-                | &SqlState::OUT_OF_MEMORY
-                | &SqlState::SYNTAX_ERROR
-                | &SqlState::T_R_SERIALIZATION_FAILURE
-                | &SqlState::INVALID_CATALOG_NAME
-                | &SqlState::INVALID_SCHEMA_NAME
-                | &SqlState::INVALID_PARAMETER_VALUE
-        )
-    }
 }
 
 impl ShouldRetry for tokio_postgres::Error {
@@ -63,15 +51,6 @@ impl ShouldRetry for tokio_postgres::Error {
             false
         }
     }
-    fn should_retry_database_address(&self) -> bool {
-        if let Some(io_err) = self.source().and_then(|x| x.downcast_ref()) {
-            io::Error::should_retry_database_address(io_err)
-        } else if let Some(db_err) = self.source().and_then(|x| x.downcast_ref()) {
-            tokio_postgres::error::DbError::should_retry_database_address(db_err)
-        } else {
-            true
-        }
-    }
 }
 
 impl ShouldRetry for compute::ConnectionError {
@@ -82,17 +61,8 @@ impl ShouldRetry for compute::ConnectionError {
             _ => false,
         }
     }
-    fn should_retry_database_address(&self) -> bool {
-        match self {
-            compute::ConnectionError::Postgres(err) => err.should_retry_database_address(),
-            compute::ConnectionError::CouldNotConnect(err) => err.should_retry_database_address(),
-            _ => true,
-        }
-    }
 }
 
-pub fn retry_after(num_retries: u32, config: RetryConfig) -> time::Duration {
-    config
-        .base_delay
-        .mul_f64(config.backoff_factor.powi((num_retries as i32) - 1))
+pub fn retry_after(num_retries: u32) -> time::Duration {
+    BASE_RETRY_WAIT_DURATION.mul_f64(RETRY_WAIT_EXPONENT_BASE.powi((num_retries as i32) - 1))
 }

proxy/src/proxy/tests.rs

@@ -10,13 +10,13 @@ use super::*;
 use crate::auth::backend::{
     ComputeCredentialKeys, ComputeCredentials, ComputeUserInfo, MaybeOwned, TestBackend,
 };
-use crate::config::{CertResolver, RetryConfig};
+use crate::config::CertResolver;
 use crate::console::caches::NodeInfoCache;
 use crate::console::messages::MetricsAuxInfo;
 use crate::console::provider::{CachedAllowedIps, CachedRoleSecret, ConsoleBackend};
 use crate::console::{self, CachedNodeInfo, NodeInfo};
 use crate::error::ErrorKind;
-use crate::proxy::retry::retry_after;
+use crate::proxy::retry::{retry_after, NUM_RETRIES_CONNECT};
 use crate::{http, sasl, scram, BranchId, EndpointId, ProjectId};
 use anyhow::{bail, Context};
 use async_trait::async_trait;
@@ -174,7 +174,7 @@ async fn dummy_proxy(
     tls: Option<TlsConfig>,
     auth: impl TestAuth + Send,
 ) -> anyhow::Result<()> {
-    let (client, _) = read_proxy_protocol(client).await?;
+    let client = WithClientIp::new(client);
     let mut stream = match handshake(client, tls.as_ref(), false).await? {
         HandshakeData::Startup(stream, _) => stream,
         HandshakeData::Cancel(_) => bail!("cancellation not supported"),
@@ -361,15 +361,11 @@ async fn scram_auth_mock() -> anyhow::Result<()> {
 #[test]
 fn connect_compute_total_wait() {
     let mut total_wait = tokio::time::Duration::ZERO;
-    let config = RetryConfig {
-        base_delay: Duration::from_secs(1),
-        max_retries: 5,
-        backoff_factor: 2.0,
-    };
-    for num_retries in 1..config.max_retries {
-        total_wait += retry_after(num_retries, config);
+    for num_retries in 1..NUM_RETRIES_CONNECT {
+        total_wait += retry_after(num_retries);
     }
-    assert!(f64::abs(total_wait.as_secs_f64() - 15.0) < 0.1);
+    assert!(total_wait < tokio::time::Duration::from_secs(12));
+    assert!(total_wait > tokio::time::Duration::from_secs(10));
 }
 
 #[derive(Clone, Copy, Debug)]
@@ -553,12 +549,7 @@ async fn connect_to_compute_success() {
     let mut ctx = RequestMonitoring::test();
     let mechanism = TestConnectMechanism::new(vec![Wake, Connect]);
     let user_info = helper_create_connect_info(&mechanism);
-    let config = RetryConfig {
-        base_delay: Duration::from_secs(1),
-        max_retries: 5,
-        backoff_factor: 2.0,
-    };
-    connect_to_compute(&mut ctx, &mechanism, &user_info, false, config, config)
+    connect_to_compute(&mut ctx, &mechanism, &user_info, false)
         .await
         .unwrap();
     mechanism.verify();
@@ -571,12 +562,7 @@ async fn connect_to_compute_retry() {
     let mut ctx = RequestMonitoring::test();
     let mechanism = TestConnectMechanism::new(vec![Wake, Retry, Wake, Connect]);
     let user_info = helper_create_connect_info(&mechanism);
-    let config = RetryConfig {
-        base_delay: Duration::from_secs(1),
-        max_retries: 5,
-        backoff_factor: 2.0,
-    };
-    connect_to_compute(&mut ctx, &mechanism, &user_info, false, config, config)
+    connect_to_compute(&mut ctx, &mechanism, &user_info, false)
         .await
         .unwrap();
     mechanism.verify();
@@ -590,12 +576,7 @@ async fn connect_to_compute_non_retry_1() {
     let mut ctx = RequestMonitoring::test();
     let mechanism = TestConnectMechanism::new(vec![Wake, Retry, Wake, Fail]);
     let user_info = helper_create_connect_info(&mechanism);
-    let config = RetryConfig {
-        base_delay: Duration::from_secs(1),
-        max_retries: 5,
-        backoff_factor: 2.0,
-    };
-    connect_to_compute(&mut ctx, &mechanism, &user_info, false, config, config)
+    connect_to_compute(&mut ctx, &mechanism, &user_info, false)
         .await
         .unwrap_err();
     mechanism.verify();
@@ -609,12 +590,7 @@ async fn connect_to_compute_non_retry_2() {
     let mut ctx = RequestMonitoring::test();
     let mechanism = TestConnectMechanism::new(vec![Wake, Fail, Wake, Connect]);
     let user_info = helper_create_connect_info(&mechanism);
-    let config = RetryConfig {
-        base_delay: Duration::from_secs(1),
-        max_retries: 5,
-        backoff_factor: 2.0,
-    };
-    connect_to_compute(&mut ctx, &mechanism, &user_info, false, config, config)
+    connect_to_compute(&mut ctx, &mechanism, &user_info, false)
         .await
         .unwrap();
     mechanism.verify();
@@ -624,32 +600,17 @@ async fn connect_to_compute_non_retry_2() {
 #[tokio::test]
 async fn connect_to_compute_non_retry_3() {
     let _ = env_logger::try_init();
-    tokio::time::pause();
+    assert_eq!(NUM_RETRIES_CONNECT, 16);
     use ConnectAction::*;
     let mut ctx = RequestMonitoring::test();
-    let mechanism =
-        TestConnectMechanism::new(vec![Wake, Retry, Wake, Retry, Retry, Retry, Retry, Retry]);
+    let mechanism = TestConnectMechanism::new(vec![
+        Wake, Retry, Wake, Retry, Retry, Retry, Retry, Retry, Retry, Retry, Retry, Retry, Retry,
+        Retry, Retry, Retry, Retry, Retry, /* the 17th time */ Retry,
+    ]);
     let user_info = helper_create_connect_info(&mechanism);
-    let wake_compute_retry_config = RetryConfig {
-        base_delay: Duration::from_secs(1),
-        max_retries: 1,
-        backoff_factor: 2.0,
-    };
-    let connect_to_compute_retry_config = RetryConfig {
-        base_delay: Duration::from_secs(1),
-        max_retries: 5,
-        backoff_factor: 2.0,
-    };
-    connect_to_compute(
-        &mut ctx,
-        &mechanism,
-        &user_info,
-        false,
-        wake_compute_retry_config,
-        connect_to_compute_retry_config,
-    )
-    .await
-    .unwrap_err();
+    connect_to_compute(&mut ctx, &mechanism, &user_info, false)
+        .await
+        .unwrap_err();
     mechanism.verify();
 }
 
@@ -661,12 +622,7 @@ async fn wake_retry() {
     let mut ctx = RequestMonitoring::test();
     let mechanism = TestConnectMechanism::new(vec![WakeRetry, Wake, Connect]);
     let user_info = helper_create_connect_info(&mechanism);
-    let config = RetryConfig {
-        base_delay: Duration::from_secs(1),
-        max_retries: 5,
-        backoff_factor: 2.0,
-    };
-    connect_to_compute(&mut ctx, &mechanism, &user_info, false, config, config)
+    connect_to_compute(&mut ctx, &mechanism, &user_info, false)
         .await
         .unwrap();
     mechanism.verify();
@@ -680,12 +636,7 @@ async fn wake_non_retry() {
     let mut ctx = RequestMonitoring::test();
     let mechanism = TestConnectMechanism::new(vec![WakeRetry, WakeFail]);
     let user_info = helper_create_connect_info(&mechanism);
-    let config = RetryConfig {
-        base_delay: Duration::from_secs(1),
-        max_retries: 5,
-        backoff_factor: 2.0,
-    };
-    connect_to_compute(&mut ctx, &mechanism, &user_info, false, config, config)
+    connect_to_compute(&mut ctx, &mechanism, &user_info, false)
         .await
         .unwrap_err();
     mechanism.verify();

proxy/src/proxy/wake_compute.rs

@@ -1,14 +1,10 @@
-use crate::config::RetryConfig;
 use crate::console::{errors::WakeComputeError, provider::CachedNodeInfo};
 use crate::context::RequestMonitoring;
-use crate::metrics::{
-    ConnectOutcome, ConnectionFailuresBreakdownGroup, Metrics, RetriesMetricGroup, RetryType,
-    WakeupFailureKind,
-};
+use crate::metrics::{ConnectionFailuresBreakdownGroup, Metrics, WakeupFailureKind};
 use crate::proxy::retry::retry_after;
 use hyper::StatusCode;
 use std::ops::ControlFlow;
-use tracing::{error, info, warn};
+use tracing::{error, warn};
 
 use super::connect_compute::ComputeConnectBackend;
 use super::retry::ShouldRetry;
@@ -17,48 +13,25 @@ pub async fn wake_compute<B: ComputeConnectBackend>(
     num_retries: &mut u32,
     ctx: &mut RequestMonitoring,
     api: &B,
-    config: RetryConfig,
 ) -> Result<CachedNodeInfo, WakeComputeError> {
-    let retry_type = RetryType::WakeCompute;
     loop {
         let wake_res = api.wake_compute(ctx).await;
-        match handle_try_wake(wake_res, *num_retries, config) {
+        match handle_try_wake(wake_res, *num_retries) {
             Err(e) => {
                 error!(error = ?e, num_retries, retriable = false, "couldn't wake compute node");
                 report_error(&e, false);
-                Metrics::get().proxy.retries_metric.observe(
-                    RetriesMetricGroup {
-                        outcome: ConnectOutcome::Failed,
-                        retry_type,
-                    },
-                    (*num_retries).into(),
-                );
                 return Err(e);
             }
             Ok(ControlFlow::Continue(e)) => {
                 warn!(error = ?e, num_retries, retriable = true, "couldn't wake compute node");
                 report_error(&e, true);
             }
-            Ok(ControlFlow::Break(n)) => {
-                Metrics::get().proxy.retries_metric.observe(
-                    RetriesMetricGroup {
-                        outcome: ConnectOutcome::Success,
-                        retry_type,
-                    },
-                    (*num_retries).into(),
-                );
-                info!(?num_retries, "compute node woken up after");
-                return Ok(n);
-            }
+            Ok(ControlFlow::Break(n)) => return Ok(n),
         }
 
-        let wait_duration = retry_after(*num_retries, config);
+        let wait_duration = retry_after(*num_retries);
         *num_retries += 1;
-        let pause = ctx
-            .latency_timer
-            .pause(crate::metrics::Waiting::RetryTimeout);
         tokio::time::sleep(wait_duration).await;
-        drop(pause);
     }
 }
 
@@ -69,11 +42,10 @@ pub async fn wake_compute<B: ComputeConnectBackend>(
 pub fn handle_try_wake(
     result: Result<CachedNodeInfo, WakeComputeError>,
     num_retries: u32,
-    config: RetryConfig,
 ) -> Result<ControlFlow<CachedNodeInfo, WakeComputeError>, WakeComputeError> {
     match result {
         Err(err) => match &err {
-            WakeComputeError::ApiError(api) if api.should_retry(num_retries, config) => {
+            WakeComputeError::ApiError(api) if api.should_retry(num_retries) => {
                 Ok(ControlFlow::Continue(err))
             }
             _ => Err(err),
@@ -118,7 +90,6 @@ fn report_error(e: &WakeComputeError, retry: bool) {
         WakeComputeError::ApiError(ApiError::Console { .. }) => {
             WakeupFailureKind::ApiConsoleOtherError
         }
-        WakeComputeError::TooManyConnections => WakeupFailureKind::ApiConsoleLocked,
         WakeComputeError::TimeoutError => WakeupFailureKind::TimeoutError,
     };
     Metrics::get()

proxy/src/rate_limiter/limiter.rs

@@ -15,7 +15,7 @@ use rand::{rngs::StdRng, Rng, SeedableRng};
 use tokio::time::{Duration, Instant};
 use tracing::info;
 
-use crate::intern::EndpointIdInt;
+use crate::EndpointId;
 
 pub struct GlobalRateLimiter {
     data: Vec<RateBucket>,
@@ -61,7 +61,12 @@ impl GlobalRateLimiter {
 // Purposefully ignore user name and database name as clients can reconnect
 // with different names, so we'll end up sending some http requests to
 // the control plane.
-pub type EndpointRateLimiter = BucketRateLimiter<EndpointIdInt, StdRng, RandomState>;
+//
+// We also may save quite a lot of CPU (I think) by bailing out right after we
+// saw SNI, before doing TLS handshake. User-side error messages in that case
+// does not look very nice (`SSL SYSCALL error: Undefined error: 0`), so for now
+// I went with a more expensive way that yields user-friendlier error messages.
+pub type EndpointRateLimiter = BucketRateLimiter<EndpointId, StdRng, RandomState>;
 
 pub struct BucketRateLimiter<Key, Rand = StdRng, Hasher = RandomState> {
     map: DashMap<Key, Vec<RateBucket>, Hasher>,
@@ -240,7 +245,7 @@ mod tests {
     use tokio::time;
 
     use super::{BucketRateLimiter, EndpointRateLimiter};
-    use crate::{intern::EndpointIdInt, rate_limiter::RateBucketInfo, EndpointId};
+    use crate::{rate_limiter::RateBucketInfo, EndpointId};
 
     #[test]
     fn rate_bucket_rpi() {
@@ -290,40 +295,39 @@ mod tests {
         let limiter = EndpointRateLimiter::new(rates);
 
         let endpoint = EndpointId::from("ep-my-endpoint-1234");
-        let endpoint = EndpointIdInt::from(endpoint);
 
         time::pause();
 
         for _ in 0..100 {
-            assert!(limiter.check(endpoint, 1));
+            assert!(limiter.check(endpoint.clone(), 1));
         }
         // more connections fail
-        assert!(!limiter.check(endpoint, 1));
+        assert!(!limiter.check(endpoint.clone(), 1));
 
         // fail even after 500ms as it's in the same bucket
         time::advance(time::Duration::from_millis(500)).await;
-        assert!(!limiter.check(endpoint, 1));
+        assert!(!limiter.check(endpoint.clone(), 1));
 
         // after a full 1s, 100 requests are allowed again
         time::advance(time::Duration::from_millis(500)).await;
         for _ in 1..6 {
             for _ in 0..50 {
-                assert!(limiter.check(endpoint, 2));
+                assert!(limiter.check(endpoint.clone(), 2));
             }
             time::advance(time::Duration::from_millis(1000)).await;
         }
 
         // more connections after 600 will exceed the 20rps@30s limit
-        assert!(!limiter.check(endpoint, 1));
+        assert!(!limiter.check(endpoint.clone(), 1));
 
         // will still fail before the 30 second limit
         time::advance(time::Duration::from_millis(30_000 - 6_000 - 1)).await;
-        assert!(!limiter.check(endpoint, 1));
+        assert!(!limiter.check(endpoint.clone(), 1));
 
         // after the full 30 seconds, 100 requests are allowed again
         time::advance(time::Duration::from_millis(1)).await;
         for _ in 0..100 {
-            assert!(limiter.check(endpoint, 1));
+            assert!(limiter.check(endpoint.clone(), 1));
         }
     }
 

proxy/src/serverless.rs

@@ -33,8 +33,9 @@ use crate::cancellation::CancellationHandlerMain;
 use crate::config::ProxyConfig;
 use crate::context::RequestMonitoring;
 use crate::metrics::Metrics;
-use crate::protocol2::read_proxy_protocol;
+use crate::protocol2::WithClientIp;
 use crate::proxy::run_until_cancelled;
+use crate::rate_limiter::EndpointRateLimiter;
 use crate::serverless::backend::PoolingBackend;
 use crate::serverless::http_util::{api_error_into_response, json_response};
 
@@ -52,6 +53,7 @@ pub async fn task_main(
     config: &'static ProxyConfig,
     ws_listener: TcpListener,
     cancellation_token: CancellationToken,
+    endpoint_rate_limiter: Arc<EndpointRateLimiter>,
     cancellation_handler: Arc<CancellationHandlerMain>,
 ) -> anyhow::Result<()> {
     scopeguard::defer! {
@@ -115,6 +117,7 @@ pub async fn task_main(
                 backend.clone(),
                 connections.clone(),
                 cancellation_handler.clone(),
+                endpoint_rate_limiter.clone(),
                 cancellation_token.clone(),
                 server.clone(),
                 tls_acceptor.clone(),
@@ -144,6 +147,7 @@ async fn connection_handler(
     backend: Arc<PoolingBackend>,
     connections: TaskTracker,
     cancellation_handler: Arc<CancellationHandlerMain>,
+    endpoint_rate_limiter: Arc<EndpointRateLimiter>,
     cancellation_token: CancellationToken,
     server: Builder<TokioExecutor>,
     tls_acceptor: TlsAcceptor,
@@ -158,8 +162,9 @@ async fn connection_handler(
         .guard(crate::metrics::Protocol::Http);
 
     // handle PROXY protocol
-    let (conn, peer) = match read_proxy_protocol(conn).await {
-        Ok(c) => c,
+    let mut conn = WithClientIp::new(conn);
+    let peer = match conn.wait_for_addr().await {
+        Ok(peer) => peer,
         Err(e) => {
             tracing::error!(?session_id, %peer_addr, "failed to accept TCP connection: invalid PROXY protocol V2 header: {e:#}");
             return;
@@ -226,6 +231,7 @@ async fn connection_handler(
                     cancellation_handler.clone(),
                     session_id,
                     peer_addr,
+                    endpoint_rate_limiter.clone(),
                     http_request_token,
                 )
                 .in_current_span()
@@ -264,6 +270,7 @@ async fn request_handler(
     cancellation_handler: Arc<CancellationHandlerMain>,
     session_id: uuid::Uuid,
     peer_addr: IpAddr,
+    endpoint_rate_limiter: Arc<EndpointRateLimiter>,
     // used to cancel in-flight HTTP requests. not used to cancel websockets
     http_cancellation_token: CancellationToken,
 ) -> Result<Response<Full<Bytes>>, ApiError> {
@@ -291,9 +298,15 @@ async fn request_handler(
 
         ws_connections.spawn(
             async move {
-                if let Err(e) =
-                    websocket::serve_websocket(config, ctx, websocket, cancellation_handler, host)
-                        .await
+                if let Err(e) = websocket::serve_websocket(
+                    config,
+                    ctx,
+                    websocket,
+                    cancellation_handler,
+                    host,
+                    endpoint_rate_limiter,
+                )
+                .await
                 {
                     error!("error in websocket connection: {e:#}");
                 }

proxy/src/serverless/backend.rs

@@ -108,8 +108,6 @@ impl PoolingBackend {
             },
             &backend,
             false, // do not allow self signed compute for http flow
-            self.config.wake_compute_retry_config,
-            self.config.connect_to_compute_retry_config,
         )
         .await
     }
@@ -179,9 +177,7 @@ impl ConnectMechanism for TokioMechanism {
             .dbname(&self.conn_info.dbname)
             .connect_timeout(timeout);
 
-        let pause = ctx.latency_timer.pause(crate::metrics::Waiting::Compute);
         let (client, connection) = config.connect(tokio_postgres::NoTls).await?;
-        drop(pause);
 
         tracing::Span::current().record("pid", &tracing::field::display(client.get_process_id()));
         Ok(poll_client(