Add new compaction abstraction, simulator, and implementation.

This consists of three parts:

1. A refactoring and new contract for implementing and testing
compaction.

The logic is now in a separate crate, with no dependency on the
'pageserver' crate. It defines an interface that the real pageserver
must implement, in order to call the compaction algorithm. The
interface models things like delta and image layers, but just the
parts that the compaction algorithm needs to make decisions. That
makes it easier unit test the algorithm and experiment with different
implementations.

I did not convert the current code to the new abstraction, however.
When compaction algorithm is set to "Legacy", we just use the old
code. It might be worthwhile to convert the old code to the new
abstraction, so that we can compare the behavior of the new algorithm
against the old one, using the same simulated cases. If we do that,
have to be careful that the converted code really is equivalent to the
old.

This inclues only trivial changes to the main pageserver code. All the
new code is behind a tenant config option. So this should be pretty
safe to merge, even if the new implementation is buggy, as long as we
don't enable it.

2. A new compaction algorithm, implemented using the new
abstraction.

The new algorithm is tiered compaction.  It is inspired by the PoC at
PR #4539, although I did not use that code directly, as I needed the
new implementation to fit the new abstraction. The algorithm here is
less advanced, I did not implement partial image layers, for example.
I wanted to keep it simple on purpose, so that as we add bells and
whistles, we can see the effects using the included simulator.

One difference to #4539 and your typical LSM tree implementations is
how we keep track of the LSM tree levels. This PR doesn't have a
permanent concept of a level, tier or sorted run at all. There are
just delta and image layers. However, when compaction starts, we look
at the layers that exist, and arrange them into levels, depending on
their shapes. That is ephemeral: when the compaction finishes, we
forget that information. This allows the new algorithm to work without
any extra bookkeeping. That makes it easier to transition from the old
algorithm to new, and back again.

There is just a new tenant config option to choose the compaction
algorithm. The default is "Legacy", meaning the current algorithm in
'main'. If you set it to "Tiered".

3. A simulator, which implements the new abstraction.

The simulator can be used to analyze write and storage amplification,
without running a test with the full pageserver. It can also draw an
SVG animation of the simulation, to visualize how layers are created
and deleted.

To run the simulator:

    ./target/debug/compaction-simulator run-suite

.config/hakari.toml

@@ -22,11 +22,5 @@ platforms = [
     # "x86_64-pc-windows-msvc",
 ]
 
-[final-excludes]
-# vm_monitor benefits from the same Cargo.lock as the rest of our artifacts, but
-# it is built primarly in separate repo neondatabase/autoscaling and thus is excluded
-# from depending on workspace-hack because most of the dependencies are not used.
-workspace-members = ["vm_monitor"]
-
 # Write out exact versions rather than a semver range. (Defaults to false.)
 # exact-versions = true

.github/workflows/build_and_test.yml

@@ -723,7 +723,6 @@ jobs:
                            --cache-repo 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cache
                            --context .
                            --build-arg GIT_VERSION=${{ github.event.pull_request.head.sha || github.sha }}
-                           --build-arg BUILD_TAG=${{ needs.tag.outputs.build-tag }}
                            --build-arg REPOSITORY=369495373322.dkr.ecr.eu-central-1.amazonaws.com
                            --destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/neon:${{needs.tag.outputs.build-tag}}
                            --destination neondatabase/neon:${{needs.tag.outputs.build-tag}}

Cargo.lock

@@ -170,12 +170,6 @@ dependencies = [
  "backtrace",
 ]
 
-[[package]]
-name = "arc-swap"
-version = "1.6.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "bddcadddf5e9015d310179a59bb28c4d4b9920ad0f11e8e14dbadf654890c9a6"
-
 [[package]]
 name = "archery"
 version = "0.5.0"
@@ -2927,16 +2921,6 @@ dependencies = [
  "winapi",
 ]
 
-[[package]]
-name = "nu-ansi-term"
-version = "0.46.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "77a8165726e8236064dbb45459242600304b42a5ea24ee2948e18e023bf7ba84"
-dependencies = [
- "overload",
- "winapi",
-]
-
 [[package]]
 name = "num-bigint"
 version = "0.4.3"
@@ -3203,12 +3187,6 @@ version = "0.5.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4030760ffd992bef45b0ae3f10ce1aba99e33464c90d14dd7c039884963ddc7a"
 
-[[package]]
-name = "overload"
-version = "0.1.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b15813163c1d831bf4a13c3610c05c0d03b39feb07f7e09fa234dac9b15aaf39"
-
 [[package]]
 name = "pagectl"
 version = "0.1.0"
@@ -3265,6 +3243,7 @@ dependencies = [
  "num_cpus",
  "once_cell",
  "pageserver_api",
+ "pageserver_compaction",
  "pin-project-lite",
  "postgres",
  "postgres-protocol",
@@ -3294,12 +3273,10 @@ dependencies = [
  "tokio",
  "tokio-io-timeout",
  "tokio-postgres",
- "tokio-stream",
  "tokio-tar",
  "tokio-util",
  "toml_edit",
  "tracing",
- "tracing-subscriber",
  "url",
  "utils",
  "walkdir",
@@ -3325,6 +3302,52 @@ dependencies = [
  "workspace_hack",
 ]
 
+[[package]]
+name = "pageserver_compaction"
+version = "0.1.0"
+dependencies = [
+ "anyhow",
+ "async-compression",
+ "async-stream",
+ "async-trait",
+ "byteorder",
+ "bytes",
+ "chrono",
+ "clap",
+ "const_format",
+ "consumption_metrics",
+ "criterion",
+ "crossbeam-utils",
+ "either",
+ "fail",
+ "flate2",
+ "futures",
+ "git-version",
+ "hex",
+ "hex-literal",
+ "humantime",
+ "humantime-serde",
+ "itertools",
+ "metrics",
+ "once_cell",
+ "pin-project-lite",
+ "rand 0.8.5",
+ "smallvec",
+ "svg_fmt",
+ "sync_wrapper",
+ "thiserror",
+ "tokio",
+ "tokio-io-timeout",
+ "tokio-util",
+ "tracing",
+ "tracing-error",
+ "tracing-subscriber",
+ "url",
+ "utils",
+ "walkdir",
+ "workspace_hack",
+]
+
 [[package]]
 name = "parking"
 version = "2.1.1"
@@ -3574,7 +3597,7 @@ dependencies = [
 [[package]]
 name = "postgres"
 version = "0.19.4"
-source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#ef8559b5f60f5c1d2b0184a62f49035600824518"
+source = "git+https://github.com/neondatabase/rust-postgres.git?rev=ce7260db5998fe27167da42503905a12e7ad9048#ce7260db5998fe27167da42503905a12e7ad9048"
 dependencies = [
  "bytes",
  "fallible-iterator",
@@ -3587,7 +3610,7 @@ dependencies = [
 [[package]]
 name = "postgres-native-tls"
 version = "0.5.0"
-source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#ef8559b5f60f5c1d2b0184a62f49035600824518"
+source = "git+https://github.com/neondatabase/rust-postgres.git?rev=ce7260db5998fe27167da42503905a12e7ad9048#ce7260db5998fe27167da42503905a12e7ad9048"
 dependencies = [
  "native-tls",
  "tokio",
@@ -3598,7 +3621,7 @@ dependencies = [
 [[package]]
 name = "postgres-protocol"
 version = "0.6.4"
-source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#ef8559b5f60f5c1d2b0184a62f49035600824518"
+source = "git+https://github.com/neondatabase/rust-postgres.git?rev=ce7260db5998fe27167da42503905a12e7ad9048#ce7260db5998fe27167da42503905a12e7ad9048"
 dependencies = [
  "base64 0.20.0",
  "byteorder",
@@ -3616,7 +3639,7 @@ dependencies = [
 [[package]]
 name = "postgres-types"
 version = "0.2.4"
-source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#ef8559b5f60f5c1d2b0184a62f49035600824518"
+source = "git+https://github.com/neondatabase/rust-postgres.git?rev=ce7260db5998fe27167da42503905a12e7ad9048#ce7260db5998fe27167da42503905a12e7ad9048"
 dependencies = [
  "bytes",
  "fallible-iterator",
@@ -4705,16 +4728,6 @@ dependencies = [
  "serde_derive",
 ]
 
-[[package]]
-name = "serde_assert"
-version = "0.5.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "eda563240c1288b044209be1f0d38bb4d15044fb3e00dc354fbc922ab4733e80"
-dependencies = [
- "hashbrown 0.13.2",
- "serde",
-]
-
 [[package]]
 name = "serde_derive"
 version = "1.0.183"
@@ -5432,7 +5445,7 @@ dependencies = [
 [[package]]
 name = "tokio-postgres"
 version = "0.7.7"
-source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#ef8559b5f60f5c1d2b0184a62f49035600824518"
+source = "git+https://github.com/neondatabase/rust-postgres.git?rev=ce7260db5998fe27167da42503905a12e7ad9048#ce7260db5998fe27167da42503905a12e7ad9048"
 dependencies = [
  "async-trait",
  "byteorder",
@@ -5789,7 +5802,6 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "30a651bc37f915e81f087d86e62a18eec5f79550c7faff886f7090b4ea757c77"
 dependencies = [
  "matchers",
- "nu-ansi-term",
  "once_cell",
  "regex",
  "serde",
@@ -5976,7 +5988,6 @@ name = "utils"
 version = "0.1.0"
 dependencies = [
  "anyhow",
- "arc-swap",
  "async-trait",
  "bincode",
  "byteorder",
@@ -6003,7 +6014,6 @@ dependencies = [
  "routerify",
  "sentry",
  "serde",
- "serde_assert",
  "serde_json",
  "serde_with",
  "signal-hook",
@@ -6074,6 +6084,7 @@ dependencies = [
  "tokio-util",
  "tracing",
  "tracing-subscriber",
+ "workspace_hack",
 ]
 
 [[package]]

Cargo.toml

@@ -4,6 +4,7 @@ members = [
     "compute_tools",
     "control_plane",
     "pageserver",
+    "pageserver/compaction",
     "pageserver/ctl",
     "proxy",
     "safekeeper",
@@ -36,7 +37,6 @@ license = "Apache-2.0"
 ## All dependency versions, used in the project
 [workspace.dependencies]
 anyhow = { version = "1.0", features = ["backtrace"] }
-arc-swap = "1.6"
 async-compression = { version = "0.4.0", features = ["tokio", "gzip"] }
 azure_core = "0.16"
 azure_identity = "0.16"
@@ -125,7 +125,6 @@ sentry = { version = "0.31", default-features = false, features = ["backtrace",
 serde = { version = "1.0", features = ["derive"] }
 serde_json = "1"
 serde_with = "2.0"
-serde_assert = "0.5.0"
 sha2 = "0.10.2"
 signal-hook = "0.3"
 smallvec = "1.11"
@@ -163,11 +162,11 @@ env_logger = "0.10"
 log = "0.4"
 
 ## Libraries from neondatabase/ git forks, ideally with changes to be upstreamed
-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", branch="neon" }
-postgres-native-tls = { git = "https://github.com/neondatabase/rust-postgres.git", branch="neon" }
-postgres-protocol = { git = "https://github.com/neondatabase/rust-postgres.git", branch="neon" }
-postgres-types = { git = "https://github.com/neondatabase/rust-postgres.git", branch="neon" }
-tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", branch="neon" }
+postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="ce7260db5998fe27167da42503905a12e7ad9048" }
+postgres-native-tls = { git = "https://github.com/neondatabase/rust-postgres.git", rev="ce7260db5998fe27167da42503905a12e7ad9048" }
+postgres-protocol = { git = "https://github.com/neondatabase/rust-postgres.git", rev="ce7260db5998fe27167da42503905a12e7ad9048" }
+postgres-types = { git = "https://github.com/neondatabase/rust-postgres.git", rev="ce7260db5998fe27167da42503905a12e7ad9048" }
+tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="ce7260db5998fe27167da42503905a12e7ad9048" }
 
 ## Other git libraries
 heapless = { default-features=false, features=[], git = "https://github.com/japaric/heapless.git", rev = "644653bf3b831c6bb4963be2de24804acf5e5001" } # upstream release pending
@@ -177,6 +176,7 @@ compute_api = { version = "0.1", path = "./libs/compute_api/" }
 consumption_metrics = { version = "0.1", path = "./libs/consumption_metrics/" }
 metrics = { version = "0.1", path = "./libs/metrics/" }
 pageserver_api = { version = "0.1", path = "./libs/pageserver_api/" }
+pageserver_compaction = { version = "0.1", path = "./pageserver/compaction/" }
 postgres_backend = { version = "0.1", path = "./libs/postgres_backend/" }
 postgres_connection = { version = "0.1", path = "./libs/postgres_connection/" }
 postgres_ffi = { version = "0.1", path = "./libs/postgres_ffi/" }
@@ -204,7 +204,7 @@ tonic-build = "0.9"
 
 # This is only needed for proxy's tests.
 # TODO: we should probably fork `tokio-postgres-rustls` instead.
-tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", branch="neon" }
+tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="ce7260db5998fe27167da42503905a12e7ad9048" }
 
 ################# Binary contents sections
 

Dockerfile

@@ -27,7 +27,6 @@ RUN set -e \
 FROM $REPOSITORY/$IMAGE:$TAG AS build
 WORKDIR /home/nonroot
 ARG GIT_VERSION=local
-ARG BUILD_TAG
 
 # Enable https://github.com/paritytech/cachepot to cache Rust crates' compilation results in Docker builds.
 # Set up cachepot to use an AWS S3 bucket for cache results, to reuse it between `docker build` invocations.
@@ -79,9 +78,9 @@ COPY --from=build --chown=neon:neon /home/nonroot/target/release/pg_sni_router
 COPY --from=build --chown=neon:neon /home/nonroot/target/release/pageserver          /usr/local/bin
 COPY --from=build --chown=neon:neon /home/nonroot/target/release/pagectl             /usr/local/bin
 COPY --from=build --chown=neon:neon /home/nonroot/target/release/safekeeper          /usr/local/bin
-COPY --from=build --chown=neon:neon /home/nonroot/target/release/storage_broker      /usr/local/bin
+COPY --from=build --chown=neon:neon /home/nonroot/target/release/storage_broker         /usr/local/bin
 COPY --from=build --chown=neon:neon /home/nonroot/target/release/proxy               /usr/local/bin
-COPY --from=build --chown=neon:neon /home/nonroot/target/release/neon_local          /usr/local/bin
+COPY --from=build --chown=neon:neon /home/nonroot/target/release/neon_local               /usr/local/bin
 
 COPY --from=pg-build /home/nonroot/pg_install/v14 /usr/local/v14/
 COPY --from=pg-build /home/nonroot/pg_install/v15 /usr/local/v15/

Makefile

@@ -72,10 +72,6 @@ neon: postgres-headers walproposer-lib
 #
 $(POSTGRES_INSTALL_DIR)/build/%/config.status:
 	+@echo "Configuring Postgres $* build"
-	@test -s $(ROOT_PROJECT_DIR)/vendor/postgres-$*/configure || { \
-		echo "\nPostgres submodule not found in $(ROOT_PROJECT_DIR)/vendor/postgres-$*/, execute "; \
-		echo "'git submodule update --init --recursive --depth 2 --progress .' in project root.\n"; \
-		exit 1; }
 	mkdir -p $(POSTGRES_INSTALL_DIR)/build/$*
 	(cd $(POSTGRES_INSTALL_DIR)/build/$* && \
 	env PATH="$(EXTRA_PATH_OVERRIDES):$$PATH" $(ROOT_PROJECT_DIR)/vendor/postgres-$*/configure \

compute_tools/src/compute.rs

@@ -710,12 +710,8 @@ impl ComputeNode {
     // `pg_ctl` for start / stop, so this just seems much easier to do as we already
     // have opened connection to Postgres and superuser access.
     #[instrument(skip_all)]
-    fn pg_reload_conf(&self) -> Result<()> {
-        let pgctl_bin = Path::new(&self.pgbin).parent().unwrap().join("pg_ctl");
-        Command::new(pgctl_bin)
-            .args(["reload", "-D", &self.pgdata])
-            .output()
-            .expect("cannot run pg_ctl process");
+    fn pg_reload_conf(&self, client: &mut Client) -> Result<()> {
+        client.simple_query("SELECT pg_reload_conf()")?;
         Ok(())
     }
 
@@ -728,9 +724,9 @@ impl ComputeNode {
         // Write new config
         let pgdata_path = Path::new(&self.pgdata);
         config::write_postgres_conf(&pgdata_path.join("postgresql.conf"), &spec, None)?;
-        self.pg_reload_conf()?;
 
         let mut client = Client::connect(self.connstr.as_str(), NoTls)?;
+        self.pg_reload_conf(&mut client)?;
 
         // Proceed with post-startup configuration. Note, that order of operations is important.
         // Disable DDL forwarding because control plane already knows about these roles/databases.

compute_tools/src/lib.rs

@@ -1,7 +1,7 @@
+//!
 //! Various tools and helpers to handle cluster / compute node (Postgres)
 //! configuration.
-#![deny(unsafe_code)]
-#![deny(clippy::undocumented_unsafe_blocks)]
+//!
 pub mod checker;
 pub mod config;
 pub mod configurator;

control_plane/src/attachment_service.rs

@@ -2,6 +2,7 @@ use crate::{background_process, local_env::LocalEnv};
 use anyhow::anyhow;
 use camino::Utf8PathBuf;
 use serde::{Deserialize, Serialize};
+use serde_with::{serde_as, DisplayFromStr};
 use std::{path::PathBuf, process::Child};
 use utils::id::{NodeId, TenantId};
 
@@ -13,8 +14,10 @@ pub struct AttachmentService {
 
 const COMMAND: &str = "attachment_service";
 
+#[serde_as]
 #[derive(Serialize, Deserialize)]
 pub struct AttachHookRequest {
+    #[serde_as(as = "DisplayFromStr")]
     pub tenant_id: TenantId,
     pub node_id: Option<NodeId>,
 }

control_plane/src/background_process.rs

@@ -262,7 +262,7 @@ where
     P: Into<Utf8PathBuf>,
 {
     let path: Utf8PathBuf = path.into();
-    // SAFETY:
+    // SAFETY
     // pre_exec is marked unsafe because it runs between fork and exec.
     // Why is that dangerous in various ways?
     // Long answer:  https://github.com/rust-lang/rust/issues/39575

control_plane/src/endpoint.rs

@@ -46,6 +46,7 @@ use std::time::Duration;
 
 use anyhow::{anyhow, bail, Context, Result};
 use serde::{Deserialize, Serialize};
+use serde_with::{serde_as, DisplayFromStr};
 use utils::id::{NodeId, TenantId, TimelineId};
 
 use crate::local_env::LocalEnv;
@@ -56,10 +57,13 @@ use compute_api::responses::{ComputeState, ComputeStatus};
 use compute_api::spec::{Cluster, ComputeMode, ComputeSpec};
 
 // contents of a endpoint.json file
+#[serde_as]
 #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Debug)]
 pub struct EndpointConf {
     endpoint_id: String,
+    #[serde_as(as = "DisplayFromStr")]
     tenant_id: TenantId,
+    #[serde_as(as = "DisplayFromStr")]
     timeline_id: TimelineId,
     mode: ComputeMode,
     pg_port: u16,

control_plane/src/lib.rs

@@ -1,10 +1,11 @@
-//! Local control plane.
-//!
-//! Can start, configure and stop postgres instances running as a local processes.
-//!
-//! Intended to be used in integration tests and in CLI tools for
-//! local installations.
-#![deny(clippy::undocumented_unsafe_blocks)]
+//
+// Local control plane.
+//
+// Can start, configure and stop postgres instances running as a local processes.
+//
+// Intended to be used in integration tests and in CLI tools for
+// local installations.
+//
 
 pub mod attachment_service;
 mod background_process;

control_plane/src/local_env.rs

@@ -8,6 +8,7 @@ use anyhow::{bail, ensure, Context};
 use postgres_backend::AuthType;
 use reqwest::Url;
 use serde::{Deserialize, Serialize};
+use serde_with::{serde_as, DisplayFromStr};
 use std::collections::HashMap;
 use std::env;
 use std::fs;
@@ -32,6 +33,7 @@ pub const DEFAULT_PG_VERSION: u32 = 15;
 // to 'neon_local init --config=<path>' option. See control_plane/simple.conf for
 // an example.
 //
+#[serde_as]
 #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Debug)]
 pub struct LocalEnv {
     // Base directory for all the nodes (the pageserver, safekeepers and
@@ -57,6 +59,7 @@ pub struct LocalEnv {
     // Default tenant ID to use with the 'neon_local' command line utility, when
     // --tenant_id is not explicitly specified.
     #[serde(default)]
+    #[serde_as(as = "Option<DisplayFromStr>")]
     pub default_tenant_id: Option<TenantId>,
 
     // used to issue tokens during e.g pg start
@@ -81,6 +84,7 @@ pub struct LocalEnv {
     // A `HashMap<String, HashMap<TenantId, TimelineId>>` would be more appropriate here,
     // but deserialization into a generic toml object as `toml::Value::try_from` fails with an error.
     // https://toml.io/en/v1.0.0 does not contain a concept of "a table inside another table".
+    #[serde_as(as = "HashMap<_, Vec<(DisplayFromStr, DisplayFromStr)>>")]
     branch_name_mappings: HashMap<String, Vec<(TenantId, TimelineId)>>,
 }
 

control_plane/src/pageserver.rs

@@ -345,6 +345,11 @@ impl PageServerNode {
                 .remove("compaction_threshold")
                 .map(|x| x.parse::<usize>())
                 .transpose()?,
+            compaction_algorithm: settings
+                .remove("compaction_algorithm")
+                .map(serde_json::from_str)
+                .transpose()
+                .context("Failed to parse 'compaction_algorithm' json")?,
             gc_horizon: settings
                 .remove("gc_horizon")
                 .map(|x| x.parse::<u64>())
@@ -440,6 +445,11 @@ impl PageServerNode {
                     .map(|x| x.parse::<usize>())
                     .transpose()
                     .context("Failed to parse 'compaction_threshold' as an integer")?,
+                compaction_algorithm: settings
+                    .remove("compactin_algorithm")
+                    .map(serde_json::from_str)
+                    .transpose()
+                    .context("Failed to parse 'compaction_algorithm' json")?,
                 gc_horizon: settings
                     .remove("gc_horizon")
                     .map(|x| x.parse::<u64>())

libs/compute_api/src/lib.rs

@@ -1,5 +1,3 @@
-#![deny(unsafe_code)]
-#![deny(clippy::undocumented_unsafe_blocks)]
 pub mod requests;
 pub mod responses;
 pub mod spec;

libs/compute_api/src/spec.rs

@@ -6,6 +6,7 @@
 use std::collections::HashMap;
 
 use serde::{Deserialize, Serialize};
+use serde_with::{serde_as, DisplayFromStr};
 use utils::id::{TenantId, TimelineId};
 use utils::lsn::Lsn;
 
@@ -18,6 +19,7 @@ pub type PgIdent = String;
 
 /// Cluster spec or configuration represented as an optional number of
 /// delta operations + final cluster state description.
+#[serde_as]
 #[derive(Clone, Debug, Default, Deserialize, Serialize)]
 pub struct ComputeSpec {
     pub format_version: f32,
@@ -48,12 +50,12 @@ pub struct ComputeSpec {
     // these, and instead set the "neon.tenant_id", "neon.timeline_id",
     // etc. GUCs in cluster.settings. TODO: Once the control plane has been
     // updated to fill these fields, we can make these non optional.
+    #[serde_as(as = "Option<DisplayFromStr>")]
     pub tenant_id: Option<TenantId>,
-
+    #[serde_as(as = "Option<DisplayFromStr>")]
     pub timeline_id: Option<TimelineId>,
-
+    #[serde_as(as = "Option<DisplayFromStr>")]
     pub pageserver_connstring: Option<String>,
-
     #[serde(default)]
     pub safekeeper_connstrings: Vec<String>,
 
@@ -138,13 +140,14 @@ impl RemoteExtSpec {
     }
 }
 
+#[serde_as]
 #[derive(Clone, Copy, Debug, Default, Eq, PartialEq, Deserialize, Serialize)]
 pub enum ComputeMode {
     /// A read-write node
     #[default]
     Primary,
     /// A read-only node, pinned at a particular LSN
-    Static(Lsn),
+    Static(#[serde_as(as = "DisplayFromStr")] Lsn),
     /// A read-only node that follows the tip of the branch in hot standby mode
     ///
     /// Future versions may want to distinguish between replicas with hot standby

libs/consumption_metrics/src/lib.rs

@@ -1,6 +1,6 @@
+//!
 //! Shared code for consumption metics collection
-#![deny(unsafe_code)]
-#![deny(clippy::undocumented_unsafe_blocks)]
+//!
 use chrono::{DateTime, Utc};
 use rand::Rng;
 use serde::{Deserialize, Serialize};

libs/metrics/src/lib.rs

@@ -2,7 +2,6 @@
 //! make sure that we use the same dep version everywhere.
 //! Otherwise, we might not see all metrics registered via
 //! a default registry.
-#![deny(clippy::undocumented_unsafe_blocks)]
 use once_cell::sync::Lazy;
 use prometheus::core::{AtomicU64, Collector, GenericGauge, GenericGaugeVec};
 pub use prometheus::opts;

libs/pageserver_api/src/control_api.rs

@@ -4,6 +4,7 @@
 //! See docs/rfcs/025-generation-numbers.md
 
 use serde::{Deserialize, Serialize};
+use serde_with::{serde_as, DisplayFromStr};
 use utils::id::{NodeId, TenantId};
 
 #[derive(Serialize, Deserialize)]
@@ -11,8 +12,10 @@ pub struct ReAttachRequest {
     pub node_id: NodeId,
 }
 
+#[serde_as]
 #[derive(Serialize, Deserialize)]
 pub struct ReAttachResponseTenant {
+    #[serde_as(as = "DisplayFromStr")]
     pub id: TenantId,
     pub gen: u32,
 }
@@ -22,8 +25,10 @@ pub struct ReAttachResponse {
     pub tenants: Vec<ReAttachResponseTenant>,
 }
 
+#[serde_as]
 #[derive(Serialize, Deserialize)]
 pub struct ValidateRequestTenant {
+    #[serde_as(as = "DisplayFromStr")]
     pub id: TenantId,
     pub gen: u32,
 }
@@ -38,8 +43,10 @@ pub struct ValidateResponse {
     pub tenants: Vec<ValidateResponseTenant>,
 }
 
+#[serde_as]
 #[derive(Serialize, Deserialize)]
 pub struct ValidateResponseTenant {
+    #[serde_as(as = "DisplayFromStr")]
     pub id: TenantId,
     pub valid: bool,
 }

libs/pageserver_api/src/lib.rs

@@ -1,5 +1,3 @@
-#![deny(unsafe_code)]
-#![deny(clippy::undocumented_unsafe_blocks)]
 use const_format::formatcp;
 
 /// Public API types

libs/pageserver_api/src/models.rs

@@ -6,7 +6,7 @@ use std::{
 
 use byteorder::{BigEndian, ReadBytesExt};
 use serde::{Deserialize, Serialize};
-use serde_with::serde_as;
+use serde_with::{serde_as, DisplayFromStr};
 use strum_macros;
 use utils::{
     completion,
@@ -18,7 +18,7 @@ use utils::{
 
 use crate::reltag::RelTag;
 use anyhow::bail;
-use bytes::{Buf, BufMut, Bytes, BytesMut};
+use bytes::{BufMut, Bytes, BytesMut};
 
 /// The state of a tenant in this pageserver.
 ///
@@ -174,19 +174,25 @@ pub enum TimelineState {
     Broken { reason: String, backtrace: String },
 }
 
+#[serde_as]
 #[derive(Serialize, Deserialize)]
 pub struct TimelineCreateRequest {
+    #[serde_as(as = "DisplayFromStr")]
     pub new_timeline_id: TimelineId,
     #[serde(default)]
+    #[serde_as(as = "Option<DisplayFromStr>")]
     pub ancestor_timeline_id: Option<TimelineId>,
     #[serde(default)]
+    #[serde_as(as = "Option<DisplayFromStr>")]
     pub ancestor_start_lsn: Option<Lsn>,
     pub pg_version: Option<u32>,
 }
 
+#[serde_as]
 #[derive(Serialize, Deserialize, Debug)]
 #[serde(deny_unknown_fields)]
 pub struct TenantCreateRequest {
+    #[serde_as(as = "DisplayFromStr")]
     pub new_tenant_id: TenantId,
     #[serde(default)]
     #[serde(skip_serializing_if = "Option::is_none")]
@@ -195,6 +201,7 @@ pub struct TenantCreateRequest {
     pub config: TenantConfig, // as we have a flattened field, we should reject all unknown fields in it
 }
 
+#[serde_as]
 #[derive(Deserialize, Debug)]
 #[serde(deny_unknown_fields)]
 pub struct TenantLoadRequest {
@@ -220,6 +227,8 @@ pub struct TenantConfig {
     pub compaction_target_size: Option<u64>,
     pub compaction_period: Option<String>,
     pub compaction_threshold: Option<usize>,
+    // defer parsing compaction_algorithm, like eviction_policy
+    pub compaction_algorithm: Option<serde_json::Value>,
     pub gc_horizon: Option<u64>,
     pub gc_period: Option<String>,
     pub image_creation_threshold: Option<usize>,
@@ -271,26 +280,31 @@ pub struct LocationConfig {
     pub tenant_conf: TenantConfig,
 }
 
+#[serde_as]
 #[derive(Serialize, Deserialize)]
 #[serde(transparent)]
-pub struct TenantCreateResponse(pub TenantId);
+pub struct TenantCreateResponse(#[serde_as(as = "DisplayFromStr")] pub TenantId);
 
 #[derive(Serialize)]
 pub struct StatusResponse {
     pub id: NodeId,
 }
 
+#[serde_as]
 #[derive(Serialize, Deserialize, Debug)]
 #[serde(deny_unknown_fields)]
 pub struct TenantLocationConfigRequest {
+    #[serde_as(as = "DisplayFromStr")]
     pub tenant_id: TenantId,
     #[serde(flatten)]
     pub config: LocationConfig, // as we have a flattened field, we should reject all unknown fields in it
 }
 
+#[serde_as]
 #[derive(Serialize, Deserialize, Debug)]
 #[serde(deny_unknown_fields)]
 pub struct TenantConfigRequest {
+    #[serde_as(as = "DisplayFromStr")]
     pub tenant_id: TenantId,
     #[serde(flatten)]
     pub config: TenantConfig, // as we have a flattened field, we should reject all unknown fields in it
@@ -312,6 +326,7 @@ impl TenantConfigRequest {
             compaction_target_size: None,
             compaction_period: None,
             compaction_threshold: None,
+            compaction_algorithm: None,
             gc_horizon: None,
             gc_period: None,
             image_creation_threshold: None,
@@ -362,8 +377,10 @@ pub enum TenantAttachmentStatus {
     Failed { reason: String },
 }
 
+#[serde_as]
 #[derive(Serialize, Deserialize, Clone)]
 pub struct TenantInfo {
+    #[serde_as(as = "DisplayFromStr")]
     pub id: TenantId,
     // NB: intentionally not part of OpenAPI, we don't want to commit to a specific set of TenantState's
     pub state: TenantState,
@@ -374,22 +391,33 @@ pub struct TenantInfo {
 }
 
 /// This represents the output of the "timeline_detail" and "timeline_list" API calls.
+#[serde_as]
 #[derive(Debug, Serialize, Deserialize, Clone)]
 pub struct TimelineInfo {
+    #[serde_as(as = "DisplayFromStr")]
     pub tenant_id: TenantId,
+    #[serde_as(as = "DisplayFromStr")]
     pub timeline_id: TimelineId,
 
+    #[serde_as(as = "Option<DisplayFromStr>")]
     pub ancestor_timeline_id: Option<TimelineId>,
+    #[serde_as(as = "Option<DisplayFromStr>")]
     pub ancestor_lsn: Option<Lsn>,
+    #[serde_as(as = "DisplayFromStr")]
     pub last_record_lsn: Lsn,
+    #[serde_as(as = "Option<DisplayFromStr>")]
     pub prev_record_lsn: Option<Lsn>,
+    #[serde_as(as = "DisplayFromStr")]
     pub latest_gc_cutoff_lsn: Lsn,
+    #[serde_as(as = "DisplayFromStr")]
     pub disk_consistent_lsn: Lsn,
 
     /// The LSN that we have succesfully uploaded to remote storage
+    #[serde_as(as = "DisplayFromStr")]
     pub remote_consistent_lsn: Lsn,
 
     /// The LSN that we are advertizing to safekeepers
+    #[serde_as(as = "DisplayFromStr")]
     pub remote_consistent_lsn_visible: Lsn,
 
     pub current_logical_size: Option<u64>, // is None when timeline is Unloaded
@@ -401,6 +429,7 @@ pub struct TimelineInfo {
     pub timeline_dir_layer_file_size_sum: Option<u64>,
 
     pub wal_source_connstr: Option<String>,
+    #[serde_as(as = "Option<DisplayFromStr>")]
     pub last_received_msg_lsn: Option<Lsn>,
     /// the timestamp (in microseconds) of the last received message
     pub last_received_msg_ts: Option<u128>,
@@ -497,13 +526,23 @@ pub struct LayerAccessStats {
     pub residence_events_history: HistoryBufferWithDropCounter<LayerResidenceEvent, 16>,
 }
 
+#[serde_as]
 #[derive(Debug, Clone, Serialize)]
 #[serde(tag = "kind")]
 pub enum InMemoryLayerInfo {
-    Open { lsn_start: Lsn },
-    Frozen { lsn_start: Lsn, lsn_end: Lsn },
+    Open {
+        #[serde_as(as = "DisplayFromStr")]
+        lsn_start: Lsn,
+    },
+    Frozen {
+        #[serde_as(as = "DisplayFromStr")]
+        lsn_start: Lsn,
+        #[serde_as(as = "DisplayFromStr")]
+        lsn_end: Lsn,
+    },
 }
 
+#[serde_as]
 #[derive(Debug, Clone, Serialize)]
 #[serde(tag = "kind")]
 pub enum HistoricLayerInfo {
@@ -511,7 +550,9 @@ pub enum HistoricLayerInfo {
         layer_file_name: String,
         layer_file_size: u64,
 
+        #[serde_as(as = "DisplayFromStr")]
         lsn_start: Lsn,
+        #[serde_as(as = "DisplayFromStr")]
         lsn_end: Lsn,
         remote: bool,
         access_stats: LayerAccessStats,
@@ -520,6 +561,7 @@ pub enum HistoricLayerInfo {
         layer_file_name: String,
         layer_file_size: u64,
 
+        #[serde_as(as = "DisplayFromStr")]
         lsn_start: Lsn,
         remote: bool,
         access_stats: LayerAccessStats,
@@ -572,18 +614,15 @@ pub enum PagestreamFeMessage {
     Nblocks(PagestreamNblocksRequest),
     GetPage(PagestreamGetPageRequest),
     DbSize(PagestreamDbSizeRequest),
-    NoOp,
 }
 
 // Wrapped in libpq CopyData
-#[derive(Debug)]
 pub enum PagestreamBeMessage {
     Exists(PagestreamExistsResponse),
     Nblocks(PagestreamNblocksResponse),
     GetPage(PagestreamGetPageResponse),
     Error(PagestreamErrorResponse),
     DbSize(PagestreamDbSizeResponse),
-    NoOp,
 }
 
 #[derive(Debug, PartialEq, Eq)]
@@ -682,10 +721,6 @@ impl PagestreamFeMessage {
                 bytes.put_u64(req.lsn.0);
                 bytes.put_u32(req.dbnode);
             }
-
-            Self::NoOp => {
-                bytes.put_u8(4);
-            }
         }
 
         bytes.into()
@@ -736,7 +771,6 @@ impl PagestreamFeMessage {
                 lsn: Lsn::from(body.read_u64::<BigEndian>()?),
                 dbnode: body.read_u32::<BigEndian>()?,
             })),
-            4 => Ok(PagestreamFeMessage::NoOp),
             _ => bail!("unknown smgr message tag: {:?}", msg_tag),
         }
     }
@@ -771,46 +805,10 @@ impl PagestreamBeMessage {
                 bytes.put_u8(104); /* tag from pagestore_client.h */
                 bytes.put_i64(resp.db_size);
             }
-            Self::NoOp => {
-                bytes.put_u8(105);
-            }
         }
 
         bytes.into()
     }
-
-    pub fn deserialize(buf: Bytes) -> anyhow::Result<Self> {
-        let mut buf = buf.reader();
-        let msg_tag = buf.read_u8()?;
-        match msg_tag {
-            100 => todo!(),
-            101 => todo!(),
-            102 => {
-                let buf = buf.get_ref();
-                /* TODO use constant */
-                if buf.len() == 8192 {
-                    Ok(PagestreamBeMessage::GetPage(PagestreamGetPageResponse {
-                        page: buf.clone(),
-                    }))
-                } else {
-                    anyhow::bail!("invalid page size: {}", buf.len());
-                }
-            }
-            103 => {
-                let buf = buf.get_ref();
-                let cstr = std::ffi::CStr::from_bytes_until_nul(buf)?;
-                let rust_str = cstr.to_str()?;
-                Ok(PagestreamBeMessage::Error(PagestreamErrorResponse {
-                    message: rust_str.to_owned(),
-                }))
-            }
-            104 => todo!(),
-            105 => {
-                Ok(PagestreamBeMessage::NoOp)
-            },
-            _ => bail!("unknown tag: {:?}", msg_tag),
-        }
-    }
 }
 
 #[cfg(test)]

libs/postgres_backend/src/lib.rs

@@ -2,8 +2,6 @@
 //! To use, create PostgresBackend and run() it, passing the Handler
 //! implementation determining how to process the queries. Currently its API
 //! is rather narrow, but we can extend it once required.
-#![deny(unsafe_code)]
-#![deny(clippy::undocumented_unsafe_blocks)]
 use anyhow::Context;
 use bytes::Bytes;
 use futures::pin_mut;
@@ -730,17 +728,12 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> PostgresBackend<IO> {
 
                 trace!("got query {query_string:?}");
                 if let Err(e) = handler.process_query(self, query_string).await {
-                    match e {
-                        QueryError::Shutdown => return Ok(ProcessMsgResult::Break),
-                        e => {
-                            log_query_error(query_string, &e);
-                            let short_error = short_error(&e);
-                            self.write_message_noflush(&BeMessage::ErrorResponse(
-                                &short_error,
-                                Some(e.pg_error_code()),
-                            ))?;
-                        }
-                    }
+                    log_query_error(query_string, &e);
+                    let short_error = short_error(&e);
+                    self.write_message_noflush(&BeMessage::ErrorResponse(
+                        &short_error,
+                        Some(e.pg_error_code()),
+                    ))?;
                 }
                 self.write_message_noflush(&BeMessage::ReadyForQuery)?;
             }

libs/postgres_connection/src/lib.rs

@@ -1,5 +1,3 @@
-#![deny(unsafe_code)]
-#![deny(clippy::undocumented_unsafe_blocks)]
 use anyhow::{bail, Context};
 use itertools::Itertools;
 use std::borrow::Cow;

libs/postgres_ffi/src/lib.rs

@@ -8,7 +8,6 @@
 // modules included with the postgres_ffi macro depend on the types of the specific version's
 // types, and trigger a too eager lint.
 #![allow(clippy::duplicate_mod)]
-#![deny(clippy::undocumented_unsafe_blocks)]
 
 use bytes::Bytes;
 use utils::bin_ser::SerializeError;
@@ -21,7 +20,6 @@ macro_rules! postgres_ffi {
             pub mod bindings {
                 // bindgen generates bindings for a lot of stuff we don't need
                 #![allow(dead_code)]
-                #![allow(clippy::undocumented_unsafe_blocks)]
 
                 use serde::{Deserialize, Serialize};
                 include!(concat!(

libs/pq_proto/src/lib.rs

@@ -1,7 +1,6 @@
 //! Postgres protocol messages serialization-deserialization. See
 //! <https://www.postgresql.org/docs/devel/protocol-message-formats.html>
 //! on message formats.
-#![deny(clippy::undocumented_unsafe_blocks)]
 
 pub mod framed;
 

libs/remote_storage/src/azure_blob.rs

@@ -1,18 +1,21 @@
 //! Azure Blob Storage wrapper
 
-use std::collections::HashMap;
 use std::env;
 use std::num::NonZeroU32;
 use std::sync::Arc;
-use std::{borrow::Cow, io::Cursor};
+use std::{borrow::Cow, collections::HashMap, io::Cursor};
 
 use super::REMOTE_STORAGE_PREFIX_SEPARATOR;
 use anyhow::Result;
 use azure_core::request_options::{MaxResults, Metadata, Range};
+use azure_core::Header;
 use azure_identity::DefaultAzureCredential;
 use azure_storage::StorageCredentials;
 use azure_storage_blobs::prelude::ClientBuilder;
-use azure_storage_blobs::{blob::operations::GetBlobBuilder, prelude::ContainerClient};
+use azure_storage_blobs::{
+    blob::operations::GetBlobBuilder,
+    prelude::{BlobClient, ContainerClient},
+};
 use futures_util::StreamExt;
 use http_types::StatusCode;
 use tokio::io::AsyncRead;
@@ -109,19 +112,16 @@ impl AzureBlobStorage {
 
     async fn download_for_builder(
         &self,
+        metadata: StorageMetadata,
         builder: GetBlobBuilder,
     ) -> Result<Download, DownloadError> {
         let mut response = builder.into_stream();
 
-        let mut metadata = HashMap::new();
         // TODO give proper streaming response instead of buffering into RAM
         // https://github.com/neondatabase/neon/issues/5563
         let mut buf = Vec::new();
         while let Some(part) = response.next().await {
             let part = part.map_err(to_download_error)?;
-            if let Some(blob_meta) = part.blob.metadata {
-                metadata.extend(blob_meta.iter().map(|(k, v)| (k.to_owned(), v.to_owned())));
-            }
             let data = part
                 .data
                 .collect()
@@ -131,9 +131,28 @@ impl AzureBlobStorage {
         }
         Ok(Download {
             download_stream: Box::pin(Cursor::new(buf)),
-            metadata: Some(StorageMetadata(metadata)),
+            metadata: Some(metadata),
         })
     }
+    // TODO get rid of this function once we have metadata included in the response
+    // https://github.com/Azure/azure-sdk-for-rust/issues/1439
+    async fn get_metadata(
+        &self,
+        blob_client: &BlobClient,
+    ) -> Result<StorageMetadata, DownloadError> {
+        let builder = blob_client.get_metadata();
+
+        let response = builder.into_future().await.map_err(to_download_error)?;
+        let mut map = HashMap::new();
+
+        for md in response.metadata.iter() {
+            map.insert(
+                md.name().as_str().to_string(),
+                md.value().as_str().to_string(),
+            );
+        }
+        Ok(StorageMetadata(map))
+    }
 
     async fn permit(&self, kind: RequestKind) -> tokio::sync::SemaphorePermit<'_> {
         self.concurrency_limiter
@@ -250,9 +269,11 @@ impl RemoteStorage for AzureBlobStorage {
         let _permit = self.permit(RequestKind::Get).await;
         let blob_client = self.client.blob_client(self.relative_path_to_name(from));
 
+        let metadata = self.get_metadata(&blob_client).await?;
+
         let builder = blob_client.get();
 
-        self.download_for_builder(builder).await
+        self.download_for_builder(metadata, builder).await
     }
 
     async fn download_byte_range(
@@ -264,6 +285,8 @@ impl RemoteStorage for AzureBlobStorage {
         let _permit = self.permit(RequestKind::Get).await;
         let blob_client = self.client.blob_client(self.relative_path_to_name(from));
 
+        let metadata = self.get_metadata(&blob_client).await?;
+
         let mut builder = blob_client.get();
 
         if let Some(end_exclusive) = end_exclusive {
@@ -278,7 +301,7 @@ impl RemoteStorage for AzureBlobStorage {
             builder = builder.range(Range::new(start_inclusive, end_exclusive));
         }
 
-        self.download_for_builder(builder).await
+        self.download_for_builder(metadata, builder).await
     }
 
     async fn delete(&self, path: &RemotePath) -> anyhow::Result<()> {

libs/remote_storage/src/lib.rs

@@ -6,8 +6,6 @@
 //!   * [`s3_bucket`] uses AWS S3 bucket as an external storage
 //!   * [`azure_blob`] allows to use Azure Blob storage as an external storage
 //!
-#![deny(unsafe_code)]
-#![deny(clippy::undocumented_unsafe_blocks)]
 
 mod azure_blob;
 mod local_fs;
@@ -114,7 +112,7 @@ impl RemotePath {
         self.0.file_name()
     }
 
-    pub fn join<P: AsRef<Utf8Path>>(&self, segment: P) -> Self {
+    pub fn join(&self, segment: &Utf8Path) -> Self {
         Self(self.0.join(segment))
     }
 

libs/safekeeper_api/src/lib.rs

@@ -1,5 +1,3 @@
-#![deny(unsafe_code)]
-#![deny(clippy::undocumented_unsafe_blocks)]
 use const_format::formatcp;
 
 /// Public API types

libs/safekeeper_api/src/models.rs

@@ -1,18 +1,23 @@
 use serde::{Deserialize, Serialize};
+use serde_with::{serde_as, DisplayFromStr};
 
 use utils::{
     id::{NodeId, TenantId, TimelineId},
     lsn::Lsn,
 };
 
+#[serde_as]
 #[derive(Serialize, Deserialize)]
 pub struct TimelineCreateRequest {
+    #[serde_as(as = "DisplayFromStr")]
     pub tenant_id: TenantId,
+    #[serde_as(as = "DisplayFromStr")]
     pub timeline_id: TimelineId,
     pub peer_ids: Option<Vec<NodeId>>,
     pub pg_version: u32,
     pub system_id: Option<u64>,
     pub wal_seg_size: Option<u32>,
+    #[serde_as(as = "DisplayFromStr")]
     pub commit_lsn: Lsn,
     // If not passed, it is assigned to the beginning of commit_lsn segment.
     pub local_start_lsn: Option<Lsn>,
@@ -23,6 +28,7 @@ fn lsn_invalid() -> Lsn {
 }
 
 /// Data about safekeeper's timeline, mirrors broker.proto.
+#[serde_as]
 #[derive(Debug, Clone, Deserialize, Serialize)]
 pub struct SkTimelineInfo {
     /// Term.
@@ -30,19 +36,25 @@ pub struct SkTimelineInfo {
     /// Term of the last entry.
     pub last_log_term: Option<u64>,
     /// LSN of the last record.
+    #[serde_as(as = "DisplayFromStr")]
     #[serde(default = "lsn_invalid")]
     pub flush_lsn: Lsn,
     /// Up to which LSN safekeeper regards its WAL as committed.
+    #[serde_as(as = "DisplayFromStr")]
     #[serde(default = "lsn_invalid")]
     pub commit_lsn: Lsn,
     /// LSN up to which safekeeper has backed WAL.
+    #[serde_as(as = "DisplayFromStr")]
     #[serde(default = "lsn_invalid")]
     pub backup_lsn: Lsn,
     /// LSN of last checkpoint uploaded by pageserver.
+    #[serde_as(as = "DisplayFromStr")]
     #[serde(default = "lsn_invalid")]
     pub remote_consistent_lsn: Lsn,
+    #[serde_as(as = "DisplayFromStr")]
     #[serde(default = "lsn_invalid")]
     pub peer_horizon_lsn: Lsn,
+    #[serde_as(as = "DisplayFromStr")]
     #[serde(default = "lsn_invalid")]
     pub local_start_lsn: Lsn,
     /// A connection string to use for WAL receiving.

libs/tenant_size_model/src/lib.rs

@@ -1,6 +1,4 @@
 //! Synthetic size calculation
-#![deny(unsafe_code)]
-#![deny(clippy::undocumented_unsafe_blocks)]
 
 mod calculation;
 pub mod svg;

libs/tracing-utils/src/lib.rs

@@ -32,8 +32,6 @@
 //!         .init();
 //! }
 //! ```
-#![deny(unsafe_code)]
-#![deny(clippy::undocumented_unsafe_blocks)]
 
 use opentelemetry::sdk::Resource;
 use opentelemetry::KeyValue;

libs/utils/Cargo.toml

@@ -5,7 +5,6 @@ edition.workspace = true
 license.workspace = true
 
 [dependencies]
-arc-swap.workspace = true
 sentry.workspace = true
 async-trait.workspace = true
 anyhow.workspace = true
@@ -56,7 +55,6 @@ bytes.workspace = true
 criterion.workspace = true
 hex-literal.workspace = true
 camino-tempfile.workspace = true
-serde_assert.workspace = true
 
 [[bench]]
 name = "benchmarks"

libs/utils/src/auth.rs

@@ -1,8 +1,7 @@
 // For details about authentication see docs/authentication.md
 
-use arc_swap::ArcSwap;
 use serde;
-use std::{fs, sync::Arc};
+use std::fs;
 
 use anyhow::Result;
 use camino::Utf8Path;
@@ -10,6 +9,7 @@ use jsonwebtoken::{
     decode, encode, Algorithm, DecodingKey, EncodingKey, Header, TokenData, Validation,
 };
 use serde::{Deserialize, Serialize};
+use serde_with::{serde_as, DisplayFromStr};
 
 use crate::id::TenantId;
 
@@ -32,9 +32,11 @@ pub enum Scope {
 }
 
 /// JWT payload. See docs/authentication.md for the format
+#[serde_as]
 #[derive(Debug, Serialize, Deserialize, Clone, PartialEq)]
 pub struct Claims {
     #[serde(default)]
+    #[serde_as(as = "Option<DisplayFromStr>")]
     pub tenant_id: Option<TenantId>,
     pub scope: Scope,
 }
@@ -45,88 +47,31 @@ impl Claims {
     }
 }
 
-pub struct SwappableJwtAuth(ArcSwap<JwtAuth>);
-
-impl SwappableJwtAuth {
-    pub fn new(jwt_auth: JwtAuth) -> Self {
-        SwappableJwtAuth(ArcSwap::new(Arc::new(jwt_auth)))
-    }
-    pub fn swap(&self, jwt_auth: JwtAuth) {
-        self.0.swap(Arc::new(jwt_auth));
-    }
-    pub fn decode(&self, token: &str) -> Result<TokenData<Claims>> {
-        self.0.load().decode(token)
-    }
-}
-
-impl std::fmt::Debug for SwappableJwtAuth {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        write!(f, "Swappable({:?})", self.0.load())
-    }
-}
-
 pub struct JwtAuth {
-    decoding_keys: Vec<DecodingKey>,
+    decoding_key: DecodingKey,
     validation: Validation,
 }
 
 impl JwtAuth {
-    pub fn new(decoding_keys: Vec<DecodingKey>) -> Self {
+    pub fn new(decoding_key: DecodingKey) -> Self {
         let mut validation = Validation::default();
         validation.algorithms = vec![STORAGE_TOKEN_ALGORITHM];
         // The default 'required_spec_claims' is 'exp'. But we don't want to require
         // expiration.
         validation.required_spec_claims = [].into();
         Self {
-            decoding_keys,
+            decoding_key,
             validation,
         }
     }
 
     pub fn from_key_path(key_path: &Utf8Path) -> Result<Self> {
-        let metadata = key_path.metadata()?;
-        let decoding_keys = if metadata.is_dir() {
-            let mut keys = Vec::new();
-            for entry in fs::read_dir(key_path)? {
-                let path = entry?.path();
-                if !path.is_file() {
-                    // Ignore directories (don't recurse)
-                    continue;
-                }
-                let public_key = fs::read(path)?;
-                keys.push(DecodingKey::from_ed_pem(&public_key)?);
-            }
-            keys
-        } else if metadata.is_file() {
-            let public_key = fs::read(key_path)?;
-            vec![DecodingKey::from_ed_pem(&public_key)?]
-        } else {
-            anyhow::bail!("path is neither a directory or a file")
-        };
-        if decoding_keys.is_empty() {
-            anyhow::bail!("Configured for JWT auth with zero decoding keys. All JWT gated requests would be rejected.");
-        }
-        Ok(Self::new(decoding_keys))
+        let public_key = fs::read(key_path)?;
+        Ok(Self::new(DecodingKey::from_ed_pem(&public_key)?))
     }
 
-    /// Attempt to decode the token with the internal decoding keys.
-    ///
-    /// The function tries the stored decoding keys in succession,
-    /// and returns the first yielding a successful result.
-    /// If there is no working decoding key, it returns the last error.
     pub fn decode(&self, token: &str) -> Result<TokenData<Claims>> {
-        let mut res = None;
-        for decoding_key in &self.decoding_keys {
-            res = Some(decode(token, decoding_key, &self.validation));
-            if let Some(Ok(res)) = res {
-                return Ok(res);
-            }
-        }
-        if let Some(res) = res {
-            res.map_err(anyhow::Error::new)
-        } else {
-            anyhow::bail!("no JWT decoding keys configured")
-        }
+        Ok(decode(token, &self.decoding_key, &self.validation)?)
     }
 }
 
@@ -187,7 +132,7 @@ MC4CAQAwBQYDK2VwBCIEID/Drmc1AA6U/znNRWpF3zEGegOATQxfkdWxitcOMsIH
         let encoded_eddsa = "eyJhbGciOiJFZERTQSIsInR5cCI6IkpXVCJ9.eyJzY29wZSI6InRlbmFudCIsInRlbmFudF9pZCI6IjNkMWY3NTk1YjQ2ODIzMDMwNGUwYjczY2VjYmNiMDgxIiwiaXNzIjoibmVvbi5jb250cm9scGxhbmUiLCJleHAiOjE3MDkyMDA4NzksImlhdCI6MTY3ODQ0MjQ3OX0.U3eA8j-uU-JnhzeO3EDHRuXLwkAUFCPxtGHEgw6p7Ccc3YRbFs2tmCdbD9PZEXP-XsxSeBQi1FY0YPcT3NXADw";
 
         // Check it can be validated with the public key
-        let auth = JwtAuth::new(vec![DecodingKey::from_ed_pem(TEST_PUB_KEY_ED25519)?]);
+        let auth = JwtAuth::new(DecodingKey::from_ed_pem(TEST_PUB_KEY_ED25519)?);
         let claims_from_token = auth.decode(encoded_eddsa)?.claims;
         assert_eq!(claims_from_token, expected_claims);
 
@@ -204,7 +149,7 @@ MC4CAQAwBQYDK2VwBCIEID/Drmc1AA6U/znNRWpF3zEGegOATQxfkdWxitcOMsIH
         let encoded = encode_from_key_file(&claims, TEST_PRIV_KEY_ED25519)?;
 
         // decode it back
-        let auth = JwtAuth::new(vec![DecodingKey::from_ed_pem(TEST_PUB_KEY_ED25519)?]);
+        let auth = JwtAuth::new(DecodingKey::from_ed_pem(TEST_PUB_KEY_ED25519)?);
         let decoded = auth.decode(&encoded)?;
 
         assert_eq!(decoded.claims, claims);

libs/utils/src/hex.rs

@@ -1,41 +0,0 @@
-/// Useful type for asserting that expected bytes match reporting the bytes more readable
-/// array-syntax compatible hex bytes.
-///
-/// # Usage
-///
-/// ```
-/// use utils::Hex;
-///
-/// let actual = serialize_something();
-/// let expected = [0x68, 0x65, 0x6c, 0x6c, 0x6f, 0x20, 0x77, 0x6f, 0x72, 0x6c, 0x64];
-///
-/// // the type implements PartialEq and on mismatch, both sides are printed in 16 wide multiline
-/// // output suffixed with an array style length for easier comparisons.
-/// assert_eq!(Hex(&actual), Hex(&expected));
-///
-/// // with `let expected = [0x68];` the error would had been:
-/// // assertion `left == right` failed
-/// //  left: [0x68, 0x65, 0x6c, 0x6c, 0x6f, 0x20, 0x77, 0x6f, 0x72, 0x6c, 0x64; 11]
-/// // right: [0x68; 1]
-/// # fn serialize_something() -> Vec<u8> { "hello world".as_bytes().to_vec() }
-/// ```
-#[derive(PartialEq)]
-pub struct Hex<'a>(pub &'a [u8]);
-
-impl std::fmt::Debug for Hex<'_> {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        write!(f, "[")?;
-        for (i, c) in self.0.chunks(16).enumerate() {
-            if i > 0 && !c.is_empty() {
-                writeln!(f, ", ")?;
-            }
-            for (j, b) in c.iter().enumerate() {
-                if j > 0 {
-                    write!(f, ", ")?;
-                }
-                write!(f, "0x{b:02x}")?;
-            }
-        }
-        write!(f, "; {}]", self.0.len())
-    }
-}

libs/utils/src/http/endpoint.rs

@@ -1,4 +1,4 @@
-use crate::auth::{Claims, SwappableJwtAuth};
+use crate::auth::{Claims, JwtAuth};
 use crate::http::error::{api_error_handler, route_error_handler, ApiError};
 use anyhow::Context;
 use hyper::header::{HeaderName, AUTHORIZATION};
@@ -389,7 +389,7 @@ fn parse_token(header_value: &str) -> Result<&str, ApiError> {
 }
 
 pub fn auth_middleware<B: hyper::body::HttpBody + Send + Sync + 'static>(
-    provide_auth: fn(&Request<Body>) -> Option<&SwappableJwtAuth>,
+    provide_auth: fn(&Request<Body>) -> Option<&JwtAuth>,
 ) -> Middleware<B, ApiError> {
     Middleware::pre(move |req| async move {
         if let Some(auth) = provide_auth(&req) {

libs/utils/src/id.rs

@@ -3,7 +3,6 @@ use std::{fmt, str::FromStr};
 use anyhow::Context;
 use hex::FromHex;
 use rand::Rng;
-use serde::de::Visitor;
 use serde::{Deserialize, Serialize};
 use thiserror::Error;
 
@@ -18,74 +17,12 @@ pub enum IdError {
 ///
 /// NOTE: It (de)serializes as an array of hex bytes, so the string representation would look
 /// like `[173,80,132,115,129,226,72,254,170,201,135,108,199,26,228,24]`.
-#[derive(Clone, Copy, PartialEq, Eq, Hash, PartialOrd, Ord)]
+///
+/// Use `#[serde_as(as = "DisplayFromStr")]` to (de)serialize it as hex string instead: `ad50847381e248feaac9876cc71ae418`.
+/// Check the `serde_with::serde_as` documentation for options for more complex types.
+#[derive(Clone, Copy, PartialEq, Eq, Hash, Serialize, Deserialize, PartialOrd, Ord)]
 struct Id([u8; 16]);
 
-impl Serialize for Id {
-    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
-    where
-        S: serde::Serializer,
-    {
-        if serializer.is_human_readable() {
-            serializer.collect_str(self)
-        } else {
-            self.0.serialize(serializer)
-        }
-    }
-}
-
-impl<'de> Deserialize<'de> for Id {
-    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
-    where
-        D: serde::Deserializer<'de>,
-    {
-        struct IdVisitor {
-            is_human_readable_deserializer: bool,
-        }
-
-        impl<'de> Visitor<'de> for IdVisitor {
-            type Value = Id;
-
-            fn expecting(&self, formatter: &mut fmt::Formatter) -> fmt::Result {
-                if self.is_human_readable_deserializer {
-                    formatter.write_str("value in form of hex string")
-                } else {
-                    formatter.write_str("value in form of integer array([u8; 16])")
-                }
-            }
-
-            fn visit_seq<A>(self, seq: A) -> Result<Self::Value, A::Error>
-            where
-                A: serde::de::SeqAccess<'de>,
-            {
-                let s = serde::de::value::SeqAccessDeserializer::new(seq);
-                let id: [u8; 16] = Deserialize::deserialize(s)?;
-                Ok(Id::from(id))
-            }
-
-            fn visit_str<E>(self, v: &str) -> Result<Self::Value, E>
-            where
-                E: serde::de::Error,
-            {
-                Id::from_str(v).map_err(E::custom)
-            }
-        }
-
-        if deserializer.is_human_readable() {
-            deserializer.deserialize_str(IdVisitor {
-                is_human_readable_deserializer: true,
-            })
-        } else {
-            deserializer.deserialize_tuple(
-                16,
-                IdVisitor {
-                    is_human_readable_deserializer: false,
-                },
-            )
-        }
-    }
-}
-
 impl Id {
     pub fn get_from_buf(buf: &mut impl bytes::Buf) -> Id {
         let mut arr = [0u8; 16];
@@ -120,8 +57,6 @@ impl Id {
             chunk[0] = HEX[((b >> 4) & 0xf) as usize];
             chunk[1] = HEX[(b & 0xf) as usize];
         }
-
-        // SAFETY: vec constructed out of `HEX`, it can only be ascii
         unsafe { String::from_utf8_unchecked(buf) }
     }
 }
@@ -373,112 +308,3 @@ impl fmt::Display for NodeId {
         write!(f, "{}", self.0)
     }
 }
-
-#[cfg(test)]
-mod tests {
-    use serde_assert::{Deserializer, Serializer, Token, Tokens};
-
-    use crate::bin_ser::BeSer;
-
-    use super::*;
-
-    #[test]
-    fn test_id_serde_non_human_readable() {
-        let original_id = Id([
-            173, 80, 132, 115, 129, 226, 72, 254, 170, 201, 135, 108, 199, 26, 228, 24,
-        ]);
-        let expected_tokens = Tokens(vec![
-            Token::Tuple { len: 16 },
-            Token::U8(173),
-            Token::U8(80),
-            Token::U8(132),
-            Token::U8(115),
-            Token::U8(129),
-            Token::U8(226),
-            Token::U8(72),
-            Token::U8(254),
-            Token::U8(170),
-            Token::U8(201),
-            Token::U8(135),
-            Token::U8(108),
-            Token::U8(199),
-            Token::U8(26),
-            Token::U8(228),
-            Token::U8(24),
-            Token::TupleEnd,
-        ]);
-
-        let serializer = Serializer::builder().is_human_readable(false).build();
-        let serialized_tokens = original_id.serialize(&serializer).unwrap();
-        assert_eq!(serialized_tokens, expected_tokens);
-
-        let mut deserializer = Deserializer::builder()
-            .is_human_readable(false)
-            .tokens(serialized_tokens)
-            .build();
-        let deserialized_id = Id::deserialize(&mut deserializer).unwrap();
-        assert_eq!(deserialized_id, original_id);
-    }
-
-    #[test]
-    fn test_id_serde_human_readable() {
-        let original_id = Id([
-            173, 80, 132, 115, 129, 226, 72, 254, 170, 201, 135, 108, 199, 26, 228, 24,
-        ]);
-        let expected_tokens = Tokens(vec![Token::Str(String::from(
-            "ad50847381e248feaac9876cc71ae418",
-        ))]);
-
-        let serializer = Serializer::builder().is_human_readable(true).build();
-        let serialized_tokens = original_id.serialize(&serializer).unwrap();
-        assert_eq!(serialized_tokens, expected_tokens);
-
-        let mut deserializer = Deserializer::builder()
-            .is_human_readable(true)
-            .tokens(Tokens(vec![Token::Str(String::from(
-                "ad50847381e248feaac9876cc71ae418",
-            ))]))
-            .build();
-        assert_eq!(Id::deserialize(&mut deserializer).unwrap(), original_id);
-    }
-
-    macro_rules! roundtrip_type {
-        ($type:ty, $expected_bytes:expr) => {{
-            let expected_bytes: [u8; 16] = $expected_bytes;
-            let original_id = <$type>::from(expected_bytes);
-
-            let ser_bytes = original_id.ser().unwrap();
-            assert_eq!(ser_bytes, expected_bytes);
-
-            let des_id = <$type>::des(&ser_bytes).unwrap();
-            assert_eq!(des_id, original_id);
-        }};
-    }
-
-    #[test]
-    fn test_id_bincode_serde() {
-        let expected_bytes = [
-            173, 80, 132, 115, 129, 226, 72, 254, 170, 201, 135, 108, 199, 26, 228, 24,
-        ];
-
-        roundtrip_type!(Id, expected_bytes);
-    }
-
-    #[test]
-    fn test_tenant_id_bincode_serde() {
-        let expected_bytes = [
-            173, 80, 132, 115, 129, 226, 72, 254, 170, 201, 135, 108, 199, 26, 228, 24,
-        ];
-
-        roundtrip_type!(TenantId, expected_bytes);
-    }
-
-    #[test]
-    fn test_timeline_id_bincode_serde() {
-        let expected_bytes = [
-            173, 80, 132, 115, 129, 226, 72, 254, 170, 201, 135, 108, 199, 26, 228, 24,
-        ];
-
-        roundtrip_type!(TimelineId, expected_bytes);
-    }
-}

libs/utils/src/lib.rs

@@ -1,6 +1,5 @@
 //! `utils` is intended to be a place to put code that is shared
 //! between other crates in this repository.
-#![deny(clippy::undocumented_unsafe_blocks)]
 
 pub mod backoff;
 
@@ -25,10 +24,6 @@ pub mod auth;
 
 // utility functions and helper traits for unified unique id generation/serialization etc.
 pub mod id;
-
-mod hex;
-pub use hex::Hex;
-
 // http endpoint utils
 pub mod http;
 
@@ -78,9 +73,6 @@ pub mod completion;
 /// Reporting utilities
 pub mod error;
 
-/// async timeout helper
-pub mod timeout;
-
 pub mod sync;
 
 /// This is a shortcut to embed git sha into binaries and avoid copying the same build script to all packages

libs/utils/src/lsn.rs

@@ -1,7 +1,7 @@
 #![warn(missing_docs)]
 
 use camino::Utf8Path;
-use serde::{de::Visitor, Deserialize, Serialize};
+use serde::{Deserialize, Serialize};
 use std::fmt;
 use std::ops::{Add, AddAssign};
 use std::str::FromStr;
@@ -13,114 +13,10 @@ use crate::seqwait::MonotonicCounter;
 pub const XLOG_BLCKSZ: u32 = 8192;
 
 /// A Postgres LSN (Log Sequence Number), also known as an XLogRecPtr
-#[derive(Clone, Copy, Eq, Ord, PartialEq, PartialOrd, Hash)]
+#[derive(Clone, Copy, Eq, Ord, PartialEq, PartialOrd, Hash, Serialize, Deserialize)]
+#[serde(transparent)]
 pub struct Lsn(pub u64);
 
-impl Serialize for Lsn {
-    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
-    where
-        S: serde::Serializer,
-    {
-        if serializer.is_human_readable() {
-            serializer.collect_str(self)
-        } else {
-            self.0.serialize(serializer)
-        }
-    }
-}
-
-impl<'de> Deserialize<'de> for Lsn {
-    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
-    where
-        D: serde::Deserializer<'de>,
-    {
-        struct LsnVisitor {
-            is_human_readable_deserializer: bool,
-        }
-
-        impl<'de> Visitor<'de> for LsnVisitor {
-            type Value = Lsn;
-
-            fn expecting(&self, formatter: &mut fmt::Formatter) -> fmt::Result {
-                if self.is_human_readable_deserializer {
-                    formatter.write_str(
-                        "value in form of hex string({upper_u32_hex}/{lower_u32_hex}) representing u64 integer",
-                    )
-                } else {
-                    formatter.write_str("value in form of integer(u64)")
-                }
-            }
-
-            fn visit_u64<E>(self, v: u64) -> Result<Self::Value, E>
-            where
-                E: serde::de::Error,
-            {
-                Ok(Lsn(v))
-            }
-
-            fn visit_str<E>(self, v: &str) -> Result<Self::Value, E>
-            where
-                E: serde::de::Error,
-            {
-                Lsn::from_str(v).map_err(|e| E::custom(e))
-            }
-        }
-
-        if deserializer.is_human_readable() {
-            deserializer.deserialize_str(LsnVisitor {
-                is_human_readable_deserializer: true,
-            })
-        } else {
-            deserializer.deserialize_u64(LsnVisitor {
-                is_human_readable_deserializer: false,
-            })
-        }
-    }
-}
-
-/// Allows (de)serialization of an `Lsn` always as `u64`.
-///
-/// ### Example
-///
-/// ```rust
-/// # use serde::{Serialize, Deserialize};
-/// use utils::lsn::Lsn;
-///
-/// #[derive(PartialEq, Serialize, Deserialize, Debug)]
-/// struct Foo {
-///   #[serde(with = "utils::lsn::serde_as_u64")]
-///   always_u64: Lsn,
-/// }
-///
-/// let orig = Foo { always_u64: Lsn(1234) };
-///
-/// let res = serde_json::to_string(&orig).unwrap();
-/// assert_eq!(res, r#"{"always_u64":1234}"#);
-///
-/// let foo = serde_json::from_str::<Foo>(&res).unwrap();
-/// assert_eq!(foo, orig);
-/// ```
-///
-pub mod serde_as_u64 {
-    use super::Lsn;
-
-    /// Serializes the Lsn as u64 disregarding the human readability of the format.
-    ///
-    /// Meant to be used via `#[serde(with = "...")]` or `#[serde(serialize_with = "...")]`.
-    pub fn serialize<S: serde::Serializer>(lsn: &Lsn, serializer: S) -> Result<S::Ok, S::Error> {
-        use serde::Serialize;
-        lsn.0.serialize(serializer)
-    }
-
-    /// Deserializes the Lsn as u64 disregarding the human readability of the format.
-    ///
-    /// Meant to be used via `#[serde(with = "...")]` or `#[serde(deserialize_with = "...")]`.
-    pub fn deserialize<'de, D: serde::Deserializer<'de>>(deserializer: D) -> Result<Lsn, D::Error> {
-        use serde::Deserialize;
-        u64::deserialize(deserializer).map(Lsn)
-    }
-}
-
 /// We tried to parse an LSN from a string, but failed
 #[derive(Debug, PartialEq, Eq, thiserror::Error)]
 #[error("LsnParseError")]
@@ -368,13 +264,8 @@ impl MonotonicCounter<Lsn> for RecordLsn {
 
 #[cfg(test)]
 mod tests {
-    use crate::bin_ser::BeSer;
-
     use super::*;
 
-    use serde::ser::Serialize;
-    use serde_assert::{Deserializer, Serializer, Token, Tokens};
-
     #[test]
     fn test_lsn_strings() {
         assert_eq!("12345678/AAAA5555".parse(), Ok(Lsn(0x12345678AAAA5555)));
@@ -450,95 +341,4 @@ mod tests {
         assert_eq!(lsn.fetch_max(Lsn(6000)), Lsn(5678));
         assert_eq!(lsn.fetch_max(Lsn(5000)), Lsn(6000));
     }
-
-    #[test]
-    fn test_lsn_serde() {
-        let original_lsn = Lsn(0x0123456789abcdef);
-        let expected_readable_tokens = Tokens(vec![Token::U64(0x0123456789abcdef)]);
-        let expected_non_readable_tokens =
-            Tokens(vec![Token::Str(String::from("1234567/89ABCDEF"))]);
-
-        // Testing human_readable ser/de
-        let serializer = Serializer::builder().is_human_readable(false).build();
-        let readable_ser_tokens = original_lsn.serialize(&serializer).unwrap();
-        assert_eq!(readable_ser_tokens, expected_readable_tokens);
-
-        let mut deserializer = Deserializer::builder()
-            .is_human_readable(false)
-            .tokens(readable_ser_tokens)
-            .build();
-        let des_lsn = Lsn::deserialize(&mut deserializer).unwrap();
-        assert_eq!(des_lsn, original_lsn);
-
-        // Testing NON human_readable ser/de
-        let serializer = Serializer::builder().is_human_readable(true).build();
-        let non_readable_ser_tokens = original_lsn.serialize(&serializer).unwrap();
-        assert_eq!(non_readable_ser_tokens, expected_non_readable_tokens);
-
-        let mut deserializer = Deserializer::builder()
-            .is_human_readable(true)
-            .tokens(non_readable_ser_tokens)
-            .build();
-        let des_lsn = Lsn::deserialize(&mut deserializer).unwrap();
-        assert_eq!(des_lsn, original_lsn);
-
-        // Testing mismatching ser/de
-        let serializer = Serializer::builder().is_human_readable(false).build();
-        let non_readable_ser_tokens = original_lsn.serialize(&serializer).unwrap();
-
-        let mut deserializer = Deserializer::builder()
-            .is_human_readable(true)
-            .tokens(non_readable_ser_tokens)
-            .build();
-        Lsn::deserialize(&mut deserializer).unwrap_err();
-
-        let serializer = Serializer::builder().is_human_readable(true).build();
-        let readable_ser_tokens = original_lsn.serialize(&serializer).unwrap();
-
-        let mut deserializer = Deserializer::builder()
-            .is_human_readable(false)
-            .tokens(readable_ser_tokens)
-            .build();
-        Lsn::deserialize(&mut deserializer).unwrap_err();
-    }
-
-    #[test]
-    fn test_lsn_ensure_roundtrip() {
-        let original_lsn = Lsn(0xaaaabbbb);
-
-        let serializer = Serializer::builder().is_human_readable(false).build();
-        let ser_tokens = original_lsn.serialize(&serializer).unwrap();
-
-        let mut deserializer = Deserializer::builder()
-            .is_human_readable(false)
-            .tokens(ser_tokens)
-            .build();
-
-        let des_lsn = Lsn::deserialize(&mut deserializer).unwrap();
-        assert_eq!(des_lsn, original_lsn);
-    }
-
-    #[test]
-    fn test_lsn_bincode_serde() {
-        let lsn = Lsn(0x0123456789abcdef);
-        let expected_bytes = [0x01, 0x23, 0x45, 0x67, 0x89, 0xab, 0xcd, 0xef];
-
-        let ser_bytes = lsn.ser().unwrap();
-        assert_eq!(ser_bytes, expected_bytes);
-
-        let des_lsn = Lsn::des(&ser_bytes).unwrap();
-        assert_eq!(des_lsn, lsn);
-    }
-
-    #[test]
-    fn test_lsn_bincode_ensure_roundtrip() {
-        let original_lsn = Lsn(0x01_02_03_04_05_06_07_08);
-        let expected_bytes = vec![0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08];
-
-        let ser_bytes = original_lsn.ser().unwrap();
-        assert_eq!(ser_bytes, expected_bytes);
-
-        let des_lsn = Lsn::des(&ser_bytes).unwrap();
-        assert_eq!(des_lsn, original_lsn);
-    }
 }

libs/utils/src/pageserver_feedback.rs

@@ -3,6 +3,7 @@ use std::time::{Duration, SystemTime};
 use bytes::{Buf, BufMut, Bytes, BytesMut};
 use pq_proto::{read_cstr, PG_EPOCH};
 use serde::{Deserialize, Serialize};
+use serde_with::{serde_as, DisplayFromStr};
 use tracing::{trace, warn};
 
 use crate::lsn::Lsn;
@@ -14,17 +15,21 @@ use crate::lsn::Lsn;
 ///
 /// serde Serialize is used only for human readable dump to json (e.g. in
 /// safekeepers debug_dump).
+#[serde_as]
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
 pub struct PageserverFeedback {
     /// Last known size of the timeline. Used to enforce timeline size limit.
     pub current_timeline_size: u64,
     /// LSN last received and ingested by the pageserver. Controls backpressure.
+    #[serde_as(as = "DisplayFromStr")]
     pub last_received_lsn: Lsn,
     /// LSN up to which data is persisted by the pageserver to its local disc.
     /// Controls backpressure.
+    #[serde_as(as = "DisplayFromStr")]
     pub disk_consistent_lsn: Lsn,
     /// LSN up to which data is persisted by the pageserver on s3; safekeepers
     /// consider WAL before it can be removed.
+    #[serde_as(as = "DisplayFromStr")]
     pub remote_consistent_lsn: Lsn,
     // Serialize with RFC3339 format.
     #[serde(with = "serde_systemtime")]

libs/utils/src/shutdown.rs

@@ -1,7 +1,6 @@
 /// Immediately terminate the calling process without calling
 /// atexit callbacks, C runtime destructors etc. We mainly use
 /// this to protect coverage data from concurrent writes.
-pub fn exit_now(code: u8) -> ! {
-    // SAFETY: exiting is safe, the ffi is not safe
+pub fn exit_now(code: u8) {
     unsafe { nix::libc::_exit(code as _) };
 }

libs/utils/src/sync.rs

@@ -1,3 +1 @@
 pub mod heavier_once_cell;
-
-pub mod gate;

libs/utils/src/sync/gate.rs

@@ -1,158 +0,0 @@
-use std::{sync::Arc, time::Duration};
-
-/// Gates are a concurrency helper, primarily used for implementing safe shutdown.
-///
-/// Users of a resource call `enter()` to acquire a GateGuard, and the owner of
-/// the resource calls `close()` when they want to ensure that all holders of guards
-/// have released them, and that no future guards will be issued.
-pub struct Gate {
-    /// Each caller of enter() takes one unit from the semaphore. In close(), we
-    /// take all the units to ensure all GateGuards are destroyed.
-    sem: Arc<tokio::sync::Semaphore>,
-
-    /// For observability only: a name that will be used to log warnings if a particular
-    /// gate is holding up shutdown
-    name: String,
-}
-
-/// RAII guard for a [`Gate`]: as long as this exists, calls to [`Gate::close`] will
-/// not complete.
-#[derive(Debug)]
-pub struct GateGuard(tokio::sync::OwnedSemaphorePermit);
-
-/// Observability helper: every `warn_period`, emit a log warning that we're still waiting on this gate
-async fn warn_if_stuck<Fut: std::future::Future>(
-    fut: Fut,
-    name: &str,
-    warn_period: std::time::Duration,
-) -> <Fut as std::future::Future>::Output {
-    let started = std::time::Instant::now();
-
-    let mut fut = std::pin::pin!(fut);
-
-    loop {
-        match tokio::time::timeout(warn_period, &mut fut).await {
-            Ok(ret) => return ret,
-            Err(_) => {
-                tracing::warn!(
-                    gate = name,
-                    elapsed_ms = started.elapsed().as_millis(),
-                    "still waiting, taking longer than expected..."
-                );
-            }
-        }
-    }
-}
-
-#[derive(Debug)]
-pub enum GateError {
-    GateClosed,
-}
-
-impl Gate {
-    const MAX_UNITS: u32 = u32::MAX;
-
-    pub fn new(name: String) -> Self {
-        Self {
-            sem: Arc::new(tokio::sync::Semaphore::new(Self::MAX_UNITS as usize)),
-            name,
-        }
-    }
-
-    /// Acquire a guard that will prevent close() calls from completing. If close()
-    /// was already called, this will return an error which should be interpreted
-    /// as "shutting down".
-    ///
-    /// This function would typically be used from e.g. request handlers. While holding
-    /// the guard returned from this function, it is important to respect a CancellationToken
-    /// to avoid blocking close() indefinitely: typically types that contain a Gate will
-    /// also contain a CancellationToken.
-    pub fn enter(&self) -> Result<GateGuard, GateError> {
-        self.sem
-            .clone()
-            .try_acquire_owned()
-            .map(GateGuard)
-            .map_err(|_| GateError::GateClosed)
-    }
-
-    /// Types with a shutdown() method and a gate should call this method at the
-    /// end of shutdown, to ensure that all GateGuard holders are done.
-    ///
-    /// This will wait for all guards to be destroyed.  For this to complete promptly, it is
-    /// important that the holders of such guards are respecting a CancellationToken which has
-    /// been cancelled before entering this function.
-    pub async fn close(&self) {
-        warn_if_stuck(self.do_close(), &self.name, Duration::from_millis(1000)).await
-    }
-
-    /// Check if [`Self::close()`] has finished waiting for all [`Self::enter()`] users to finish.  This
-    /// is usually analoguous for "Did shutdown finish?" for types that include a Gate, whereas checking
-    /// the CancellationToken on such types is analogous to "Did shutdown start?"
-    pub fn close_complete(&self) -> bool {
-        self.sem.is_closed()
-    }
-
-    async fn do_close(&self) {
-        tracing::debug!(gate = self.name, "Closing Gate...");
-        match self.sem.acquire_many(Self::MAX_UNITS).await {
-            Ok(_units) => {
-                // While holding all units, close the semaphore.  All subsequent calls to enter() will fail.
-                self.sem.close();
-            }
-            Err(_) => {
-                // Semaphore closed: we are the only function that can do this, so it indicates a double-call.
-                // This is legal.  Timeline::shutdown for example is not protected from being called more than
-                // once.
-                tracing::debug!(gate = self.name, "Double close")
-            }
-        }
-        tracing::debug!(gate = self.name, "Closed Gate.")
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use futures::FutureExt;
-
-    use super::*;
-
-    #[tokio::test]
-    async fn test_idle_gate() {
-        // Having taken no gates, we should not be blocked in close
-        let gate = Gate::new("test".to_string());
-        gate.close().await;
-
-        // If a guard is dropped before entering, close should not be blocked
-        let gate = Gate::new("test".to_string());
-        let guard = gate.enter().unwrap();
-        drop(guard);
-        gate.close().await;
-
-        // Entering a closed guard fails
-        gate.enter().expect_err("enter should fail after close");
-    }
-
-    #[tokio::test]
-    async fn test_busy_gate() {
-        let gate = Gate::new("test".to_string());
-
-        let guard = gate.enter().unwrap();
-
-        let mut close_fut = std::pin::pin!(gate.close());
-
-        // Close should be blocked
-        assert!(close_fut.as_mut().now_or_never().is_none());
-
-        // Attempting to enter() should fail, even though close isn't done yet.
-        gate.enter()
-            .expect_err("enter should fail after entering close");
-
-        drop(guard);
-
-        // Guard is gone, close should finish
-        assert!(close_fut.as_mut().now_or_never().is_some());
-
-        // Attempting to enter() is still forbidden
-        gate.enter().expect_err("enter should fail finishing close");
-    }
-}

libs/utils/src/timeout.rs

@@ -1,37 +0,0 @@
-use std::time::Duration;
-
-use tokio_util::sync::CancellationToken;
-
-pub enum TimeoutCancellableError {
-    Timeout,
-    Cancelled,
-}
-
-/// Wrap [`tokio::time::timeout`] with a CancellationToken.
-///
-/// This wrapper is appropriate for any long running operation in a task
-/// that ought to respect a CancellationToken (which means most tasks).
-///
-/// The only time you should use a bare tokio::timeout is when the future `F`
-/// itself respects a CancellationToken: otherwise, always use this wrapper
-/// with your CancellationToken to ensure that your task does not hold up
-/// graceful shutdown.
-pub async fn timeout_cancellable<F>(
-    duration: Duration,
-    cancel: &CancellationToken,
-    future: F,
-) -> Result<F::Output, TimeoutCancellableError>
-where
-    F: std::future::Future,
-{
-    tokio::select!(
-        r = tokio::time::timeout(duration, future) => {
-            r.map_err(|_| TimeoutCancellableError::Timeout)
-
-        },
-        _ = cancel.cancelled() => {
-            Err(TimeoutCancellableError::Cancelled)
-
-        }
-    )
-}

libs/vm_monitor/Cargo.toml

@@ -19,12 +19,13 @@ inotify.workspace = true
 serde.workspace = true
 serde_json.workspace = true
 sysinfo.workspace = true
-tokio = { workspace = true, features = ["rt-multi-thread"] }
+tokio.workspace = true
 tokio-postgres.workspace = true
 tokio-stream.workspace = true
 tokio-util.workspace = true
 tracing.workspace = true
 tracing-subscriber.workspace = true
+workspace_hack = { version = "0.1", path = "../../workspace_hack" }
 
 [target.'cfg(target_os = "linux")'.dependencies]
 cgroups-rs = "0.3.3"

libs/vm_monitor/src/lib.rs

@@ -1,5 +1,3 @@
-#![deny(unsafe_code)]
-#![deny(clippy::undocumented_unsafe_blocks)]
 #![cfg(target_os = "linux")]
 
 use anyhow::Context;

pageserver/Cargo.toml

@@ -68,6 +68,7 @@ url.workspace = true
 walkdir.workspace = true
 metrics.workspace = true
 pageserver_api.workspace = true
+pageserver_compaction.workspace = true
 postgres_connection.workspace = true
 postgres_ffi.workspace = true
 pq_proto.workspace = true
@@ -82,8 +83,6 @@ enum-map.workspace = true
 enumset.workspace = true
 strum.workspace = true
 strum_macros.workspace = true
-tokio-stream.workspace = true
-tracing-subscriber = { version = "0.3.17", features = ["env-filter"] }
 
 [dev-dependencies]
 criterion.workspace = true

pageserver/compaction/Cargo.toml

@@ -0,0 +1,53 @@
+[package]
+name = "pageserver_compaction"
+version = "0.1.0"
+edition.workspace = true
+license.workspace = true
+
+[features]
+default = []
+
+[dependencies]
+anyhow.workspace = true
+async-compression.workspace = true
+async-stream.workspace = true
+async-trait.workspace = true
+byteorder.workspace = true
+bytes.workspace = true
+chrono = { workspace = true, features = ["serde"] }
+clap = { workspace = true, features = ["string"] }
+const_format.workspace = true
+consumption_metrics.workspace = true
+crossbeam-utils.workspace = true
+either.workspace = true
+flate2.workspace = true
+fail.workspace = true
+futures.workspace = true
+git-version.workspace = true
+hex.workspace = true
+humantime.workspace = true
+humantime-serde.workspace = true
+itertools.workspace = true
+once_cell.workspace = true
+pin-project-lite.workspace = true
+rand.workspace = true
+smallvec = { workspace = true, features = ["write"] }
+svg_fmt.workspace = true
+sync_wrapper.workspace = true
+thiserror.workspace = true
+tokio = { workspace = true, features = ["process", "sync", "fs", "rt", "io-util", "time"] }
+tokio-io-timeout.workspace = true
+tokio-util.workspace = true
+tracing.workspace = true
+tracing-error.workspace = true
+tracing-subscriber.workspace = true
+url.workspace = true
+walkdir.workspace = true
+metrics.workspace = true
+utils.workspace = true
+workspace_hack.workspace = true
+
+[dev-dependencies]
+criterion.workspace = true
+hex-literal.workspace = true
+tokio = { workspace = true, features = ["process", "sync", "fs", "rt", "io-util", "time", "test-util"] }

pageserver/compaction/TODO.md

@@ -0,0 +1,49 @@
+# TODO
+
+- If the key space can be perfectly partitioned at some key, perform planning on each
+  partition separately. For example, if we are compacting a level with layers like this:
+
+              :
+  +--+ +----+ :  +------+
+  |  | |    | :  |      |
+  +--+ +----+ :  +------+
+              :
+  +-----+ +-+ : +--------+
+  |     | | | : |        |
+  +-----+ +-+ : +--------+
+              :
+
+  At the dotted line, there is a natural split in the key space, such that all
+  layers are either on the left or the right of it. We can compact the
+  partitions separately.  We could choose to create image layers for one
+  partition but not the other one, for example.
+
+- All the layers don't have to be exactly the same size, we can choose to cut a
+  layer short or stretch it a little larger than the target size, if it helps
+  the overall system. We can help perfect partitions (see previous bullet point)
+  to happen more frequently, by choosing the cut points wisely. For example, try
+  to cut layers at boundaries of underlying image layers. And "snap to grid",
+  i.e. don't cut layers at any key, but e.g. only when key % 10000 = 0.
+
+- Avoid rewriting layers when we'd just create an identical layer to an input
+  layer.
+
+- Parallelism. The code is already split up into planning and execution, so that
+  we first split up the compaction work into "Jobs", and then execute them.
+  It would be straightforward to execute multiple jobs in parallel.
+
+- Materialize extra pages in delta layers during compaction. This would reduce
+  read amplification. There has been the idea of partial image layers. Materializing
+  extra pages in the delta layers achieve the same goal, without introducing a new
+  concept.
+
+## Simulator
+
+- Expand the simulator for more workloads
+- Automate a test suite that runs the simluator with different workloads and
+  spits out a table of results
+- Model read amplification
+- More sanity checking. One idea is to keep a reference count of each
+  MockRecord, i.e. use Arc<MockRecord> instead of plain MockRecord, and panic if
+  a MockRecord that is newer than PITR horizon is completely dropped. That would
+  indicate that the record was lost.

pageserver/compaction/src/bin/compaction-simulator.rs

@@ -0,0 +1,214 @@
+use clap::{Parser, Subcommand};
+use pageserver_compaction::simulator::MockTimeline;
+use rand::Rng;
+use std::io::Write;
+use std::path::{Path, PathBuf};
+use std::sync::OnceLock;
+
+use utils::project_git_version;
+
+project_git_version!(GIT_VERSION);
+
+#[derive(Parser)]
+#[command(
+    version = GIT_VERSION,
+    about = "Neon Pageserver compaction simulator",
+    long_about = "A developer tool to visualize and test compaction"
+)]
+#[command(propagate_version = true)]
+struct CliOpts {
+    #[command(subcommand)]
+    command: Commands,
+}
+
+#[derive(Subcommand)]
+enum Commands {
+    RunSuite,
+    Simulate(SimulateCmd),
+}
+
+#[derive(Clone, clap::ValueEnum)]
+enum Distribution {
+    Uniform,
+    HotCold,
+}
+
+/// Read and update pageserver metadata file
+#[derive(Parser)]
+struct SimulateCmd {
+    distribution: Distribution,
+
+    /// Number of records to digest
+    num_records: u64,
+    /// Record length
+    record_len: u64,
+
+    // Logical database size in MB
+    logical_size: u64,
+}
+
+async fn simulate(cmd: &SimulateCmd, results_path: &Path) -> anyhow::Result<()> {
+    let mut executor = MockTimeline::new();
+
+    // Convert the logical size in MB into a key range.
+    let key_range = 0..((cmd.logical_size * 1024 * 1024) / 8192);
+    //let key_range = u64::MIN..u64::MAX;
+    println!(
+        "starting simulation with key range {:016X}-{:016X}",
+        key_range.start, key_range.end
+    );
+
+    // helper function to print progress indicator
+    let print_progress = |i| -> anyhow::Result<()> {
+        if i == 0 || (i + 1) % 10000 == 0 || i == cmd.num_records - 1 {
+            print!(
+                "\ringested {} / {} records, {} MiB / {} MiB...",
+                i + 1,
+                cmd.num_records,
+                (i + 1) * cmd.record_len / (1_000_000),
+                cmd.num_records * cmd.record_len / (1_000_000),
+            );
+            std::io::stdout().flush()?;
+        }
+        Ok(())
+    };
+
+    match cmd.distribution {
+        Distribution::Uniform => {
+            for i in 0..cmd.num_records {
+                executor.ingest_uniform(1, cmd.record_len, &key_range)?;
+                executor.compact_if_needed().await?;
+
+                print_progress(i)?;
+            }
+        }
+        Distribution::HotCold => {
+            let splitpoint = key_range.end / 10;
+            let hot_key_range = 0..splitpoint;
+            let cold_key_range = splitpoint..key_range.end;
+
+            for i in 0..cmd.num_records {
+                let chosen_range = if rand::thread_rng().gen_bool(0.9) {
+                    &hot_key_range
+                } else {
+                    &cold_key_range
+                };
+                executor.ingest_uniform(1, cmd.record_len, chosen_range)?;
+                executor.compact_if_needed().await?;
+
+                print_progress(i)?;
+            }
+        }
+    }
+    println!("done!");
+    executor.flush_l0();
+    executor.compact_if_needed().await?;
+    let stats = executor.print_stats()?;
+
+    // Print the stats to stdout, and also to a file
+    print!("{}", stats);
+    std::fs::write(results_path.join("stats.txt"), stats)?;
+
+    let animation_path = results_path.join("compaction-animation.html");
+    executor.draw_history(std::fs::File::create(&animation_path)?)?;
+    println!(
+        "animation: file://{}",
+        animation_path.canonicalize()?.display()
+    );
+
+    Ok(())
+}
+
+async fn run_suite_cmd(results_path: &Path, workload: &SimulateCmd) -> anyhow::Result<()> {
+    std::fs::create_dir(results_path)?;
+
+    set_log_file(File::create(results_path.join("log"))?);
+    let result = simulate(workload, results_path).await;
+    set_log_stdout();
+    result
+}
+
+async fn run_suite() -> anyhow::Result<()> {
+    let top_results_path = PathBuf::from(format!(
+        "compaction-suite-results.{}",
+        std::time::SystemTime::UNIX_EPOCH.elapsed()?.as_secs()
+    ));
+    std::fs::create_dir(&top_results_path)?;
+
+    let workload = SimulateCmd {
+        distribution: Distribution::Uniform,
+        // Generate 20 GB of WAL
+        record_len: 1_000,
+        num_records: 20_000_000,
+        // Logical size 5 GB
+        logical_size: 5_000,
+    };
+
+    run_suite_cmd(&top_results_path.join("uniform-20GB-5GB"), &workload).await?;
+
+    println!(
+        "All tests finished. Results in {}",
+        top_results_path.display()
+    );
+    Ok(())
+}
+
+use std::fs::File;
+use std::io::Stdout;
+use std::sync::Mutex;
+use tracing_subscriber::fmt::writer::EitherWriter;
+use tracing_subscriber::fmt::MakeWriter;
+
+static LOG_FILE: OnceLock<Mutex<EitherWriter<File, Stdout>>> = OnceLock::new();
+fn get_log_output() -> &'static Mutex<EitherWriter<File, Stdout>> {
+    LOG_FILE.get_or_init(|| std::sync::Mutex::new(EitherWriter::B(std::io::stdout())))
+}
+
+fn set_log_file(f: File) {
+    *get_log_output().lock().unwrap() = EitherWriter::A(f);
+}
+
+fn set_log_stdout() {
+    *get_log_output().lock().unwrap() = EitherWriter::B(std::io::stdout());
+}
+
+fn init_logging() -> anyhow::Result<()> {
+    // We fall back to printing all spans at info-level or above if
+    // the RUST_LOG environment variable is not set.
+    let rust_log_env_filter = || {
+        tracing_subscriber::EnvFilter::try_from_default_env()
+            .unwrap_or_else(|_| tracing_subscriber::EnvFilter::new("info"))
+    };
+
+    // NB: the order of the with() calls does not matter.
+    // See https://docs.rs/tracing-subscriber/0.3.16/tracing_subscriber/layer/index.html#per-layer-filtering
+    use tracing_subscriber::prelude::*;
+    tracing_subscriber::registry()
+        .with({
+            let log_layer = tracing_subscriber::fmt::layer()
+                .with_target(false)
+                .with_ansi(false)
+                .with_writer(|| get_log_output().make_writer());
+            log_layer.with_filter(rust_log_env_filter())
+        })
+        .init();
+
+    Ok(())
+}
+
+#[tokio::main]
+async fn main() -> anyhow::Result<()> {
+    let cli = CliOpts::parse();
+
+    init_logging()?;
+
+    match cli.command {
+        Commands::Simulate(cmd) => {
+            simulate(&cmd, &PathBuf::from("/tmp/compactions.html")).await?;
+        }
+        Commands::RunSuite => {
+            run_suite().await?;
+        }
+    };
+    Ok(())
+}

pageserver/compaction/src/compact_tiered.rs

@@ -0,0 +1,870 @@
+//! # Tiered compaction algorithm.
+//!
+//! Read all the input delta files, and write a new set of delta files that
+//! include all the input WAL records. See retile_deltas().
+//!
+//! In a "normal" LSM tree, you get to remove any values that are overwritten by
+//! later values, but in our system, we keep all the history. So the reshuffling
+//! doesn't remove any garbage, it just reshuffles the records to reduce read
+//! amplification, i.e. the number of files that you need to access to find the
+//! WAL records for a given key.
+//!
+//! If the new delta files would be very "narrow", i.e. each file would cover
+//! only a narrow key range, then we create a new set of image files
+//! instead. The current threshold is that if the estimated total size of the
+//! image layers is smaller than the size of the deltas, then we create image
+//! layers. That amounts to 2x storage amplification, and it means that the
+//! distance of image layers in LSN dimension is roughly equal to the logical
+//! database size. For example, if the logical database size is 10 GB, we would
+//! generate new image layers every 10 GB of WAL.
+//!
+use futures::StreamExt;
+use tracing::{debug, info};
+
+use std::collections::{HashSet, VecDeque};
+use std::ops::Range;
+
+use crate::helpers::{accum_key_values, keyspace_total_size, merge_delta_keys, overlaps_with};
+use crate::interface::*;
+use utils::lsn::Lsn;
+
+use crate::identify_levels::identify_level;
+
+/// Main entry point to compaction.
+///
+/// The starting point is a cutoff LSN (`end_lsn`). The compaction is run on
+/// everything below that point, that needs compaction. The cutoff LSN must
+/// partition the layers so that there are no layers that span across that
+/// LSN. To start compaction at the top of the tree, pass the end LSN of the
+/// written last L0 layer.
+pub async fn compact_tiered<E: CompactionJobExecutor>(
+    executor: &mut E,
+    end_lsn: Lsn,
+    target_file_size: u64,
+    fanout: u64,
+    ctx: &E::RequestContext,
+) -> anyhow::Result<()> {
+    assert!(fanout >= 2);
+    // Start at L0
+    let mut current_level_no = 0;
+    let mut current_level_target_height = target_file_size;
+    loop {
+        // end LSN +1 to include possible image layers exactly at 'end_lsn'.
+        let all_layers = executor
+            .get_layers(
+                &(E::Key::MIN..E::Key::MAX),
+                &(Lsn(u64::MIN)..end_lsn + 1),
+                ctx,
+            )
+            .await?;
+        info!(
+            "Compacting L{}, total # of layers: {}",
+            current_level_no,
+            all_layers.len()
+        );
+
+        // Identify the range of LSNs that belong to this level. We assume that
+        // each file in this level span an LSN range up to 1.75x target file
+        // size. That should give us enough slop that if we created a slightly
+        // oversized L0 layer, e.g. because flushing the in-memory layer was
+        // delayed for some reason, we don't consider the oversized layer to
+        // belong to L1. But not too much slop, that we don't accidentally
+        // "skip" levels.
+        let max_height = (current_level_target_height as f64 * 1.75) as u64;
+        let Some(level) = identify_level(all_layers, end_lsn, max_height).await? else {
+            break;
+        };
+
+        // Calculate the height of this level. If the # of tiers exceeds the
+        // fanout parameter, it's time to compact it.
+        let depth = level.depth();
+        info!(
+            "Level {} identified as LSN range {}-{}: depth {}",
+            current_level_no, level.lsn_range.start, level.lsn_range.end, depth
+        );
+        for l in &level.layers {
+            debug!("LEVEL {} layer: {}", current_level_no, l.short_id());
+        }
+        if depth < fanout {
+            debug!(
+                level = current_level_no,
+                depth = depth,
+                fanout,
+                "too few deltas to compact"
+            );
+            break;
+        }
+
+        compact_level(
+            &level.lsn_range,
+            &level.layers,
+            executor,
+            target_file_size,
+            ctx,
+        )
+        .await?;
+        if target_file_size == u64::MAX {
+            break;
+        }
+        current_level_no += 1;
+        current_level_target_height = current_level_target_height.saturating_mul(fanout);
+    }
+    Ok(())
+}
+
+async fn compact_level<E: CompactionJobExecutor>(
+    lsn_range: &Range<Lsn>,
+    layers: &[E::Layer],
+    executor: &mut E,
+    target_file_size: u64,
+    ctx: &E::RequestContext,
+) -> anyhow::Result<bool> {
+    let mut layer_fragments = Vec::new();
+    for l in layers {
+        layer_fragments.push(LayerFragment::new(l.clone()));
+    }
+
+    let mut state = LevelCompactionState {
+        target_file_size,
+        _lsn_range: lsn_range.clone(),
+        layers: layer_fragments,
+        jobs: Vec::new(),
+        job_queue: Vec::new(),
+        next_level: false,
+        executor,
+    };
+
+    let first_job = CompactionJob {
+        key_range: E::Key::MIN..E::Key::MAX,
+        lsn_range: lsn_range.clone(),
+        strategy: CompactionStrategy::Divide,
+        input_layers: state
+            .layers
+            .iter()
+            .enumerate()
+            .map(|i| LayerId(i.0))
+            .collect(),
+        completed: false,
+    };
+
+    state.jobs.push(first_job);
+    state.job_queue.push(JobId(0));
+    state.execute(ctx).await?;
+
+    info!(
+        "compaction completed! Need to process next level: {}",
+        state.next_level
+    );
+
+    Ok(state.next_level)
+}
+
+/// Blackboard that keeps track of the state of all the jobs and work remaining
+struct LevelCompactionState<'a, E>
+where
+    E: CompactionJobExecutor,
+{
+    // parameters
+    target_file_size: u64,
+
+    _lsn_range: Range<Lsn>,
+    layers: Vec<LayerFragment<E>>,
+
+    // job queue
+    jobs: Vec<CompactionJob<E>>,
+    job_queue: Vec<JobId>,
+
+    /// If false, no need to compact levels below this
+    next_level: bool,
+
+    /// Interface to the outside world
+    executor: &'a mut E,
+}
+
+#[derive(Debug, Clone, Copy, Hash, PartialEq, Eq)]
+struct LayerId(usize);
+#[derive(Debug, Clone, Copy, Hash, PartialEq, Eq)]
+struct JobId(usize);
+
+struct PendingJobSet {
+    pending: HashSet<JobId>,
+    completed: HashSet<JobId>,
+}
+
+impl PendingJobSet {
+    fn new() -> Self {
+        PendingJobSet {
+            pending: HashSet::new(),
+            completed: HashSet::new(),
+        }
+    }
+
+    fn complete_job(&mut self, job_id: JobId) {
+        self.pending.remove(&job_id);
+        self.completed.insert(job_id);
+    }
+
+    fn all_completed(&self) -> bool {
+        self.pending.is_empty()
+    }
+}
+
+// When we decide to rewrite a set of layers, LayerFragment is used to keep
+// track which new layers supersede an old layer. When all the stakeholder jobs
+// have completed, this layer can be deleted.
+struct LayerFragment<E>
+where
+    E: CompactionJobExecutor,
+{
+    layer: E::Layer,
+
+    // If we will write new layers to replace this one, this keeps track of the
+    // jobs that need to complete before this layer can be deleted. As the jobs
+    // complete, they are moved from 'pending' to 'completed' set. Once the
+    // 'pending' set becomes empty, the layer can be deleted.
+    //
+    // If None, this layer is not rewritten and must not be deleted.
+    deletable_after: Option<PendingJobSet>,
+
+    deleted: bool,
+}
+
+impl<E> LayerFragment<E>
+where
+    E: CompactionJobExecutor,
+{
+    fn new(layer: E::Layer) -> Self {
+        LayerFragment {
+            layer,
+            deletable_after: None,
+            deleted: false,
+        }
+    }
+}
+
+#[derive(PartialEq)]
+enum CompactionStrategy {
+    Divide,
+    CreateDelta,
+    CreateImage,
+}
+
+#[allow(dead_code)] // Todo
+struct CompactionJob<E: CompactionJobExecutor> {
+    key_range: Range<E::Key>,
+    lsn_range: Range<Lsn>,
+
+    strategy: CompactionStrategy,
+
+    input_layers: Vec<LayerId>,
+
+    completed: bool,
+}
+
+impl<'a, E> LevelCompactionState<'a, E>
+where
+    E: CompactionJobExecutor,
+{
+    /// Main loop of the executor.
+    ///
+    /// In each iteration, we take the next job from the queue, and execute it.
+    /// The execution might add new jobs to the queue. Keep going until the
+    /// queue is empty.
+    ///
+    /// Initially, the job queue consists of one Divide job over the whole
+    /// level. On first call, it is divided into smaller jobs.
+    ///
+    async fn execute(&mut self, ctx: &E::RequestContext) -> anyhow::Result<()> {
+        // TODO: this would be pretty straightforward to parallelize with FuturesUnordered
+        while let Some(next_job_id) = self.job_queue.pop() {
+            info!("executing job {}", next_job_id.0);
+            self.execute_job(next_job_id, ctx).await?;
+        }
+
+        // all done!
+        Ok(())
+    }
+
+    async fn execute_job(&mut self, job_id: JobId, ctx: &E::RequestContext) -> anyhow::Result<()> {
+        let job = &self.jobs[job_id.0];
+        match job.strategy {
+            CompactionStrategy::Divide => {
+                self.divide_job(job_id, ctx).await?;
+                Ok(())
+            }
+            CompactionStrategy::CreateDelta => {
+                let mut deltas: Vec<E::DeltaLayer> = Vec::new();
+                let mut layer_ids: Vec<LayerId> = Vec::new();
+                for layer_id in &job.input_layers {
+                    let layer = &self.layers[layer_id.0].layer;
+                    if let Some(dl) = self.executor.downcast_delta_layer(layer).await? {
+                        deltas.push(dl.clone());
+                        layer_ids.push(*layer_id);
+                    }
+                }
+
+                self.executor
+                    .create_delta(&job.lsn_range, &job.key_range, &deltas, ctx)
+                    .await?;
+                self.jobs[job_id.0].completed = true;
+
+                // did we complete any fragments?
+                for layer_id in layer_ids {
+                    let l = &mut self.layers[layer_id.0];
+                    if let Some(deletable_after) = l.deletable_after.as_mut() {
+                        deletable_after.complete_job(job_id);
+                        if deletable_after.all_completed() {
+                            self.executor.delete_layer(&l.layer, ctx).await?;
+                            l.deleted = true;
+                        }
+                    }
+                }
+
+                self.next_level = true;
+
+                Ok(())
+            }
+            CompactionStrategy::CreateImage => {
+                self.executor
+                    .create_image(job.lsn_range.end, &job.key_range, ctx)
+                    .await?;
+                self.jobs[job_id.0].completed = true;
+
+                // TODO: we could check if any layers < PITR horizon became deletable
+                Ok(())
+            }
+        }
+    }
+
+    fn push_job(&mut self, job: CompactionJob<E>) -> JobId {
+        let job_id = JobId(self.jobs.len());
+        self.jobs.push(job);
+        self.job_queue.push(job_id);
+        job_id
+    }
+
+    ///
+    /// Take a partition of the key space, and decide how to compact it.
+    ///
+    /// TODO: Currently, this is called exactly once for the level, and we
+    /// decide whether to create new image layers to cover the whole level, or
+    /// write a new set of delta. In the future, this should try to partition
+    /// the key space, and make the decision separately for each partition.
+    ///
+    async fn divide_job(&mut self, job_id: JobId, ctx: &E::RequestContext) -> anyhow::Result<()> {
+        let job = &self.jobs[job_id.0];
+        assert!(job.strategy == CompactionStrategy::Divide);
+
+        // Check for dummy cases
+        if job.input_layers.is_empty() {
+            return Ok(());
+        }
+
+        let job = &self.jobs[job_id.0];
+        assert!(job.strategy == CompactionStrategy::Divide);
+
+        // Would it be better to create images for this partition?
+        // Decide based on the average density of the level
+        let keyspace_size = keyspace_total_size(
+            &self
+                .executor
+                .get_keyspace(&job.key_range, job.lsn_range.end, ctx)
+                .await?,
+        ) * 8192;
+
+        let wal_size = job
+            .input_layers
+            .iter()
+            .filter(|layer_id| self.layers[layer_id.0].layer.is_delta())
+            .map(|layer_id| self.layers[layer_id.0].layer.file_size())
+            .sum::<u64>();
+        if keyspace_size < wal_size {
+            // seems worth it
+            info!(
+                "covering with images, because keyspace_size is {}, size of deltas between {}-{} is {}",
+                keyspace_size, job.lsn_range.start, job.lsn_range.end, wal_size
+            );
+            self.cover_with_images(job_id, ctx).await
+        } else {
+            // do deltas
+            info!(
+                "coverage not worth it, keyspace_size {}, wal_size {}",
+                keyspace_size, wal_size
+            );
+            self.retile_deltas(job_id, ctx).await
+        }
+    }
+
+    // LSN
+    //  ^
+    //  |
+    //  |                          ###|###|#####
+    //  | +--+-----+--+            +--+-----+--+
+    //  | |  |     |  |            |  |     |  |
+    //  | +--+--+--+--+            +--+--+--+--+
+    //  | |     |     |            |     |     |
+    //  | +---+-+-+---+     ==>    +---+-+-+---+
+    //  | |   |   |   |            |   |   |   |
+    //  | +---+-+-++--+            +---+-+-++--+
+    //  | |     |  |  |            |     |  |  |
+    //  | +-----+--+--+            +-----+--+--+
+    //  |
+    //  +--------------> key
+    //
+    async fn cover_with_images(
+        &mut self,
+        job_id: JobId,
+        ctx: &E::RequestContext,
+    ) -> anyhow::Result<()> {
+        let job = &self.jobs[job_id.0];
+        assert!(job.strategy == CompactionStrategy::Divide);
+
+        // XXX: do we still need the "holes" stuff?
+
+        let mut new_jobs = Vec::new();
+
+        // Slide a window through the keyspace
+        let keyspace = self
+            .executor
+            .get_keyspace(&job.key_range, job.lsn_range.end, ctx)
+            .await?;
+
+        let mut window = KeyspaceWindow::new(
+            E::Key::MIN..E::Key::MAX,
+            keyspace,
+            self.target_file_size / 8192,
+        );
+        while let Some(key_range) = window.choose_next_image() {
+            new_jobs.push(CompactionJob::<E> {
+                key_range,
+                lsn_range: job.lsn_range.clone(),
+                strategy: CompactionStrategy::CreateImage,
+                input_layers: Vec::new(), // XXX: Is it OK for  this to be empty for image layer?
+                completed: false,
+            });
+        }
+
+        for j in new_jobs.into_iter().rev() {
+            let _job_id = self.push_job(j);
+
+            // TODO: image layers don't let us delete anything. unless < PITR horizon
+            //let j = &self.jobs[job_id.0];
+            // for layer_id in j.input_layers.iter() {
+            //    self.layers[layer_id.0].pending_stakeholders.insert(job_id);
+            //}
+        }
+
+        Ok(())
+    }
+
+    // Merge the contents of all the input delta layers into a new set
+    // of delta layers, based on the current partitioning.
+    //
+    // We split the new delta layers on the key dimension. We iterate through
+    // the key space, and for each key, check if including the next key to the
+    // current output layer we're building would cause the layer to become too
+    // large. If so, dump the current output layer and start new one.  It's
+    // possible that there is a single key with so many page versions that
+    // storing all of them in a single layer file would be too large. In that
+    // case, we also split on the LSN dimension.
+    //
+    // LSN
+    //  ^
+    //  |
+    //  | +-----------+            +--+--+--+--+
+    //  | |           |            |  |  |  |  |
+    //  | +-----------+            |  |  |  |  |
+    //  | |           |            |  |  |  |  |
+    //  | +-----------+     ==>    |  |  |  |  |
+    //  | |           |            |  |  |  |  |
+    //  | +-----------+            |  |  |  |  |
+    //  | |           |            |  |  |  |  |
+    //  | +-----------+            +--+--+--+--+
+    //  |
+    //  +--------------> key
+    //
+    //
+    // If one key (X) has a lot of page versions:
+    //
+    // LSN
+    //  ^
+    //  |                                 (X)
+    //  | +-----------+            +--+--+--+--+
+    //  | |           |            |  |  |  |  |
+    //  | +-----------+            |  |  +--+  |
+    //  | |           |            |  |  |  |  |
+    //  | +-----------+     ==>    |  |  |  |  |
+    //  | |           |            |  |  +--+  |
+    //  | +-----------+            |  |  |  |  |
+    //  | |           |            |  |  |  |  |
+    //  | +-----------+            +--+--+--+--+
+    //  |
+    //  +--------------> key
+    //
+    // TODO: this actually divides the layers into fixed-size chunks, not
+    // based on the partitioning.
+    //
+    // TODO: we should also opportunistically materialize and
+    // garbage collect what we can.
+    async fn retile_deltas(
+        &mut self,
+        job_id: JobId,
+        ctx: &E::RequestContext,
+    ) -> anyhow::Result<()> {
+        let job = &self.jobs[job_id.0];
+        assert!(job.strategy == CompactionStrategy::Divide);
+
+        // Sweep the key space left to right, running an estimate of how much
+        // disk size and keyspace we have accumulated
+        //
+        // Once the disk size reaches the target threshold, stop and think.
+        // If we have accumulated only a narrow band of keyspace, create an
+        // image layer. Otherwise write a delta layer.
+
+        // FIXME: deal with the case of lots of values for same key
+
+        // FIXME: we are ignoring images here. Did we already divide the work
+        // so that we won't encounter them here?
+
+        let mut deltas: Vec<E::DeltaLayer> = Vec::new();
+        for layer_id in &job.input_layers {
+            let l = &self.layers[layer_id.0];
+            if let Some(dl) = self.executor.downcast_delta_layer(&l.layer).await? {
+                deltas.push(dl.clone());
+            }
+        }
+        // Open stream
+        let key_value_stream = std::pin::pin!(merge_delta_keys::<E>(deltas.as_slice(), ctx));
+        let mut new_jobs = Vec::new();
+
+        // Slide a window through the keyspace
+        let mut key_accum = std::pin::pin!(accum_key_values(key_value_stream));
+        let mut all_in_window: bool = false;
+        let mut window = Window::new();
+        loop {
+            if all_in_window && window.elems.is_empty() {
+                // All done!
+                break;
+            }
+            if let Some(key_range) = window.choose_next_delta(self.target_file_size, !all_in_window)
+            {
+                let batch_layers: Vec<LayerId> = job
+                    .input_layers
+                    .iter()
+                    .filter(|layer_id| {
+                        overlaps_with(self.layers[layer_id.0].layer.key_range(), &key_range)
+                    })
+                    .cloned()
+                    .collect();
+                assert!(!batch_layers.is_empty());
+                new_jobs.push(CompactionJob {
+                    key_range,
+                    lsn_range: job.lsn_range.clone(),
+                    strategy: CompactionStrategy::CreateDelta,
+                    input_layers: batch_layers,
+                    completed: false,
+                });
+            } else {
+                assert!(!all_in_window);
+                if let Some(next_key) = key_accum.next().await.transpose()? {
+                    window.feed(next_key.key, next_key.size);
+                } else {
+                    all_in_window = true;
+                }
+            }
+        }
+
+        // All the input files are rewritten. Set up the tracking for when they can
+        // be deleted.
+        for layer_id in job.input_layers.iter() {
+            let l = &mut self.layers[layer_id.0];
+            assert!(l.deletable_after.is_none());
+            l.deletable_after = Some(PendingJobSet::new());
+        }
+        for j in new_jobs.into_iter().rev() {
+            let job_id = self.push_job(j);
+            let j = &self.jobs[job_id.0];
+            for layer_id in j.input_layers.iter() {
+                self.layers[layer_id.0]
+                    .deletable_after
+                    .as_mut()
+                    .unwrap()
+                    .pending
+                    .insert(job_id);
+            }
+        }
+
+        Ok(())
+    }
+}
+
+// Sliding window through keyspace and values
+// This is used by over_with_images to decide on good split points
+struct KeyspaceWindow<K> {
+    head: KeyspaceWindowHead<K>,
+
+    start_pos: KeyspaceWindowPos<K>,
+}
+struct KeyspaceWindowHead<K> {
+    // overall key range to cover
+    key_range: Range<K>,
+
+    keyspace: Vec<Range<K>>,
+    target_keysize: u64,
+}
+
+#[derive(Clone)]
+struct KeyspaceWindowPos<K> {
+    end_key: K,
+
+    keyspace_idx: usize,
+
+    accum_keysize: u64,
+}
+impl<K: CompactionKey> KeyspaceWindowPos<K> {
+    fn reached_end(&self, w: &KeyspaceWindowHead<K>) -> bool {
+        self.keyspace_idx == w.keyspace.len()
+    }
+
+    // Advance the cursor until it reaches 'target_keysize'.
+    fn advance_until_size(&mut self, w: &KeyspaceWindowHead<K>, max_size: u64) {
+        while self.accum_keysize < max_size && !self.reached_end(w) {
+            let curr_range = &w.keyspace[self.keyspace_idx];
+            if self.end_key < curr_range.start {
+                // skip over any unused space
+                self.end_key = curr_range.start;
+            }
+
+            // We're now within 'curr_range'. Can we advance past it completely?
+            let distance = K::key_range_size(&(self.end_key..curr_range.end));
+            if (self.accum_keysize + distance as u64) < max_size {
+                // oh yeah, it fits
+                self.end_key = curr_range.end;
+                self.keyspace_idx += 1;
+                self.accum_keysize += distance as u64;
+            } else {
+                // advance within the range
+                let skip_key = self.end_key.skip_some();
+                let distance = K::key_range_size(&(self.end_key..skip_key));
+                if (self.accum_keysize + distance as u64) < max_size {
+                    self.end_key = skip_key;
+                    self.accum_keysize += distance as u64;
+                } else {
+                    self.end_key = self.end_key.next();
+                    self.accum_keysize += 1;
+                }
+            }
+        }
+    }
+}
+
+impl<K> KeyspaceWindow<K>
+where
+    K: CompactionKey,
+{
+    fn new(key_range: Range<K>, keyspace: CompactionKeySpace<K>, target_keysize: u64) -> Self {
+        assert!(keyspace.first().unwrap().start >= key_range.start);
+
+        let start_key = key_range.start;
+        let start_pos = KeyspaceWindowPos::<K> {
+            end_key: start_key,
+            keyspace_idx: 0,
+            accum_keysize: 0,
+        };
+        Self {
+            head: KeyspaceWindowHead::<K> {
+                key_range,
+                keyspace,
+                target_keysize,
+            },
+            start_pos,
+        }
+    }
+
+    fn choose_next_image(&mut self) -> Option<Range<K>> {
+        if self.start_pos.keyspace_idx == self.head.keyspace.len() {
+            // we've reached the end
+            return None;
+        }
+
+        let mut next_pos = self.start_pos.clone();
+        next_pos.advance_until_size(
+            &self.head,
+            self.start_pos.accum_keysize + self.head.target_keysize,
+        );
+
+        // See if we can gobble up the rest of the keyspace if we stretch out the layer, up to
+        // 1.25x target size
+        let mut end_pos = next_pos.clone();
+        end_pos.advance_until_size(
+            &self.head,
+            self.start_pos.accum_keysize + (self.head.target_keysize * 5 / 4),
+        );
+        if end_pos.reached_end(&self.head) {
+            // gobble up any unused keyspace between the last used key and end of the range
+            assert!(end_pos.end_key <= self.head.key_range.end);
+            end_pos.end_key = self.head.key_range.end;
+            next_pos = end_pos;
+        }
+
+        let start_key = self.start_pos.end_key;
+        self.start_pos = next_pos;
+        Some(start_key..self.start_pos.end_key)
+    }
+}
+
+// Sliding window through keyspace and values
+//
+// This is used to decide what layer to write next, from the beginning of the window.
+//
+// Candidates:
+//
+// 1. Create an image layer, snapping to previous images
+// 2. Create a delta layer, snapping to previous images
+// 3. Create an image layer, snapping to
+//
+//
+
+// Take previous partitioning, based on the image layers below.
+//
+// Candidate is at the front:
+//
+// Consider stretching an image layer to next divider? If it's close enough,
+// that's the image candidate
+//
+// If it's too far, consider splitting at a reasonable point
+//
+// Is the image candidate smaller than the equivalent delta? If so,
+// split off the image. Otherwise, split off one delta.
+// Try to snap off the delta at a reasonable point
+
+struct WindowElement<K> {
+    start_key: K, // inclusive
+    last_key: K,  // inclusive
+    accum_size: u64,
+}
+struct Window<K> {
+    elems: VecDeque<WindowElement<K>>,
+
+    // last key that was split off, inclusive
+    splitoff_key: Option<K>,
+    splitoff_size: u64,
+}
+
+impl<K> Window<K>
+where
+    K: CompactionKey,
+{
+    fn new() -> Self {
+        Self {
+            elems: VecDeque::new(),
+            splitoff_key: None,
+            splitoff_size: 0,
+        }
+    }
+
+    fn feed(&mut self, key: K, size: u64) {
+        let last_size;
+        if let Some(last) = self.elems.back_mut() {
+            assert!(last.last_key <= key);
+            if key == last.last_key {
+                last.accum_size += size;
+                return;
+            }
+            last_size = last.accum_size;
+        } else {
+            last_size = 0;
+        }
+        // This is a new key.
+        let elem = WindowElement {
+            start_key: key,
+            last_key: key,
+            accum_size: last_size + size,
+        };
+        self.elems.push_back(elem);
+    }
+
+    fn remain_size(&self) -> u64 {
+        self.elems.back().unwrap().accum_size - self.splitoff_size
+    }
+
+    fn peek_size(&self) -> u64 {
+        self.elems.front().unwrap().accum_size - self.splitoff_size
+    }
+
+    fn commit_upto(&mut self, mut upto: usize) {
+        while upto > 1 {
+            let popped = self.elems.pop_front().unwrap();
+            self.elems.front_mut().unwrap().start_key = popped.start_key;
+            upto -= 1;
+        }
+    }
+
+    fn find_size_split(&self, target_size: u64) -> usize {
+        self.elems
+            .partition_point(|elem| elem.accum_size - self.splitoff_size < target_size)
+    }
+
+    fn pop(&mut self) {
+        let first = self.elems.pop_front().unwrap();
+        self.splitoff_size = first.accum_size;
+
+        self.splitoff_key = Some(first.last_key);
+    }
+
+    // the difference between delta and image is that an image covers
+    // any unused keyspace before and after, while a delta tries to
+    // minimize that. TODO: difference not implemented
+    fn pop_delta(&mut self) -> Range<K> {
+        let first = self.elems.front().unwrap();
+        let key_range = first.start_key..first.last_key.next();
+
+        self.pop();
+        key_range
+    }
+
+    // Prerequisite: we have enough input in the window
+    //
+    // On return None, the caller should feed more data and call again
+    fn choose_next_delta(&mut self, target_size: u64, has_more: bool) -> Option<Range<K>> {
+        if has_more && self.elems.is_empty() {
+            // Starting up
+            return None;
+        }
+
+        // If we still have an undersized candidate, just keep going
+        while self.peek_size() < target_size {
+            if self.elems.len() > 1 {
+                self.commit_upto(2);
+            } else if has_more {
+                return None;
+            } else {
+                break;
+            }
+        }
+
+        // Ensure we have enough input in the window to make a good decision
+        if has_more && self.remain_size() < target_size * 5 / 4 {
+            return None;
+        }
+
+        // The candidate on the front is now large enough, for a delta.
+        // And we have enough data in the window to decide.
+
+        // If we're willing to stretch it up to 1.25 target size, could we
+        // gobble up the rest of the work? This avoids creating very small
+        // "tail" layers at the end of the keyspace
+        if !has_more && self.remain_size() < target_size * 5 / 3 {
+            self.commit_upto(self.elems.len());
+        } else {
+            let delta_split_at = self.find_size_split(target_size);
+            self.commit_upto(delta_split_at);
+
+            // If it's still not large enough, request the caller to fill the window
+            if self.elems.len() == 1 && has_more {
+                return None;
+            }
+        }
+        Some(self.pop_delta())
+    }
+}

pageserver/compaction/src/helpers.rs

@@ -0,0 +1,251 @@
+//! This file contains generic utility functions over the interface types,
+//! which could be handy for any compaction implementation.
+use crate::interface::*;
+
+use futures::future::BoxFuture;
+use futures::{Stream, StreamExt};
+use itertools::Itertools;
+use pin_project_lite::pin_project;
+use std::cmp::Ord;
+use std::collections::BinaryHeap;
+use std::collections::VecDeque;
+use std::future::Future;
+use std::ops::{DerefMut, Range};
+use std::pin::Pin;
+use std::task::Poll;
+
+pub fn keyspace_total_size<K>(keyspace: &CompactionKeySpace<K>) -> u64
+where
+    K: CompactionKey,
+{
+    let mut total = 0;
+    for r in keyspace.iter() {
+        total += K::key_range_size(r) as u64;
+    }
+    total
+}
+
+pub fn overlaps_with<T: Ord>(a: &Range<T>, b: &Range<T>) -> bool {
+    !(a.end <= b.start || b.end <= a.start)
+}
+
+pub fn union_to_keyspace<K: Ord>(a: &mut CompactionKeySpace<K>, b: CompactionKeySpace<K>) {
+    let x = std::mem::take(a);
+    let mut all_ranges_iter = [x.into_iter(), b.into_iter()]
+        .into_iter()
+        .kmerge_by(|a, b| a.start < b.start);
+    let mut ranges = Vec::new();
+    if let Some(first) = all_ranges_iter.next() {
+        let (mut start, mut end) = (first.start, first.end);
+
+        for r in all_ranges_iter {
+            assert!(r.start >= start);
+            if r.start > end {
+                ranges.push(start..end);
+                start = r.start;
+                end = r.end;
+            } else if r.end > end {
+                end = r.end;
+            }
+        }
+        ranges.push(start..end);
+    }
+    *a = ranges
+}
+
+pub fn intersect_keyspace<K: Ord + Clone + Copy>(
+    a: &CompactionKeySpace<K>,
+    r: &Range<K>,
+) -> CompactionKeySpace<K> {
+    let mut ranges: Vec<Range<K>> = Vec::new();
+
+    for x in a.iter() {
+        if x.end <= r.start {
+            continue;
+        }
+        if x.start >= r.end {
+            break;
+        }
+        ranges.push(x.clone())
+    }
+
+    // trim the ends
+    if let Some(first) = ranges.first_mut() {
+        first.start = std::cmp::max(first.start, r.start);
+    }
+    if let Some(last) = ranges.last_mut() {
+        last.end = std::cmp::min(last.end, r.end);
+    }
+    ranges
+}
+
+/// Create a stream that iterates through all DeltaEntrys among all input
+/// layers, in key-lsn order.
+///
+/// This is public because the create_delta() implementation likely wants to use this too
+/// TODO: move to a more shared place
+pub fn merge_delta_keys<'a, E: CompactionJobExecutor>(
+    layers: &'a [E::DeltaLayer],
+    ctx: &'a E::RequestContext,
+) -> MergeDeltaKeys<'a, E> {
+    // Use a binary heap to merge the layers. Each input layer is initially
+    // represented by a LazyLoadLayer::Unloaded element, which uses the start of
+    // the layer's key range as the key. The first time a layer reaches the top
+    // of the heap, all the keys of the layer are loaded into a sorted vector.
+    //
+    // This helps to keep the memory usage reasonable: we only need to hold in
+    // memory the DeltaEntrys of the layers that overlap with the "current" key.
+    let mut heap: BinaryHeap<LazyLoadLayer<'a, E>> = BinaryHeap::new();
+    for l in layers {
+        heap.push(LazyLoadLayer::Unloaded(l));
+    }
+    MergeDeltaKeys {
+        heap,
+        ctx,
+        load_future: None,
+    }
+}
+
+enum LazyLoadLayer<'a, E: CompactionJobExecutor> {
+    Loaded(VecDeque<<E::DeltaLayer as CompactionDeltaLayer<E>>::DeltaEntry<'a>>),
+    Unloaded(&'a E::DeltaLayer),
+}
+impl<'a, E: CompactionJobExecutor> LazyLoadLayer<'a, E> {
+    fn key(&self) -> E::Key {
+        match self {
+            Self::Loaded(entries) => entries.front().unwrap().key(),
+            Self::Unloaded(dl) => dl.key_range().start,
+        }
+    }
+}
+impl<'a, E: CompactionJobExecutor> PartialOrd for LazyLoadLayer<'a, E> {
+    fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
+        // reverse order so that we get a min-heap
+        other.key().partial_cmp(&self.key())
+    }
+}
+impl<'a, E: CompactionJobExecutor> Ord for LazyLoadLayer<'a, E> {
+    fn cmp(&self, other: &Self) -> std::cmp::Ordering {
+        // reverse order so that we get a min-heap
+        other.key().cmp(&self.key())
+    }
+}
+impl<'a, E: CompactionJobExecutor> PartialEq for LazyLoadLayer<'a, E> {
+    fn eq(&self, other: &Self) -> bool {
+        self.key().eq(&other.key())
+    }
+}
+impl<'a, E: CompactionJobExecutor> Eq for LazyLoadLayer<'a, E> {}
+
+type LoadFuture<'a, E> = BoxFuture<'a, anyhow::Result<Vec<E>>>;
+
+// Stream returned by `merge_delta_keys`
+pin_project! {
+#[allow(clippy::type_complexity)]
+pub struct MergeDeltaKeys<'a, E: CompactionJobExecutor> {
+    heap: BinaryHeap<LazyLoadLayer<'a, E>>,
+
+    #[pin]
+    load_future: Option<LoadFuture<'a, <E::DeltaLayer as CompactionDeltaLayer<E>>::DeltaEntry<'a>>>,
+
+    ctx: &'a E::RequestContext,
+}
+}
+
+impl<'a, E> Stream for MergeDeltaKeys<'a, E>
+where
+    E: CompactionJobExecutor + 'a,
+{
+    type Item = anyhow::Result<<E::DeltaLayer as CompactionDeltaLayer<E>>::DeltaEntry<'a>>;
+
+    fn poll_next(
+        self: Pin<&mut Self>,
+        cx: &mut std::task::Context<'_>,
+    ) -> Poll<std::option::Option<<Self as futures::Stream>::Item>> {
+        let mut this = self.project();
+        loop {
+            if let Some(mut load_future) = this.load_future.as_mut().as_pin_mut() {
+                // We are waiting for loading the keys to finish
+                match load_future.as_mut().poll(cx) {
+                    Poll::Ready(Ok(entries)) => {
+                        this.load_future.set(None);
+                        *this.heap.peek_mut().unwrap() =
+                            LazyLoadLayer::Loaded(VecDeque::from(entries));
+                    }
+                    Poll::Ready(Err(e)) => {
+                        return Poll::Ready(Some(Err(e)));
+                    }
+                    Poll::Pending => {
+                        return Poll::Pending;
+                    }
+                }
+            }
+
+            // If the topmost layer in the heap hasn't been loaded yet, start
+            // loading it. Otherwise return the next entry from it and update
+            // the layer's position in the heap (this decreaseKey operation is
+            // performed implicitly when `top` is dropped).
+            if let Some(mut top) = this.heap.peek_mut() {
+                match top.deref_mut() {
+                    LazyLoadLayer::Unloaded(ref mut l) => {
+                        let fut = l.load_keys(this.ctx);
+                        this.load_future.set(Some(Box::pin(fut)));
+                        continue;
+                    }
+                    LazyLoadLayer::Loaded(ref mut entries) => {
+                        let result = entries.pop_front().unwrap();
+                        if entries.is_empty() {
+                            std::collections::binary_heap::PeekMut::pop(top);
+                        }
+                        return Poll::Ready(Some(Ok(result)));
+                    }
+                }
+            } else {
+                return Poll::Ready(None);
+            }
+        }
+    }
+}
+
+// Accumulate values at key boundaries
+pub struct KeySize<K> {
+    pub key: K,
+    pub num_values: u64,
+    pub size: u64,
+}
+
+pub fn accum_key_values<'a, I, K, D, E>(input: I) -> impl Stream<Item = Result<KeySize<K>, E>>
+where
+    K: Eq,
+    I: Stream<Item = Result<D, E>>,
+    D: CompactionDeltaEntry<'a, K>,
+{
+    async_stream::try_stream! {
+        // Initialize the state from the first value
+        let mut input = std::pin::pin!(input);
+
+        if let Some(first) = input.next().await {
+            let first = first?;
+            let mut accum: KeySize<K> = KeySize {
+                key: first.key(),
+                num_values: 1,
+                size: first.size(),
+            };
+            while let Some(this) = input.next().await {
+                let this = this?;
+                if this.key() == accum.key {
+                    accum.size += this.size();
+                    accum.num_values += 1;
+                } else {
+                    yield accum;
+                    accum = KeySize {
+                        key: this.key(),
+                        num_values: 1,
+                        size: this.size(),
+                    };
+                }
+            }
+            yield accum;
+        }
+    }
+}

pageserver/compaction/src/identify_levels.rs

@@ -0,0 +1,376 @@
+//! An LSM tree consists of multiple levels, each exponential larger than the
+//! previous level. And each level consists of be multiple "tiers". With tiered
+//! compaction, a level is compacted when it has accumulated more than N tiers,
+//! forming one tier on the next level.
+//!
+//! In the pageserver, we don't explicitly track the levels and tiers. Instead,
+//! we identify them by looking at the shapes of the layers. It's an easy task
+//! for a human, but it's not straightforward to come up with the exact
+//! rules. Especially if there are cases like interrupted, half-finished
+//! compactions, or highly skewed data distributions that have let us "skip"
+//! some levels. It's not critical to classify all cases correctly; at worst we
+//! delay some compaction work, and suffer from more read amplification, or we
+//! perform some unnecessary compaction work.
+//!
+//! `identify_level` performs that shape-matching.
+//!
+//! It returns a Level struct, which has `depth()` function to count the number
+//! of "tiers" in the level. The tier count is the max depth of stacked layers
+//! within the level. That's a good measure, because the point of compacting is
+//! to reduce read amplification, and the depth is what determines that.
+//!
+//! One interesting effect of this is that if we generate very small delta
+//! layers at L0, e.g. because the L0 layers are flushed by timeout rather than
+//! because they reach the target size, the L0 compaction will combine them to
+//! one larger file. But if the combined file is still smaller than the target
+//! file size, the file will still be considered to be part of L0 at the next
+//! iteration.
+
+use anyhow::bail;
+use std::collections::BTreeSet;
+use std::ops::Range;
+use utils::lsn::Lsn;
+
+use crate::interface::*;
+
+use tracing::{info, trace};
+
+pub struct Level<L> {
+    pub lsn_range: Range<Lsn>,
+    pub layers: Vec<L>,
+}
+
+/// Identify an LSN > `end_lsn` that partitions the LSN space, so that there are
+/// no layers that cross the boundary LSN.
+///
+/// A further restriction is that all layers in the returned partition cover at
+/// most 'lsn_max_size' LSN bytes.
+pub async fn identify_level<K, L>(
+    all_layers: Vec<L>,
+    end_lsn: Lsn,
+    lsn_max_size: u64,
+) -> anyhow::Result<Option<Level<L>>>
+where
+    K: CompactionKey,
+    L: CompactionLayer<K> + Clone,
+{
+    // filter out layers that are above the `end_lsn`, they are completely irrelevant.
+    let mut layers = Vec::new();
+    for l in all_layers {
+        if l.lsn_range().start < end_lsn && l.lsn_range().end > end_lsn {
+            // shouldn't happen. Indicates that the caller passed a bogus
+            // end_lsn.
+            bail!("identify_level() called with end_lsn that does not partition the LSN space: end_lsn {} intersects with layer {}", end_lsn, l.short_id());
+        }
+        // include image layers sitting exacty at `end_lsn`.
+        let is_image = !l.is_delta();
+        if (is_image && l.lsn_range().start > end_lsn)
+            || (!is_image && l.lsn_range().start >= end_lsn)
+        {
+            continue;
+        }
+        layers.push(l);
+    }
+    // All the remaining layers either belong to this level, or are below it.
+    info!(
+        "identify level at {}, size {}, num layers below: {}",
+        end_lsn,
+        lsn_max_size,
+        layers.len()
+    );
+    if layers.is_empty() {
+        return Ok(None);
+    }
+
+    // Walk the ranges in LSN order.
+    //
+    // ----- end_lsn
+    //  |
+    //  |
+    //  v
+    //
+    layers.sort_by_key(|l| l.lsn_range().end);
+    let mut candidate_start_lsn = end_lsn;
+    let mut candidate_layers: Vec<L> = Vec::new();
+    let mut current_best_start_lsn = end_lsn;
+    let mut current_best_layers: Vec<L> = Vec::new();
+    let mut iter = layers.into_iter();
+    loop {
+        let Some(l) = iter.next_back() else {
+            // Reached end. Accept the last candidate
+            current_best_start_lsn = candidate_start_lsn;
+            current_best_layers.extend_from_slice(&std::mem::take(&mut candidate_layers));
+            break;
+        };
+        trace!(
+            "inspecting {} for candidate {}, current best {}",
+            l.short_id(),
+            candidate_start_lsn,
+            current_best_start_lsn
+        );
+
+        let r = l.lsn_range();
+
+        // Image layers don't restrict our choice of cutoff LSN
+        if l.is_delta() {
+            // Is this candidate workable? In other words, are there any
+            // delta layers that span across this LSN
+            //
+            // Valid:                 Not valid:
+            //  +                     +
+            //  |                     | +
+            //  +  <- candidate       + |   <- candidate
+            //     +                    +
+            //     |
+            //     +
+            if r.end <= candidate_start_lsn {
+                // Hooray, there are no crossing LSNs. And we have visited
+                // through all the layers within candidate..end_lsn. The
+                // current candidate can be accepted.
+                current_best_start_lsn = r.end;
+                current_best_layers.extend_from_slice(&std::mem::take(&mut candidate_layers));
+                candidate_start_lsn = r.start;
+            }
+
+            // Is it small enough to be considered part of this level?
+            if r.end.0 - r.start.0 > lsn_max_size {
+                // Too large, this layer belongs to next level. Stop.
+                trace!(
+                    "too large {}, size {} vs {}",
+                    l.short_id(),
+                    r.end.0 - r.start.0,
+                    lsn_max_size
+                );
+                break;
+            }
+
+            // If this crosses the candidate lsn, push it down.
+            if r.start < candidate_start_lsn {
+                trace!(
+                    "layer {} prevents from stopping at {}",
+                    l.short_id(),
+                    candidate_start_lsn
+                );
+                candidate_start_lsn = r.start;
+            }
+        }
+
+        // Include this layer in our candidate
+        candidate_layers.push(l);
+    }
+
+    Ok(if current_best_start_lsn == end_lsn {
+        // empty level
+        None
+    } else {
+        Some(Level {
+            lsn_range: current_best_start_lsn..end_lsn,
+            layers: current_best_layers,
+        })
+    })
+}
+
+// helper struct used in depth()
+struct Event<K> {
+    key: K,
+    layer_idx: usize,
+    start: bool,
+}
+
+impl<L> Level<L> {
+    /// Count the number of deltas stacked on each other.
+    pub fn depth<K>(&self) -> u64
+    where
+        K: CompactionKey,
+        L: CompactionLayer<K>,
+    {
+        let mut events: Vec<Event<K>> = Vec::new();
+        for (idx, l) in self.layers.iter().enumerate() {
+            events.push(Event {
+                key: l.key_range().start,
+                layer_idx: idx,
+                start: true,
+            });
+            events.push(Event {
+                key: l.key_range().end,
+                layer_idx: idx,
+                start: false,
+            });
+        }
+        events.sort_by_key(|e| (e.key, e.start));
+
+        // Sweep the key space left to right. Stop at each distinct key, and
+        // count the number of deltas on top of the highest image at that key.
+        //
+        // This is a little enefficient, as we walk through the active_set on
+        // every key. We could increment/decrement a counter on each step
+        // instead, but that'd require a bit more complex bookkeeping.
+        let mut active_set: BTreeSet<(Lsn, bool, usize)> = BTreeSet::new();
+        let mut max_depth = 0;
+        let mut events_iter = events.iter().peekable();
+        while let Some(e) = events_iter.next() {
+            let l = &self.layers[e.layer_idx];
+            let is_image = !l.is_delta();
+
+            // update the active set
+            if e.start {
+                active_set.insert((l.lsn_range().end, is_image, e.layer_idx));
+            } else {
+                active_set.remove(&(l.lsn_range().end, is_image, e.layer_idx));
+            }
+
+            // recalculate depth if this was the last event at this point
+            let more_events_at_this_key = events_iter
+                .peek()
+                .map_or(false, |next_e| next_e.key == e.key);
+            if !more_events_at_this_key {
+                let mut active_depth = 0;
+                for (_end_lsn, is_image, _idx) in active_set.iter().rev() {
+                    if *is_image {
+                        break;
+                    }
+                    active_depth += 1;
+                }
+                if active_depth > max_depth {
+                    max_depth = active_depth;
+                }
+            }
+        }
+        max_depth
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::simulator::{Key, MockDeltaLayer, MockImageLayer, MockLayer};
+    use std::sync::{Arc, Mutex};
+
+    fn delta(key_range: Range<Key>, lsn_range: Range<Lsn>) -> MockLayer {
+        MockLayer::Delta(Arc::new(MockDeltaLayer {
+            key_range,
+            lsn_range,
+            // identify_level() doesn't pay attention to the rest of the fields
+            file_size: 0,
+            deleted: Mutex::new(false),
+            records: vec![],
+        }))
+    }
+
+    fn image(key_range: Range<Key>, lsn: Lsn) -> MockLayer {
+        MockLayer::Image(Arc::new(MockImageLayer {
+            key_range,
+            lsn_range: lsn..(lsn + 1),
+            // identify_level() doesn't pay attention to the rest of the fields
+            file_size: 0,
+            deleted: Mutex::new(false),
+        }))
+    }
+
+    #[tokio::test]
+    async fn test_identify_level() -> anyhow::Result<()> {
+        let layers = vec![
+            delta(Key::MIN..Key::MAX, Lsn(0x8000)..Lsn(0x9000)),
+            delta(Key::MIN..Key::MAX, Lsn(0x5000)..Lsn(0x7000)),
+            delta(Key::MIN..Key::MAX, Lsn(0x4000)..Lsn(0x5000)),
+            delta(Key::MIN..Key::MAX, Lsn(0x3000)..Lsn(0x4000)),
+            delta(Key::MIN..Key::MAX, Lsn(0x2000)..Lsn(0x3000)),
+            delta(Key::MIN..Key::MAX, Lsn(0x1000)..Lsn(0x2000)),
+        ];
+
+        // All layers fit in the max file size
+        let level = identify_level(layers.clone(), Lsn(0x10000), 0x2000)
+            .await?
+            .unwrap();
+        assert_eq!(level.depth(), 6);
+
+        // Same LSN with smaller max file size. The second layer from the top is larger
+        // and belongs to next level.
+        let level = identify_level(layers.clone(), Lsn(0x10000), 0x1000)
+            .await?
+            .unwrap();
+        assert_eq!(level.depth(), 1);
+
+        // Call with a smaller LSN
+        let level = identify_level(layers.clone(), Lsn(0x3000), 0x1000)
+            .await?
+            .unwrap();
+        assert_eq!(level.depth(), 2);
+
+        // Call with an LSN that doesn't partition the space
+        let result = identify_level(layers, Lsn(0x6000), 0x1000).await;
+        assert!(result.is_err());
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn test_overlapping_lsn_ranges() -> anyhow::Result<()> {
+        // The files LSN ranges overlap, so even though there are more files that
+        // fit under the file size, they are not included in the level because they
+        // overlap so that we'd need to include the oldest file, too, which is
+        // larger
+        let layers = vec![
+            delta(Key::MIN..Key::MAX, Lsn(0x4000)..Lsn(0x5000)),
+            delta(Key::MIN..Key::MAX, Lsn(0x3000)..Lsn(0x4000)), // overlap
+            delta(Key::MIN..Key::MAX, Lsn(0x2500)..Lsn(0x3500)), // overlap
+            delta(Key::MIN..Key::MAX, Lsn(0x2000)..Lsn(0x3000)), // overlap
+            delta(Key::MIN..Key::MAX, Lsn(0x1000)..Lsn(0x2500)), // larger
+        ];
+
+        let level = identify_level(layers.clone(), Lsn(0x10000), 0x1000)
+            .await?
+            .unwrap();
+        assert_eq!(level.depth(), 1);
+
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn test_depth_nonoverlapping() -> anyhow::Result<()> {
+        // The key ranges don't overlap, so depth is only 1.
+        let layers = vec![
+            delta(4000..5000, Lsn(0x6000)..Lsn(0x7000)),
+            delta(3000..4000, Lsn(0x7000)..Lsn(0x8000)),
+            delta(1000..2000, Lsn(0x8000)..Lsn(0x9000)),
+        ];
+
+        let level = identify_level(layers.clone(), Lsn(0x10000), 0x2000)
+            .await?
+            .unwrap();
+        assert_eq!(level.layers.len(), 3);
+        assert_eq!(level.depth(), 1);
+
+        // Staggered. The 1st and 3rd layer don't overlap with each other.
+        let layers = vec![
+            delta(1000..2000, Lsn(0x8000)..Lsn(0x9000)),
+            delta(1500..2500, Lsn(0x7000)..Lsn(0x8000)),
+            delta(2000..3000, Lsn(0x6000)..Lsn(0x7000)),
+        ];
+
+        let level = identify_level(layers.clone(), Lsn(0x10000), 0x2000)
+            .await?
+            .unwrap();
+        assert_eq!(level.layers.len(), 3);
+        assert_eq!(level.depth(), 2);
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn test_depth_images() -> anyhow::Result<()> {
+        let layers: Vec<MockLayer> = vec![
+            delta(1000..2000, Lsn(0x8000)..Lsn(0x9000)),
+            delta(1500..2500, Lsn(0x7000)..Lsn(0x8000)),
+            delta(2000..3000, Lsn(0x6000)..Lsn(0x7000)),
+            // This covers the same key range as the 2nd delta layer. The depth
+            // in that key range is therefore 0.
+            image(1500..2500, Lsn(0x9000)),
+        ];
+
+        let level = identify_level(layers.clone(), Lsn(0x10000), 0x2000)
+            .await?
+            .unwrap();
+        assert_eq!(level.layers.len(), 4);
+        assert_eq!(level.depth(), 1);
+        Ok(())
+    }
+}

pageserver/compaction/src/interface.rs

@@ -0,0 +1,152 @@
+//! This is what the compaction implementation needs to know about
+//! layers, keyspace etc.
+//!
+//! All the heavy lifting is done by the create_image and create_delta
+//! functions that the implementor provides.
+//!
+use async_trait::async_trait;
+use std::ops::Range;
+use utils::lsn::Lsn;
+
+/// Public interface. This is the main thing that the implementor needs to provide
+#[async_trait]
+pub trait CompactionJobExecutor {
+    // Type system.
+    //
+    // We assume that there are two kinds of layers, deltas and images. The
+    // compaction doesn't distinguish whether they are stored locally or
+    // remotely.
+    //
+    // The keyspace is defined by CompactionKey trait.
+    //
+    type Key: CompactionKey;
+
+    type Layer: CompactionLayer<Self::Key> + Clone;
+    type DeltaLayer: CompactionDeltaLayer<Self> + Clone;
+    type ImageLayer: CompactionImageLayer<Self> + Clone;
+
+    // This is passed through to all the interface functions. The compaction
+    // implementation doesn't do anything with it, but it might be useful for
+    // the interface implementation.
+    type RequestContext: CompactionRequestContext;
+
+    // ----
+    // Functions that the planner uses to support its decisions
+    // ----
+
+    /// Return all layers that overlap the given bounding box.
+    async fn get_layers(
+        &mut self,
+        key_range: &Range<Self::Key>,
+        lsn_range: &Range<Lsn>,
+        ctx: &Self::RequestContext,
+    ) -> anyhow::Result<Vec<Self::Layer>>;
+
+    async fn get_keyspace(
+        &mut self,
+        key_range: &Range<Self::Key>,
+        lsn: Lsn,
+        ctx: &Self::RequestContext,
+    ) -> anyhow::Result<CompactionKeySpace<Self::Key>>;
+
+    /// NB: This is a pretty expensive operation. In the real pageserver
+    /// implementation, it downloads the layer, and keeps it resident
+    /// until the DeltaLayer is dropped.
+    async fn downcast_delta_layer(
+        &self,
+        layer: &Self::Layer,
+    ) -> anyhow::Result<Option<Self::DeltaLayer>>;
+
+    // ----
+    // Functions to execute the plan
+    // ----
+
+    /// Create a new image layer, materializing all the values in the key range,
+    /// at given 'lsn'.
+    async fn create_image(
+        &mut self,
+        lsn: Lsn,
+        key_range: &Range<Self::Key>,
+        ctx: &Self::RequestContext,
+    ) -> anyhow::Result<()>;
+
+    /// Create a new delta layer, containing all the values from 'input_layers'
+    /// in the given key and LSN range.
+    async fn create_delta(
+        &mut self,
+        lsn_range: &Range<Lsn>,
+        key_range: &Range<Self::Key>,
+        input_layers: &[Self::DeltaLayer],
+        ctx: &Self::RequestContext,
+    ) -> anyhow::Result<()>;
+
+    /// Delete a layer. The compaction implementation will call this only after
+    /// all the create_image() or create_delta() calls that deletion of this
+    /// layer depends on have finished. But if the implementor has extra lazy
+    /// background tasks, like uploading the index json file to remote storage,
+    /// it is the implemenation's responsibility to track those.
+    async fn delete_layer(
+        &mut self,
+        layer: &Self::Layer,
+        ctx: &Self::RequestContext,
+    ) -> anyhow::Result<()>;
+}
+
+pub trait CompactionKey: std::cmp::Ord + Clone + Copy + std::fmt::Display {
+    const MIN: Self;
+    const MAX: Self;
+
+    /// Calculate distance between key_range.start and key_range.end.
+    ///
+    /// This returns u32, for compatibility with Repository::key. If the
+    /// distance is larger, return u32::MAX.
+    fn key_range_size(key_range: &Range<Self>) -> u32;
+
+    // return "self + 1"
+    fn next(&self) -> Self;
+
+    // return "self + <some decent amount to skip>". The amount to skip
+    // is left to the implementation.
+    // FIXME: why not just "add(u32)" ?  This is hard to use
+    fn skip_some(&self) -> Self;
+}
+
+/// Contiguous ranges of keys that belong to the key space. In key order, and
+/// with no overlap.
+pub type CompactionKeySpace<K> = Vec<Range<K>>;
+
+/// Functions needed from all layers.
+pub trait CompactionLayer<K: CompactionKey + ?Sized> {
+    fn key_range(&self) -> &Range<K>;
+    fn lsn_range(&self) -> &Range<Lsn>;
+
+    fn file_size(&self) -> u64;
+
+    /// For debugging, short human-readable representation of the layer. E.g. filename.
+    fn short_id(&self) -> String;
+
+    fn is_delta(&self) -> bool;
+}
+
+#[async_trait]
+pub trait CompactionDeltaLayer<E: CompactionJobExecutor + ?Sized>: CompactionLayer<E::Key> {
+    type DeltaEntry<'a>: CompactionDeltaEntry<'a, E::Key>
+    where
+        Self: 'a;
+
+    /// Return all keys in this delta layer.
+    async fn load_keys<'a>(
+        &self,
+        ctx: &E::RequestContext,
+    ) -> anyhow::Result<Vec<Self::DeltaEntry<'_>>>;
+}
+
+pub trait CompactionImageLayer<E: CompactionJobExecutor + ?Sized>: CompactionLayer<E::Key> {}
+
+pub trait CompactionDeltaEntry<'a, K> {
+    fn key(&self) -> K;
+    fn lsn(&self) -> Lsn;
+    fn size(&self) -> u64;
+}
+
+pub trait CompactionRequestContext {}

pageserver/compaction/src/lib.rs

@@ -0,0 +1,12 @@
+// The main module implementing the compaction algorithm
+pub mod compact_tiered;
+pub(crate) mod identify_levels;
+
+// Traits that the caller of the compaction needs to implement
+pub mod interface;
+
+// Utility functions, useful for the implementation
+pub mod helpers;
+
+// A simulator with mock implementations of 'interface'
+pub mod simulator;

pageserver/compaction/src/simulator.rs

@@ -0,0 +1,613 @@
+mod draw;
+
+use draw::{LayerTraceEvent, LayerTraceFile, LayerTraceOp};
+
+use async_trait::async_trait;
+use futures::StreamExt;
+use rand::Rng;
+use tracing::info;
+
+use utils::lsn::Lsn;
+
+use std::fmt::Write;
+use std::ops::Range;
+use std::sync::Arc;
+use std::sync::Mutex;
+
+use crate::helpers::{merge_delta_keys, overlaps_with};
+
+use crate::interface;
+use crate::interface::CompactionLayer;
+
+//
+// Implementation for the CompactionExecutor interface
+//
+pub struct MockTimeline {
+    // Parameters for the compaction algorithm
+    pub target_file_size: u64,
+    tiers_per_level: u64,
+
+    num_l0_flushes: u64,
+    last_compact_at_flush: u64,
+    last_flush_lsn: Lsn,
+
+    // In-memory layer
+    records: Vec<MockRecord>,
+    total_len: u64,
+    start_lsn: Lsn,
+    end_lsn: Lsn,
+
+    // Current keyspace at `end_lsn`. This is updated on every ingested record.
+    keyspace: KeySpace,
+
+    // historic keyspaces
+    old_keyspaces: Vec<(Lsn, KeySpace)>,
+
+    // "on-disk" layers
+    pub live_layers: Vec<MockLayer>,
+
+    num_deleted_layers: u64,
+
+    // Statistics
+    wal_ingested: u64,
+    bytes_written: u64,
+    bytes_deleted: u64,
+    layers_created: u64,
+    layers_deleted: u64,
+
+    // All the events - creation and deletion of files - are collected
+    // in 'history'. It is used to draw the SVG animation at the end.
+    time: u64,
+    history: Vec<draw::LayerTraceEvent>,
+}
+
+type KeySpace = interface::CompactionKeySpace<Key>;
+
+pub struct MockRequestContext {}
+impl interface::CompactionRequestContext for MockRequestContext {}
+
+pub type Key = u64;
+
+impl interface::CompactionKey for Key {
+    const MIN: Self = u64::MIN;
+    const MAX: Self = u64::MAX;
+
+    fn key_range_size(key_range: &Range<Self>) -> u32 {
+        std::cmp::min(key_range.end - key_range.start, u32::MAX as u64) as u32
+    }
+
+    fn next(&self) -> Self {
+        self + 1
+    }
+    fn skip_some(&self) -> Self {
+        // round up to next xx
+        self + 100
+    }
+}
+
+#[derive(Clone)]
+pub struct MockRecord {
+    lsn: Lsn,
+    key: Key,
+    len: u64,
+}
+
+impl interface::CompactionDeltaEntry<'_, Key> for MockRecord {
+    fn key(&self) -> Key {
+        self.key
+    }
+    fn lsn(&self) -> Lsn {
+        self.lsn
+    }
+    fn size(&self) -> u64 {
+        self.len
+    }
+}
+
+pub struct MockDeltaLayer {
+    pub key_range: Range<Key>,
+    pub lsn_range: Range<Lsn>,
+
+    pub file_size: u64,
+
+    pub deleted: Mutex<bool>,
+
+    pub records: Vec<MockRecord>,
+}
+
+impl interface::CompactionLayer<Key> for Arc<MockDeltaLayer> {
+    fn key_range(&self) -> &Range<Key> {
+        &self.key_range
+    }
+    fn lsn_range(&self) -> &Range<Lsn> {
+        &self.lsn_range
+    }
+
+    fn file_size(&self) -> u64 {
+        self.file_size
+    }
+
+    fn short_id(&self) -> String {
+        format!(
+            "{:016X}-{:016X}__{:08X}-{:08X}",
+            self.key_range.start, self.key_range.end, self.lsn_range.start.0, self.lsn_range.end.0
+        )
+    }
+
+    fn is_delta(&self) -> bool {
+        true
+    }
+}
+
+#[async_trait]
+impl interface::CompactionDeltaLayer<MockTimeline> for Arc<MockDeltaLayer> {
+    type DeltaEntry<'a> = MockRecord;
+
+    async fn load_keys<'a>(&self, _ctx: &MockRequestContext) -> anyhow::Result<Vec<MockRecord>> {
+        Ok(self.records.clone())
+    }
+}
+
+pub struct MockImageLayer {
+    pub key_range: Range<Key>,
+    pub lsn_range: Range<Lsn>,
+
+    pub file_size: u64,
+
+    pub deleted: Mutex<bool>,
+}
+
+impl interface::CompactionImageLayer<MockTimeline> for Arc<MockImageLayer> {}
+
+impl interface::CompactionLayer<Key> for Arc<MockImageLayer> {
+    fn key_range(&self) -> &Range<Key> {
+        &self.key_range
+    }
+    fn lsn_range(&self) -> &Range<Lsn> {
+        &self.lsn_range
+    }
+
+    fn file_size(&self) -> u64 {
+        self.file_size
+    }
+
+    fn short_id(&self) -> String {
+        format!(
+            "{:016X}-{:016X}__{:08X}",
+            self.key_range.start, self.key_range.end, self.lsn_range.start.0,
+        )
+    }
+
+    fn is_delta(&self) -> bool {
+        false
+    }
+}
+
+impl MockTimeline {
+    pub fn new() -> Self {
+        MockTimeline {
+            target_file_size: 256 * 1024 * 1024,
+            tiers_per_level: 4,
+
+            num_l0_flushes: 0,
+            last_compact_at_flush: 0,
+            last_flush_lsn: Lsn(0),
+
+            records: Vec::new(),
+            total_len: 0,
+            start_lsn: Lsn(1000),
+            end_lsn: Lsn(1000),
+            keyspace: KeySpace::new(),
+
+            old_keyspaces: vec![],
+
+            live_layers: vec![],
+
+            num_deleted_layers: 0,
+
+            wal_ingested: 0,
+            bytes_written: 0,
+            bytes_deleted: 0,
+            layers_created: 0,
+            layers_deleted: 0,
+
+            time: 0,
+            history: Vec::new(),
+        }
+    }
+
+    pub async fn compact(&mut self) -> anyhow::Result<()> {
+        let ctx = MockRequestContext {};
+
+        crate::compact_tiered::compact_tiered(
+            self,
+            self.last_flush_lsn,
+            self.target_file_size,
+            self.tiers_per_level,
+            &ctx,
+        )
+        .await?;
+
+        Ok(())
+    }
+
+    // Ingest one record to the timeline
+    pub fn ingest_record(&mut self, key: Key, len: u64) {
+        self.records.push(MockRecord {
+            lsn: self.end_lsn,
+            key,
+            len,
+        });
+        self.total_len += len;
+        self.end_lsn += len;
+
+        if self.total_len > self.target_file_size {
+            self.flush_l0();
+        }
+    }
+
+    pub async fn compact_if_needed(&mut self) -> anyhow::Result<()> {
+        if self.num_l0_flushes - self.last_compact_at_flush >= self.tiers_per_level {
+            self.compact().await?;
+            self.last_compact_at_flush = self.num_l0_flushes;
+        }
+        Ok(())
+    }
+
+    pub fn flush_l0(&mut self) {
+        if self.records.is_empty() {
+            return;
+        }
+
+        let mut records = std::mem::take(&mut self.records);
+        records.sort_by_key(|rec| rec.key);
+
+        let lsn_range = self.start_lsn..self.end_lsn;
+        let new_layer = Arc::new(MockDeltaLayer {
+            key_range: Key::MIN..Key::MAX,
+            lsn_range: lsn_range.clone(),
+            file_size: self.total_len,
+            records,
+            deleted: Mutex::new(false),
+        });
+        info!("flushed L0 layer {}", new_layer.short_id());
+        self.live_layers.push(MockLayer::from(&new_layer));
+
+        // reset L0
+        self.start_lsn = self.end_lsn;
+        self.total_len = 0;
+        self.records = Vec::new();
+
+        self.layers_created += 1;
+        self.bytes_written += new_layer.file_size;
+
+        self.time += 1;
+        self.history.push(LayerTraceEvent {
+            time_rel: self.time,
+            op: LayerTraceOp::Flush,
+            file: LayerTraceFile {
+                filename: new_layer.short_id(),
+                key_range: new_layer.key_range.clone(),
+                lsn_range: new_layer.lsn_range.clone(),
+            },
+        });
+
+        self.num_l0_flushes += 1;
+        self.last_flush_lsn = self.end_lsn;
+    }
+
+    // Ingest `num_records' records to the timeline, with random keys
+    // uniformly distributed in `key_range`
+    pub fn ingest_uniform(
+        &mut self,
+        num_records: u64,
+        len: u64,
+        key_range: &Range<Key>,
+    ) -> anyhow::Result<()> {
+        crate::helpers::union_to_keyspace(&mut self.keyspace, vec![key_range.clone()]);
+        let mut rng = rand::thread_rng();
+        for _ in 0..num_records {
+            self.ingest_record(rng.gen_range(key_range.clone()), len);
+            self.wal_ingested += len;
+        }
+        Ok(())
+    }
+
+    pub fn print_stats(&self) -> anyhow::Result<String> {
+        let mut s = String::new();
+
+        writeln!(s, "STATISTICS:")?;
+        writeln!(
+            s,
+            "WAL ingested:   {:>10} MB",
+            self.wal_ingested / (1024 * 1024)
+        )?;
+        writeln!(
+            s,
+            "size created:   {:>10} MB",
+            self.bytes_written / (1024 * 1024)
+        )?;
+        writeln!(
+            s,
+            "size deleted:   {:>10} MB",
+            self.bytes_deleted / (1024 * 1024)
+        )?;
+        writeln!(s, "files created:     {:>10}", self.layers_created)?;
+        writeln!(s, "files deleted:     {:>10}", self.layers_deleted)?;
+        writeln!(
+            s,
+            "write amp:         {:>10.2}",
+            self.bytes_written as f64 / self.wal_ingested as f64
+        )?;
+        writeln!(
+            s,
+            "storage amp:       {:>10.2}",
+            (self.bytes_written - self.bytes_deleted) as f64 / self.wal_ingested as f64
+        )?;
+
+        Ok(s)
+    }
+
+    pub fn draw_history<W: std::io::Write>(&self, output: W) -> anyhow::Result<()> {
+        draw::draw_history(&self.history, output)
+    }
+}
+
+impl Default for MockTimeline {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+#[derive(Clone)]
+pub enum MockLayer {
+    Delta(Arc<MockDeltaLayer>),
+    Image(Arc<MockImageLayer>),
+}
+
+impl interface::CompactionLayer<Key> for MockLayer {
+    fn key_range(&self) -> &Range<Key> {
+        match self {
+            MockLayer::Delta(this) => this.key_range(),
+            MockLayer::Image(this) => this.key_range(),
+        }
+    }
+    fn lsn_range(&self) -> &Range<Lsn> {
+        match self {
+            MockLayer::Delta(this) => this.lsn_range(),
+            MockLayer::Image(this) => this.lsn_range(),
+        }
+    }
+    fn file_size(&self) -> u64 {
+        match self {
+            MockLayer::Delta(this) => this.file_size(),
+            MockLayer::Image(this) => this.file_size(),
+        }
+    }
+    fn short_id(&self) -> String {
+        match self {
+            MockLayer::Delta(this) => this.short_id(),
+            MockLayer::Image(this) => this.short_id(),
+        }
+    }
+
+    fn is_delta(&self) -> bool {
+        match self {
+            MockLayer::Delta(_) => true,
+            MockLayer::Image(_) => false,
+        }
+    }
+}
+
+impl MockLayer {
+    fn is_deleted(&self) -> bool {
+        let guard = match self {
+            MockLayer::Delta(this) => this.deleted.lock().unwrap(),
+            MockLayer::Image(this) => this.deleted.lock().unwrap(),
+        };
+        *guard
+    }
+    fn mark_deleted(&self) {
+        let mut deleted_guard = match self {
+            MockLayer::Delta(this) => this.deleted.lock().unwrap(),
+            MockLayer::Image(this) => this.deleted.lock().unwrap(),
+        };
+        assert!(!*deleted_guard, "layer already deleted");
+        *deleted_guard = true;
+    }
+}
+
+impl From<&Arc<MockDeltaLayer>> for MockLayer {
+    fn from(l: &Arc<MockDeltaLayer>) -> Self {
+        MockLayer::Delta(l.clone())
+    }
+}
+
+impl From<&Arc<MockImageLayer>> for MockLayer {
+    fn from(l: &Arc<MockImageLayer>) -> Self {
+        MockLayer::Image(l.clone())
+    }
+}
+
+#[async_trait]
+impl interface::CompactionJobExecutor for MockTimeline {
+    type Key = Key;
+    type Layer = MockLayer;
+    type DeltaLayer = Arc<MockDeltaLayer>;
+    type ImageLayer = Arc<MockImageLayer>;
+    type RequestContext = MockRequestContext;
+
+    async fn get_layers(
+        &mut self,
+        key_range: &Range<Self::Key>,
+        lsn_range: &Range<Lsn>,
+        _ctx: &Self::RequestContext,
+    ) -> anyhow::Result<Vec<Self::Layer>> {
+        // Clear any deleted layers from our vec
+        self.live_layers.retain(|l| !l.is_deleted());
+
+        let layers: Vec<MockLayer> = self
+            .live_layers
+            .iter()
+            .filter(|l| {
+                overlaps_with(l.lsn_range(), lsn_range) && overlaps_with(l.key_range(), key_range)
+            })
+            .cloned()
+            .collect();
+
+        Ok(layers)
+    }
+
+    async fn get_keyspace(
+        &mut self,
+        key_range: &Range<Self::Key>,
+        _lsn: Lsn,
+        _ctx: &Self::RequestContext,
+    ) -> anyhow::Result<interface::CompactionKeySpace<Key>> {
+        // find it in the levels
+        if self.old_keyspaces.is_empty() {
+            Ok(crate::helpers::intersect_keyspace(
+                &self.keyspace,
+                key_range,
+            ))
+        } else {
+            // not implemented
+
+            // The mock implementation only allows requesting the
+            // keyspace at the level's end LSN. That's all that the
+            // current implementation needs.
+            panic!("keyspace not available for requested lsn");
+        }
+    }
+
+    async fn downcast_delta_layer(
+        &self,
+        layer: &MockLayer,
+    ) -> anyhow::Result<Option<Arc<MockDeltaLayer>>> {
+        Ok(match layer {
+            MockLayer::Delta(l) => Some(l.clone()),
+            MockLayer::Image(_) => None,
+        })
+    }
+
+    async fn create_image(
+        &mut self,
+        lsn: Lsn,
+        key_range: &Range<Key>,
+        ctx: &MockRequestContext,
+    ) -> anyhow::Result<()> {
+        let keyspace = self.get_keyspace(key_range, lsn, ctx).await?;
+
+        let mut accum_size: u64 = 0;
+        for r in keyspace {
+            accum_size += r.end - r.start;
+        }
+
+        let new_layer = Arc::new(MockImageLayer {
+            key_range: key_range.clone(),
+            lsn_range: lsn..lsn,
+            file_size: accum_size * 8192,
+            deleted: Mutex::new(false),
+        });
+        info!(
+            "created image layer, size {}: {}",
+            new_layer.file_size,
+            new_layer.short_id()
+        );
+        self.live_layers.push(MockLayer::Image(new_layer.clone()));
+
+        // update stats
+        self.bytes_written += new_layer.file_size;
+        self.layers_created += 1;
+
+        self.time += 1;
+        self.history.push(LayerTraceEvent {
+            time_rel: self.time,
+            op: LayerTraceOp::CreateImage,
+            file: LayerTraceFile {
+                filename: new_layer.short_id(),
+                key_range: new_layer.key_range.clone(),
+                lsn_range: new_layer.lsn_range.clone(),
+            },
+        });
+
+        Ok(())
+    }
+
+    async fn create_delta(
+        &mut self,
+        lsn_range: &Range<Lsn>,
+        key_range: &Range<Key>,
+        input_layers: &[Arc<MockDeltaLayer>],
+        ctx: &MockRequestContext,
+    ) -> anyhow::Result<()> {
+        let mut key_value_stream =
+            std::pin::pin!(merge_delta_keys::<MockTimeline>(input_layers, ctx));
+        let mut records: Vec<MockRecord> = Vec::new();
+        let mut total_len = 2;
+        while let Some(delta_entry) = key_value_stream.next().await {
+            let delta_entry: MockRecord = delta_entry?;
+            if key_range.contains(&delta_entry.key) && lsn_range.contains(&delta_entry.lsn) {
+                total_len += delta_entry.len;
+                records.push(delta_entry);
+            }
+        }
+        let total_records = records.len();
+        let new_layer = Arc::new(MockDeltaLayer {
+            key_range: key_range.clone(),
+            lsn_range: lsn_range.clone(),
+            file_size: total_len,
+            records,
+            deleted: Mutex::new(false),
+        });
+        info!(
+            "created delta layer, recs {}, size {}: {}",
+            total_records,
+            total_len,
+            new_layer.short_id()
+        );
+        self.live_layers.push(MockLayer::Delta(new_layer.clone()));
+
+        // update stats
+        self.bytes_written += total_len;
+        self.layers_created += 1;
+
+        self.time += 1;
+        self.history.push(LayerTraceEvent {
+            time_rel: self.time,
+            op: LayerTraceOp::CreateDelta,
+            file: LayerTraceFile {
+                filename: new_layer.short_id(),
+                key_range: new_layer.key_range.clone(),
+                lsn_range: new_layer.lsn_range.clone(),
+            },
+        });
+
+        Ok(())
+    }
+
+    async fn delete_layer(
+        &mut self,
+        layer: &Self::Layer,
+        _ctx: &MockRequestContext,
+    ) -> anyhow::Result<()> {
+        let layer = std::pin::pin!(layer);
+        info!("deleting layer: {}", layer.short_id());
+        self.num_deleted_layers += 1;
+        self.bytes_deleted += layer.file_size();
+        layer.mark_deleted();
+
+        self.time += 1;
+        self.history.push(LayerTraceEvent {
+            time_rel: self.time,
+            op: LayerTraceOp::Delete,
+            file: LayerTraceFile {
+                filename: layer.short_id(),
+                key_range: layer.key_range().clone(),
+                lsn_range: layer.lsn_range().clone(),
+            },
+        });
+
+        Ok(())
+    }
+}

pageserver/compaction/src/simulator/draw.rs

@@ -0,0 +1,411 @@
+use super::Key;
+use anyhow::Result;
+use std::cmp::Ordering;
+use std::{
+    collections::{BTreeMap, BTreeSet, HashSet},
+    fmt::Write,
+    ops::Range,
+};
+use svg_fmt::{rgb, BeginSvg, EndSvg, Fill, Stroke, Style};
+use utils::lsn::Lsn;
+
+// Map values to their compressed coordinate - the index the value
+// would have in a sorted and deduplicated list of all values.
+struct CoordinateMap<T: Ord + Copy> {
+    map: BTreeMap<T, usize>,
+    stretch: f32,
+}
+
+impl<T: Ord + Copy> CoordinateMap<T> {
+    fn new(coords: Vec<T>, stretch: f32) -> Self {
+        let set: BTreeSet<T> = coords.into_iter().collect();
+
+        let mut map: BTreeMap<T, usize> = BTreeMap::new();
+        for (i, e) in set.iter().enumerate() {
+            map.insert(*e, i);
+        }
+
+        Self { map, stretch }
+    }
+
+    // This assumes that the map contains an exact point for this.
+    // Use map_inexact for values inbetween
+    fn map(&self, val: T) -> f32 {
+        *self.map.get(&val).unwrap() as f32 * self.stretch
+    }
+
+    // the value is still assumed to be within the min/max bounds
+    // (this is currently unused)
+    fn _map_inexact(&self, val: T) -> f32 {
+        let prev = *self.map.range(..=val).next().unwrap().1;
+        let next = *self.map.range(val..).next().unwrap().1;
+
+        // interpolate
+        (prev as f32 + (next - prev) as f32) * self.stretch
+    }
+
+    fn max(&self) -> f32 {
+        self.map.len() as f32 * self.stretch
+    }
+}
+
+#[derive(PartialEq, Hash, Eq)]
+pub enum LayerTraceOp {
+    Flush,
+    CreateDelta,
+    CreateImage,
+    Delete,
+}
+
+impl std::fmt::Display for LayerTraceOp {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> Result<(), std::fmt::Error> {
+        let op_str = match self {
+            LayerTraceOp::Flush => "flush",
+            LayerTraceOp::CreateDelta => "create_delta",
+            LayerTraceOp::CreateImage => "create_image",
+            LayerTraceOp::Delete => "delete",
+        };
+        f.write_str(op_str)
+    }
+}
+
+#[derive(PartialEq, Hash, Eq, Clone)]
+pub struct LayerTraceFile {
+    pub filename: String,
+    pub key_range: Range<Key>,
+    pub lsn_range: Range<Lsn>,
+}
+
+impl LayerTraceFile {
+    fn is_image(&self) -> bool {
+        self.lsn_range.end == self.lsn_range.start
+    }
+}
+
+pub struct LayerTraceEvent {
+    pub time_rel: u64,
+    pub op: LayerTraceOp,
+    pub file: LayerTraceFile,
+}
+
+pub fn draw_history<W: std::io::Write>(history: &[LayerTraceEvent], mut output: W) -> Result<()> {
+    let mut files: Vec<LayerTraceFile> = Vec::new();
+
+    for event in history {
+        files.push(event.file.clone());
+    }
+    let last_time_rel = history.last().unwrap().time_rel;
+
+    // Collect all coordinates
+    let mut keys: Vec<Key> = vec![];
+    let mut lsns: Vec<Lsn> = vec![];
+    for f in files.iter() {
+        keys.push(f.key_range.start);
+        keys.push(f.key_range.end);
+        lsns.push(f.lsn_range.start);
+        lsns.push(f.lsn_range.end);
+    }
+
+    // Analyze
+    let key_map = CoordinateMap::new(keys, 2.0);
+    // Stretch out vertically for better visibility
+    let lsn_map = CoordinateMap::new(lsns, 3.0);
+
+    let mut svg = String::new();
+
+    // Draw
+    writeln!(
+        svg,
+        "{}",
+        BeginSvg {
+            w: key_map.max(),
+            h: lsn_map.max(),
+        }
+    )?;
+    let lsn_max = lsn_map.max();
+
+    // Sort the files by LSN, but so that image layers go after all delta layers
+    // The SVG is painted in the order the elements appear, and we want to draw
+    // image layers on top of the delta layers if they overlap
+    //
+    // (This could also be implemented via z coordinates: image layers get one z
+    // coord, delta layers get another z coord.)
+    let mut files_sorted: Vec<LayerTraceFile> = files.into_iter().collect();
+    files_sorted.sort_by(|a, b| {
+        if a.is_image() && !b.is_image() {
+            Ordering::Greater
+        } else if !a.is_image() && b.is_image() {
+            Ordering::Less
+        } else {
+            a.lsn_range.end.cmp(&b.lsn_range.end)
+        }
+    });
+
+    writeln!(svg, "<!-- layers -->")?;
+    let mut files_seen = HashSet::new();
+    for f in files_sorted {
+        if files_seen.contains(&f) {
+            continue;
+        }
+        let key_start = key_map.map(f.key_range.start);
+        let key_end = key_map.map(f.key_range.end);
+        let key_diff = key_end - key_start;
+
+        if key_start >= key_end {
+            panic!("Invalid key range {}-{}", key_start, key_end);
+        }
+
+        let lsn_start = lsn_map.map(f.lsn_range.start);
+        let lsn_end = lsn_map.map(f.lsn_range.end);
+
+        // Fill in and thicken rectangle if it's an
+        // image layer so that we can see it.
+        let mut style = Style::default();
+        style.fill = Fill::Color(rgb(0x80, 0x80, 0x80));
+        style.stroke = Stroke::Color(rgb(0, 0, 0), 0.5);
+
+        let y_start = lsn_max - lsn_start;
+        let y_end = lsn_max - lsn_end;
+
+        let x_margin = 0.25;
+        let y_margin = 0.5;
+
+        match f.lsn_range.start.cmp(&f.lsn_range.end) {
+            Ordering::Less => {
+                write!(
+                    svg,
+                    r#"    <rect id="layer_{}" x="{}" y="{}" width="{}" height="{}" ry="{}" style="{}">"#,
+                    f.filename,
+                    key_start + x_margin,
+                    y_end + y_margin,
+                    key_diff - x_margin * 2.0,
+                    y_start - y_end - y_margin * 2.0,
+                    1.0, // border_radius,
+                    style,
+                )?;
+                write!(svg, "<title>{}</title>", f.filename)?;
+                writeln!(svg, "</rect>")?;
+            }
+            Ordering::Equal => {
+                //lsn_diff = 0.3;
+                //lsn_offset = -lsn_diff / 2.0;
+                //margin = 0.05;
+                style.fill = Fill::Color(rgb(0x80, 0, 0x80));
+                style.stroke = Stroke::Color(rgb(0x80, 0, 0x80), 3.0);
+                write!(
+                    svg,
+                    r#"    <line id="layer_{}" x1="{}" y1="{}" x2="{}" y2="{}" style="{}">"#,
+                    f.filename,
+                    key_start + x_margin,
+                    y_end,
+                    key_end - x_margin,
+                    y_end,
+                    style,
+                )?;
+                write!(
+                    svg,
+                    "<title>{}<br>{} - {}</title>",
+                    f.filename, lsn_end, y_end
+                )?;
+                writeln!(svg, "</line>")?;
+            }
+            Ordering::Greater => panic!("Invalid lsn range {}-{}", lsn_start, lsn_end),
+        }
+        files_seen.insert(f);
+    }
+
+    let mut record_style = Style::default();
+    record_style.fill = Fill::Color(rgb(0x80, 0x80, 0x80));
+    record_style.stroke = Stroke::None;
+
+    writeln!(svg, "{}", EndSvg)?;
+
+    let mut layer_events_str = String::new();
+    let mut first = true;
+    for e in history {
+        if !first {
+            writeln!(layer_events_str, ",")?;
+        }
+        write!(
+            layer_events_str,
+            r#"  {{"time_rel": {}, "filename": "{}", "op": "{}"}}"#,
+            e.time_rel, e.file.filename, e.op
+        )?;
+        first = false;
+    }
+    writeln!(layer_events_str)?;
+
+    writeln!(
+        output,
+        r#"<!DOCTYPE html>
+<html>
+<head>
+<style>
+/* Keep the slider pinned at top */
+.topbar {{
+  display: block;
+  overflow: hidden;
+  background-color: lightgrey;
+  position: fixed;
+  top: 0;
+  width: 100%;
+/*  width: 500px; */
+}}
+.slidercontainer {{
+  float: left;
+  width: 50%;
+  margin-right: 200px;
+}}
+.slider {{
+  float: left;
+  width: 100%;
+}}
+.legend {{
+  width: 200px;
+  float: right;
+}}
+
+/* Main content */
+.main {{
+  margin-top: 50px; /* Add a top margin to avoid content overlay */
+}}
+</style>
+</head>
+
+  <body onload="init()">
+    <script type="text/javascript">
+
+      var layer_events = [{layer_events_str}]
+
+      let ticker;
+
+      function init() {{
+          for (let i = 0; i < layer_events.length; i++) {{
+              var layer = document.getElementById("layer_" + layer_events[i].filename);
+              layer.style.visibility = "hidden";
+          }}
+          last_layer_event = -1;
+          moveSlider(last_slider_pos)
+      }}
+
+      function startAnimation() {{
+          ticker = setInterval(animateStep, 100);
+      }}
+      function stopAnimation() {{
+          clearInterval(ticker);
+      }}
+
+      function animateStep() {{
+          if (last_layer_event < layer_events.length - 1) {{
+              var slider = document.getElementById("time-slider");
+              let prevPos = slider.value
+              let nextEvent = last_layer_event + 1
+              while (nextEvent <= layer_events.length - 1) {{
+                  if (layer_events[nextEvent].time_rel > prevPos) {{
+                      break;
+                  }}
+                  nextEvent += 1;
+              }}
+              let nextPos = layer_events[nextEvent].time_rel
+              slider.value = nextPos
+              moveSlider(nextPos)
+          }}
+      }}
+
+      function redoLayerEvent(n, dir) {{
+          var layer = document.getElementById("layer_" + layer_events[n].filename);
+          switch (layer_events[n].op) {{
+              case "flush":
+                  layer.style.visibility = "visible";
+                  break;
+              case "create_delta":
+                  layer.style.visibility = "visible";
+                  break;
+              case "create_image":
+                  layer.style.visibility = "visible";
+                  break;
+              case "delete":
+                  layer.style.visibility = "hidden";
+                  break;
+          }}
+      }}
+      function undoLayerEvent(n) {{
+          var layer = document.getElementById("layer_" + layer_events[n].filename);
+          switch (layer_events[n].op) {{
+              case "flush":
+                  layer.style.visibility = "hidden";
+                  break;
+              case "create_delta":
+                  layer.style.visibility = "hidden";
+                  break;
+              case "create_image":
+                  layer.style.visibility = "hidden";
+                  break;
+              case "delete":
+                  layer.style.visibility = "visible";
+                  break;
+          }}
+      }}
+
+      var last_slider_pos = 0
+      var last_layer_event = 0
+
+      var moveSlider = function(new_pos) {{
+          if (new_pos > last_slider_pos) {{
+              while (last_layer_event < layer_events.length - 1) {{
+                  if (layer_events[last_layer_event + 1].time_rel > new_pos) {{
+                      break;
+                  }}
+                  last_layer_event += 1;
+                  redoLayerEvent(last_layer_event)
+              }}
+          }}
+          if (new_pos < last_slider_pos) {{
+              while (last_layer_event >= 0) {{
+                  if (layer_events[last_layer_event].time_rel <= new_pos) {{
+                      break;
+                  }}
+                  undoLayerEvent(last_layer_event)
+                  last_layer_event -= 1;
+              }}
+          }}
+          last_slider_pos = new_pos;
+          document.getElementById("debug_pos").textContent=new_pos;
+          if (last_layer_event >= 0) {{
+              document.getElementById("debug_layer_event").textContent=last_layer_event + " " + layer_events[last_layer_event].time_rel + " " + layer_events[last_layer_event].op;
+          }} else {{
+              document.getElementById("debug_layer_event").textContent="begin";
+          }}
+      }}
+    </script>
+
+    <div class="topbar">
+      <div class="slidercontainer">
+        <label for="time-slider">TIME</label>:
+        <input id="time-slider" class="slider" type="range" min="0" max="{last_time_rel}" value="0" oninput="moveSlider(this.value)"><br>
+
+        pos: <span id="debug_pos"></span><br>
+        event: <span id="debug_layer_event"></span><br>
+        gc: <span id="debug_gc_event"></span><br>
+      </div>
+
+      <button onclick="startAnimation()">Play</button>
+      <button onclick="stopAnimation()">Stop</button>
+
+      <svg class="legend">
+        <rect x=5 y=0 width=20 height=20 style="fill:rgb(128,128,128);stroke:rgb(0,0,0);stroke-width:0.5;fill-opacity:1;stroke-opacity:1;"/>
+        <line x1=5 y1=30 x2=25 y2=30 style="fill:rgb(128,0,128);stroke:rgb(128,0,128);stroke-width:3;fill-opacity:1;stroke-opacity:1;"/>
+        <line x1=0 y1=40 x2=30 y2=40 style="fill:none;stroke:rgb(255,0,0);stroke-width:0.5;fill-opacity:1;stroke-opacity:1;"/>
+      </svg>
+    </div>
+
+    <div class="main">
+{svg}
+    </div>
+  </body>
+</html>
+"#
+    )?;
+
+    Ok(())
+}

pageserver/compaction/tests/tests.rs

@@ -0,0 +1,37 @@
+use pageserver_compaction::interface::CompactionLayer;
+use pageserver_compaction::simulator::MockTimeline;
+
+/// Test the extreme case that there are so many updates for a single key that
+/// even if we produce an extremely narrow delta layer, spanning just that one
+/// key, we still too many records to fit in the target file size. We need to
+/// split in the LSN dimension too in that case.
+///
+/// TODO: The code to avoid this problem has not been implemented yet! So the
+/// assertion currently fails, but we need to make it not fail.
+#[ignore]
+#[tokio::test]
+async fn test_many_updates_for_single_key() -> anyhow::Result<()> {
+    let mut executor = MockTimeline::new();
+    executor.target_file_size = 10_000_000; // 10 MB
+
+    // Ingest 100 MB of updates to a single key.
+    for _ in 1..1000 {
+        executor.ingest_uniform(100, 10, &(0..100_000))?;
+        executor.ingest_uniform(10_000, 10, &(0..1))?;
+        executor.compact().await?;
+    }
+
+    // Check that all the layers are smaller than the target size (with some slop)
+    for l in executor.live_layers.iter() {
+        println!("layer {}: {}", l.short_id(), l.file_size());
+    }
+    for l in executor.live_layers.iter() {
+        assert!(l.file_size() < executor.target_file_size * 2);
+        // sanity check that none of the delta layers are stupidly small either
+        if l.is_delta() {
+            assert!(l.file_size() > executor.target_file_size / 2);
+        }
+    }
+
+    Ok(())
+}

pageserver/src/bin/getpage_bench_http.rs

@@ -1,245 +0,0 @@
-use clap::Parser;
-use hyper::client::conn::Parts;
-use hyper::client::HttpConnector;
-use hyper::{Body, Client, Uri};
-use pageserver::{repository, tenant};
-use rand::prelude::*;
-use std::env::args;
-use std::future::Future;
-use std::str::FromStr;
-use std::sync::atomic::{AtomicU64, Ordering};
-use std::sync::{Arc, Mutex};
-use std::thread;
-use tokio::sync::mpsc::{channel, Sender};
-use tokio::sync::Mutex as AsyncMutex;
-use tokio::task::JoinHandle;
-
-struct Key(repository::Key);
-
-impl std::str::FromStr for Key {
-    type Err = anyhow::Error;
-
-    fn from_str(s: &str) -> std::result::Result<Self, Self::Err> {
-        repository::Key::from_hex(s).map(Key)
-    }
-}
-
-struct KeyRange {
-    start: Key,
-    end: Key,
-}
-
-impl KeyRange {
-    fn len(&self) -> i128 {
-        self.end.0.to_i128() - self.start.0.to_i128()
-    }
-}
-
-#[derive(clap::Parser)]
-struct Args {
-    #[clap(long, default_value = "http://localhost:9898")]
-    ps_endpoint: String,
-    // tenant_id: String,
-    // timeline_id: String,
-    num_tasks: usize,
-    num_requests: usize,
-    tenants: Option<Vec<String>>,
-    #[clap(long)]
-    pick_n_tenants: Option<usize>,
-}
-
-#[derive(Debug, Default)]
-struct Stats {
-    completed_requests: AtomicU64,
-}
-
-impl Stats {
-    fn inc(&self) {
-        self.completed_requests.fetch_add(1, Ordering::Relaxed);
-    }
-}
-
-#[tokio::main]
-async fn main() {
-    let args: &'static Args = Box::leak(Box::new(Args::parse()));
-
-    let client = Client::new();
-
-    let tenants = if let Some(tenants) = &args.tenants {
-        tenants.clone()
-    } else {
-        // let tenant_id = "b97965931096047b2d54958756baee7b";
-        // let timeline_id = "2868f84a8d166779e4c651b116c45059";
-
-        let resp = client
-            .get(Uri::try_from(&format!("{}/v1/tenant", args.ps_endpoint)).unwrap())
-            .await
-            .unwrap();
-
-        let body = hyper::body::to_bytes(resp).await.unwrap();
-        let tenants: serde_json::Value = serde_json::from_slice(&body).unwrap();
-        let mut out = Vec::new();
-        for t in tenants.as_array().unwrap() {
-            if let Some(limit) = args.pick_n_tenants {
-                if out.len() >= limit {
-                    break;
-                }
-            }
-            out.push(t.get("id").unwrap().as_str().unwrap().to_owned());
-        }
-        if let Some(limit) = args.pick_n_tenants {
-            assert_eq!(out.len(), limit);
-        }
-        out
-    };
-
-    let mut tenant_timelines = Vec::new();
-    for tenant_id in tenants {
-        let resp = client
-            .get(
-                Uri::try_from(&format!(
-                    "{}/v1/tenant/{}/timeline",
-                    args.ps_endpoint, tenant_id
-                ))
-                .unwrap(),
-            )
-            .await
-            .unwrap();
-
-        let body = hyper::body::to_bytes(resp).await.unwrap();
-        let timelines: serde_json::Value = serde_json::from_slice(&body).unwrap();
-        for t in timelines.as_array().unwrap() {
-            let timeline_id = t.get("timeline_id").unwrap().as_str().unwrap().to_owned();
-            tenant_timelines.push((tenant_id.clone(), timeline_id));
-        }
-    }
-    println!("tenant_timelines:\n{:?}", tenant_timelines);
-
-    let mut stats = Arc::new(Stats::default());
-
-    tokio::spawn({
-        let stats = Arc::clone(&stats);
-        async move {
-            loop {
-                let start = std::time::Instant::now();
-                tokio::time::sleep(std::time::Duration::from_secs(1)).await;
-                let completed_requests = stats.completed_requests.swap(0, Ordering::Relaxed);
-                let elapsed = start.elapsed();
-                println!(
-                    "RPS: {:.0}",
-                    completed_requests as f64 / elapsed.as_secs_f64()
-                );
-            }
-        }
-    });
-
-    let mut tasks = Vec::new();
-    for (tenant_id, timeline_id) in tenant_timelines {
-        let t = tokio::spawn(timeline(
-            args,
-            client.clone(),
-            tenant_id,
-            timeline_id,
-            Arc::clone(&stats),
-        ));
-        tasks.push(t);
-    }
-
-    for t in tasks {
-        t.await.unwrap();
-    }
-}
-
-fn timeline(
-    args: &'static Args,
-    client: Client<HttpConnector, Body>,
-    tenant_id: String,
-    timeline_id: String,
-    stats: Arc<Stats>,
-) -> impl Future<Output = ()> {
-    async move {
-        let mut resp = client
-            .get(
-                Uri::try_from(&format!(
-                    "{}/v1/tenant/{}/timeline/{}/keyspace",
-                    args.ps_endpoint, tenant_id, timeline_id
-                ))
-                .unwrap(),
-            )
-            .await
-            .unwrap();
-        if !resp.status().is_success() {
-            panic!("Failed to get keyspace: {resp:?}");
-        }
-        let body = hyper::body::to_bytes(resp).await.unwrap();
-        let keyspace: serde_json::Value = serde_json::from_slice(&body).unwrap();
-
-        let lsn = Arc::new(keyspace["at_lsn"].as_str().unwrap().to_owned());
-
-        let ranges = keyspace["keys"]
-            .as_array()
-            .unwrap()
-            .iter()
-            .map(|r| {
-                let r = r.as_array().unwrap();
-                assert_eq!(r.len(), 2);
-                let start = Key::from_str(r[0].as_str().unwrap()).unwrap();
-                let end = Key::from_str(r[1].as_str().unwrap()).unwrap();
-                KeyRange { start, end }
-            })
-            .collect::<Vec<_>>();
-
-        // weighted ranges
-        let weights = ranges.iter().map(|r| r.len()).collect::<Vec<_>>();
-
-        let ranges = Arc::new(ranges);
-        let weights = Arc::new(weights);
-
-        let (tx, mut rx) = channel::<i32>(1000);
-        let tx = Arc::new(AsyncMutex::new(tx));
-
-        let mut tasks = Vec::<JoinHandle<()>>::new();
-
-        let start = std::time::Instant::now();
-
-        for i in 0..args.num_tasks {
-            let ranges = ranges.clone();
-            let weights = weights.clone();
-            let lsn = lsn.clone();
-            let client = client.clone();
-            let tenant_id = tenant_id.clone();
-            let timeline_id = timeline_id.clone();
-            let stats = Arc::clone(&stats);
-            let task = tokio::spawn(async move {
-                for i in 0..args.num_requests {
-                    let key = {
-                        let mut rng = rand::thread_rng();
-                        let r = ranges.choose_weighted(&mut rng, |r| r.len()).unwrap();
-                        let key = rng.gen_range((r.start.0.to_i128()..r.end.0.to_i128()));
-                        key
-                    };
-                    let url = format!(
-                        "{}/v1/tenant/{}/timeline/{}/getpage?key={:036x}&lsn={}",
-                        args.ps_endpoint, tenant_id, timeline_id, key, lsn
-                    );
-                    let uri = url.parse::<Uri>().unwrap();
-                    let resp = client.get(uri).await.unwrap();
-                    stats.inc();
-                }
-            });
-            tasks.push(task);
-        }
-
-        drop(tx);
-
-        for task in tasks {
-            task.await.unwrap();
-        }
-
-        let elapsed = start.elapsed();
-        println!(
-            "RPS: {:.0}",
-            (args.num_requests * args.num_tasks) as f64 / elapsed.as_secs_f64()
-        );
-    }
-}

pageserver/src/bin/getpage_bench_libpq.rs

@@ -1,411 +0,0 @@
-use anyhow::Context;
-use clap::Parser;
-use futures::{SinkExt, TryStreamExt};
-use hyper::client::conn::Parts;
-use hyper::client::HttpConnector;
-use hyper::{Client, Uri};
-use pageserver::page_cache::PAGE_SZ;
-use pageserver::pgdatadir_mapping::{is_rel_block_key, key_to_rel_block};
-use pageserver::{repository, tenant};
-use pageserver_api::models::{
-    PagestreamBeMessage, PagestreamFeMessage, PagestreamGetPageRequest, PagestreamGetPageResponse,
-};
-use pageserver_api::reltag::RelTag;
-use rand::prelude::*;
-use scopeguard::defer;
-use std::env::args;
-use std::future::Future;
-use std::str::FromStr;
-use std::sync::atomic::{AtomicU64, Ordering};
-use std::sync::{Arc, Mutex};
-use std::thread;
-use tokio::sync::mpsc::{channel, Sender};
-use tokio::sync::Mutex as AsyncMutex;
-use tokio::task::JoinHandle;
-use tokio_stream::{Stream, StreamExt};
-use utils::completion;
-use utils::lsn::Lsn;
-
-struct Key(repository::Key);
-
-impl std::str::FromStr for Key {
-    type Err = anyhow::Error;
-
-    fn from_str(s: &str) -> std::result::Result<Self, Self::Err> {
-        repository::Key::from_hex(s).map(Key)
-    }
-}
-
-struct KeyRange {
-    start: i128,
-    end: i128,
-}
-
-impl KeyRange {
-    fn len(&self) -> i128 {
-        self.end - self.start
-    }
-}
-
-struct RelTagBlockNo {
-    rel_tag: RelTag,
-    block_no: u32,
-}
-
-#[derive(clap::Parser)]
-struct Args {
-    #[clap(long, default_value = "http://localhost:9898")]
-    mgmt_api_endpoint: String,
-    #[clap(long, default_value = "postgres://postgres@localhost:64000")]
-    page_service_connstring: String,
-    // tenant_id: String,
-    // timeline_id: String,
-    num_tasks: usize,
-    num_requests: usize,
-    tenants: Option<Vec<String>>,
-    #[clap(long)]
-    pick_n_tenants: Option<usize>,
-    #[clap(subcommand)]
-    mode: Mode,
-}
-
-#[derive(clap::Parser, Clone)]
-enum Mode {
-    GetPage,
-    NoOp,
-}
-
-#[derive(Debug, Default)]
-struct Stats {
-    completed_requests: AtomicU64,
-}
-
-impl Stats {
-    fn inc(&self) {
-        self.completed_requests.fetch_add(1, Ordering::Relaxed);
-    }
-}
-
-#[tokio::main]
-async fn main() {
-    let args: &'static Args = Box::leak(Box::new(Args::parse()));
-
-    // std::env::set_var("RUST_LOG", "info,tokio_postgres=trace");
-    // tracing_subscriber::fmt::init();
-
-    let client = Client::new();
-
-    let tenants = if let Some(tenants) = &args.tenants {
-        tenants.clone()
-    } else {
-        // let tenant_id = "b97965931096047b2d54958756baee7b";
-        // let timeline_id = "2868f84a8d166779e4c651b116c45059";
-
-        let resp = client
-            .get(Uri::try_from(&format!("{}/v1/tenant", args.mgmt_api_endpoint)).unwrap())
-            .await
-            .unwrap();
-
-        let body = hyper::body::to_bytes(resp).await.unwrap();
-        let tenants: serde_json::Value = serde_json::from_slice(&body).unwrap();
-        let mut out = Vec::new();
-        for t in tenants.as_array().unwrap() {
-            if let Some(limit) = args.pick_n_tenants {
-                if out.len() >= limit {
-                    break;
-                }
-            }
-            out.push(t.get("id").unwrap().as_str().unwrap().to_owned());
-        }
-        if let Some(limit) = args.pick_n_tenants {
-            assert_eq!(out.len(), limit);
-        }
-        out
-    };
-
-    let mut tenant_timelines = Vec::new();
-    for tenant_id in tenants {
-        let resp = client
-            .get(
-                Uri::try_from(&format!(
-                    "{}/v1/tenant/{}/timeline",
-                    args.mgmt_api_endpoint, tenant_id
-                ))
-                .unwrap(),
-            )
-            .await
-            .unwrap();
-
-        let body = hyper::body::to_bytes(resp).await.unwrap();
-        let timelines: serde_json::Value = serde_json::from_slice(&body).unwrap();
-        for t in timelines.as_array().unwrap() {
-            let timeline_id = t.get("timeline_id").unwrap().as_str().unwrap().to_owned();
-            tenant_timelines.push((tenant_id.clone(), timeline_id));
-        }
-    }
-    println!("tenant_timelines:\n{:?}", tenant_timelines);
-
-    let mut stats = Arc::new(Stats::default());
-
-    tokio::spawn({
-        let stats = Arc::clone(&stats);
-        async move {
-            loop {
-                let start = std::time::Instant::now();
-                tokio::time::sleep(std::time::Duration::from_secs(1)).await;
-                let completed_requests = stats.completed_requests.swap(0, Ordering::Relaxed);
-                let elapsed = start.elapsed();
-                println!(
-                    "RPS: {:.0}",
-                    completed_requests as f64 / elapsed.as_secs_f64()
-                );
-            }
-        }
-    });
-
-    let mut tasks = Vec::new();
-    for (tenant_id, timeline_id) in tenant_timelines {
-        let stats = Arc::clone(&stats);
-        let t = tokio::spawn(timeline(
-            args,
-            client.clone(),
-            tenant_id,
-            timeline_id,
-            stats,
-        ));
-        tasks.push(t);
-    }
-
-    for t in tasks {
-        t.await.unwrap();
-    }
-}
-
-fn timeline(
-    args: &'static Args,
-    http_client: Client<HttpConnector, hyper::Body>,
-    tenant_id: String,
-    timeline_id: String,
-    stats: Arc<Stats>,
-) -> impl Future<Output = ()> + Send + Sync {
-    async move {
-        let mut resp = http_client
-            .get(
-                Uri::try_from(&format!(
-                    "{}/v1/tenant/{}/timeline/{}/keyspace",
-                    args.mgmt_api_endpoint, tenant_id, timeline_id
-                ))
-                .unwrap(),
-            )
-            .await
-            .unwrap();
-        if !resp.status().is_success() {
-            panic!("Failed to get keyspace: {resp:?}");
-        }
-        let body = hyper::body::to_bytes(resp).await.unwrap();
-        let keyspace: serde_json::Value = serde_json::from_slice(&body).unwrap();
-        let lsn: Lsn = keyspace["at_lsn"].as_str().unwrap().parse().unwrap();
-
-        let ranges = keyspace["keys"]
-            .as_array()
-            .unwrap()
-            .iter()
-            .filter_map(|r| {
-                let r = r.as_array().unwrap();
-                assert_eq!(r.len(), 2);
-                let start = Key::from_str(r[0].as_str().unwrap()).unwrap();
-                let end = Key::from_str(r[1].as_str().unwrap()).unwrap();
-                // filter out non-relblock keys
-                match (is_rel_block_key(start.0), is_rel_block_key(end.0)) {
-                    (true, true) => Some(KeyRange {
-                        start: start.0.to_i128(),
-                        end: end.0.to_i128(),
-                    }),
-                    (true, false) | (false, true) => {
-                        unimplemented!("split up range")
-                    }
-                    (false, false) => None,
-                }
-            })
-            .collect::<Vec<_>>();
-
-        // weighted ranges
-        let weights = ranges.iter().map(|r| r.len()).collect::<Vec<_>>();
-
-        let ranges = Arc::new(ranges);
-        let weights = Arc::new(weights);
-
-        let mut tasks = Vec::<JoinHandle<()>>::new();
-
-        let start = std::time::Instant::now();
-
-        for i in 0..args.num_tasks {
-            let ranges = ranges.clone();
-            let weights = weights.clone();
-            let client = http_client.clone();
-            let tenant_id = tenant_id.clone();
-            let timeline_id = timeline_id.clone();
-            let task = tokio::spawn({
-                let stats = Arc::clone(&stats);
-                async move {
-                    let mut client = getpage_client::Client::new(
-                        args.page_service_connstring.clone(),
-                        tenant_id.clone(),
-                        timeline_id.clone(),
-                    )
-                    .await
-                    .unwrap();
-                    for i in 0..args.num_requests {
-                        match args.mode {
-                            Mode::GetPage => {
-                                let key = {
-                                    let mut rng = rand::thread_rng();
-                                    let r = ranges.choose_weighted(&mut rng, |r| r.len()).unwrap();
-                                    let key: i128 = rng.gen_range((r.start..r.end));
-                                    let key = repository::Key::from_i128(key);
-                                    // XXX filter these out when we iterate the keyspace
-                                    assert!(
-                                        is_rel_block_key(key),
-                                        "we filter non-relblock keys out above"
-                                    );
-                                    let (rel_tag, block_no) =
-                                        key_to_rel_block(key).expect("we just checked");
-                                    RelTagBlockNo { rel_tag, block_no }
-                                };
-                                client
-                                    .getpage(key, lsn)
-                                    .await
-                                    .with_context(|| {
-                                        format!(
-                                            "getpage for tenant {} timeline {}",
-                                            tenant_id, timeline_id
-                                        )
-                                    })
-                                    .unwrap();
-                            }
-                            Mode::NoOp => {
-                                client.noop().await.unwrap();
-                            }
-                        }
-                        stats.inc();
-                    }
-                    client.shutdown().await;
-                }
-            });
-            tasks.push(task);
-        }
-
-        for task in tasks {
-            task.await.unwrap();
-        }
-    }
-}
-
-mod getpage_client {
-    use std::pin::Pin;
-
-    use futures::SinkExt;
-    use pageserver_api::models::{
-        PagestreamBeMessage, PagestreamFeMessage, PagestreamGetPageRequest,
-        PagestreamGetPageResponse,
-    };
-    use tokio::task::JoinHandle;
-    use tokio_stream::StreamExt;
-    use tokio_util::sync::CancellationToken;
-    use utils::lsn::Lsn;
-
-    use crate::RelTagBlockNo;
-
-    pub(crate) struct Client {
-        copy_both: Pin<Box<tokio_postgres::CopyBothDuplex<bytes::Bytes>>>,
-        cancel_on_client_drop: Option<tokio_util::sync::DropGuard>,
-        conn_task: JoinHandle<()>,
-    }
-
-    impl Client {
-        pub fn new(
-            connstring: String,
-            tenant_id: String,
-            timeline_id: String,
-        ) -> impl std::future::Future<Output = anyhow::Result<Self>> + Send {
-            async move {
-                let (client, connection) =
-                    tokio_postgres::connect(&connstring, postgres::NoTls).await?;
-
-                let conn_task_cancel = CancellationToken::new();
-                let conn_task = tokio::spawn({
-                    let conn_task_cancel = conn_task_cancel.clone();
-                    async move {
-                        tokio::select! {
-                            _ = conn_task_cancel.cancelled() => {
-                                return;
-                            }
-                            res = connection => {
-                                res.unwrap();
-                            }
-                        }
-                    }
-                });
-
-                let copy_both: tokio_postgres::CopyBothDuplex<bytes::Bytes> = client
-                    .copy_both_simple(&format!("pagestream {tenant_id} {timeline_id}"))
-                    .await?;
-
-                Ok(Self {
-                    copy_both: Box::pin(copy_both),
-                    conn_task,
-                    cancel_on_client_drop: Some(conn_task_cancel.drop_guard()),
-                })
-            }
-        }
-
-        pub async fn shutdown(mut self) {
-            let _ = self.cancel_on_client_drop.take();
-            self.conn_task.await.unwrap();
-        }
-
-        pub async fn getpage(
-            &mut self,
-            key: RelTagBlockNo,
-            lsn: Lsn,
-        ) -> anyhow::Result<PagestreamGetPageResponse> {
-            let req = PagestreamGetPageRequest {
-                latest: false,
-                rel: key.rel_tag,
-                blkno: key.block_no,
-                lsn,
-            };
-            let req = PagestreamFeMessage::GetPage(req);
-            match self.do_request(req).await? {
-                PagestreamBeMessage::GetPage(p) => Ok(p),
-                x => anyhow::bail!("Unexpected response: {:?}", x),
-            }
-        }
-
-        pub async fn noop(&mut self) -> anyhow::Result<()> {
-            match self.do_request(PagestreamFeMessage::NoOp).await? {
-                PagestreamBeMessage::NoOp => Ok(()),
-                x => anyhow::bail!("Unexpected response: {:?}", x),
-            }
-        }
-
-        async fn do_request(
-            &mut self,
-            req: PagestreamFeMessage,
-        ) -> Result<PagestreamBeMessage, anyhow::Error> {
-            let req: bytes::Bytes = req.serialize();
-            // let mut req = tokio_util::io::ReaderStream::new(&req);
-            let mut req = tokio_stream::once(Ok(req));
-
-            self.copy_both.send_all(&mut req).await?;
-
-            let next: Option<Result<bytes::Bytes, _>> = self.copy_both.next().await;
-            let next = next.unwrap().unwrap();
-
-            match PagestreamBeMessage::deserialize(next)? {
-                PagestreamBeMessage::Error(e) => anyhow::bail!("Error: {:?}", e),
-                x => Ok(x),
-            }
-        }
-    }
-}

pageserver/src/bin/noop_server.rs

@@ -1,109 +0,0 @@
-use anyhow::Context;
-use bytes::Buf;
-use clap::Parser;
-use pageserver_api::models::{PagestreamBeMessage, PagestreamErrorResponse, PagestreamFeMessage};
-use postgres_backend::{AuthType, PostgresBackend, QueryError};
-use pq_proto::{BeMessage, FeMessage};
-use tokio::io::{AsyncRead, AsyncWrite};
-use tokio_util::sync::CancellationToken;
-
-#[derive(clap::Parser)]
-struct Args {
-    bind: String,
-}
-
-#[tokio::main]
-async fn main() {
-    let args = Args::parse();
-
-    let listener = tokio::net::TcpListener::bind(&args.bind).await.unwrap();
-    loop {
-        let (socket, _) = listener.accept().await.unwrap();
-        tokio::spawn(async move {
-            handle_connection(socket).await.unwrap();
-        });
-    }
-}
-
-async fn handle_connection(socket: tokio::net::TcpStream) -> anyhow::Result<()> {
-    socket
-        .set_nodelay(true)
-        .context("could not set TCP_NODELAY")?;
-
-    let peer_addr = socket.peer_addr().context("get peer address")?;
-    let socket = tokio_io_timeout::TimeoutReader::new(socket);
-    tokio::pin!(socket);
-    let pgbackend = PostgresBackend::new_from_io(socket, peer_addr, AuthType::Trust, None)?;
-    let mut conn_handler = NoOpHandler;
-    let cancel = CancellationToken::new();
-    pgbackend
-        .run(&mut conn_handler, || {
-            let cancel = cancel.clone();
-            async move { cancel.cancelled().await }
-        })
-        .await?;
-    anyhow::Ok(())
-}
-
-struct NoOpHandler;
-
-#[async_trait::async_trait]
-impl<IO> postgres_backend::Handler<IO> for NoOpHandler
-where
-    IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
-{
-    fn startup(
-        &mut self,
-        _pgb: &mut PostgresBackend<IO>,
-        _sm: &pq_proto::FeStartupPacket,
-    ) -> Result<(), QueryError> {
-        Ok(())
-    }
-
-    async fn process_query(
-        &mut self,
-        pgb: &mut PostgresBackend<IO>,
-        query_string: &str,
-    ) -> Result<(), QueryError> {
-        if !query_string.starts_with("pagestream ") {
-            return Err(QueryError::Other(anyhow::anyhow!("not a pagestream query")));
-        }
-
-        // switch client to COPYBOTH
-        pgb.write_message_noflush(&BeMessage::CopyBothResponse)?;
-        pgb.flush().await?;
-
-        loop {
-            let msg = pgb.read_message().await?;
-
-            let copy_data_bytes = match msg {
-                Some(FeMessage::CopyData(bytes)) => bytes,
-                Some(FeMessage::Terminate) => return Ok(()),
-                Some(m) => {
-                    return Err(QueryError::Other(anyhow::anyhow!(
-                        "unexpected message: {m:?} during COPY"
-                    )));
-                }
-                None => return Ok(()), // client disconnected
-            };
-
-            let neon_fe_msg = PagestreamFeMessage::parse(&mut copy_data_bytes.reader())?;
-
-            let response = match neon_fe_msg {
-                PagestreamFeMessage::NoOp => Ok(PagestreamBeMessage::NoOp),
-                x => Err(QueryError::Other(anyhow::anyhow!(
-                    "this server only supports no-op: {x:?}"
-                ))),
-            };
-
-            let response = response.unwrap_or_else(|e| {
-                PagestreamBeMessage::Error(PagestreamErrorResponse {
-                    message: e.to_string(),
-                })
-            });
-
-            pgb.write_message_noflush(&BeMessage::CopyData(&response.serialize()))?;
-            pgb.flush().await?;
-        }
-    }
-}

pageserver/src/bin/pageserver.rs

@@ -34,11 +34,8 @@ use postgres_backend::AuthType;
 use utils::logging::TracingErrorLayerEnablement;
 use utils::signals::ShutdownSignals;
 use utils::{
-    auth::{JwtAuth, SwappableJwtAuth},
-    logging, project_build_tag, project_git_version,
-    sentry_init::init_sentry,
-    signals::Signal,
-    tcp_listener,
+    auth::JwtAuth, logging, project_build_tag, project_git_version, sentry_init::init_sentry,
+    signals::Signal, tcp_listener,
 };
 
 project_git_version!(GIT_VERSION);
@@ -324,12 +321,13 @@ fn start_pageserver(
     let http_auth;
     let pg_auth;
     if conf.http_auth_type == AuthType::NeonJWT || conf.pg_auth_type == AuthType::NeonJWT {
-        // unwrap is ok because check is performed when creating config, so path is set and exists
+        // unwrap is ok because check is performed when creating config, so path is set and file exists
         let key_path = conf.auth_validation_public_key_path.as_ref().unwrap();
-        info!("Loading public key(s) for verifying JWT tokens from {key_path:?}");
-
-        let jwt_auth = JwtAuth::from_key_path(key_path)?;
-        let auth: Arc<SwappableJwtAuth> = Arc::new(SwappableJwtAuth::new(jwt_auth));
+        info!(
+            "Loading public key for verifying JWT tokens from {:#?}",
+            key_path
+        );
+        let auth: Arc<JwtAuth> = Arc::new(JwtAuth::from_key_path(key_path)?);
 
         http_auth = match &conf.http_auth_type {
             AuthType::Trust => None,
@@ -412,7 +410,7 @@ fn start_pageserver(
 
     // Scan the local 'tenants/' directory and start loading the tenants
     let deletion_queue_client = deletion_queue.new_client();
-    let tenant_manager = BACKGROUND_RUNTIME.block_on(mgr::init_tenant_mgr(
+    BACKGROUND_RUNTIME.block_on(mgr::init_tenant_mgr(
         conf,
         TenantSharedResources {
             broker_client: broker_client.clone(),
@@ -422,7 +420,6 @@ fn start_pageserver(
         order,
         shutdown_pageserver.clone(),
     ))?;
-    let tenant_manager = Arc::new(tenant_manager);
 
     BACKGROUND_RUNTIME.spawn({
         let init_done_rx = init_done_rx;
@@ -551,7 +548,6 @@ fn start_pageserver(
         let router_state = Arc::new(
             http::routes::State::new(
                 conf,
-                tenant_manager,
                 http_auth.clone(),
                 remote_storage.clone(),
                 broker_client.clone(),

pageserver/src/bin/tokio_tcp_bench.rs

@@ -1,130 +0,0 @@
-use std::env::args;
-
-use clap::Parser;
-use tokio::io::{AsyncReadExt, AsyncWriteExt};
-
-#[derive(clap::Parser)]
-struct Args {
-    #[clap(subcommand)]
-    mode: Mode,
-}
-
-#[derive(clap::Parser)]
-enum Mode {
-    Client(Client),
-    Server(Server),
-}
-
-#[derive(clap::Parser)]
-struct Client {
-    num_tasks: usize,
-}
-#[derive(clap::Parser)]
-struct Server {}
-
-#[tokio::main]
-async fn main() {
-    let args: &'static _ = Box::leak(Box::new(Args::parse()));
-
-    match &args.mode {
-        Mode::Client(x) => client::client(x).await,
-        Mode::Server(x) => server::server(x).await,
-    }
-}
-
-mod client {
-    use std::sync::{atomic::{Ordering, AtomicU64}, Arc};
-
-    use tokio::io::{AsyncReadExt, AsyncWriteExt};
-
-    use super::Client;
-
-    #[derive(Debug, Default)]
-    struct Stats {
-        completed_requests: AtomicU64,
-    }
-
-    impl Stats {
-        fn inc(&self) {
-            self.completed_requests.fetch_add(1, Ordering::Relaxed);
-        }
-    }
-    pub(crate) async fn client(args: &'static Client) {
-        let mut stats = Arc::new(Stats::default());
-
-        tokio::spawn({
-            let stats = Arc::clone(&stats);
-            async move {
-                loop {
-                    let start = std::time::Instant::now();
-                    tokio::time::sleep(std::time::Duration::from_secs(1)).await;
-                    let completed_requests = stats.completed_requests.swap(0, Ordering::Relaxed);
-                    let elapsed = start.elapsed();
-                    println!(
-                        "RPS: {:.0} RPS/client: {:.2}",
-                        completed_requests as f64 / elapsed.as_secs_f64(),
-                        completed_requests as f64 / elapsed.as_secs_f64() / args.num_tasks as f64,
-                    );
-                }
-            }
-        });
-
-        let mut tasks = Vec::new();
-        for _  in 0..args.num_tasks {
-            let stats = Arc::clone(&stats);
-            let t = tokio::spawn(client_task(args, stats));
-            tasks.push(t);
-        }
-
-        for t in tasks {
-            t.await.unwrap();
-        }
-    }
-
-    async fn client_task(args: &'static Client, stats: Arc<Stats>) -> anyhow::Result<()> {
-        let mut conn = tokio::net::TcpStream::connect("localhost:65000").await?;
-        conn.set_nodelay(true)?;
-
-        loop {
-            let mut buf = [0u8; 1];
-            conn.write_all(&buf).await?;
-            conn.read_exact(&mut buf).await?;
-            stats.inc();
-        }
-    }
-}
-
-mod server {
-
-    use anyhow::Context;
-    use tokio::io::{AsyncReadExt, AsyncWriteExt};
-
-    use super::Server;
-
-    pub(crate) async fn server(args: &'static Server) {
-        let listener = tokio::net::TcpListener::bind("localhost:65000").await.unwrap();
-        loop {
-            let (socket, _) = listener.accept().await.unwrap();
-            tokio::spawn(async move {
-                server_handle_connection(args, socket).await.unwrap();
-            });
-        }
-    }
-
-    async fn server_handle_connection(
-        args: &'static Server,
-        socket: tokio::net::TcpStream,
-    ) -> anyhow::Result<()> {
-        socket
-            .set_nodelay(true)
-            .context("could not set TCP_NODELAY")?;
-        // let socket = tokio_io_timeout::TimeoutReader::new(socket);
-        tokio::pin!(socket);
-
-        loop {
-            let mut buf = [0u8; 4096];
-            socket.read_exact(&mut buf).await?;
-            socket.write_all(&buf).await?;
-        }
-    }
-}

pageserver/src/config.rs

@@ -161,7 +161,7 @@ pub struct PageServerConf {
     pub http_auth_type: AuthType,
     /// authentication method for libpq connections from compute
     pub pg_auth_type: AuthType,
-    /// Path to a file or directory containing public key(s) for verifying JWT tokens.
+    /// Path to a file containing public key for verifying JWT tokens.
     /// Used for both mgmt and compute auth, if enabled.
     pub auth_validation_public_key_path: Option<Utf8PathBuf>,
 
@@ -880,6 +880,13 @@ impl PageServerConf {
             );
         }
 
+        if let Some(compaction_algorithm) = item.get("compaction_algorithm") {
+            t_conf.compaction_algorithm = Some(
+                deserialize_from_item("compaction_algorithm", compaction_algorithm)
+                    .context("parse compaction_algorithm")?,
+            );
+        }
+
         if let Some(gc_horizon) = item.get("gc_horizon") {
             t_conf.gc_horizon = Some(parse_toml_u64("gc_horizon", gc_horizon)?);
         }

pageserver/src/consumption_metrics.rs

@@ -16,7 +16,7 @@ use tracing::*;
 use utils::id::NodeId;
 
 mod metrics;
-use metrics::MetricsKey;
+use crate::consumption_metrics::metrics::MetricsKey;
 mod disk_cache;
 mod upload;
 
@@ -266,7 +266,7 @@ async fn calculate_synthetic_size_worker(
                 continue;
             }
 
-            if let Ok(tenant) = mgr::get_tenant(tenant_id, true) {
+            if let Ok(tenant) = mgr::get_tenant(tenant_id, true).await {
                 // TODO should we use concurrent_background_tasks_rate_limit() here, like the other background tasks?
                 // We can put in some prioritization for consumption metrics.
                 // Same for the loop that fetches computed metrics.

pageserver/src/consumption_metrics/metrics.rs

@@ -3,6 +3,7 @@ use anyhow::Context;
 use chrono::{DateTime, Utc};
 use consumption_metrics::EventType;
 use futures::stream::StreamExt;
+use serde_with::serde_as;
 use std::{sync::Arc, time::SystemTime};
 use utils::{
     id::{TenantId, TimelineId},
@@ -41,10 +42,13 @@ pub(super) enum Name {
 ///
 /// This is a denormalization done at the MetricsKey const methods; these should not be constructed
 /// elsewhere.
+#[serde_with::serde_as]
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, serde::Serialize, serde::Deserialize)]
 pub(crate) struct MetricsKey {
+    #[serde_as(as = "serde_with::DisplayFromStr")]
     pub(super) tenant_id: TenantId,
 
+    #[serde_as(as = "Option<serde_with::DisplayFromStr>")]
     #[serde(skip_serializing_if = "Option::is_none")]
     pub(super) timeline_id: Option<TimelineId>,
 
@@ -202,6 +206,7 @@ pub(super) async fn collect_all_metrics(
             None
         } else {
             crate::tenant::mgr::get_tenant(id, true)
+                .await
                 .ok()
                 .map(|tenant| (id, tenant))
         }

pageserver/src/consumption_metrics/upload.rs

@@ -1,4 +1,5 @@
 use consumption_metrics::{Event, EventChunk, IdempotencyKey, CHUNK_SIZE};
+use serde_with::serde_as;
 use tokio_util::sync::CancellationToken;
 use tracing::Instrument;
 
@@ -6,9 +7,12 @@ use super::{metrics::Name, Cache, MetricsKey, RawMetric};
 use utils::id::{TenantId, TimelineId};
 
 /// How the metrics from pageserver are identified.
+#[serde_with::serde_as]
 #[derive(serde::Serialize, serde::Deserialize, Debug, Clone, Copy, PartialEq)]
 struct Ids {
+    #[serde_as(as = "serde_with::DisplayFromStr")]
     pub(super) tenant_id: TenantId,
+    #[serde_as(as = "Option<serde_with::DisplayFromStr>")]
     #[serde(skip_serializing_if = "Option::is_none")]
     pub(super) timeline_id: Option<TimelineId>,
 }

pageserver/src/deletion_queue.rs

@@ -18,6 +18,7 @@ use hex::FromHex;
 use remote_storage::{GenericRemoteStorage, RemotePath};
 use serde::Deserialize;
 use serde::Serialize;
+use serde_with::serde_as;
 use thiserror::Error;
 use tokio;
 use tokio_util::sync::CancellationToken;
@@ -214,6 +215,7 @@ where
 /// during recovery as startup.
 const TEMP_SUFFIX: &str = "tmp";
 
+#[serde_as]
 #[derive(Debug, Serialize, Deserialize)]
 struct DeletionList {
     /// Serialization version, for future use
@@ -242,6 +244,7 @@ struct DeletionList {
     validated: bool,
 }
 
+#[serde_as]
 #[derive(Debug, Serialize, Deserialize)]
 struct DeletionHeader {
     /// Serialization version, for future use
@@ -345,7 +348,7 @@ impl DeletionList {
                 result.extend(
                     timeline_layers
                         .into_iter()
-                        .map(|l| timeline_remote_path.join(Utf8PathBuf::from(l))),
+                        .map(|l| timeline_remote_path.join(&Utf8PathBuf::from(l))),
                 );
             }
         }

pageserver/src/deletion_queue/deleter.rs

@@ -55,24 +55,21 @@ impl Deleter {
 
     /// Wrap the remote `delete_objects` with a failpoint
     async fn remote_delete(&self) -> Result<(), anyhow::Error> {
+        fail::fail_point!("deletion-queue-before-execute", |_| {
+            info!("Skipping execution, failpoint set");
+            metrics::DELETION_QUEUE
+                .remote_errors
+                .with_label_values(&["failpoint"])
+                .inc();
+            Err(anyhow::anyhow!("failpoint hit"))
+        });
+
         // A backoff::retry is used here for two reasons:
         // - To provide a backoff rather than busy-polling the API on errors
         // - To absorb transient 429/503 conditions without hitting our error
         //   logging path for issues deleting objects.
         backoff::retry(
-            || async {
-                fail::fail_point!("deletion-queue-before-execute", |_| {
-                    info!("Skipping execution, failpoint set");
-
-                    metrics::DELETION_QUEUE
-                        .remote_errors
-                        .with_label_values(&["failpoint"])
-                        .inc();
-                    Err(anyhow::anyhow!("failpoint: deletion-queue-before-execute"))
-                });
-
-                self.remote_storage.delete_objects(&self.accumulator).await
-            },
+            || async { self.remote_storage.delete_objects(&self.accumulator).await },
             |_| false,
             3,
             10,

pageserver/src/disk_usage_eviction_task.rs

@@ -403,7 +403,7 @@ pub async fn disk_usage_eviction_task_iteration_impl<U: Usage>(
                     return (evicted_bytes, evictions_failed);
                 };
 
-                let results = timeline.evict_layers(&batch).await;
+                let results = timeline.evict_layers(&batch, &cancel).await;
 
                 match results {
                     Ok(results) => {
@@ -545,7 +545,7 @@ async fn collect_eviction_candidates(
         if cancel.is_cancelled() {
             return Ok(EvictionCandidates::Cancelled);
         }
-        let tenant = match tenant::mgr::get_tenant(*tenant_id, true) {
+        let tenant = match tenant::mgr::get_tenant(*tenant_id, true).await {
             Ok(tenant) => tenant,
             Err(e) => {
                 // this can happen if tenant has lifecycle transition after we fetched it
@@ -554,11 +554,6 @@ async fn collect_eviction_candidates(
             }
         };
 
-        if tenant.cancel.is_cancelled() {
-            info!(%tenant_id, "Skipping tenant for eviction, it is shutting down");
-            continue;
-        }
-
         // collect layers from all timelines in this tenant
         //
         // If one of the timelines becomes `!is_active()` during the iteration,

pageserver/src/http/openapi_spec.yml

@@ -52,31 +52,6 @@ paths:
               schema:
                 type: object
 
-  /v1/reload_auth_validation_keys:
-    post:
-      description: Reloads the JWT public keys from their pre-configured location on disk.
-      responses:
-        "200":
-          description: The reload completed successfully.
-        "401":
-          description: Unauthorized Error
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/UnauthorizedError"
-        "403":
-          description: Forbidden Error
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ForbiddenError"
-        "500":
-          description: Generic operation error (also hits if no keys were found)
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/Error"
-
   /v1/tenant/{tenant_id}:
     parameters:
       - name: tenant_id

pageserver/src/http/routes.rs

@@ -17,10 +17,10 @@ use pageserver_api::models::{
     TenantLoadRequest, TenantLocationConfigRequest,
 };
 use remote_storage::GenericRemoteStorage;
+use serde_with::{serde_as, DisplayFromStr};
 use tenant_size_model::{SizeResult, StorageModel};
 use tokio_util::sync::CancellationToken;
 use tracing::*;
-use utils::auth::JwtAuth;
 use utils::http::endpoint::request_span;
 use utils::http::json::json_request_or_empty_body;
 use utils::http::request::{get_request_param, must_get_query_param, parse_query_param};
@@ -36,8 +36,7 @@ use crate::pgdatadir_mapping::LsnForTimestamp;
 use crate::task_mgr::TaskKind;
 use crate::tenant::config::{LocationConf, TenantConfOpt};
 use crate::tenant::mgr::{
-    GetTenantError, SetNewTenantConfigError, TenantManager, TenantMapError, TenantMapInsertError,
-    TenantSlotError, TenantSlotUpsertError, TenantStateError,
+    GetTenantError, SetNewTenantConfigError, TenantMapInsertError, TenantStateError,
 };
 use crate::tenant::size::ModelInputs;
 use crate::tenant::storage_layer::LayerAccessStatsReset;
@@ -46,7 +45,7 @@ use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError, TenantSha
 use crate::{config::PageServerConf, tenant::mgr};
 use crate::{disk_usage_eviction_task, tenant};
 use utils::{
-    auth::SwappableJwtAuth,
+    auth::JwtAuth,
     generation::Generation,
     http::{
         endpoint::{self, attach_openapi_ui, auth_middleware, check_permission_with},
@@ -64,8 +63,7 @@ use super::models::ConfigureFailpointsRequest;
 
 pub struct State {
     conf: &'static PageServerConf,
-    tenant_manager: Arc<TenantManager>,
-    auth: Option<Arc<SwappableJwtAuth>>,
+    auth: Option<Arc<JwtAuth>>,
     allowlist_routes: Vec<Uri>,
     remote_storage: Option<GenericRemoteStorage>,
     broker_client: storage_broker::BrokerClientChannel,
@@ -76,8 +74,7 @@ pub struct State {
 impl State {
     pub fn new(
         conf: &'static PageServerConf,
-        tenant_manager: Arc<TenantManager>,
-        auth: Option<Arc<SwappableJwtAuth>>,
+        auth: Option<Arc<JwtAuth>>,
         remote_storage: Option<GenericRemoteStorage>,
         broker_client: storage_broker::BrokerClientChannel,
         disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
@@ -89,7 +86,6 @@ impl State {
             .collect::<Vec<_>>();
         Ok(Self {
             conf,
-            tenant_manager,
             auth,
             allowlist_routes,
             remote_storage,
@@ -151,59 +147,28 @@ impl From<PageReconstructError> for ApiError {
 impl From<TenantMapInsertError> for ApiError {
     fn from(tmie: TenantMapInsertError) -> ApiError {
         match tmie {
-            TenantMapInsertError::SlotError(e) => e.into(),
-            TenantMapInsertError::SlotUpsertError(e) => e.into(),
+            TenantMapInsertError::StillInitializing | TenantMapInsertError::ShuttingDown => {
+                ApiError::ResourceUnavailable(format!("{tmie}").into())
+            }
+            TenantMapInsertError::TenantAlreadyExists(id, state) => {
+                ApiError::Conflict(format!("tenant {id} already exists, state: {state:?}"))
+            }
+            TenantMapInsertError::TenantExistsSecondary(id) => {
+                ApiError::Conflict(format!("tenant {id} already exists as secondary"))
+            }
             TenantMapInsertError::Other(e) => ApiError::InternalServerError(e),
         }
     }
 }
 
-impl From<TenantSlotError> for ApiError {
-    fn from(e: TenantSlotError) -> ApiError {
-        use TenantSlotError::*;
-        match e {
-            NotFound(tenant_id) => {
-                ApiError::NotFound(anyhow::anyhow!("NotFound: tenant {tenant_id}").into())
-            }
-            e @ (AlreadyExists(_, _) | Conflict(_)) => ApiError::Conflict(format!("{e}")),
-            InProgress => {
-                ApiError::ResourceUnavailable("Tenant is being modified concurrently".into())
-            }
-            MapState(e) => e.into(),
-        }
-    }
-}
-
-impl From<TenantSlotUpsertError> for ApiError {
-    fn from(e: TenantSlotUpsertError) -> ApiError {
-        use TenantSlotUpsertError::*;
-        match e {
-            InternalError(e) => ApiError::InternalServerError(anyhow::anyhow!("{e}")),
-            MapState(e) => e.into(),
-        }
-    }
-}
-
-impl From<TenantMapError> for ApiError {
-    fn from(e: TenantMapError) -> ApiError {
-        use TenantMapError::*;
-        match e {
-            StillInitializing | ShuttingDown => {
-                ApiError::ResourceUnavailable(format!("{e}").into())
-            }
-        }
-    }
-}
-
 impl From<TenantStateError> for ApiError {
     fn from(tse: TenantStateError) -> ApiError {
         match tse {
+            TenantStateError::NotFound(tid) => ApiError::NotFound(anyhow!("tenant {}", tid).into()),
             TenantStateError::IsStopping(_) => {
                 ApiError::ResourceUnavailable("Tenant is stopping".into())
             }
-            TenantStateError::SlotError(e) => e.into(),
-            TenantStateError::SlotUpsertError(e) => e.into(),
-            TenantStateError::Other(e) => ApiError::InternalServerError(anyhow!(e)),
+            _ => ApiError::InternalServerError(anyhow::Error::new(tse)),
         }
     }
 }
@@ -224,7 +189,6 @@ impl From<GetTenantError> for ApiError {
                 // (We can produce this variant only in `mgr::get_tenant(..., active=true)` calls).
                 ApiError::ResourceUnavailable("Tenant not yet active".into())
             }
-            GetTenantError::MapState(e) => ApiError::ResourceUnavailable(format!("{e}").into()),
         }
     }
 }
@@ -279,9 +243,6 @@ impl From<crate::tenant::delete::DeleteTenantError> for ApiError {
             Get(g) => ApiError::from(g),
             e @ AlreadyInProgress => ApiError::Conflict(e.to_string()),
             Timeline(t) => ApiError::from(t),
-            NotAttached => ApiError::NotFound(anyhow::anyhow!("Tenant is not attached").into()),
-            SlotError(e) => e.into(),
-            SlotUpsertError(e) => e.into(),
             Other(o) => ApiError::InternalServerError(o),
             e @ InvalidState(_) => ApiError::PreconditionFailed(e.to_string().into_boxed_str()),
         }
@@ -393,32 +354,6 @@ async fn status_handler(
     json_response(StatusCode::OK, StatusResponse { id: config.id })
 }
 
-async fn reload_auth_validation_keys_handler(
-    request: Request<Body>,
-    _cancel: CancellationToken,
-) -> Result<Response<Body>, ApiError> {
-    check_permission(&request, None)?;
-    let config = get_config(&request);
-    let state = get_state(&request);
-    let Some(shared_auth) = &state.auth else {
-        return json_response(StatusCode::BAD_REQUEST, ());
-    };
-    // unwrap is ok because check is performed when creating config, so path is set and exists
-    let key_path = config.auth_validation_public_key_path.as_ref().unwrap();
-    info!("Reloading public key(s) for verifying JWT tokens from {key_path:?}");
-
-    match JwtAuth::from_key_path(key_path) {
-        Ok(new_auth) => {
-            shared_auth.swap(new_auth);
-            json_response(StatusCode::OK, ())
-        }
-        Err(e) => {
-            warn!("Error reloading public keys from {key_path:?}: {e:}");
-            json_response(StatusCode::INTERNAL_SERVER_ERROR, ())
-        }
-    }
-}
-
 async fn timeline_create_handler(
     mut request: Request<Body>,
     _cancel: CancellationToken,
@@ -434,7 +369,7 @@ async fn timeline_create_handler(
     let state = get_state(&request);
 
     async {
-        let tenant = mgr::get_tenant(tenant_id, true)?;
+        let tenant = mgr::get_tenant(tenant_id, true).await?;
         match tenant.create_timeline(
             new_timeline_id,
             request_data.ancestor_timeline_id.map(TimelineId::from),
@@ -462,9 +397,6 @@ async fn timeline_create_handler(
             Err(e @ tenant::CreateTimelineError::AncestorNotActive) => {
                 json_response(StatusCode::SERVICE_UNAVAILABLE, HttpErrorBody::from_msg(e.to_string()))
             }
-            Err(tenant::CreateTimelineError::ShuttingDown) => {
-                json_response(StatusCode::SERVICE_UNAVAILABLE,HttpErrorBody::from_msg("tenant shutting down".to_string()))
-            }
             Err(tenant::CreateTimelineError::Other(err)) => Err(ApiError::InternalServerError(err)),
         }
     }
@@ -484,7 +416,7 @@ async fn timeline_list_handler(
     let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
 
     let response_data = async {
-        let tenant = mgr::get_tenant(tenant_id, true)?;
+        let tenant = mgr::get_tenant(tenant_id, true).await?;
         let timelines = tenant.list_timelines();
 
         let mut response_data = Vec::with_capacity(timelines.len());
@@ -523,7 +455,7 @@ async fn timeline_detail_handler(
     let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
 
     let timeline_info = async {
-        let tenant = mgr::get_tenant(tenant_id, true)?;
+        let tenant = mgr::get_tenant(tenant_id, true).await?;
 
         let timeline = tenant
             .get_timeline(timeline_id, false)
@@ -567,8 +499,10 @@ async fn get_lsn_by_timestamp_handler(
     let result = timeline.find_lsn_for_timestamp(timestamp_pg, &ctx).await?;
 
     if version.unwrap_or(0) > 1 {
+        #[serde_as]
         #[derive(serde::Serialize)]
         struct Result {
+            #[serde_as(as = "DisplayFromStr")]
             lsn: Lsn,
             kind: &'static str,
         }
@@ -747,46 +681,6 @@ async fn tenant_ignore_handler(
     json_response(StatusCode::OK, ())
 }
 
-async fn tenant_duplicate_handler(
-    mut request: Request<Body>,
-    cancel: CancellationToken,
-) -> Result<Response<Body>, ApiError> {
-    let src_tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
-
-    let request_data: TenantCreateRequest = json_request(&mut request).await?;
-    let new_tenant_id = request_data.new_tenant_id;
-    check_permission(&request, None)?;
-
-    let _timer = STORAGE_TIME_GLOBAL
-        .get_metric_with_label_values(&[StorageTimeOperation::DuplicateTenant.into()])
-        .expect("bug")
-        .start_timer();
-
-    let tenant_conf =
-        TenantConfOpt::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
-
-    let state = get_state(&request);
-
-    let generation = get_request_generation(state, request_data.generation)?;
-
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
-
-    mgr::duplicate_tenant(
-        state.conf,
-        tenant_conf,
-        src_tenant_id,
-        new_tenant_id,
-        generation,
-        state.tenant_resources(),
-        &ctx,
-        &cancel,
-    )
-    .instrument(info_span!("tenant_duplicate", %src_tenant_id, tenant_id = %new_tenant_id))
-    .await?;
-
-    json_response(StatusCode::CREATED, TenantCreateResponse(new_tenant_id))
-}
-
 async fn tenant_list_handler(
     request: Request<Body>,
     _cancel: CancellationToken,
@@ -819,7 +713,7 @@ async fn tenant_status(
     check_permission(&request, Some(tenant_id))?;
 
     let tenant_info = async {
-        let tenant = mgr::get_tenant(tenant_id, false)?;
+        let tenant = mgr::get_tenant(tenant_id, false).await?;
 
         // Calculate total physical size of all timelines
         let mut current_physical_size = 0;
@@ -882,7 +776,7 @@ async fn tenant_size_handler(
     let headers = request.headers();
 
     let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
-    let tenant = mgr::get_tenant(tenant_id, true)?;
+    let tenant = mgr::get_tenant(tenant_id, true).await?;
 
     // this can be long operation
     let inputs = tenant
@@ -917,8 +811,10 @@ async fn tenant_size_handler(
     }
 
     /// The type resides in the pageserver not to expose `ModelInputs`.
+    #[serde_with::serde_as]
     #[derive(serde::Serialize)]
     struct TenantHistorySize {
+        #[serde_as(as = "serde_with::DisplayFromStr")]
         id: TenantId,
         /// Size is a mixture of WAL and logical size, so the unit is bytes.
         ///
@@ -1139,7 +1035,7 @@ async fn get_tenant_config_handler(
     let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
     check_permission(&request, Some(tenant_id))?;
 
-    let tenant = mgr::get_tenant(tenant_id, false)?;
+    let tenant = mgr::get_tenant(tenant_id, false).await?;
 
     let response = HashMap::from([
         (
@@ -1198,7 +1094,7 @@ async fn put_tenant_location_config_handler(
             .await
         {
             match e {
-                TenantStateError::SlotError(TenantSlotError::NotFound(_)) => {
+                TenantStateError::NotFound(_) => {
                     // This API is idempotent: a NotFound on a detach is fine.
                 }
                 _ => return Err(e.into()),
@@ -1210,14 +1106,20 @@ async fn put_tenant_location_config_handler(
     let location_conf =
         LocationConf::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
 
-    state
-        .tenant_manager
-        .upsert_location(tenant_id, location_conf, &ctx)
-        .await
-        // TODO: badrequest assumes the caller was asking for something unreasonable, but in
-        // principle we might have hit something like concurrent API calls to the same tenant,
-        // which is not a 400 but a 409.
-        .map_err(ApiError::BadRequest)?;
+    mgr::upsert_location(
+        state.conf,
+        tenant_id,
+        location_conf,
+        state.broker_client.clone(),
+        state.remote_storage.clone(),
+        state.deletion_queue_client.clone(),
+        &ctx,
+    )
+    .await
+    // TODO: badrequest assumes the caller was asking for something unreasonable, but in
+    // principle we might have hit something like concurrent API calls to the same tenant,
+    // which is not a 400 but a 409.
+    .map_err(ApiError::BadRequest)?;
 
     json_response(StatusCode::OK, ())
 }
@@ -1230,6 +1132,7 @@ async fn handle_tenant_break(
     let tenant_id: TenantId = parse_request_param(&r, "tenant_id")?;
 
     let tenant = crate::tenant::mgr::get_tenant(tenant_id, true)
+        .await
         .map_err(|_| ApiError::Conflict(String::from("no active tenant found")))?;
 
     tenant.set_broken("broken from test".to_owned()).await;
@@ -1534,7 +1437,7 @@ async fn active_timeline_of_active_tenant(
     tenant_id: TenantId,
     timeline_id: TimelineId,
 ) -> Result<Arc<Timeline>, ApiError> {
-    let tenant = mgr::get_tenant(tenant_id, true)?;
+    let tenant = mgr::get_tenant(tenant_id, true).await?;
     tenant
         .get_timeline(timeline_id, true)
         .map_err(|e| ApiError::NotFound(e.into()))
@@ -1759,7 +1662,7 @@ where
 pub fn make_router(
     state: Arc<State>,
     launch_ts: &'static LaunchTimestamp,
-    auth: Option<Arc<SwappableJwtAuth>>,
+    auth: Option<Arc<JwtAuth>>,
 ) -> anyhow::Result<RouterBuilder<hyper::Body, ApiError>> {
     let spec = include_bytes!("openapi_spec.yml");
     let mut router = attach_openapi_ui(endpoint::make_router(), spec, "/swagger.yml", "/v1/doc");
@@ -1788,9 +1691,6 @@ pub fn make_router(
         .put("/v1/failpoints", |r| {
             testing_api_handler("manage failpoints", r, failpoints_handler)
         })
-        .post("/v1/reload_auth_validation_keys", |r| {
-            api_handler(r, reload_auth_validation_keys_handler)
-        })
         .get("/v1/tenant", |r| api_handler(r, tenant_list_handler))
         .post("/v1/tenant", |r| api_handler(r, tenant_create_handler))
         .get("/v1/tenant/:tenant_id", |r| api_handler(r, tenant_status))
@@ -1827,9 +1727,6 @@ pub fn make_router(
         .post("/v1/tenant/:tenant_id/ignore", |r| {
             api_handler(r, tenant_ignore_handler)
         })
-        .post("/v1/tenant/:tenant_id/duplicate", |r| {
-            api_handler(r, tenant_duplicate_handler)
-        })
         .get("/v1/tenant/:tenant_id/timeline/:timeline_id", |r| {
             api_handler(r, timeline_detail_handler)
         })

pageserver/src/lib.rs

@@ -1,5 +1,3 @@
-#![deny(clippy::undocumented_unsafe_blocks)]
-
 mod auth;
 pub mod basebackup;
 pub mod config;
@@ -63,6 +61,14 @@ pub async fn shutdown_pageserver(deletion_queue: Option<DeletionQueue>, exit_cod
     )
     .await;
 
+    // Shut down any page service tasks.
+    timed(
+        task_mgr::shutdown_tasks(Some(TaskKind::PageRequestHandler), None, None),
+        "shutdown PageRequestHandlers",
+        Duration::from_secs(1),
+    )
+    .await;
+
     // Shut down all the tenants. This flushes everything to disk and kills
     // the checkpoint and GC tasks.
     timed(
@@ -72,15 +78,6 @@ pub async fn shutdown_pageserver(deletion_queue: Option<DeletionQueue>, exit_cod
     )
     .await;
 
-    // Shut down any page service tasks: any in-progress work for particular timelines or tenants
-    // should already have been canclled via mgr::shutdown_all_tenants
-    timed(
-        task_mgr::shutdown_tasks(Some(TaskKind::PageRequestHandler), None, None),
-        "shutdown PageRequestHandlers",
-        Duration::from_secs(1),
-    )
-    .await;
-
     // Best effort to persist any outstanding deletions, to avoid leaking objects
     if let Some(mut deletion_queue) = deletion_queue {
         deletion_queue.shutdown(Duration::from_secs(5)).await;

pageserver/src/metrics.rs

@@ -51,9 +51,6 @@ pub enum StorageTimeOperation {
 
     #[strum(serialize = "create tenant")]
     CreateTenant,
-
-    #[strum(serialize = "duplicate tenant")]
-    DuplicateTenant,
 }
 
 pub static STORAGE_TIME_SUM_PER_TIMELINE: Lazy<CounterVec> = Lazy::new(|| {
@@ -760,7 +757,6 @@ pub enum SmgrQueryType {
     GetRelSize,
     GetPageAtLsn,
     GetDbSize,
-    NoOp,
 }
 
 #[derive(Debug)]
@@ -966,32 +962,6 @@ static REMOTE_TIMELINE_CLIENT_BYTES_FINISHED_COUNTER: Lazy<IntCounterVec> = Lazy
     .expect("failed to define a metric")
 });
 
-pub(crate) struct TenantManagerMetrics {
-    pub(crate) tenant_slots: UIntGauge,
-    pub(crate) tenant_slot_writes: IntCounter,
-    pub(crate) unexpected_errors: IntCounter,
-}
-
-pub(crate) static TENANT_MANAGER: Lazy<TenantManagerMetrics> = Lazy::new(|| {
-    TenantManagerMetrics {
-    tenant_slots: register_uint_gauge!(
-        "pageserver_tenant_manager_slots",
-        "How many slots currently exist, including all attached, secondary and in-progress operations",
-    )
-    .expect("failed to define a metric"),
-    tenant_slot_writes: register_int_counter!(
-        "pageserver_tenant_manager_slot_writes",
-        "Writes to a tenant slot, including all of create/attach/detach/delete"
-    )
-    .expect("failed to define a metric"),
-    unexpected_errors: register_int_counter!(
-        "pageserver_tenant_manager_unexpected_errors_total",
-        "Number of unexpected conditions encountered: nonzero value indicates a non-fatal bug."
-    )
-    .expect("failed to define a metric"),
-}
-});
-
 pub(crate) struct DeletionQueueMetrics {
     pub(crate) keys_submitted: IntCounter,
     pub(crate) keys_dropped: IntCounter,
@@ -1914,9 +1884,6 @@ pub fn preinitialize_metrics() {
     // Deletion queue stats
     Lazy::force(&DELETION_QUEUE);
 
-    // Tenant manager stats
-    Lazy::force(&TENANT_MANAGER);
-
     // countervecs
     [&BACKGROUND_LOOP_PERIOD_OVERRUN_COUNT]
         .into_iter()

pageserver/src/page_service.rs

@@ -40,7 +40,7 @@ use tracing::field;
 use tracing::*;
 use utils::id::ConnectionId;
 use utils::{
-    auth::{Claims, Scope, SwappableJwtAuth},
+    auth::{Claims, JwtAuth, Scope},
     id::{TenantId, TimelineId},
     lsn::Lsn,
     simple_rcu::RcuReadGuard,
@@ -55,20 +55,16 @@ use crate::metrics;
 use crate::metrics::LIVE_CONNECTIONS_COUNT;
 use crate::task_mgr;
 use crate::task_mgr::TaskKind;
+use crate::tenant;
 use crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id;
 use crate::tenant::mgr;
-use crate::tenant::mgr::get_active_tenant_with_timeout;
-use crate::tenant::mgr::GetActiveTenantError;
-use crate::tenant::Timeline;
+use crate::tenant::mgr::GetTenantError;
+use crate::tenant::{Tenant, Timeline};
 use crate::trace::Tracer;
 
 use postgres_ffi::pg_constants::DEFAULTTABLESPACE_OID;
 use postgres_ffi::BLCKSZ;
 
-// How long we may block waiting for a [`TenantSlot::InProgress`]` and/or a [`Tenant`] which
-// is not yet in state [`TenantState::Active`].
-const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(5000);
-
 /// Read the end of a tar archive.
 ///
 /// A tar archive normally ends with two consecutive blocks of zeros, 512 bytes each.
@@ -122,7 +118,7 @@ async fn read_tar_eof(mut reader: (impl AsyncRead + Unpin)) -> anyhow::Result<()
 pub async fn libpq_listener_main(
     conf: &'static PageServerConf,
     broker_client: storage_broker::BrokerClientChannel,
-    auth: Option<Arc<SwappableJwtAuth>>,
+    auth: Option<Arc<JwtAuth>>,
     listener: TcpListener,
     auth_type: AuthType,
     listener_ctx: RequestContext,
@@ -190,7 +186,7 @@ pub async fn libpq_listener_main(
 async fn page_service_conn_main(
     conf: &'static PageServerConf,
     broker_client: storage_broker::BrokerClientChannel,
-    auth: Option<Arc<SwappableJwtAuth>>,
+    auth: Option<Arc<JwtAuth>>,
     socket: tokio::net::TcpStream,
     auth_type: AuthType,
     connection_ctx: RequestContext,
@@ -227,7 +223,13 @@ async fn page_service_conn_main(
     // and create a child per-query context when it invokes process_query.
     // But it's in a shared crate, so, we store connection_ctx inside PageServerHandler
     // and create the per-query context in process_query ourselves.
-    let mut conn_handler = PageServerHandler::new(conf, broker_client, auth, connection_ctx);
+    let mut conn_handler = PageServerHandler::new(
+        conf,
+        broker_client,
+        auth,
+        connection_ctx,
+        task_mgr::shutdown_token(),
+    );
     let pgbackend = PostgresBackend::new_from_io(socket, peer_addr, auth_type, None)?;
 
     match pgbackend
@@ -253,7 +255,7 @@ async fn page_service_conn_main(
 struct PageServerHandler {
     _conf: &'static PageServerConf,
     broker_client: storage_broker::BrokerClientChannel,
-    auth: Option<Arc<SwappableJwtAuth>>,
+    auth: Option<Arc<JwtAuth>>,
     claims: Option<Claims>,
 
     /// The context created for the lifetime of the connection
@@ -261,14 +263,19 @@ struct PageServerHandler {
     /// For each query received over the connection,
     /// `process_query` creates a child context from this one.
     connection_ctx: RequestContext,
+
+    /// A token that should fire when the tenant transitions from
+    /// attached state, or when the pageserver is shutting down.
+    cancel: CancellationToken,
 }
 
 impl PageServerHandler {
     pub fn new(
         conf: &'static PageServerConf,
         broker_client: storage_broker::BrokerClientChannel,
-        auth: Option<Arc<SwappableJwtAuth>>,
+        auth: Option<Arc<JwtAuth>>,
         connection_ctx: RequestContext,
+        cancel: CancellationToken,
     ) -> Self {
         PageServerHandler {
             _conf: conf,
@@ -276,6 +283,7 @@ impl PageServerHandler {
             auth,
             claims: None,
             connection_ctx,
+            cancel,
         }
     }
 
@@ -283,11 +291,7 @@ impl PageServerHandler {
     /// this rather than naked flush() in order to shut down promptly.  Without this, we would
     /// block shutdown of a tenant if a postgres client was failing to consume bytes we send
     /// in the flush.
-    async fn flush_cancellable<IO>(
-        &self,
-        pgb: &mut PostgresBackend<IO>,
-        cancel: &CancellationToken,
-    ) -> Result<(), QueryError>
+    async fn flush_cancellable<IO>(&self, pgb: &mut PostgresBackend<IO>) -> Result<(), QueryError>
     where
         IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
     {
@@ -295,7 +299,7 @@ impl PageServerHandler {
             flush_r = pgb.flush() => {
                 Ok(flush_r?)
             },
-            _ = cancel.cancelled() => {
+            _ = self.cancel.cancelled() => {
                 Err(QueryError::Shutdown)
             }
         )
@@ -304,7 +308,6 @@ impl PageServerHandler {
     fn copyin_stream<'a, IO>(
         &'a self,
         pgb: &'a mut PostgresBackend<IO>,
-        cancel: &'a CancellationToken,
     ) -> impl Stream<Item = io::Result<Bytes>> + 'a
     where
         IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
@@ -314,7 +317,7 @@ impl PageServerHandler {
                 let msg = tokio::select! {
                     biased;
 
-                    _ = cancel.cancelled() => {
+                    _ = self.cancel.cancelled() => {
                         // We were requested to shut down.
                         let msg = "pageserver is shutting down";
                         let _ = pgb.write_message_noflush(&BeMessage::ErrorResponse(msg, None));
@@ -354,7 +357,7 @@ impl PageServerHandler {
                         let query_error = QueryError::Disconnected(ConnectionError::Io(io::Error::new(io::ErrorKind::ConnectionReset, msg)));
                         // error can't happen here, ErrorResponse serialization should be always ok
                         pgb.write_message_noflush(&BeMessage::ErrorResponse(msg, Some(query_error.pg_error_code()))).map_err(|e| e.into_io_error())?;
-                        self.flush_cancellable(pgb, cancel).await.map_err(|e| io::Error::new(io::ErrorKind::Other, e.to_string()))?;
+                        self.flush_cancellable(pgb).await.map_err(|e| io::Error::new(io::ErrorKind::Other, e.to_string()))?;
                         Err(io::Error::new(io::ErrorKind::ConnectionReset, msg))?;
                     }
                     Err(QueryError::Disconnected(ConnectionError::Io(io_error))) => {
@@ -381,13 +384,12 @@ impl PageServerHandler {
     {
         debug_assert_current_span_has_tenant_and_timeline_id();
 
+        // NOTE: pagerequests handler exits when connection is closed,
+        //       so there is no need to reset the association
+        task_mgr::associate_with(Some(tenant_id), Some(timeline_id));
+
         // Make request tracer if needed
-        let tenant = mgr::get_active_tenant_with_timeout(
-            tenant_id,
-            ACTIVE_TENANT_TIMEOUT,
-            &task_mgr::shutdown_token(),
-        )
-        .await?;
+        let tenant = get_active_tenant_with_timeout(tenant_id, &ctx).await?;
         let mut tracer = if tenant.get_trace_read_requests() {
             let connection_id = ConnectionId::generate();
             let path = tenant
@@ -403,14 +405,9 @@ impl PageServerHandler {
             .get_timeline(timeline_id, true)
             .map_err(|e| anyhow::anyhow!(e))?;
 
-        // Avoid starting new requests if the timeline has already started shutting down,
-        // and block timeline shutdown until this request is complete, or drops out due
-        // to cancellation.
-        let _timeline_guard = timeline.gate.enter().map_err(|_| QueryError::Shutdown)?;
-
         // switch client to COPYBOTH
         pgb.write_message_noflush(&BeMessage::CopyBothResponse)?;
-        self.flush_cancellable(pgb, &timeline.cancel).await?;
+        self.flush_cancellable(pgb).await?;
 
         let metrics = metrics::SmgrQueryTimePerTimeline::new(&tenant_id, &timeline_id);
 
@@ -418,7 +415,7 @@ impl PageServerHandler {
             let msg = tokio::select! {
                 biased;
 
-                _ = timeline.cancel.cancelled() => {
+                _ = self.cancel.cancelled() => {
                     // We were requested to shut down.
                     info!("shutdown request received in page handler");
                     return Err(QueryError::Shutdown)
@@ -491,27 +488,11 @@ impl PageServerHandler {
                         span,
                     )
                 }
-                PagestreamFeMessage::NoOp => {
-                    let _timer = metrics.start_timer(metrics::SmgrQueryType::NoOp);
-                    let span = tracing::info_span!("no_op");
-                    (Ok(PagestreamBeMessage::NoOp), span)
-                }
             };
 
-            if let Err(e) = &response {
-                if timeline.cancel.is_cancelled() {
-                    // If we fail to fulfil a request during shutdown, which may be _because_ of
-                    // shutdown, then do not send the error to the client.  Instead just drop the
-                    // connection.
-                    span.in_scope(|| info!("dropped response during shutdown: {e:#}"));
-                    return Err(QueryError::Shutdown);
-                }
-            }
-
             let response = response.unwrap_or_else(|e| {
                 // print the all details to the log with {:#}, but for the client the
-                // error message is enough.  Do not log if shutting down, as the anyhow::Error
-                // here includes cancellation which is not an error.
+                // error message is enough
                 span.in_scope(|| error!("error reading relation or page version: {:#}", e));
                 PagestreamBeMessage::Error(PagestreamErrorResponse {
                     message: e.to_string(),
@@ -519,7 +500,7 @@ impl PageServerHandler {
             });
 
             pgb.write_message_noflush(&BeMessage::CopyData(&response.serialize()))?;
-            self.flush_cancellable(pgb, &timeline.cancel).await?;
+            self.flush_cancellable(pgb).await?;
         }
         Ok(())
     }
@@ -541,14 +522,10 @@ impl PageServerHandler {
     {
         debug_assert_current_span_has_tenant_and_timeline_id();
 
+        task_mgr::associate_with(Some(tenant_id), Some(timeline_id));
         // Create empty timeline
         info!("creating new timeline");
-        let tenant = get_active_tenant_with_timeout(
-            tenant_id,
-            ACTIVE_TENANT_TIMEOUT,
-            &task_mgr::shutdown_token(),
-        )
-        .await?;
+        let tenant = get_active_tenant_with_timeout(tenant_id, &ctx).await?;
         let timeline = tenant
             .create_empty_timeline(timeline_id, base_lsn, pg_version, &ctx)
             .await?;
@@ -566,9 +543,9 @@ impl PageServerHandler {
         // Import basebackup provided via CopyData
         info!("importing basebackup");
         pgb.write_message_noflush(&BeMessage::CopyInResponse)?;
-        self.flush_cancellable(pgb, &tenant.cancel).await?;
+        self.flush_cancellable(pgb).await?;
 
-        let mut copyin_reader = pin!(StreamReader::new(self.copyin_stream(pgb, &tenant.cancel)));
+        let mut copyin_reader = pin!(StreamReader::new(self.copyin_stream(pgb)));
         timeline
             .import_basebackup_from_tar(
                 &mut copyin_reader,
@@ -605,10 +582,9 @@ impl PageServerHandler {
         IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
     {
         debug_assert_current_span_has_tenant_and_timeline_id();
+        task_mgr::associate_with(Some(tenant_id), Some(timeline_id));
 
-        let timeline = self
-            .get_active_tenant_timeline(tenant_id, timeline_id)
-            .await?;
+        let timeline = get_active_tenant_timeline(tenant_id, timeline_id, &ctx).await?;
         let last_record_lsn = timeline.get_last_record_lsn();
         if last_record_lsn != start_lsn {
             return Err(QueryError::Other(
@@ -622,8 +598,8 @@ impl PageServerHandler {
         // Import wal provided via CopyData
         info!("importing wal");
         pgb.write_message_noflush(&BeMessage::CopyInResponse)?;
-        self.flush_cancellable(pgb, &timeline.cancel).await?;
-        let mut copyin_reader = pin!(StreamReader::new(self.copyin_stream(pgb, &timeline.cancel)));
+        self.flush_cancellable(pgb).await?;
+        let mut copyin_reader = pin!(StreamReader::new(self.copyin_stream(pgb)));
         import_wal_from_tar(&timeline, &mut copyin_reader, start_lsn, end_lsn, &ctx).await?;
         info!("wal import complete");
 
@@ -816,9 +792,7 @@ impl PageServerHandler {
         let started = std::time::Instant::now();
 
         // check that the timeline exists
-        let timeline = self
-            .get_active_tenant_timeline(tenant_id, timeline_id)
-            .await?;
+        let timeline = get_active_tenant_timeline(tenant_id, timeline_id, &ctx).await?;
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
         if let Some(lsn) = lsn {
             // Backup was requested at a particular LSN. Wait for it to arrive.
@@ -833,7 +807,7 @@ impl PageServerHandler {
 
         // switch client to COPYOUT
         pgb.write_message_noflush(&BeMessage::CopyOutResponse)?;
-        self.flush_cancellable(pgb, &timeline.cancel).await?;
+        self.flush_cancellable(pgb).await?;
 
         // Send a tarball of the latest layer on the timeline. Compress if not
         // fullbackup. TODO Compress in that case too (tests need to be updated)
@@ -885,7 +859,7 @@ impl PageServerHandler {
         }
 
         pgb.write_message_noflush(&BeMessage::CopyDone)?;
-        self.flush_cancellable(pgb, &timeline.cancel).await?;
+        self.flush_cancellable(pgb).await?;
 
         let basebackup_after = started
             .elapsed()
@@ -917,25 +891,6 @@ impl PageServerHandler {
             .expect("claims presence already checked");
         check_permission(claims, tenant_id)
     }
-
-    /// Shorthand for getting a reference to a Timeline of an Active tenant.
-    async fn get_active_tenant_timeline(
-        &self,
-        tenant_id: TenantId,
-        timeline_id: TimelineId,
-    ) -> Result<Arc<Timeline>, GetActiveTimelineError> {
-        let tenant = get_active_tenant_with_timeout(
-            tenant_id,
-            ACTIVE_TENANT_TIMEOUT,
-            &task_mgr::shutdown_token(),
-        )
-        .await
-        .map_err(GetActiveTimelineError::Tenant)?;
-        let timeline = tenant
-            .get_timeline(timeline_id, true)
-            .map_err(|e| GetActiveTimelineError::Timeline(anyhow::anyhow!(e)))?;
-        Ok(timeline)
-    }
 }
 
 #[async_trait::async_trait]
@@ -1093,9 +1048,7 @@ where
                 .record("timeline_id", field::display(timeline_id));
 
             self.check_permission(Some(tenant_id))?;
-            let timeline = self
-                .get_active_tenant_timeline(tenant_id, timeline_id)
-                .await?;
+            let timeline = get_active_tenant_timeline(tenant_id, timeline_id, &ctx).await?;
 
             let end_of_timeline = timeline.get_last_record_rlsn();
 
@@ -1279,12 +1232,7 @@ where
 
             self.check_permission(Some(tenant_id))?;
 
-            let tenant = get_active_tenant_with_timeout(
-                tenant_id,
-                ACTIVE_TENANT_TIMEOUT,
-                &task_mgr::shutdown_token(),
-            )
-            .await?;
+            let tenant = get_active_tenant_with_timeout(tenant_id, &ctx).await?;
             pgb.write_message_noflush(&BeMessage::RowDescription(&[
                 RowDescriptor::int8_col(b"checkpoint_distance"),
                 RowDescriptor::int8_col(b"checkpoint_timeout"),
@@ -1330,16 +1278,67 @@ where
     }
 }
 
+#[derive(thiserror::Error, Debug)]
+enum GetActiveTenantError {
+    #[error(
+        "Timed out waiting {wait_time:?} for tenant active state. Latest state: {latest_state:?}"
+    )]
+    WaitForActiveTimeout {
+        latest_state: TenantState,
+        wait_time: Duration,
+    },
+    #[error(transparent)]
+    NotFound(GetTenantError),
+    #[error(transparent)]
+    WaitTenantActive(tenant::WaitToBecomeActiveError),
+}
+
 impl From<GetActiveTenantError> for QueryError {
     fn from(e: GetActiveTenantError) -> Self {
         match e {
             GetActiveTenantError::WaitForActiveTimeout { .. } => QueryError::Disconnected(
                 ConnectionError::Io(io::Error::new(io::ErrorKind::TimedOut, e.to_string())),
             ),
-            GetActiveTenantError::WillNotBecomeActive(TenantState::Stopping { .. }) => {
-                QueryError::Shutdown
+            GetActiveTenantError::WaitTenantActive(e) => QueryError::Other(anyhow::Error::new(e)),
+            GetActiveTenantError::NotFound(e) => QueryError::Other(anyhow::Error::new(e)),
+        }
+    }
+}
+
+/// Get active tenant.
+///
+/// If the tenant is Loading, waits for it to become Active, for up to 30 s. That
+/// ensures that queries don't fail immediately after pageserver startup, because
+/// all tenants are still loading.
+async fn get_active_tenant_with_timeout(
+    tenant_id: TenantId,
+    _ctx: &RequestContext, /* require get a context to support cancellation in the future */
+) -> Result<Arc<Tenant>, GetActiveTenantError> {
+    let tenant = match mgr::get_tenant(tenant_id, false).await {
+        Ok(tenant) => tenant,
+        Err(e @ GetTenantError::NotFound(_)) => return Err(GetActiveTenantError::NotFound(e)),
+        Err(GetTenantError::NotActive(_)) => {
+            unreachable!("we're calling get_tenant with active_only=false")
+        }
+        Err(GetTenantError::Broken(_)) => {
+            unreachable!("we're calling get_tenant with active_only=false")
+        }
+    };
+    let wait_time = Duration::from_secs(30);
+    match tokio::time::timeout(wait_time, tenant.wait_to_become_active()).await {
+        Ok(Ok(())) => Ok(tenant),
+        // no .context(), the error message is good enough and some tests depend on it
+        Ok(Err(e)) => Err(GetActiveTenantError::WaitTenantActive(e)),
+        Err(_) => {
+            let latest_state = tenant.current_state();
+            if latest_state == TenantState::Active {
+                Ok(tenant)
+            } else {
+                Err(GetActiveTenantError::WaitForActiveTimeout {
+                    latest_state,
+                    wait_time,
+                })
             }
-            e => QueryError::Other(anyhow::anyhow!(e)),
         }
     }
 }
@@ -1360,3 +1359,18 @@ impl From<GetActiveTimelineError> for QueryError {
         }
     }
 }
+
+/// Shorthand for getting a reference to a Timeline of an Active tenant.
+async fn get_active_tenant_timeline(
+    tenant_id: TenantId,
+    timeline_id: TimelineId,
+    ctx: &RequestContext,
+) -> Result<Arc<Timeline>, GetActiveTimelineError> {
+    let tenant = get_active_tenant_with_timeout(tenant_id, ctx)
+        .await
+        .map_err(GetActiveTimelineError::Tenant)?;
+    let timeline = tenant
+        .get_timeline(timeline_id, true)
+        .map_err(|e| GetActiveTimelineError::Timeline(anyhow::anyhow!(e)))?;
+    Ok(timeline)
+}

pageserver/src/pgdatadir_mapping.rs

@@ -44,17 +44,6 @@ pub enum CalculateLogicalSizeError {
     Other(#[from] anyhow::Error),
 }
 
-impl From<PageReconstructError> for CalculateLogicalSizeError {
-    fn from(pre: PageReconstructError) -> Self {
-        match pre {
-            PageReconstructError::AncestorStopping(_) | PageReconstructError::Cancelled => {
-                Self::Cancelled
-            }
-            _ => Self::Other(pre.into()),
-        }
-    }
-}
-
 #[derive(Debug, thiserror::Error)]
 pub enum RelationError {
     #[error("Relation Already Exists")]
@@ -584,17 +573,24 @@ impl Timeline {
         crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
 
         // Fetch list of database dirs and iterate them
-        let buf = self.get(DBDIR_KEY, lsn, ctx).await?;
+        let buf = self.get(DBDIR_KEY, lsn, ctx).await.context("read dbdir")?;
         let dbdir = DbDirectory::des(&buf).context("deserialize db directory")?;
 
         let mut total_size: u64 = 0;
         for (spcnode, dbnode) in dbdir.dbdirs.keys() {
-            for rel in self.list_rels(*spcnode, *dbnode, lsn, ctx).await? {
+            for rel in self
+                .list_rels(*spcnode, *dbnode, lsn, ctx)
+                .await
+                .context("list rels")?
+            {
                 if cancel.is_cancelled() {
                     return Err(CalculateLogicalSizeError::Cancelled);
                 }
                 let relsize_key = rel_size_to_key(rel);
-                let mut buf = self.get(relsize_key, lsn, ctx).await?;
+                let mut buf = self
+                    .get(relsize_key, lsn, ctx)
+                    .await
+                    .with_context(|| format!("read relation size of {rel:?}"))?;
                 let relsize = buf.get_u32_le();
 
                 total_size += relsize as u64;
@@ -1700,7 +1696,6 @@ const AUX_FILES_KEY: Key = Key {
 // Reverse mappings for a few Keys.
 // These are needed by WAL redo manager.
 
-/// Guaranteed to return `Ok()` if [[is_rel_block_key]] returns `true` for `key`.
 pub fn key_to_rel_block(key: Key) -> anyhow::Result<(RelTag, BlockNumber)> {
     Ok(match key.field1 {
         0x00 => (
@@ -1716,8 +1711,7 @@ pub fn key_to_rel_block(key: Key) -> anyhow::Result<(RelTag, BlockNumber)> {
     })
 }
 
-/// See [[key_to_rel_block]].
-pub fn is_rel_block_key(key: Key) -> bool {
+fn is_rel_block_key(key: Key) -> bool {
     key.field1 == 0x00 && key.field4 != 0
 }
 

pageserver/src/repository.rs

@@ -172,6 +172,21 @@ impl Key {
     }
 }
 
+impl pageserver_compaction::interface::CompactionKey for Key {
+    const MIN: Self = Self::MIN;
+    const MAX: Self = Self::MAX;
+
+    fn key_range_size(r: &std::ops::Range<Self>) -> u32 {
+        key_range_size(r)
+    }
+    fn next(&self) -> Key {
+        (self as &Key).next()
+    }
+    fn skip_some(&self) -> Key {
+        self.add(128)
+    }
+}
+
 /// A 'value' stored for a one Key.
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub enum Value {

pageserver/src/task_mgr.rs

@@ -299,6 +299,10 @@ pub enum TaskKind {
 
 #[derive(Default)]
 struct MutableTaskState {
+    /// Tenant and timeline that this task is associated with.
+    tenant_id: Option<TenantId>,
+    timeline_id: Option<TimelineId>,
+
     /// Handle for waiting for the task to exit. It can be None, if the
     /// the task has already exited.
     join_handle: Option<JoinHandle<()>>,
@@ -315,11 +319,6 @@ struct PageServerTask {
     // To request task shutdown, just cancel this token.
     cancel: CancellationToken,
 
-    /// Tasks may optionally be launched for a particular tenant/timeline, enabling
-    /// later cancelling tasks for that tenant/timeline in [`shutdown_tasks`]
-    tenant_id: Option<TenantId>,
-    timeline_id: Option<TimelineId>,
-
     mutable: Mutex<MutableTaskState>,
 }
 
@@ -345,9 +344,11 @@ where
         kind,
         name: name.to_string(),
         cancel: cancel.clone(),
-        tenant_id,
-        timeline_id,
-        mutable: Mutex::new(MutableTaskState { join_handle: None }),
+        mutable: Mutex::new(MutableTaskState {
+            tenant_id,
+            timeline_id,
+            join_handle: None,
+        }),
     });
 
     TASKS.lock().unwrap().insert(task_id, Arc::clone(&task));
@@ -417,6 +418,8 @@ async fn task_finish(
 
     let mut shutdown_process = false;
     {
+        let task_mut = task.mutable.lock().unwrap();
+
         match result {
             Ok(Ok(())) => {
                 debug!("Task '{}' exited normally", task_name);
@@ -425,13 +428,13 @@ async fn task_finish(
                 if shutdown_process_on_error {
                     error!(
                         "Shutting down: task '{}' tenant_id: {:?}, timeline_id: {:?} exited with error: {:?}",
-                        task_name, task.tenant_id, task.timeline_id, err
+                        task_name, task_mut.tenant_id, task_mut.timeline_id, err
                     );
                     shutdown_process = true;
                 } else {
                     error!(
                         "Task '{}' tenant_id: {:?}, timeline_id: {:?} exited with error: {:?}",
-                        task_name, task.tenant_id, task.timeline_id, err
+                        task_name, task_mut.tenant_id, task_mut.timeline_id, err
                     );
                 }
             }
@@ -439,13 +442,13 @@ async fn task_finish(
                 if shutdown_process_on_error {
                     error!(
                         "Shutting down: task '{}' tenant_id: {:?}, timeline_id: {:?} panicked: {:?}",
-                        task_name, task.tenant_id, task.timeline_id, err
+                        task_name, task_mut.tenant_id, task_mut.timeline_id, err
                     );
                     shutdown_process = true;
                 } else {
                     error!(
                         "Task '{}' tenant_id: {:?}, timeline_id: {:?} panicked: {:?}",
-                        task_name, task.tenant_id, task.timeline_id, err
+                        task_name, task_mut.tenant_id, task_mut.timeline_id, err
                     );
                 }
             }
@@ -457,6 +460,17 @@ async fn task_finish(
     }
 }
 
+// expected to be called from the task of the given id.
+pub fn associate_with(tenant_id: Option<TenantId>, timeline_id: Option<TimelineId>) {
+    CURRENT_TASK.with(|ct| {
+        let mut task_mut = ct.mutable.lock().unwrap();
+        task_mut.tenant_id = tenant_id;
+        task_mut.timeline_id = timeline_id;
+    });
+}
+
+/// Is there a task running that matches the criteria
+
 /// Signal and wait for tasks to shut down.
 ///
 ///
@@ -479,16 +493,17 @@ pub async fn shutdown_tasks(
     {
         let tasks = TASKS.lock().unwrap();
         for task in tasks.values() {
+            let task_mut = task.mutable.lock().unwrap();
             if (kind.is_none() || Some(task.kind) == kind)
-                && (tenant_id.is_none() || task.tenant_id == tenant_id)
-                && (timeline_id.is_none() || task.timeline_id == timeline_id)
+                && (tenant_id.is_none() || task_mut.tenant_id == tenant_id)
+                && (timeline_id.is_none() || task_mut.timeline_id == timeline_id)
             {
                 task.cancel.cancel();
                 victim_tasks.push((
                     Arc::clone(task),
                     task.kind,
-                    task.tenant_id,
-                    task.timeline_id,
+                    task_mut.tenant_id,
+                    task_mut.timeline_id,
                 ));
             }
         }

pageserver/src/tenant.rs

@@ -26,7 +26,6 @@ use tracing::*;
 use utils::completion;
 use utils::crashsafe::path_with_suffix_extension;
 use utils::fs_ext;
-use utils::sync::gate::Gate;
 
 use std::cmp::min;
 use std::collections::hash_map::Entry;
@@ -55,8 +54,6 @@ use self::config::TenantConf;
 use self::delete::DeleteTenantFlow;
 use self::metadata::LoadMetadataError;
 use self::metadata::TimelineMetadata;
-use self::mgr::GetActiveTenantError;
-use self::mgr::GetTenantError;
 use self::mgr::TenantsMap;
 use self::remote_timeline_client::RemoteTimelineClient;
 use self::timeline::uninit::TimelineUninitMark;
@@ -255,20 +252,6 @@ pub struct Tenant {
     eviction_task_tenant_state: tokio::sync::Mutex<EvictionTaskTenantState>,
 
     pub(crate) delete_progress: Arc<tokio::sync::Mutex<DeleteTenantFlow>>,
-
-    // Cancellation token fires when we have entered shutdown().  This is a parent of
-    // Timelines' cancellation token.
-    pub(crate) cancel: CancellationToken,
-
-    // Users of the Tenant such as the page service must take this Gate to avoid
-    // trying to use a Tenant which is shutting down.
-    pub(crate) gate: Gate,
-}
-
-impl std::fmt::Debug for Tenant {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        write!(f, "{} ({})", self.tenant_id, self.current_state())
-    }
 }
 
 pub(crate) enum WalRedoManager {
@@ -376,6 +359,34 @@ impl Debug for SetStoppingError {
     }
 }
 
+#[derive(Debug, thiserror::Error)]
+pub(crate) enum WaitToBecomeActiveError {
+    WillNotBecomeActive {
+        tenant_id: TenantId,
+        state: TenantState,
+    },
+    TenantDropped {
+        tenant_id: TenantId,
+    },
+}
+
+impl std::fmt::Display for WaitToBecomeActiveError {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        match self {
+            WaitToBecomeActiveError::WillNotBecomeActive { tenant_id, state } => {
+                write!(
+                    f,
+                    "Tenant {} will not become active. Current state: {:?}",
+                    tenant_id, state
+                )
+            }
+            WaitToBecomeActiveError::TenantDropped { tenant_id } => {
+                write!(f, "Tenant {tenant_id} will not become active (dropped)")
+            }
+        }
+    }
+}
+
 #[derive(thiserror::Error, Debug)]
 pub enum CreateTimelineError {
     #[error("a timeline with the given ID already exists")]
@@ -384,8 +395,6 @@ pub enum CreateTimelineError {
     AncestorLsn(anyhow::Error),
     #[error("ancestor timeline is not active")]
     AncestorNotActive,
-    #[error("tenant shutting down")]
-    ShuttingDown,
     #[error(transparent)]
     Other(#[from] anyhow::Error),
 }
@@ -517,7 +526,7 @@ impl Tenant {
         resources: TenantSharedResources,
         attached_conf: AttachedTenantConf,
         init_order: Option<InitializationOrder>,
-        tenants: &'static std::sync::RwLock<TenantsMap>,
+        tenants: &'static tokio::sync::RwLock<TenantsMap>,
         mode: SpawnMode,
         ctx: &RequestContext,
     ) -> anyhow::Result<Arc<Tenant>> {
@@ -1515,11 +1524,6 @@ impl Tenant {
             )));
         }
 
-        let _gate = self
-            .gate
-            .enter()
-            .map_err(|_| CreateTimelineError::ShuttingDown)?;
-
         if let Ok(existing) = self.get_timeline(new_timeline_id, false) {
             debug!("timeline {new_timeline_id} already exists");
 
@@ -1804,7 +1808,6 @@ impl Tenant {
         freeze_and_flush: bool,
     ) -> Result<(), completion::Barrier> {
         span::debug_assert_current_span_has_tenant_id();
-
         // Set tenant (and its timlines) to Stoppping state.
         //
         // Since we can only transition into Stopping state after activation is complete,
@@ -1830,7 +1833,6 @@ impl Tenant {
             }
             Err(SetStoppingError::AlreadyStopping(other)) => {
                 // give caller the option to wait for this this shutdown
-                info!("Tenant::shutdown: AlreadyStopping");
                 return Err(other);
             }
         };
@@ -1844,7 +1846,6 @@ impl Tenant {
                 js.spawn(async move { timeline.shutdown(freeze_and_flush).instrument(span).await });
             })
         };
-        tracing::info!("Waiting for timelines...");
         while let Some(res) = js.join_next().await {
             match res {
                 Ok(()) => {}
@@ -1854,21 +1855,12 @@ impl Tenant {
             }
         }
 
-        // We cancel the Tenant's cancellation token _after_ the timelines have all shut down.  This permits
-        // them to continue to do work during their shutdown methods, e.g. flushing data.
-        tracing::debug!("Cancelling CancellationToken");
-        self.cancel.cancel();
-
         // shutdown all tenant and timeline tasks: gc, compaction, page service
         // No new tasks will be started for this tenant because it's in `Stopping` state.
         //
         // this will additionally shutdown and await all timeline tasks.
-        tracing::debug!("Waiting for tasks...");
         task_mgr::shutdown_tasks(None, Some(self.tenant_id), None).await;
 
-        // Wait for any in-flight operations to complete
-        self.gate.close().await;
-
         Ok(())
     }
 
@@ -2029,7 +2021,7 @@ impl Tenant {
         self.state.subscribe()
     }
 
-    pub(crate) async fn wait_to_become_active(&self) -> Result<(), GetActiveTenantError> {
+    pub(crate) async fn wait_to_become_active(&self) -> Result<(), WaitToBecomeActiveError> {
         let mut receiver = self.state.subscribe();
         loop {
             let current_state = receiver.borrow_and_update().clone();
@@ -2037,9 +2029,11 @@ impl Tenant {
                 TenantState::Loading | TenantState::Attaching | TenantState::Activating(_) => {
                     // in these states, there's a chance that we can reach ::Active
                     receiver.changed().await.map_err(
-                        |_e: tokio::sync::watch::error::RecvError|
-                            // Tenant existed but was dropped: report it as non-existent
-                            GetActiveTenantError::NotFound(GetTenantError::NotFound(self.tenant_id))
+                        |_e: tokio::sync::watch::error::RecvError| {
+                            WaitToBecomeActiveError::TenantDropped {
+                                tenant_id: self.tenant_id,
+                            }
+                        },
                     )?;
                 }
                 TenantState::Active { .. } => {
@@ -2047,7 +2041,10 @@ impl Tenant {
                 }
                 TenantState::Broken { .. } | TenantState::Stopping { .. } => {
                     // There's no chance the tenant can transition back into ::Active
-                    return Err(GetActiveTenantError::WillNotBecomeActive(current_state));
+                    return Err(WaitToBecomeActiveError::WillNotBecomeActive {
+                        tenant_id: self.tenant_id,
+                        state: current_state,
+                    });
                 }
             }
         }
@@ -2113,9 +2110,6 @@ where
 }
 
 impl Tenant {
-    pub fn get_tenant_id(&self) -> TenantId {
-        self.tenant_id
-    }
     pub fn tenant_specific_overrides(&self) -> TenantConfOpt {
         self.tenant_conf.read().unwrap().tenant_conf
     }
@@ -2273,7 +2267,6 @@ impl Tenant {
             initial_logical_size_can_start.cloned(),
             initial_logical_size_attempt.cloned().flatten(),
             state,
-            self.cancel.child_token(),
         );
 
         Ok(timeline)
@@ -2363,8 +2356,6 @@ impl Tenant {
             cached_synthetic_tenant_size: Arc::new(AtomicU64::new(0)),
             eviction_task_tenant_state: tokio::sync::Mutex::new(EvictionTaskTenantState::default()),
             delete_progress: Arc::new(tokio::sync::Mutex::new(DeleteTenantFlow::default())),
-            cancel: CancellationToken::default(),
-            gate: Gate::new(format!("Tenant<{tenant_id}>")),
         }
     }
 
@@ -3452,6 +3443,7 @@ pub(crate) mod harness {
                 compaction_target_size: Some(tenant_conf.compaction_target_size),
                 compaction_period: Some(tenant_conf.compaction_period),
                 compaction_threshold: Some(tenant_conf.compaction_threshold),
+                compaction_algorithm: Some(tenant_conf.compaction_algorithm),
                 gc_horizon: Some(tenant_conf.gc_horizon),
                 gc_period: Some(tenant_conf.gc_period),
                 image_creation_threshold: Some(tenant_conf.image_creation_threshold),
@@ -3701,7 +3693,7 @@ mod tests {
     use tokio_util::sync::CancellationToken;
 
     static TEST_KEY: Lazy<Key> =
-        Lazy::new(|| Key::from_slice(&hex!("010000000033333333444444445500000001")));
+        Lazy::new(|| Key::from_slice(&hex!("112222222233333333444444445500000001")));
 
     #[tokio::test]
     async fn test_basic() -> anyhow::Result<()> {
@@ -3797,9 +3789,9 @@ mod tests {
         let writer = tline.writer().await;
 
         #[allow(non_snake_case)]
-        let TEST_KEY_A: Key = Key::from_hex("110000000033333333444444445500000001").unwrap();
+        let TEST_KEY_A: Key = Key::from_hex("112222222233333333444444445500000001").unwrap();
         #[allow(non_snake_case)]
-        let TEST_KEY_B: Key = Key::from_hex("110000000033333333444444445500000002").unwrap();
+        let TEST_KEY_B: Key = Key::from_hex("112222222233333333444444445500000002").unwrap();
 
         // Insert a value on the timeline
         writer
@@ -4245,7 +4237,11 @@ mod tests {
         metadata_bytes[8] ^= 1;
         std::fs::write(metadata_path, metadata_bytes)?;
 
-        let err = harness.try_load_local(&ctx).await.expect_err("should fail");
+        let err = harness
+            .try_load_local(&ctx)
+            .await
+            .err()
+            .expect("should fail");
         // get all the stack with all .context, not only the last one
         let message = format!("{err:#}");
         let expected = "failed to load metadata";
@@ -4379,7 +4375,7 @@ mod tests {
 
         let mut keyspace = KeySpaceAccum::new();
 
-        let mut test_key = Key::from_hex("010000000033333333444444445500000000").unwrap();
+        let mut test_key = Key::from_hex("012222222233333333444444445500000000").unwrap();
         let mut blknum = 0;
         for _ in 0..50 {
             for _ in 0..10000 {
@@ -4425,7 +4421,7 @@ mod tests {
 
         const NUM_KEYS: usize = 1000;
 
-        let mut test_key = Key::from_hex("010000000033333333444444445500000000").unwrap();
+        let mut test_key = Key::from_hex("012222222233333333444444445500000000").unwrap();
 
         let mut keyspace = KeySpaceAccum::new();
 
@@ -4506,7 +4502,7 @@ mod tests {
 
         const NUM_KEYS: usize = 1000;
 
-        let mut test_key = Key::from_hex("010000000033333333444444445500000000").unwrap();
+        let mut test_key = Key::from_hex("012222222233333333444444445500000000").unwrap();
 
         let mut keyspace = KeySpaceAccum::new();
 
@@ -4597,7 +4593,7 @@ mod tests {
         const NUM_KEYS: usize = 100;
         const NUM_TLINES: usize = 50;
 
-        let mut test_key = Key::from_hex("010000000033333333444444445500000000").unwrap();
+        let mut test_key = Key::from_hex("012222222233333333444444445500000000").unwrap();
         // Track page mutation lsns across different timelines.
         let mut updated = [[Lsn(0); NUM_KEYS]; NUM_TLINES];
 

pageserver/src/tenant/config.rs

@@ -23,12 +23,17 @@ pub mod defaults {
     pub const DEFAULT_CHECKPOINT_DISTANCE: u64 = 256 * 1024 * 1024;
     pub const DEFAULT_CHECKPOINT_TIMEOUT: &str = "10 m";
 
+    // FIXME the below configs are only used by legacy algorithm. The new algorithm
+    // has different parameters.
+
     // Target file size, when creating image and delta layers.
     // This parameter determines L1 layer file size.
     pub const DEFAULT_COMPACTION_TARGET_SIZE: u64 = 128 * 1024 * 1024;
 
     pub const DEFAULT_COMPACTION_PERIOD: &str = "20 s";
     pub const DEFAULT_COMPACTION_THRESHOLD: usize = 10;
+    pub const DEFAULT_COMPACTION_ALGORITHM: super::CompactionAlgorithm =
+        super::CompactionAlgorithm::Legacy;
 
     pub const DEFAULT_GC_HORIZON: u64 = 64 * 1024 * 1024;
 
@@ -270,6 +275,7 @@ pub struct TenantConf {
     pub compaction_period: Duration,
     // Level0 delta layer threshold for compaction.
     pub compaction_threshold: usize,
+    pub compaction_algorithm: CompactionAlgorithm,
     // Determines how much history is retained, to allow
     // branching and read replicas at an older point in time.
     // The unit is #of bytes of WAL.
@@ -333,6 +339,10 @@ pub struct TenantConfOpt {
     #[serde(default)]
     pub compaction_threshold: Option<usize>,
 
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[serde(default)]
+    pub compaction_algorithm: Option<CompactionAlgorithm>,
+
     #[serde(skip_serializing_if = "Option::is_none")]
     #[serde(default)]
     pub gc_horizon: Option<u64>,
@@ -387,6 +397,13 @@ pub struct TenantConfOpt {
     pub gc_feedback: Option<bool>,
 }
 
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
+#[serde(tag = "kind")]
+pub enum CompactionAlgorithm {
+    Legacy,
+    Tiered,
+}
+
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
 #[serde(tag = "kind")]
 pub enum EvictionPolicy {
@@ -429,6 +446,9 @@ impl TenantConfOpt {
             compaction_threshold: self
                 .compaction_threshold
                 .unwrap_or(global_conf.compaction_threshold),
+            compaction_algorithm: self
+                .compaction_algorithm
+                .unwrap_or(global_conf.compaction_algorithm),
             gc_horizon: self.gc_horizon.unwrap_or(global_conf.gc_horizon),
             gc_period: self.gc_period.unwrap_or(global_conf.gc_period),
             image_creation_threshold: self
@@ -468,6 +488,7 @@ impl Default for TenantConf {
             compaction_period: humantime::parse_duration(DEFAULT_COMPACTION_PERIOD)
                 .expect("cannot parse default compaction period"),
             compaction_threshold: DEFAULT_COMPACTION_THRESHOLD,
+            compaction_algorithm: DEFAULT_COMPACTION_ALGORITHM,
             gc_horizon: DEFAULT_GC_HORIZON,
             gc_period: humantime::parse_duration(DEFAULT_GC_PERIOD)
                 .expect("cannot parse default gc period"),
@@ -556,6 +577,12 @@ impl TryFrom<&'_ models::TenantConfig> for TenantConfOpt {
 
         tenant_conf.compaction_target_size = request_data.compaction_target_size;
         tenant_conf.compaction_threshold = request_data.compaction_threshold;
+        if let Some(compaction_algorithm) = &request_data.compaction_algorithm {
+            tenant_conf.compaction_algorithm = Some(
+                serde::Deserialize::deserialize(compaction_algorithm)
+                    .context("parse field `compaction_algorithm`")?,
+            );
+        }
 
         if let Some(compaction_period) = &request_data.compaction_period {
             tenant_conf.compaction_period = Some(

pageserver/src/tenant/delete.rs

@@ -21,7 +21,7 @@ use crate::{
 };
 
 use super::{
-    mgr::{GetTenantError, TenantSlotError, TenantSlotUpsertError, TenantsMap},
+    mgr::{GetTenantError, TenantsMap},
     remote_timeline_client::{FAILED_REMOTE_OP_RETRIES, FAILED_UPLOAD_WARN_THRESHOLD},
     span,
     timeline::delete::DeleteTimelineFlow,
@@ -33,21 +33,12 @@ pub(crate) enum DeleteTenantError {
     #[error("GetTenant {0}")]
     Get(#[from] GetTenantError),
 
-    #[error("Tenant not attached")]
-    NotAttached,
-
     #[error("Invalid state {0}. Expected Active or Broken")]
     InvalidState(TenantState),
 
     #[error("Tenant deletion is already in progress")]
     AlreadyInProgress,
 
-    #[error("Tenant map slot error {0}")]
-    SlotError(#[from] TenantSlotError),
-
-    #[error("Tenant map slot upsert error {0}")]
-    SlotUpsertError(#[from] TenantSlotUpsertError),
-
     #[error("Timeline {0}")]
     Timeline(#[from] DeleteTimelineError),
 
@@ -282,12 +273,12 @@ impl DeleteTenantFlow {
     pub(crate) async fn run(
         conf: &'static PageServerConf,
         remote_storage: Option<GenericRemoteStorage>,
-        tenants: &'static std::sync::RwLock<TenantsMap>,
-        tenant: Arc<Tenant>,
+        tenants: &'static tokio::sync::RwLock<TenantsMap>,
+        tenant_id: TenantId,
     ) -> Result<(), DeleteTenantError> {
         span::debug_assert_current_span_has_tenant_id();
 
-        let mut guard = Self::prepare(&tenant).await?;
+        let (tenant, mut guard) = Self::prepare(tenants, tenant_id).await?;
 
         if let Err(e) = Self::run_inner(&mut guard, conf, remote_storage.as_ref(), &tenant).await {
             tenant.set_broken(format!("{e:#}")).await;
@@ -387,7 +378,7 @@ impl DeleteTenantFlow {
         guard: DeletionGuard,
         tenant: &Arc<Tenant>,
         preload: Option<TenantPreload>,
-        tenants: &'static std::sync::RwLock<TenantsMap>,
+        tenants: &'static tokio::sync::RwLock<TenantsMap>,
         init_order: Option<InitializationOrder>,
         ctx: &RequestContext,
     ) -> Result<(), DeleteTenantError> {
@@ -414,8 +405,15 @@ impl DeleteTenantFlow {
     }
 
     async fn prepare(
-        tenant: &Arc<Tenant>,
-    ) -> Result<tokio::sync::OwnedMutexGuard<Self>, DeleteTenantError> {
+        tenants: &tokio::sync::RwLock<TenantsMap>,
+        tenant_id: TenantId,
+    ) -> Result<(Arc<Tenant>, tokio::sync::OwnedMutexGuard<Self>), DeleteTenantError> {
+        let m = tenants.read().await;
+
+        let tenant = m
+            .get(&tenant_id)
+            .ok_or(GetTenantError::NotFound(tenant_id))?;
+
         // FIXME: unsure about active only. Our init jobs may not be cancellable properly,
         // so at least for now allow deletions only for active tenants. TODO recheck
         // Broken and Stopping is needed for retries.
@@ -449,14 +447,14 @@ impl DeleteTenantFlow {
             )));
         }
 
-        Ok(guard)
+        Ok((Arc::clone(tenant), guard))
     }
 
     fn schedule_background(
         guard: OwnedMutexGuard<Self>,
         conf: &'static PageServerConf,
         remote_storage: Option<GenericRemoteStorage>,
-        tenants: &'static std::sync::RwLock<TenantsMap>,
+        tenants: &'static tokio::sync::RwLock<TenantsMap>,
         tenant: Arc<Tenant>,
     ) {
         let tenant_id = tenant.tenant_id;
@@ -489,7 +487,7 @@ impl DeleteTenantFlow {
         mut guard: OwnedMutexGuard<Self>,
         conf: &PageServerConf,
         remote_storage: Option<GenericRemoteStorage>,
-        tenants: &'static std::sync::RwLock<TenantsMap>,
+        tenants: &'static tokio::sync::RwLock<TenantsMap>,
         tenant: &Arc<Tenant>,
     ) -> Result<(), DeleteTenantError> {
         // Tree sort timelines, schedule delete for them. Mention retries from the console side.
@@ -537,18 +535,10 @@ impl DeleteTenantFlow {
             .await
             .context("cleanup_remaining_fs_traces")?;
 
-        {
-            let mut locked = tenants.write().unwrap();
-            if locked.remove(&tenant.tenant_id).is_none() {
-                warn!("Tenant got removed from tenants map during deletion");
-            };
-
-            // FIXME: we should not be modifying this from outside of mgr.rs.
-            // This will go away when we simplify deletion (https://github.com/neondatabase/neon/issues/5080)
-            crate::metrics::TENANT_MANAGER
-                .tenant_slots
-                .set(locked.len() as u64);
-        }
+        let mut locked = tenants.write().await;
+        if locked.remove(&tenant.tenant_id).is_none() {
+            warn!("Tenant got removed from tenants map during deletion");
+        };
 
         *guard = Self::Finished;
 

pageserver/src/tenant/metadata.rs

@@ -406,123 +406,4 @@ mod tests {
             METADATA_OLD_FORMAT_VERSION, METADATA_FORMAT_VERSION
         );
     }
-
-    #[test]
-    fn test_metadata_bincode_serde() {
-        let original_metadata = TimelineMetadata::new(
-            Lsn(0x200),
-            Some(Lsn(0x100)),
-            Some(TIMELINE_ID),
-            Lsn(0),
-            Lsn(0),
-            Lsn(0),
-            // Any version will do here, so use the default
-            crate::DEFAULT_PG_VERSION,
-        );
-        let metadata_bytes = original_metadata
-            .to_bytes()
-            .expect("Cannot create bytes array from metadata");
-
-        let metadata_bincode_be_bytes = original_metadata
-            .ser()
-            .expect("Cannot serialize the metadata");
-
-        // 8 bytes for the length of the vector
-        assert_eq!(metadata_bincode_be_bytes.len(), 8 + metadata_bytes.len());
-
-        let expected_bincode_bytes = {
-            let mut temp = vec![];
-            let len_bytes = metadata_bytes.len().to_be_bytes();
-            temp.extend_from_slice(&len_bytes);
-            temp.extend_from_slice(&metadata_bytes);
-            temp
-        };
-        assert_eq!(metadata_bincode_be_bytes, expected_bincode_bytes);
-
-        let deserialized_metadata = TimelineMetadata::des(&metadata_bincode_be_bytes).unwrap();
-        // Deserialized metadata has the metadata header, which is different from the serialized one.
-        //   Reference: TimelineMetaData::to_bytes()
-        let expected_metadata = {
-            let mut temp_metadata = original_metadata;
-            let body_bytes = temp_metadata
-                .body
-                .ser()
-                .expect("Cannot serialize the metadata body");
-            let metadata_size = METADATA_HDR_SIZE + body_bytes.len();
-            let hdr = TimelineMetadataHeader {
-                size: metadata_size as u16,
-                format_version: METADATA_FORMAT_VERSION,
-                checksum: crc32c::crc32c(&body_bytes),
-            };
-            temp_metadata.hdr = hdr;
-            temp_metadata
-        };
-        assert_eq!(deserialized_metadata, expected_metadata);
-    }
-
-    #[test]
-    fn test_metadata_bincode_serde_ensure_roundtrip() {
-        let original_metadata = TimelineMetadata::new(
-            Lsn(0x200),
-            Some(Lsn(0x100)),
-            Some(TIMELINE_ID),
-            Lsn(0),
-            Lsn(0),
-            Lsn(0),
-            // Any version will do here, so use the default
-            crate::DEFAULT_PG_VERSION,
-        );
-        let expected_bytes = vec![
-            /* bincode length encoding bytes */
-            0, 0, 0, 0, 0, 0, 2, 0, // 8 bytes for the length of the serialized vector
-            /* TimelineMetadataHeader */
-            4, 37, 101, 34, 0, 70, 0, 4, // checksum, size, format_version (4 + 2 + 2)
-            /* TimelineMetadataBodyV2 */
-            0, 0, 0, 0, 0, 0, 2, 0, // disk_consistent_lsn (8 bytes)
-            1, 0, 0, 0, 0, 0, 0, 1, 0, // prev_record_lsn (9 bytes)
-            1, 17, 34, 51, 68, 85, 102, 119, 136, 17, 34, 51, 68, 85, 102, 119,
-            136, // ancestor_timeline (17 bytes)
-            0, 0, 0, 0, 0, 0, 0, 0, // ancestor_lsn (8 bytes)
-            0, 0, 0, 0, 0, 0, 0, 0, // latest_gc_cutoff_lsn (8 bytes)
-            0, 0, 0, 0, 0, 0, 0, 0, // initdb_lsn (8 bytes)
-            0, 0, 0, 15, // pg_version (4 bytes)
-            /* padding bytes */
-            0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-            0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-            0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-            0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-            0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-            0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-            0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-            0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-            0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-            0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-            0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-            0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-            0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-            0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-            0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-            0, 0, 0, 0, 0, 0, 0,
-        ];
-        let metadata_ser_bytes = original_metadata.ser().unwrap();
-        assert_eq!(metadata_ser_bytes, expected_bytes);
-
-        let expected_metadata = {
-            let mut temp_metadata = original_metadata;
-            let body_bytes = temp_metadata
-                .body
-                .ser()
-                .expect("Cannot serialize the metadata body");
-            let metadata_size = METADATA_HDR_SIZE + body_bytes.len();
-            let hdr = TimelineMetadataHeader {
-                size: metadata_size as u16,
-                format_version: METADATA_FORMAT_VERSION,
-                checksum: crc32c::crc32c(&body_bytes),
-            };
-            temp_metadata.hdr = hdr;
-            temp_metadata
-        };
-        let des_metadata = TimelineMetadata::des(&metadata_ser_bytes).unwrap();
-        assert_eq!(des_metadata, expected_metadata);
-    }
 }

pageserver/src/tenant/remote_timeline_client/index.rs

@@ -6,6 +6,7 @@ use std::collections::HashMap;
 
 use chrono::NaiveDateTime;
 use serde::{Deserialize, Serialize};
+use serde_with::{serde_as, DisplayFromStr};
 use utils::bin_ser::SerializeError;
 
 use crate::tenant::metadata::TimelineMetadata;
@@ -57,6 +58,7 @@ impl LayerFileMetadata {
 ///
 /// This type needs to be backwards and forwards compatible. When changing the fields,
 /// remember to add a test case for the changed version.
+#[serde_as]
 #[derive(Debug, PartialEq, Eq, Clone, Serialize, Deserialize)]
 pub struct IndexPart {
     /// Debugging aid describing the version of this type.
@@ -76,6 +78,7 @@ pub struct IndexPart {
     // 'disk_consistent_lsn' is a copy of the 'disk_consistent_lsn' in the metadata.
     // It's duplicated for convenience when reading the serialized structure, but is
     // private because internally we would read from metadata instead.
+    #[serde_as(as = "DisplayFromStr")]
     disk_consistent_lsn: Lsn,
 
     #[serde(rename = "metadata_bytes")]

pageserver/src/tenant/size.rs

@@ -29,6 +29,7 @@ use tenant_size_model::{Segment, StorageModel};
 /// needs. We will convert this into a StorageModel when it's time to perform
 /// the calculation.
 ///
+#[serde_with::serde_as]
 #[derive(Debug, serde::Serialize, serde::Deserialize)]
 pub struct ModelInputs {
     pub segments: Vec<SegmentMeta>,
@@ -36,9 +37,11 @@ pub struct ModelInputs {
 }
 
 /// A [`Segment`], with some extra information for display purposes
+#[serde_with::serde_as]
 #[derive(Debug, serde::Serialize, serde::Deserialize)]
 pub struct SegmentMeta {
     pub segment: Segment,
+    #[serde_as(as = "serde_with::DisplayFromStr")]
     pub timeline_id: TimelineId,
     pub kind: LsnKind,
 }
@@ -74,22 +77,32 @@ pub enum LsnKind {
 
 /// Collect all relevant LSNs to the inputs. These will only be helpful in the serialized form as
 /// part of [`ModelInputs`] from the HTTP api, explaining the inputs.
+#[serde_with::serde_as]
 #[derive(Debug, serde::Serialize, serde::Deserialize)]
 pub struct TimelineInputs {
+    #[serde_as(as = "serde_with::DisplayFromStr")]
     pub timeline_id: TimelineId,
 
+    #[serde_as(as = "Option<serde_with::DisplayFromStr>")]
     pub ancestor_id: Option<TimelineId>,
 
+    #[serde_as(as = "serde_with::DisplayFromStr")]
     ancestor_lsn: Lsn,
+    #[serde_as(as = "serde_with::DisplayFromStr")]
     last_record: Lsn,
+    #[serde_as(as = "serde_with::DisplayFromStr")]
     latest_gc_cutoff: Lsn,
+    #[serde_as(as = "serde_with::DisplayFromStr")]
     horizon_cutoff: Lsn,
+    #[serde_as(as = "serde_with::DisplayFromStr")]
     pitr_cutoff: Lsn,
 
     /// Cutoff point based on GC settings
+    #[serde_as(as = "serde_with::DisplayFromStr")]
     next_gc_cutoff: Lsn,
 
     /// Cutoff point calculated from the user-supplied 'max_retention_period'
+    #[serde_as(as = "Option<serde_with::DisplayFromStr>")]
     retention_param_cutoff: Option<Lsn>,
 }
 
@@ -406,12 +419,10 @@ async fn fill_logical_sizes(
                 have_any_error = true;
             }
             Ok(Ok(TimelineAtLsnSizeResult(timeline, lsn, Err(error)))) => {
-                if !matches!(error, CalculateLogicalSizeError::Cancelled) {
-                    warn!(
-                        timeline_id=%timeline.timeline_id,
-                        "failed to calculate logical size at {lsn}: {error:#}"
-                    );
-                }
+                warn!(
+                    timeline_id=%timeline.timeline_id,
+                    "failed to calculate logical size at {lsn}: {error:#}"
+                );
                 have_any_error = true;
             }
             Ok(Ok(TimelineAtLsnSizeResult(timeline, lsn, Ok(size)))) => {

pageserver/src/tenant/storage_layer/delta_layer.rs

@@ -609,49 +609,6 @@ impl Drop for DeltaLayerWriter {
     }
 }
 
-impl DeltaLayer {
-    /// Assume the file at `path` is corrupt if this function returns with an error.
-    pub(crate) async fn rewrite_tenant_timeline(
-        path: &Utf8Path,
-        new_tenant: TenantId,
-        new_timeline: TimelineId,
-        ctx: &RequestContext,
-    ) -> anyhow::Result<()> {
-        let file = VirtualFile::open_with_options(
-            path,
-            &*std::fs::OpenOptions::new().read(true).write(true),
-        )
-        .await
-        .with_context(|| format!("Failed to open file '{}'", path))?;
-        let file = FileBlockReader::new(file);
-        let summary_blk = file.read_blk(0, ctx).await?;
-        let actual_summary = Summary::des_prefix(summary_blk.as_ref())?;
-        let mut file = file.file;
-        if actual_summary.magic != DELTA_FILE_MAGIC {
-            bail!("File '{}' is not a delta layer", path);
-        }
-        let new_summary = Summary {
-            tenant_id: new_tenant,
-            timeline_id: new_timeline,
-            ..actual_summary
-        };
-
-        let mut buf = smallvec::SmallVec::<[u8; PAGE_SZ]>::new();
-        Summary::ser_into(&new_summary, &mut buf)?;
-        if buf.spilled() {
-            // The code in ImageLayerWriterInner just warn!()s for this.
-            // It should probably error out as well.
-            anyhow::bail!(
-                "Used more than one page size for summary buffer: {}",
-                buf.len()
-            );
-        }
-        file.seek(SeekFrom::Start(0)).await?;
-        file.write_all(&buf).await?;
-        Ok(())
-    }
-}
-
 impl DeltaLayerInner {
     pub(super) async fn load(
         path: &Utf8Path,
@@ -925,3 +882,15 @@ impl AsRef<DeltaLayerInner> for DeltaLayerInner {
         self
     }
 }
+
+impl<'a> pageserver_compaction::interface::CompactionDeltaEntry<'a, Key> for DeltaEntry<'a> {
+    fn key(&self) -> Key {
+        self.key
+    }
+    fn lsn(&self) -> Lsn {
+        self.lsn
+    }
+    fn size(&self) -> u64 {
+        self.size
+    }
+}

pageserver/src/tenant/storage_layer/image_layer.rs

@@ -294,49 +294,6 @@ impl ImageLayer {
     }
 }
 
-impl ImageLayer {
-    /// Assume the file at `path` is corrupt if this function returns with an error.
-    pub(crate) async fn rewrite_tenant_timeline(
-        path: &Utf8Path,
-        new_tenant: TenantId,
-        new_timeline: TimelineId,
-        ctx: &RequestContext,
-    ) -> anyhow::Result<()> {
-        let file = VirtualFile::open_with_options(
-            path,
-            &*std::fs::OpenOptions::new().read(true).write(true),
-        )
-        .await
-        .with_context(|| format!("Failed to open file '{}'", path))?;
-        let file = FileBlockReader::new(file);
-        let summary_blk = file.read_blk(0, ctx).await?;
-        let actual_summary = Summary::des_prefix(summary_blk.as_ref())?;
-        let mut file = file.file;
-        if actual_summary.magic != IMAGE_FILE_MAGIC {
-            bail!("File '{}' is not a delta layer", path);
-        }
-        let new_summary = Summary {
-            tenant_id: new_tenant,
-            timeline_id: new_timeline,
-            ..actual_summary
-        };
-
-        let mut buf = smallvec::SmallVec::<[u8; PAGE_SZ]>::new();
-        Summary::ser_into(&new_summary, &mut buf)?;
-        if buf.spilled() {
-            // The code in ImageLayerWriterInner just warn!()s for this.
-            // It should probably error out as well.
-            anyhow::bail!(
-                "Used more than one page size for summary buffer: {}",
-                buf.len()
-            );
-        }
-        file.seek(SeekFrom::Start(0)).await?;
-        file.write_all(&buf).await?;
-        Ok(())
-    }
-}
-
 impl ImageLayerInner {
     pub(super) async fn load(
         path: &Utf8Path,

pageserver/src/tenant/storage_layer/inmemory_layer.rs

@@ -345,19 +345,14 @@ impl InMemoryLayer {
 
         let cursor = inner.file.block_cursor();
 
-        // Sort the keys because delta layer writer expects them sorted.
-        //
-        // NOTE: this sort can take up significant time if the layer has millions of
-        //       keys. To speed up all the comparisons we convert the key to i128 and
-        //       keep the value as a reference.
-        let mut keys: Vec<_> = inner.index.iter().map(|(k, m)| (k.to_i128(), m)).collect();
-        keys.sort_unstable_by_key(|k| k.0);
+        let mut keys: Vec<(&Key, &VecMap<Lsn, u64>)> = inner.index.iter().collect();
+        keys.sort_by_key(|k| k.0);
 
         let ctx = RequestContextBuilder::extend(ctx)
             .page_content_kind(PageContentKind::InMemoryLayer)
             .build();
         for (key, vec_map) in keys.iter() {
-            let key = Key::from_i128(*key);
+            let key = **key;
             // Write all page versions
             for (lsn, pos) in vec_map.as_slice() {
                 cursor.read_blob_into_buf(*pos, &mut buf, &ctx).await?;

pageserver/src/tenant/timeline.rs

@@ -1,3 +1,4 @@
+mod compaction;
 pub mod delete;
 mod eviction_task;
 mod init;
@@ -23,7 +24,7 @@ use tokio::{
 };
 use tokio_util::sync::CancellationToken;
 use tracing::*;
-use utils::{id::TenantTimelineId, sync::gate::Gate};
+use utils::id::TenantTimelineId;
 
 use std::cmp::{max, min, Ordering};
 use std::collections::{BinaryHeap, HashMap, HashSet};
@@ -59,7 +60,7 @@ use crate::metrics::{
 use crate::pgdatadir_mapping::LsnForTimestamp;
 use crate::pgdatadir_mapping::{is_rel_fsm_block_key, is_rel_vm_block_key};
 use crate::pgdatadir_mapping::{BlockNumber, CalculateLogicalSizeError};
-use crate::tenant::config::{EvictionPolicy, TenantConfOpt};
+use crate::tenant::config::{CompactionAlgorithm, EvictionPolicy, TenantConfOpt};
 use pageserver_api::reltag::RelTag;
 
 use postgres_connection::PgConnectionConfig;
@@ -310,13 +311,6 @@ pub struct Timeline {
     /// Load or creation time information about the disk_consistent_lsn and when the loading
     /// happened. Used for consumption metrics.
     pub(crate) loaded_at: (Lsn, SystemTime),
-
-    /// Gate to prevent shutdown completing while I/O is still happening to this timeline's data
-    pub(crate) gate: Gate,
-
-    /// Cancellation token scoped to this timeline: anything doing long-running work relating
-    /// to the timeline should drop out when this token fires.
-    pub(crate) cancel: CancellationToken,
 }
 
 pub struct WalReceiverInfo {
@@ -704,6 +698,18 @@ impl Timeline {
             return Ok(());
         }
 
+        match self.get_compaction_algorithm() {
+            CompactionAlgorithm::Tiered => self.compact_tiered(cancel, ctx).await,
+            CompactionAlgorithm::Legacy => self.compact_legacy(cancel, ctx).await,
+        }
+    }
+
+    /// TODO: cancellation
+    async fn compact_legacy(
+        self: &Arc<Self>,
+        _cancel: &CancellationToken,
+        ctx: &RequestContext,
+    ) -> Result<(), CompactionError> {
         // High level strategy for compaction / image creation:
         //
         // 1. First, calculate the desired "partitioning" of the
@@ -793,11 +799,7 @@ impl Timeline {
                 // as an empty timeline. Also in unit tests, when we use the timeline
                 // as a simple key-value store, ignoring the datadir layout. Log the
                 // error but continue.
-                //
-                // Suppress error when it's due to cancellation
-                if !self.cancel.is_cancelled() {
-                    error!("could not compact, repartitioning keyspace failed: {err:?}");
-                }
+                error!("could not compact, repartitioning keyspace failed: {err:?}");
             }
         };
 
@@ -895,12 +897,7 @@ impl Timeline {
     pub async fn shutdown(self: &Arc<Self>, freeze_and_flush: bool) {
         debug_assert_current_span_has_tenant_and_timeline_id();
 
-        // Signal any subscribers to our cancellation token to drop out
-        tracing::debug!("Cancelling CancellationToken");
-        self.cancel.cancel();
-
         // prevent writes to the InMemoryLayer
-        tracing::debug!("Waiting for WalReceiverManager...");
         task_mgr::shutdown_tasks(
             Some(TaskKind::WalReceiverManager),
             Some(self.tenant_id),
@@ -936,16 +933,6 @@ impl Timeline {
                 warn!("failed to await for frozen and flushed uploads: {e:#}");
             }
         }
-
-        // Page request handlers might be waiting for LSN to advance: they do not respect Timeline::cancel
-        // while doing so.
-        self.last_record_lsn.shutdown();
-
-        tracing::debug!("Waiting for tasks...");
-        task_mgr::shutdown_tasks(None, Some(self.tenant_id), Some(self.timeline_id)).await;
-
-        // Finally wait until any gate-holders are complete
-        self.gate.close().await;
     }
 
     pub fn set_state(&self, new_state: TimelineState) {
@@ -1074,11 +1061,6 @@ impl Timeline {
     /// Like [`evict_layer_batch`](Self::evict_layer_batch), but for just one layer.
     /// Additional case `Ok(None)` covers the case where the layer could not be found by its `layer_file_name`.
     pub async fn evict_layer(&self, layer_file_name: &str) -> anyhow::Result<Option<bool>> {
-        let _gate = self
-            .gate
-            .enter()
-            .map_err(|_| anyhow::anyhow!("Shutting down"))?;
-
         let Some(local_layer) = self.find_layer(layer_file_name).await else {
             return Ok(None);
         };
@@ -1094,8 +1076,9 @@ impl Timeline {
             .as_ref()
             .ok_or_else(|| anyhow::anyhow!("remote storage not configured; cannot evict"))?;
 
+        let cancel = CancellationToken::new();
         let results = self
-            .evict_layer_batch(remote_client, &[local_layer])
+            .evict_layer_batch(remote_client, &[local_layer], &cancel)
             .await?;
         assert_eq!(results.len(), 1);
         let result: Option<Result<(), EvictionError>> = results.into_iter().next().unwrap();
@@ -1110,18 +1093,15 @@ impl Timeline {
     pub(crate) async fn evict_layers(
         &self,
         layers_to_evict: &[Layer],
+        cancel: &CancellationToken,
     ) -> anyhow::Result<Vec<Option<Result<(), EvictionError>>>> {
-        let _gate = self
-            .gate
-            .enter()
-            .map_err(|_| anyhow::anyhow!("Shutting down"))?;
-
         let remote_client = self
             .remote_client
             .as_ref()
             .context("timeline must have RemoteTimelineClient")?;
 
-        self.evict_layer_batch(remote_client, layers_to_evict).await
+        self.evict_layer_batch(remote_client, layers_to_evict, cancel)
+            .await
     }
 
     /// Evict multiple layers at once, continuing through errors.
@@ -1142,6 +1122,7 @@ impl Timeline {
         &self,
         remote_client: &Arc<RemoteTimelineClient>,
         layers_to_evict: &[Layer],
+        cancel: &CancellationToken,
     ) -> anyhow::Result<Vec<Option<Result<(), EvictionError>>>> {
         // ensure that the layers have finished uploading
         // (don't hold the layer_removal_cs while we do it, we're not removing anything yet)
@@ -1189,7 +1170,7 @@ impl Timeline {
         };
 
         tokio::select! {
-            _ = self.cancel.cancelled() => {},
+            _ = cancel.cancelled() => {},
             _ = join => {}
         }
 
@@ -1238,6 +1219,13 @@ impl Timeline {
             .unwrap_or(self.conf.default_tenant_conf.image_creation_threshold)
     }
 
+    fn get_compaction_algorithm(&self) -> CompactionAlgorithm {
+        let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
+        tenant_conf
+            .compaction_algorithm
+            .unwrap_or(self.conf.default_tenant_conf.compaction_algorithm)
+    }
+
     fn get_eviction_policy(&self) -> EvictionPolicy {
         let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
         tenant_conf
@@ -1299,7 +1287,6 @@ impl Timeline {
         initial_logical_size_can_start: Option<completion::Barrier>,
         initial_logical_size_attempt: Option<completion::Completion>,
         state: TimelineState,
-        cancel: CancellationToken,
     ) -> Arc<Self> {
         let disk_consistent_lsn = metadata.disk_consistent_lsn();
         let (state, _) = watch::channel(state);
@@ -1400,8 +1387,6 @@ impl Timeline {
 
                 initial_logical_size_can_start,
                 initial_logical_size_attempt: Mutex::new(initial_logical_size_attempt),
-                cancel,
-                gate: Gate::new(format!("Timeline<{tenant_id}/{timeline_id}>")),
             };
             result.repartition_threshold =
                 result.get_checkpoint_distance() / REPARTITION_FREQ_IN_CHECKPOINT_DISTANCE;
@@ -2065,10 +2050,6 @@ impl Timeline {
         let mut cont_lsn = Lsn(request_lsn.0 + 1);
 
         'outer: loop {
-            if self.cancel.is_cancelled() {
-                return Err(PageReconstructError::Cancelled);
-            }
-
             // The function should have updated 'state'
             //info!("CALLED for {} at {}: {:?} with {} records, cached {}", key, cont_lsn, result, reconstruct_state.records.len(), cached_lsn);
             match result {
@@ -2975,10 +2956,13 @@ struct CompactLevel0Phase1StatsBuilder {
     new_deltas_size: Option<u64>,
 }
 
+#[serde_as]
 #[derive(serde::Serialize)]
 struct CompactLevel0Phase1Stats {
     version: u64,
+    #[serde_as(as = "serde_with::DisplayFromStr")]
     tenant_id: TenantId,
+    #[serde_as(as = "serde_with::DisplayFromStr")]
     timeline_id: TimelineId,
     read_lock_acquisition_micros: RecordedDuration,
     read_lock_held_spawn_blocking_startup_micros: RecordedDuration,
@@ -3046,7 +3030,7 @@ impl TryFrom<CompactLevel0Phase1StatsBuilder> for CompactLevel0Phase1Stats {
 }
 
 impl Timeline {
-    /// Level0 files first phase of compaction, explained in the [`Self::compact`] comment.
+    /// Level0 files first phase of compaction, explained in the [`compact_legacy`] comment.
     ///
     /// This method takes the `_layer_removal_cs` guard to highlight it required downloads are
     /// returned as an error. If the `layer_removal_cs` boundary is changed not to be taken in the
@@ -3529,6 +3513,23 @@ impl Timeline {
             return Ok(());
         }
 
+        self.finish_compact_batch(
+            layer_removal_cs,
+            &new_layers,
+            &Vec::new(),
+            &deltas_to_compact,
+        )
+        .await?;
+        Ok(())
+    }
+
+    async fn finish_compact_batch(
+        self: &Arc<Self>,
+        layer_removal_cs: Arc<tokio::sync::OwnedMutexGuard<()>>,
+        new_deltas: &[ResidentLayer],
+        new_images: &[ResidentLayer],
+        layers_to_remove: &[Layer],
+    ) -> anyhow::Result<()> {
         // Before deleting any layers, we need to wait for their upload ops to finish.
         // See remote_timeline_client module level comment on consistency.
         // Do it here because we don't want to hold self.layers.write() while waiting.
@@ -3544,9 +3545,9 @@ impl Timeline {
 
         let mut duplicated_layers = HashSet::new();
 
-        let mut insert_layers = Vec::with_capacity(new_layers.len());
+        let mut insert_layers = Vec::with_capacity(new_deltas.len());
 
-        for l in &new_layers {
+        for l in new_deltas {
             if guard.contains(l.as_ref()) {
                 // expected in tests
                 tracing::error!(layer=%l, "duplicated L1 layer");
@@ -3557,18 +3558,22 @@ impl Timeline {
                 // because we have not implemented L0 => L0 compaction.
                 duplicated_layers.insert(l.layer_desc().key());
             } else if LayerMap::is_l0(l.layer_desc()) {
-                return Err(CompactionError::Other(anyhow!("compaction generates a L0 layer file as output, which will cause infinite compaction.")));
+                bail!("compaction generates a L0 layer file as output, which will cause infinite compaction.");
             } else {
                 insert_layers.push(l.clone());
             }
         }
 
-        let remove_layers = {
-            let mut deltas_to_compact = deltas_to_compact;
-            // only remove those inputs which were not outputs
-            deltas_to_compact.retain(|l| !duplicated_layers.contains(&l.layer_desc().key()));
-            deltas_to_compact
-        };
+        // only remove those inputs which were not outputs
+        let remove_layers: Vec<Layer> = layers_to_remove
+            .iter()
+            .filter(|l| !duplicated_layers.contains(&l.layer_desc().key()))
+            .cloned()
+            .collect();
+
+        if !new_images.is_empty() {
+            guard.track_new_image_layers(new_images, &self.metrics);
+        }
 
         // deletion will happen later, the layer file manager calls garbage_collect_on_drop
         guard.finish_compact_l0(
@@ -3579,7 +3584,7 @@ impl Timeline {
         );
 
         if let Some(remote_client) = self.remote_client.as_ref() {
-            remote_client.schedule_compaction_update(&remove_layers, &new_layers)?;
+            remote_client.schedule_compaction_update(&remove_layers, &new_deltas)?;
         }
 
         drop_wlock(guard);
@@ -4405,10 +4410,25 @@ mod tests {
             .expect("should had been resident")
             .drop_eviction_guard();
 
+        let cancel = tokio_util::sync::CancellationToken::new();
         let batch = [layer];
 
-        let first = async { timeline.evict_layer_batch(&rc, &batch).await.unwrap() };
-        let second = async { timeline.evict_layer_batch(&rc, &batch).await.unwrap() };
+        let first = {
+            let cancel = cancel.child_token();
+            async {
+                let cancel = cancel;
+                timeline
+                    .evict_layer_batch(&rc, &batch, &cancel)
+                    .await
+                    .unwrap()
+            }
+        };
+        let second = async {
+            timeline
+                .evict_layer_batch(&rc, &batch, &cancel)
+                .await
+                .unwrap()
+        };
 
         let (first, second) = tokio::join!(first, second);
 

pageserver/src/tenant/timeline/compaction.rs

@@ -0,0 +1,473 @@
+//! New compaction implementation. The algorithm itself is implemented in the
+//! compaction crate. This file implements the callbacks and structs that allow
+//! the algorithm to drive the process.
+//!
+//! The old legacy algorithm is implemented directly in `timeline.rs`.
+
+use std::ops::Range;
+use std::sync::Arc;
+
+use super::Timeline;
+
+use async_trait::async_trait;
+use fail::fail_point;
+use tokio_util::sync::CancellationToken;
+use tracing::{debug, trace, warn};
+
+use crate::context::RequestContext;
+use crate::tenant::storage_layer::{AsLayerDesc, PersistentLayerDesc};
+use crate::tenant::timeline::{is_rel_fsm_block_key, is_rel_vm_block_key};
+use crate::tenant::timeline::{DeltaLayerWriter, ImageLayerWriter};
+use crate::tenant::timeline::{Layer, ResidentLayer};
+use crate::tenant::DeltaLayer;
+use crate::tenant::PageReconstructError;
+use crate::ZERO_PAGE;
+
+use crate::keyspace::KeySpace;
+use crate::repository::Key;
+
+use utils::lsn::Lsn;
+
+use pageserver_compaction::helpers::overlaps_with;
+use pageserver_compaction::interface::*;
+
+use super::CompactionError;
+
+impl Timeline {
+    /// Entry point for new tiered compaction algorithm.
+    ///
+    /// All the real work is in the implementation in the pageserver_compaction
+    /// crate. The code here would apply to any algorithm implemented by the
+    /// same interface, but tiered is the only one at the moment.
+    ///
+    /// TODO: cancellation
+    pub(crate) async fn compact_tiered(
+        self: &Arc<Self>,
+        _cancel: &CancellationToken,
+        ctx: &RequestContext,
+    ) -> Result<(), CompactionError> {
+        let fanout = self.get_compaction_threshold() as u64;
+        let target_file_size = self.get_checkpoint_distance();
+
+        // Find the top of the historical layers
+        let end_lsn = {
+            let guard = self.layers.read().await;
+            let layers = guard.layer_map();
+
+            let l0_deltas = layers.get_level0_deltas()?;
+            drop(guard);
+
+            // As an optimization, if we find that there are too few L0 layers,
+            // bail out early. We know that the compaction algorithm would do
+            // nothing in that case.
+            if l0_deltas.len() < fanout as usize {
+                // doesn't need compacting
+                return Ok(());
+            }
+            l0_deltas.iter().map(|l| l.lsn_range.end).max().unwrap()
+        };
+
+        // now lock out layer removal (compaction, gc, timeline deletion)
+        let layer_removal_cs = Arc::new(self.layer_removal_cs.clone().lock_owned().await);
+        // Is the timeline being deleted?
+        if self.is_stopping() {
+            trace!("Dropping out of compaction on timeline shutdown");
+            return Err(CompactionError::ShuttingDown);
+        }
+
+        let keyspace = self.collect_keyspace(end_lsn, ctx).await?;
+        let mut adaptor = TimelineAdaptor::new(self, layer_removal_cs, (end_lsn, keyspace));
+        let ctx_adaptor = RequestContextAdaptor(ctx.clone());
+
+        pageserver_compaction::compact_tiered::compact_tiered(
+            &mut adaptor,
+            end_lsn,
+            target_file_size,
+            fanout,
+            &ctx_adaptor,
+        )
+        .await?;
+
+        adaptor.flush_updates().await?;
+        Ok(())
+    }
+}
+
+struct TimelineAdaptor {
+    timeline: Arc<Timeline>,
+    layer_removal_cs: Arc<tokio::sync::OwnedMutexGuard<()>>,
+
+    keyspace: (Lsn, KeySpace),
+
+    new_deltas: Vec<ResidentLayer>,
+    new_images: Vec<ResidentLayer>,
+    layers_to_delete: Vec<Arc<PersistentLayerDesc>>,
+}
+
+impl TimelineAdaptor {
+    pub fn new(
+        timeline: &Arc<Timeline>,
+        layer_removal_cs: Arc<tokio::sync::OwnedMutexGuard<()>>,
+        keyspace: (Lsn, KeySpace),
+    ) -> Self {
+        Self {
+            timeline: timeline.clone(),
+            layer_removal_cs,
+            keyspace,
+            new_images: Vec::new(),
+            new_deltas: Vec::new(),
+            layers_to_delete: Vec::new(),
+        }
+    }
+
+    pub async fn flush_updates(&mut self) -> anyhow::Result<()> {
+        let layers_to_delete = {
+            let guard = self.timeline.layers.read().await;
+            self.layers_to_delete
+                .iter()
+                .map(|x| guard.get_from_desc(x))
+                .collect::<Vec<Layer>>()
+        };
+        self.timeline
+            .finish_compact_batch(
+                self.layer_removal_cs.clone(),
+                &self.new_deltas,
+                &self.new_images,
+                &layers_to_delete,
+            )
+            .await?;
+        self.new_images.clear();
+        self.new_deltas.clear();
+        self.layers_to_delete.clear();
+        Ok(())
+    }
+}
+
+#[derive(Clone)]
+struct ResidentDeltaLayer(ResidentLayer);
+#[derive(Clone)]
+struct ResidentImageLayer(ResidentLayer);
+
+#[async_trait]
+impl CompactionJobExecutor for TimelineAdaptor {
+    type Key = crate::repository::Key;
+
+    type Layer = Arc<PersistentLayerDesc>;
+    type DeltaLayer = ResidentDeltaLayer;
+    type ImageLayer = ResidentImageLayer;
+
+    type RequestContext = RequestContextAdaptor;
+
+    async fn get_layers(
+        &mut self,
+        key_range: &Range<Key>,
+        lsn_range: &Range<Lsn>,
+        _ctx: &RequestContextAdaptor,
+    ) -> anyhow::Result<Vec<Arc<PersistentLayerDesc>>> {
+        self.flush_updates().await?;
+
+        let guard = self.timeline.layers.read().await;
+        let layer_map = guard.layer_map();
+
+        let result = layer_map
+            .iter_historic_layers()
+            .filter(|l| {
+                overlaps_with(&l.lsn_range, lsn_range) && overlaps_with(&l.key_range, key_range)
+            })
+            .collect();
+        Ok(result)
+    }
+
+    async fn get_keyspace(
+        &mut self,
+        key_range: &Range<Key>,
+        lsn: Lsn,
+        _ctx: &RequestContextAdaptor,
+    ) -> anyhow::Result<Vec<Range<Key>>> {
+        if lsn == self.keyspace.0 {
+            Ok(pageserver_compaction::helpers::intersect_keyspace(
+                &self.keyspace.1.ranges,
+                key_range,
+            ))
+        } else {
+            // The current compaction implementatin only ever requests the key space
+            // at the compaction end LSN.
+            anyhow::bail!("keyspace not available for requested lsn");
+        }
+    }
+
+    async fn downcast_delta_layer(
+        &self,
+        layer: &Arc<PersistentLayerDesc>,
+    ) -> anyhow::Result<Option<ResidentDeltaLayer>> {
+        // this is a lot more complex than a simple downcast...
+        if layer.is_delta() {
+            let l = {
+                let guard = self.timeline.layers.read().await;
+                guard.get_from_desc(layer)
+            };
+            let result = l.download_and_keep_resident().await?;
+
+            Ok(Some(ResidentDeltaLayer(result)))
+        } else {
+            Ok(None)
+        }
+    }
+
+    async fn create_image(
+        &mut self,
+        lsn: Lsn,
+        key_range: &Range<Key>,
+        ctx: &RequestContextAdaptor,
+    ) -> anyhow::Result<()> {
+        Ok(self.create_image_impl(lsn, key_range, ctx).await?)
+    }
+
+    async fn create_delta(
+        &mut self,
+        lsn_range: &Range<Lsn>,
+        key_range: &Range<Key>,
+        input_layers: &[ResidentDeltaLayer],
+        ctx: &RequestContextAdaptor,
+    ) -> anyhow::Result<()> {
+        debug!("Create new layer {}..{}", lsn_range.start, lsn_range.end);
+
+        let mut all_entries = Vec::new();
+        for dl in input_layers.iter() {
+            all_entries.extend(dl.load_keys(ctx).await?);
+        }
+
+        // The current stdlib sorting implementation is designed in a way where it is
+        // particularly fast where the slice is made up of sorted sub-ranges.
+        all_entries.sort_by_key(|DeltaEntry { key, lsn, .. }| (*key, *lsn));
+
+        let mut writer = DeltaLayerWriter::new(
+            self.timeline.conf,
+            self.timeline.timeline_id,
+            self.timeline.tenant_id,
+            key_range.start,
+            lsn_range.clone(),
+        )
+        .await?;
+
+        let mut dup_values = 0;
+
+        // This iterator walks through all key-value pairs from all the layers
+        // we're compacting, in key, LSN order.
+        let mut prev: Option<(Key, Lsn)> = None;
+        for &DeltaEntry {
+            key, lsn, ref val, ..
+        } in all_entries.iter()
+        {
+            if prev == Some((key, lsn)) {
+                // This is a duplicate. Skip it.
+                //
+                // It can happen if compaction is interrupted after writing some
+                // layers but not all, and we are compacting the range again.
+                // The calculations in the algorithm assume that there are no
+                // duplicates, so the math on targeted file size is likely off,
+                // and we will create smaller files than expected.
+                dup_values += 1;
+                continue;
+            }
+
+            let value = val.load(ctx).await?;
+
+            writer.put_value(key, lsn, value).await?;
+
+            prev = Some((key, lsn));
+        }
+
+        if dup_values > 0 {
+            warn!("delta layer created with {} duplicate values", dup_values);
+        }
+
+        fail_point!("delta-layer-writer-fail-before-finish", |_| {
+            Err(anyhow::anyhow!(
+                "failpoint delta-layer-writer-fail-before-finish"
+            ))
+        });
+
+        let new_delta_layer = writer
+            .finish(prev.unwrap().0.next(), &self.timeline)
+            .await?;
+
+        self.new_deltas.push(new_delta_layer);
+        Ok(())
+    }
+
+    async fn delete_layer(
+        &mut self,
+        layer: &Arc<PersistentLayerDesc>,
+        _ctx: &RequestContextAdaptor,
+    ) -> anyhow::Result<()> {
+        self.layers_to_delete.push(layer.clone());
+        Ok(())
+    }
+}
+
+impl TimelineAdaptor {
+    async fn create_image_impl(
+        &mut self,
+        lsn: Lsn,
+        key_range: &Range<Key>,
+        ctx: &RequestContextAdaptor,
+    ) -> Result<(), PageReconstructError> {
+        let timer = self.timeline.metrics.create_images_time_histo.start_timer();
+
+        let mut image_layer_writer = ImageLayerWriter::new(
+            self.timeline.conf,
+            self.timeline.timeline_id,
+            self.timeline.tenant_id,
+            key_range,
+            lsn,
+        )
+        .await?;
+
+        fail_point!("image-layer-writer-fail-before-finish", |_| {
+            Err(PageReconstructError::Other(anyhow::anyhow!(
+                "failpoint image-layer-writer-fail-before-finish"
+            )))
+        });
+        let keyspace_ranges = self.get_keyspace(key_range, lsn, ctx).await?;
+        for range in &keyspace_ranges {
+            let mut key = range.start;
+            while key < range.end {
+                let img = match self.timeline.get(key, lsn, ctx).await {
+                    Ok(img) => img,
+                    Err(err) => {
+                        // If we fail to reconstruct a VM or FSM page, we can zero the
+                        // page without losing any actual user data. That seems better
+                        // than failing repeatedly and getting stuck.
+                        //
+                        // We had a bug at one point, where we truncated the FSM and VM
+                        // in the pageserver, but the Postgres didn't know about that
+                        // and continued to generate incremental WAL records for pages
+                        // that didn't exist in the pageserver. Trying to replay those
+                        // WAL records failed to find the previous image of the page.
+                        // This special case allows us to recover from that situation.
+                        // See https://github.com/neondatabase/neon/issues/2601.
+                        //
+                        // Unfortunately we cannot do this for the main fork, or for
+                        // any metadata keys, keys, as that would lead to actual data
+                        // loss.
+                        if is_rel_fsm_block_key(key) || is_rel_vm_block_key(key) {
+                            warn!("could not reconstruct FSM or VM key {key}, filling with zeros: {err:?}");
+                            ZERO_PAGE.clone()
+                        } else {
+                            return Err(err);
+                        }
+                    }
+                };
+                image_layer_writer.put_image(key, &img).await?;
+                key = key.next();
+            }
+        }
+        let image_layer = image_layer_writer.finish(&self.timeline).await?;
+
+        self.new_images.push(image_layer);
+
+        timer.stop_and_record();
+
+        Ok(())
+    }
+}
+
+pub struct RequestContextAdaptor(pub RequestContext);
+
+impl std::ops::Deref for RequestContextAdaptor {
+    type Target = RequestContext;
+
+    fn deref(&self) -> &Self::Target {
+        &self.0
+    }
+}
+
+impl CompactionRequestContext for RequestContextAdaptor {}
+
+impl CompactionLayer<Key> for Arc<PersistentLayerDesc> {
+    fn key_range(&self) -> &Range<Key> {
+        &self.key_range
+    }
+    fn lsn_range(&self) -> &Range<Lsn> {
+        &self.lsn_range
+    }
+    fn file_size(&self) -> u64 {
+        self.file_size
+    }
+    fn short_id(&self) -> std::string::String {
+        self.as_ref().short_id().to_string()
+    }
+    fn is_delta(&self) -> bool {
+        self.as_ref().is_delta()
+    }
+}
+
+impl CompactionLayer<Key> for Arc<DeltaLayer> {
+    fn key_range(&self) -> &Range<Key> {
+        &self.layer_desc().key_range
+    }
+    fn lsn_range(&self) -> &Range<Lsn> {
+        &self.layer_desc().lsn_range
+    }
+    fn file_size(&self) -> u64 {
+        self.layer_desc().file_size
+    }
+    fn short_id(&self) -> std::string::String {
+        self.layer_desc().short_id().to_string()
+    }
+    fn is_delta(&self) -> bool {
+        true
+    }
+}
+
+use crate::tenant::timeline::DeltaEntry;
+
+impl CompactionLayer<Key> for ResidentDeltaLayer {
+    fn key_range(&self) -> &Range<Key> {
+        &self.0.layer_desc().key_range
+    }
+    fn lsn_range(&self) -> &Range<Lsn> {
+        &self.0.layer_desc().lsn_range
+    }
+    fn file_size(&self) -> u64 {
+        self.0.layer_desc().file_size
+    }
+    fn short_id(&self) -> std::string::String {
+        self.0.layer_desc().short_id().to_string()
+    }
+    fn is_delta(&self) -> bool {
+        true
+    }
+}
+
+#[async_trait]
+impl CompactionDeltaLayer<TimelineAdaptor> for ResidentDeltaLayer {
+    type DeltaEntry<'a> = DeltaEntry<'a>;
+
+    async fn load_keys<'a>(
+        &self,
+        ctx: &RequestContextAdaptor,
+    ) -> anyhow::Result<Vec<DeltaEntry<'_>>> {
+        self.0.load_keys(ctx).await
+    }
+}
+
+impl CompactionLayer<Key> for ResidentImageLayer {
+    fn key_range(&self) -> &Range<Key> {
+        &self.0.layer_desc().key_range
+    }
+    fn lsn_range(&self) -> &Range<Lsn> {
+        &self.0.layer_desc().lsn_range
+    }
+    fn file_size(&self) -> u64 {
+        self.0.layer_desc().file_size
+    }
+    fn short_id(&self) -> std::string::String {
+        self.0.layer_desc().short_id().to_string()
+    }
+    fn is_delta(&self) -> bool {
+        false
+    }
+}
+impl CompactionImageLayer<TimelineAdaptor> for ResidentImageLayer {}

pageserver/src/tenant/timeline/delete.rs

@@ -17,7 +17,6 @@ use crate::{
     deletion_queue::DeletionQueueClient,
     task_mgr::{self, TaskKind},
     tenant::{
-        debug_assert_current_span_has_tenant_and_timeline_id,
         metadata::TimelineMetadata,
         remote_timeline_client::{
             self, PersistIndexPartWithDeletedFlagError, RemoteTimelineClient,
@@ -31,11 +30,6 @@ use super::{Timeline, TimelineResources};
 
 /// Now that the Timeline is in Stopping state, request all the related tasks to shut down.
 async fn stop_tasks(timeline: &Timeline) -> Result<(), DeleteTimelineError> {
-    debug_assert_current_span_has_tenant_and_timeline_id();
-    // Notify any timeline work to drop out of loops/requests
-    tracing::debug!("Cancelling CancellationToken");
-    timeline.cancel.cancel();
-
     // Stop the walreceiver first.
     debug!("waiting for wal receiver to shutdown");
     let maybe_started_walreceiver = { timeline.walreceiver.lock().unwrap().take() };
@@ -80,11 +74,6 @@ async fn stop_tasks(timeline: &Timeline) -> Result<(), DeleteTimelineError> {
             "failpoint: timeline-delete-before-index-deleted-at"
         ))?
     });
-
-    tracing::debug!("Waiting for gate...");
-    timeline.gate.close().await;
-    tracing::debug!("Shutdown complete");
-
     Ok(())
 }
 

pageserver/src/tenant/timeline/eviction_task.rs

@@ -277,7 +277,10 @@ impl Timeline {
             Some(c) => c,
         };
 
-        let results = match self.evict_layer_batch(remote_client, &candidates).await {
+        let results = match self
+            .evict_layer_batch(remote_client, &candidates, cancel)
+            .await
+        {
             Err(pre_err) => {
                 stats.errors += candidates.len();
                 error!("could not do any evictions: {pre_err:#}");
@@ -341,7 +344,20 @@ 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 tenant = match crate::tenant::mgr::get_tenant(self.tenant_id, true) {
+        //
+        // It is critical we are responsive to cancellation here. Otherwise, we deadlock with
+        // tenant deletion (holds TENANTS in read mode) any other task that attempts to
+        // acquire TENANTS in write mode before we here call get_tenant.
+        // See https://github.com/neondatabase/neon/issues/5284.
+        let res = tokio::select! {
+            _ = cancel.cancelled() => {
+                return ControlFlow::Break(());
+            }
+            res = crate::tenant::mgr::get_tenant(self.tenant_id, true) => {
+                res
+            }
+        };
+        let tenant = match res {
             Ok(t) => t,
             Err(_) => {
                 return ControlFlow::Break(());

pageserver/src/tenant/timeline/walreceiver/connection_manager.rs

@@ -426,7 +426,7 @@ impl ConnectionManagerState {
                     timeline,
                     new_sk.wal_source_connconf,
                     events_sender,
-                    cancellation.clone(),
+                    cancellation,
                     connect_timeout,
                     ctx,
                     node_id,
@@ -447,14 +447,7 @@ impl ConnectionManagerState {
                             }
                             WalReceiverError::Other(e) => {
                                 // give out an error to have task_mgr give it a really verbose logging
-                                if cancellation.is_cancelled() {
-                                    // Ideally we would learn about this via some path other than Other, but
-                                    // that requires refactoring all the intermediate layers of ingest code
-                                    // that only emit anyhow::Error
-                                    Ok(())
-                                } else {
-                                    Err(e).context("walreceiver connection handling failure")
-                                }
+                                Err(e).context("walreceiver connection handling failure")
                             }
                         }
                     }

pageserver/src/walredo.rs

@@ -596,21 +596,21 @@ trait CloseFileDescriptors: CommandExt {
 
 impl<C: CommandExt> CloseFileDescriptors for C {
     fn close_fds(&mut self) -> &mut Command {
-        // SAFETY: Code executed inside pre_exec should have async-signal-safety,
-        // which means it should be safe to execute inside a signal handler.
-        // The precise meaning depends on platform. See `man signal-safety`
-        // for the linux definition.
-        //
-        // The set_fds_cloexec_threadsafe function is documented to be
-        // async-signal-safe.
-        //
-        // Aside from this function, the rest of the code is re-entrant and
-        // doesn't make any syscalls. We're just passing constants.
-        //
-        // NOTE: It's easy to indirectly cause a malloc or lock a mutex,
-        // which is not async-signal-safe. Be careful.
         unsafe {
             self.pre_exec(move || {
+                // SAFETY: Code executed inside pre_exec should have async-signal-safety,
+                // which means it should be safe to execute inside a signal handler.
+                // The precise meaning depends on platform. See `man signal-safety`
+                // for the linux definition.
+                //
+                // The set_fds_cloexec_threadsafe function is documented to be
+                // async-signal-safe.
+                //
+                // Aside from this function, the rest of the code is re-entrant and
+                // doesn't make any syscalls. We're just passing constants.
+                //
+                // NOTE: It's easy to indirectly cause a malloc or lock a mutex,
+                // which is not async-signal-safe. Be careful.
                 close_fds::set_fds_cloexec_threadsafe(3, &[]);
                 Ok(())
             })

pgxn/neon/libpagestore.c

@@ -19,10 +19,7 @@
 #include "access/xlog.h"
 #include "access/xlogutils.h"
 #include "storage/buf_internals.h"
-#include "storage/lwlock.h"
-#include "storage/ipc.h"
 #include "c.h"
-#include "postmaster/interrupt.h"
 
 #include "libpq-fe.h"
 #include "libpq/pqformat.h"
@@ -64,63 +61,23 @@ int			flush_every_n_requests = 8;
 int			n_reconnect_attempts = 0;
 int			max_reconnect_attempts = 60;
 
-#define MAX_PAGESERVER_CONNSTRING_SIZE 256
-
-typedef struct
-{
-    LWLockId lock;
-    pg_atomic_uint64 update_counter;
-    char pageserver_connstring[MAX_PAGESERVER_CONNSTRING_SIZE];
-} PagestoreShmemState;
-
-#if PG_VERSION_NUM >= 150000
-static shmem_request_hook_type prev_shmem_request_hook = NULL;
-static void walproposer_shmem_request(void);
-#endif
-static shmem_startup_hook_type prev_shmem_startup_hook;
-static PagestoreShmemState *pagestore_shared;
-static uint64 pagestore_local_counter = 0;
-static char local_pageserver_connstring[MAX_PAGESERVER_CONNSTRING_SIZE];
-
 bool	(*old_redo_read_buffer_filter) (XLogReaderState *record, uint8 block_id) = NULL;
 
 static bool pageserver_flush(void);
 static void pageserver_disconnect(void);
 
-static bool
-CheckPageserverConnstring(char **newval, void **extra, GucSource source)
-{
-    return strlen(*newval) < MAX_PAGESERVER_CONNSTRING_SIZE;
-}
+
+static pqsigfunc	 prev_signal_handler;
 
 static void
-AssignPageserverConnstring(const char *newval, void *extra)
+pageserver_sighup_handler(SIGNAL_ARGS)
 {
-    if(!pagestore_shared)
-        return;
-    LWLockAcquire(pagestore_shared->lock, LW_EXCLUSIVE);
-    strlcpy(pagestore_shared->pageserver_connstring, newval, MAX_PAGESERVER_CONNSTRING_SIZE);
-    pg_atomic_fetch_add_u64(&pagestore_shared->update_counter, 1);
-    LWLockRelease(pagestore_shared->lock);
-}
-
-static bool
-CheckConnstringUpdated()
-{
-    if(!pagestore_shared)
-        return false;
-    return pagestore_local_counter < pg_atomic_read_u64(&pagestore_shared->update_counter);
-}
-
-static void
-ReloadConnstring()
-{
-    if(!pagestore_shared)
-        return;
-    LWLockAcquire(pagestore_shared->lock, LW_SHARED);
-    strlcpy(local_pageserver_connstring, pagestore_shared->pageserver_connstring, sizeof(local_pageserver_connstring));
-    pagestore_local_counter = pg_atomic_read_u64(&pagestore_shared->update_counter);
-    LWLockRelease(pagestore_shared->lock);
+	if (prev_signal_handler)
+	{
+        	prev_signal_handler(postgres_signal_arg);
+	}
+	neon_log(LOG, "Received SIGHUP, disconnecting pageserver. New pageserver connstring is %s", page_server_connstring);
+	pageserver_disconnect();
 }
 
 static bool
@@ -134,11 +91,6 @@ pageserver_connect(int elevel)
 
 	Assert(!connected);
 
-        if(CheckConnstringUpdated())
-        {
-            ReloadConnstring();
-        }
-
 	/*
 	 * Connect using the connection string we got from the
 	 * neon.pageserver_connstring GUC. If the NEON_AUTH_TOKEN environment
@@ -158,7 +110,7 @@ pageserver_connect(int elevel)
 		n++;
 	}
 	keywords[n] = "dbname";
-	values[n] = local_pageserver_connstring;
+	values[n] = page_server_connstring;
 	n++;
 	keywords[n] = NULL;
 	values[n] = NULL;
@@ -302,12 +254,6 @@ pageserver_send(NeonRequest * request)
 {
 	StringInfoData req_buff;
 
-        if(CheckConnstringUpdated())
-        {
-            pageserver_disconnect();
-            ReloadConnstring();
-        }
-
 	/* If the connection was lost for some reason, reconnect */
 	if (connected && PQstatus(pageserver_conn) == CONNECTION_BAD)
 	{
@@ -328,7 +274,6 @@ pageserver_send(NeonRequest * request)
 	{
 		while (!pageserver_connect(n_reconnect_attempts < max_reconnect_attempts ? LOG : ERROR))
 		{
-			HandleMainLoopInterrupts();
 			n_reconnect_attempts += 1;
 			pg_usleep(RECONNECT_INTERVAL_USEC);
 		}
@@ -446,8 +391,7 @@ pageserver_flush(void)
 	return true;
 }
 
-page_server_api api =
-{
+page_server_api api = {
 	.send = pageserver_send,
 	.flush = pageserver_flush,
 	.receive = pageserver_receive
@@ -461,72 +405,12 @@ check_neon_id(char **newval, void **extra, GucSource source)
 	return **newval == '\0' || HexDecodeString(id, *newval, 16);
 }
 
-static Size
-PagestoreShmemSize(void)
-{
-    return sizeof(PagestoreShmemState);
-}
-
-static bool
-PagestoreShmemInit(void)
-{
-    bool found;
-    LWLockAcquire(AddinShmemInitLock, LW_EXCLUSIVE);
-    pagestore_shared = ShmemInitStruct("libpagestore shared state",
-                                       PagestoreShmemSize(),
-                                       &found);
-    if(!found)
-    {
-        pagestore_shared->lock = &(GetNamedLWLockTranche("neon_libpagestore")->lock);
-        pg_atomic_init_u64(&pagestore_shared->update_counter, 0);
-        AssignPageserverConnstring(page_server_connstring, NULL);
-    }
-    LWLockRelease(AddinShmemInitLock);
-    return found;
-}
-
-static void
-pagestore_shmem_startup_hook(void)
-{
-    if(prev_shmem_startup_hook)
-        prev_shmem_startup_hook();
-
-    PagestoreShmemInit();
-}
-
-static void
-pagestore_shmem_request(void)
-{
-#if PG_VERSION_NUM >= 150000
-    if(prev_shmem_request_hook)
-        prev_shmem_request_hook();
-#endif
-
-    RequestAddinShmemSpace(PagestoreShmemSize());
-    RequestNamedLWLockTranche("neon_libpagestore", 1);
-}
-
-static void
-pagestore_prepare_shmem(void)
-{
-#if PG_VERSION_NUM >= 150000
-	prev_shmem_request_hook = shmem_request_hook;
-	shmem_request_hook = pagestore_shmem_request;
-#else
-        pagestore_shmem_request();
-#endif
-	prev_shmem_startup_hook = shmem_startup_hook;
-	shmem_startup_hook = pagestore_shmem_startup_hook;
-}
-
 /*
  * Module initialization function
  */
 void
 pg_init_libpagestore(void)
 {
-        pagestore_prepare_shmem();
-
 	DefineCustomStringVariable("neon.pageserver_connstring",
 							   "connection string to the page server",
 							   NULL,
@@ -534,7 +418,7 @@ pg_init_libpagestore(void)
 							   "",
 							   PGC_SIGHUP,
 							   0,	/* no flags required */
-							   CheckPageserverConnstring, AssignPageserverConnstring, NULL);
+							   NULL, NULL, NULL);
 
 	DefineCustomStringVariable("neon.timeline_id",
 							   "Neon timeline_id the server is running on",
@@ -615,5 +499,7 @@ pg_init_libpagestore(void)
 		redo_read_buffer_filter = neon_redo_read_buffer_filter;
 	}
 
+        prev_signal_handler = pqsignal(SIGHUP, pageserver_sighup_handler);
+
 	lfc_init();
 }

pgxn/neon/pagestore_client.h

@@ -33,7 +33,6 @@ typedef enum
 	T_NeonNblocksRequest,
 	T_NeonGetPageRequest,
 	T_NeonDbSizeRequest,
-	T_NeonPagestoreNoopRequest,
 
 	/* pagestore -> pagestore_client */
 	T_NeonExistsResponse = 100,
@@ -41,7 +40,6 @@ typedef enum
 	T_NeonGetPageResponse,
 	T_NeonErrorResponse,
 	T_NeonDbSizeResponse,
-	T_NeonPagestoreNoopResponse,
 }			NeonMessageTag;
 
 /* base struct for c-style inheritance */

proxy/src/auth/credentials.rs

@@ -1,8 +1,6 @@
 //! User credentials used in authentication.
 
-use crate::{
-    auth::password_hack::parse_endpoint_param, error::UserFacingError, proxy::neon_options,
-};
+use crate::{auth::password_hack::parse_endpoint_param, error::UserFacingError};
 use itertools::Itertools;
 use pq_proto::StartupMessageParams;
 use std::collections::HashSet;
@@ -40,8 +38,6 @@ pub struct ClientCredentials<'a> {
     pub user: &'a str,
     // TODO: this is a severe misnomer! We should think of a new name ASAP.
     pub project: Option<String>,
-
-    pub cache_key: String,
 }
 
 impl ClientCredentials<'_> {
@@ -57,7 +53,6 @@ impl<'a> ClientCredentials<'a> {
         ClientCredentials {
             user: "",
             project: None,
-            cache_key: "".to_string(),
         }
     }
 
@@ -125,17 +120,7 @@ impl<'a> ClientCredentials<'a> {
 
         info!(user, project = project.as_deref(), "credentials");
 
-        let cache_key = format!(
-            "{}{}",
-            project.as_deref().unwrap_or(""),
-            neon_options(params).unwrap_or("".to_string())
-        );
-
-        Ok(Self {
-            user,
-            project,
-            cache_key,
-        })
+        Ok(Self { user, project })
     }
 }
 
@@ -191,7 +176,6 @@ mod tests {
         let creds = ClientCredentials::parse(&options, sni, common_names)?;
         assert_eq!(creds.user, "john_doe");
         assert_eq!(creds.project.as_deref(), Some("foo"));
-        assert_eq!(creds.cache_key, "foo");
 
         Ok(())
     }
@@ -319,23 +303,4 @@ mod tests {
             _ => panic!("bad error: {err:?}"),
         }
     }
-
-    #[test]
-    fn parse_neon_options() -> anyhow::Result<()> {
-        let options = StartupMessageParams::new([
-            ("user", "john_doe"),
-            ("options", "neon_lsn:0/2 neon_endpoint_type:read_write"),
-        ]);
-
-        let sni = Some("project.localhost");
-        let common_names = Some(["localhost".into()].into());
-        let creds = ClientCredentials::parse(&options, sni, common_names)?;
-        assert_eq!(creds.project.as_deref(), Some("project"));
-        assert_eq!(
-            creds.cache_key,
-            "projectneon_endpoint_type:read_write neon_lsn:0/2"
-        );
-
-        Ok(())
-    }
 }

proxy/src/compute.rs

@@ -3,7 +3,6 @@ use crate::{
     cancellation::CancelClosure,
     console::errors::WakeComputeError,
     error::{io_error, UserFacingError},
-    proxy::is_neon_param,
 };
 use futures::{FutureExt, TryFutureExt};
 use itertools::Itertools;
@@ -279,7 +278,7 @@ fn filtered_options(params: &StartupMessageParams) -> Option<String> {
     #[allow(unstable_name_collisions)]
     let options: String = params
         .options_raw()?
-        .filter(|opt| parse_endpoint_param(opt).is_none() && !is_neon_param(opt))
+        .filter(|opt| parse_endpoint_param(opt).is_none())
         .intersperse(" ") // TODO: use impl from std once it's stabilized
         .collect();
 
@@ -314,11 +313,5 @@ mod tests {
 
         let params = StartupMessageParams::new([("options", "project = foo")]);
         assert_eq!(filtered_options(&params).as_deref(), Some("project = foo"));
-
-        let params = StartupMessageParams::new([(
-            "options",
-            "project = foo neon_endpoint_type:read_write   neon_lsn:0/2",
-        )]);
-        assert_eq!(filtered_options(&params).as_deref(), Some("project = foo"));
     }
 }

proxy/src/console/provider.rs

@@ -178,7 +178,6 @@ pub struct ConsoleReqExtra<'a> {
     pub session_id: uuid::Uuid,
     /// Name of client application, if set.
     pub application_name: Option<&'a str>,
-    pub options: Option<&'a str>,
 }
 
 /// Auth secret which is managed by the cloud.

proxy/src/console/provider/neon.rs

@@ -99,7 +99,6 @@ impl Api {
                 .query(&[
                     ("application_name", extra.application_name),
                     ("project", Some(project)),
-                    ("options", extra.options),
                 ])
                 .build()?;
 
@@ -152,7 +151,7 @@ impl super::Api for Api {
         extra: &ConsoleReqExtra<'_>,
         creds: &ClientCredentials,
     ) -> Result<CachedNodeInfo, WakeComputeError> {
-        let key: &str = &creds.cache_key;
+        let key = creds.project().expect("impossible");
 
         // Every time we do a wakeup http request, the compute node will stay up
         // for some time (highly depends on the console's scale-to-zero policy);