Bump Postgres versions

part of https://github.com/neondatabase/neon/issues/8140

The blob writer path now uses `maybe_fatal_err`

Signed-off-by: Alex Chi Z <chi@neon.tech>

.github/actions/run-python-test-set/action.yml

@@ -43,7 +43,7 @@ inputs:
   pg_version:
     description: 'Postgres version to use for tests'
     required: false
-    default: 'v14'
+    default: 'v16'
   benchmark_durations:
     description: 'benchmark durations JSON'
     required: false

.github/workflows/benchmarking.yml

@@ -222,13 +222,20 @@ jobs:
       id: create-allure-report
       if: ${{ !cancelled() }}
       uses: ./.github/actions/allure-report-generate
+      with:
+        store-test-results-into-db: true
+      env:
+        REGRESS_TEST_RESULT_CONNSTR_NEW: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}
 
     - name: Post to a Slack channel
       if: ${{ github.event.schedule && failure() }}
       uses: slackapi/slack-github-action@v1
       with:
-        channel-id: "C033QLM5P7D" # dev-staging-stream
-        slack-message: "Periodic replication testing: ${{ job.status }}\n${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+        channel-id: "C06T9AMNDQQ" # on-call-compute-staging-stream
+        slack-message: |
+          Periodic replication testing: ${{ job.status }}
+          <${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}|GitHub Run>
+          <${{ steps.create-allure-report.outputs.report-url }}|Allure report>
       env:
         SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
 
@@ -330,7 +337,7 @@ jobs:
   prepare_AWS_RDS_databases:
     uses: ./.github/workflows/_benchmarking_preparation.yml
     secrets: inherit
-  
+
   pgbench-compare:
     if: ${{ github.event.inputs.run_only_pgvector_tests == 'false' || github.event.inputs.run_only_pgvector_tests == null }}
     needs: [ generate-matrices, prepare_AWS_RDS_databases ]

.github/workflows/build_and_test.yml

@@ -280,6 +280,7 @@ jobs:
           save_perf_report: ${{ github.ref_name == 'main' }}
           extra_params: --splits 5 --group ${{ matrix.pytest_split_group }}
           benchmark_durations: ${{ needs.get-benchmarks-durations.outputs.json }}
+          pg_version: v16
         env:
           VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
           PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
@@ -985,10 +986,10 @@ jobs:
           GH_TOKEN: ${{ secrets.CI_ACCESS_TOKEN }}
         run: |
           if [[ "$GITHUB_REF_NAME" == "main" ]]; then
-            gh workflow --repo neondatabase/aws run deploy-dev.yml --ref main -f branch=main -f dockerTag=${{needs.tag.outputs.build-tag}} -f deployPreprodRegion=false
+            gh workflow --repo neondatabase/infra run deploy-dev.yml --ref main -f branch=main -f dockerTag=${{needs.tag.outputs.build-tag}} -f deployPreprodRegion=false
             gh workflow --repo neondatabase/azure run deploy.yml -f dockerTag=${{needs.tag.outputs.build-tag}}
           elif [[ "$GITHUB_REF_NAME" == "release" ]]; then
-            gh workflow --repo neondatabase/aws run deploy-dev.yml --ref main \
+            gh workflow --repo neondatabase/infra run deploy-dev.yml --ref main \
               -f deployPgSniRouter=false \
               -f deployProxy=false \
               -f deployStorage=true \
@@ -998,14 +999,14 @@ jobs:
               -f dockerTag=${{needs.tag.outputs.build-tag}} \
               -f deployPreprodRegion=true
 
-            gh workflow --repo neondatabase/aws run deploy-prod.yml --ref main \
+            gh workflow --repo neondatabase/infra run deploy-prod.yml --ref main \
               -f deployStorage=true \
               -f deployStorageBroker=true \
               -f deployStorageController=true \
               -f branch=main \
               -f dockerTag=${{needs.tag.outputs.build-tag}}
           elif [[ "$GITHUB_REF_NAME" == "release-proxy" ]]; then
-            gh workflow --repo neondatabase/aws run deploy-dev.yml --ref main \
+            gh workflow --repo neondatabase/infra run deploy-dev.yml --ref main \
               -f deployPgSniRouter=true \
               -f deployProxy=true \
               -f deployStorage=false \
@@ -1015,7 +1016,7 @@ jobs:
               -f dockerTag=${{needs.tag.outputs.build-tag}} \
               -f deployPreprodRegion=true
 
-            gh workflow --repo neondatabase/aws run deploy-proxy-prod.yml --ref main \
+            gh workflow --repo neondatabase/infra run deploy-proxy-prod.yml --ref main \
               -f deployPgSniRouter=true \
               -f deployProxy=true \
               -f branch=main \

.github/workflows/label-for-external-users.yml

@@ -4,7 +4,7 @@ on:
   issues:
     types:
       - opened
-  pull_request:
+  pull_request_target:
     types:
       - opened
 
@@ -25,7 +25,7 @@ jobs:
     - name: Check whether `${{ github.actor }}` is a member of `${{ github.repository_owner }}`
       id: check-user
       env:
-        GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        GH_TOKEN: ${{ secrets.CI_ACCESS_TOKEN }}
       run: |
         if gh api -H "Accept: application/vnd.github+json" -H "X-GitHub-Api-Version: 2022-11-28" "/orgs/${GITHUB_REPOSITORY_OWNER}/members/${GITHUB_ACTOR}"; then
           is_member=true
@@ -45,10 +45,10 @@ jobs:
       issues: write        # for `gh issue edit`
 
     steps:
-    - name: Label new ${{ github.event_name }}
+    - name: Add `${{ env.LABEL }}` label
       env:
         GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        ITEM_NUMBER: ${{ github.event[github.event_name == 'pull_request' && 'pull_request' || 'issue'].number }}
-        GH_CLI_COMMAND: ${{ github.event_name == 'pull_request' && 'pr' || 'issue' }}
+        ITEM_NUMBER: ${{ github.event[github.event_name == 'pull_request_target' && 'pull_request' || 'issue'].number }}
+        GH_CLI_COMMAND: ${{ github.event_name == 'pull_request_target' && 'pr' || 'issue' }}
       run: |
         gh ${GH_CLI_COMMAND} --repo ${GITHUB_REPOSITORY} edit --add-label=${LABEL} ${ITEM_NUMBER}

README.md

@@ -262,7 +262,7 @@ By default, this runs both debug and release modes, and all supported postgres v
 testing locally, it is convenient to run just one set of permutations, like this:
 
 ```sh
-DEFAULT_PG_VERSION=15 BUILD_TYPE=release ./scripts/pytest
+DEFAULT_PG_VERSION=16 BUILD_TYPE=release ./scripts/pytest
 ```
 
 ## Flamegraphs

control_plane/src/background_process.rs

@@ -379,7 +379,7 @@ where
     }
 }
 
-fn process_has_stopped(pid: Pid) -> anyhow::Result<bool> {
+pub(crate) fn process_has_stopped(pid: Pid) -> anyhow::Result<bool> {
     match kill(pid, None) {
         // Process exists, keep waiting
         Ok(_) => Ok(false),

control_plane/src/bin/neon_local.rs

@@ -15,7 +15,9 @@ use control_plane::local_env::{
 };
 use control_plane::pageserver::PageServerNode;
 use control_plane::safekeeper::SafekeeperNode;
-use control_plane::storage_controller::StorageController;
+use control_plane::storage_controller::{
+    NeonStorageControllerStartArgs, NeonStorageControllerStopArgs, StorageController,
+};
 use control_plane::{broker, local_env};
 use pageserver_api::config::{
     DEFAULT_HTTP_LISTEN_PORT as DEFAULT_PAGESERVER_HTTP_PORT,
@@ -52,7 +54,7 @@ const DEFAULT_PAGESERVER_ID: NodeId = NodeId(1);
 const DEFAULT_BRANCH_NAME: &str = "main";
 project_git_version!(GIT_VERSION);
 
-const DEFAULT_PG_VERSION: &str = "15";
+const DEFAULT_PG_VERSION: &str = "16";
 
 const DEFAULT_PAGESERVER_CONTROL_PLANE_API: &str = "http://127.0.0.1:1234/upcall/v1/";
 
@@ -1052,6 +1054,36 @@ fn get_start_timeout(args: &ArgMatches) -> &Duration {
     humantime_duration.as_ref()
 }
 
+fn storage_controller_start_args(args: &ArgMatches) -> NeonStorageControllerStartArgs {
+    let maybe_instance_id = args.get_one::<u8>("instance-id");
+
+    let base_port = args.get_one::<u16>("base-port");
+
+    if maybe_instance_id.is_some() && base_port.is_none() {
+        panic!("storage-controller start specificied instance-id but did not provide base-port");
+    }
+
+    let start_timeout = args
+        .get_one::<humantime::Duration>("start-timeout")
+        .expect("invalid value for start-timeout");
+
+    NeonStorageControllerStartArgs {
+        instance_id: maybe_instance_id.copied().unwrap_or(1),
+        base_port: base_port.copied(),
+        start_timeout: *start_timeout,
+    }
+}
+
+fn storage_controller_stop_args(args: &ArgMatches) -> NeonStorageControllerStopArgs {
+    let maybe_instance_id = args.get_one::<u8>("instance-id");
+    let immediate = args.get_one::<String>("stop-mode").map(|s| s.as_str()) == Some("immediate");
+
+    NeonStorageControllerStopArgs {
+        instance_id: maybe_instance_id.copied().unwrap_or(1),
+        immediate,
+    }
+}
+
 async fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
     match sub_match.subcommand() {
         Some(("start", subcommand_args)) => {
@@ -1113,19 +1145,14 @@ async fn handle_storage_controller(
     let svc = StorageController::from_env(env);
     match sub_match.subcommand() {
         Some(("start", start_match)) => {
-            if let Err(e) = svc.start(get_start_timeout(start_match)).await {
+            if let Err(e) = svc.start(storage_controller_start_args(start_match)).await {
                 eprintln!("start failed: {e}");
                 exit(1);
             }
         }
 
         Some(("stop", stop_match)) => {
-            let immediate = stop_match
-                .get_one::<String>("stop-mode")
-                .map(|s| s.as_str())
-                == Some("immediate");
-
-            if let Err(e) = svc.stop(immediate).await {
+            if let Err(e) = svc.stop(storage_controller_stop_args(stop_match)).await {
                 eprintln!("stop failed: {}", e);
                 exit(1);
             }
@@ -1228,7 +1255,12 @@ async fn handle_start_all(
     // Only start the storage controller if the pageserver is configured to need it
     if env.control_plane_api.is_some() {
         let storage_controller = StorageController::from_env(env);
-        if let Err(e) = storage_controller.start(retry_timeout).await {
+        if let Err(e) = storage_controller
+            .start(NeonStorageControllerStartArgs::with_default_instance_id(
+                (*retry_timeout).into(),
+            ))
+            .await
+        {
             eprintln!("storage_controller start failed: {:#}", e);
             try_stop_all(env, true).await;
             exit(1);
@@ -1358,10 +1390,21 @@ async fn try_stop_all(env: &local_env::LocalEnv, immediate: bool) {
         eprintln!("neon broker stop failed: {e:#}");
     }
 
-    if env.control_plane_api.is_some() {
+    // Stop all storage controller instances. In the most common case there's only one,
+    // but iterate though the base data directory in order to discover the instances.
+    let storcon_instances = env
+        .storage_controller_instances()
+        .await
+        .expect("Must inspect data dir");
+    for (instance_id, _instance_dir_path) in storcon_instances {
         let storage_controller = StorageController::from_env(env);
-        if let Err(e) = storage_controller.stop(immediate).await {
-            eprintln!("storage controller stop failed: {e:#}");
+        let stop_args = NeonStorageControllerStopArgs {
+            instance_id,
+            immediate,
+        };
+
+        if let Err(e) = storage_controller.stop(stop_args).await {
+            eprintln!("Storage controller instance {instance_id} stop failed: {e:#}");
         }
     }
 }
@@ -1501,6 +1544,18 @@ fn cli() -> Command {
         .action(ArgAction::SetTrue)
         .required(false);
 
+    let instance_id = Arg::new("instance-id")
+        .long("instance-id")
+        .help("Identifier used to distinguish storage controller instances (default 1)")
+        .value_parser(value_parser!(u8))
+        .required(false);
+
+    let base_port = Arg::new("base-port")
+        .long("base-port")
+        .help("Base port for the storage controller instance idenfified by instance-id (defaults to pagserver cplane api)")
+        .value_parser(value_parser!(u16))
+        .required(false);
+
     Command::new("Neon CLI")
         .arg_required_else_help(true)
         .version(GIT_VERSION)
@@ -1609,9 +1664,12 @@ fn cli() -> Command {
                 .arg_required_else_help(true)
                 .about("Manage storage_controller")
                 .subcommand(Command::new("start").about("Start storage controller")
-                            .arg(timeout_arg.clone()))
+                            .arg(timeout_arg.clone())
+                            .arg(instance_id.clone())
+                            .arg(base_port))
                 .subcommand(Command::new("stop").about("Stop storage controller")
-                            .arg(stop_mode_arg.clone()))
+                            .arg(stop_mode_arg.clone())
+                            .arg(instance_id))
         )
         .subcommand(
             Command::new("safekeeper")

control_plane/src/local_env.rs

@@ -27,7 +27,7 @@ use crate::pageserver::PageServerNode;
 use crate::pageserver::PAGESERVER_REMOTE_STORAGE_DIR;
 use crate::safekeeper::SafekeeperNode;
 
-pub const DEFAULT_PG_VERSION: u32 = 15;
+pub const DEFAULT_PG_VERSION: u32 = 16;
 
 //
 // This data structures represents neon_local CLI config
@@ -156,6 +156,11 @@ pub struct NeonStorageControllerConf {
     #[serde(with = "humantime_serde")]
     pub max_warming_up: Duration,
 
+    pub start_as_candidate: bool,
+
+    /// Database url used when running multiple storage controller instances
+    pub database_url: Option<SocketAddr>,
+
     /// Threshold for auto-splitting a tenant into shards
     pub split_threshold: Option<u64>,
 
@@ -174,6 +179,8 @@ impl Default for NeonStorageControllerConf {
         Self {
             max_offline: Self::DEFAULT_MAX_OFFLINE_INTERVAL,
             max_warming_up: Self::DEFAULT_MAX_WARMING_UP_INTERVAL,
+            start_as_candidate: false,
+            database_url: None,
             split_threshold: None,
             max_secondary_lag_bytes: None,
         }
@@ -392,6 +399,36 @@ impl LocalEnv {
         }
     }
 
+    /// Inspect the base data directory and extract the instance id and instance directory path
+    /// for all storage controller instances
+    pub async fn storage_controller_instances(&self) -> std::io::Result<Vec<(u8, PathBuf)>> {
+        let mut instances = Vec::default();
+
+        let dir = std::fs::read_dir(self.base_data_dir.clone())?;
+        for dentry in dir {
+            let dentry = dentry?;
+            let is_dir = dentry.metadata()?.is_dir();
+            let filename = dentry.file_name().into_string().unwrap();
+            let parsed_instance_id = match filename.strip_prefix("storage_controller_") {
+                Some(suffix) => suffix.parse::<u8>().ok(),
+                None => None,
+            };
+
+            let is_instance_dir = is_dir && parsed_instance_id.is_some();
+
+            if !is_instance_dir {
+                continue;
+            }
+
+            instances.push((
+                parsed_instance_id.expect("Checked previously"),
+                dentry.path(),
+            ));
+        }
+
+        Ok(instances)
+    }
+
     pub fn register_branch_mapping(
         &mut self,
         branch_name: String,

control_plane/src/storage_controller.rs

@@ -3,6 +3,8 @@ use crate::{
     local_env::{LocalEnv, NeonStorageControllerConf},
 };
 use camino::{Utf8Path, Utf8PathBuf};
+use hyper::Uri;
+use nix::unistd::Pid;
 use pageserver_api::{
     controller_api::{
         NodeConfigureRequest, NodeDescribeResponse, NodeRegisterRequest, TenantCreateRequest,
@@ -18,7 +20,7 @@ use pageserver_client::mgmt_api::ResponseErrorMessageExt;
 use postgres_backend::AuthType;
 use reqwest::Method;
 use serde::{de::DeserializeOwned, Deserialize, Serialize};
-use std::{fs, str::FromStr, time::Duration};
+use std::{fs, net::SocketAddr, path::PathBuf, str::FromStr, sync::OnceLock};
 use tokio::process::Command;
 use tracing::instrument;
 use url::Url;
@@ -29,12 +31,14 @@ use utils::{
 
 pub struct StorageController {
     env: LocalEnv,
-    listen: String,
     private_key: Option<Vec<u8>>,
     public_key: Option<String>,
-    postgres_port: u16,
     client: reqwest::Client,
     config: NeonStorageControllerConf,
+
+    // The listen addresses is learned when starting the storage controller,
+    // hence the use of OnceLock to init it at the right time.
+    listen: OnceLock<SocketAddr>,
 }
 
 const COMMAND: &str = "storage_controller";
@@ -43,6 +47,36 @@ const STORAGE_CONTROLLER_POSTGRES_VERSION: u32 = 16;
 
 const DB_NAME: &str = "storage_controller";
 
+pub struct NeonStorageControllerStartArgs {
+    pub instance_id: u8,
+    pub base_port: Option<u16>,
+    pub start_timeout: humantime::Duration,
+}
+
+impl NeonStorageControllerStartArgs {
+    pub fn with_default_instance_id(start_timeout: humantime::Duration) -> Self {
+        Self {
+            instance_id: 1,
+            base_port: None,
+            start_timeout,
+        }
+    }
+}
+
+pub struct NeonStorageControllerStopArgs {
+    pub instance_id: u8,
+    pub immediate: bool,
+}
+
+impl NeonStorageControllerStopArgs {
+    pub fn with_default_instance_id(immediate: bool) -> Self {
+        Self {
+            instance_id: 1,
+            immediate,
+        }
+    }
+}
+
 #[derive(Serialize, Deserialize)]
 pub struct AttachHookRequest {
     pub tenant_shard_id: TenantShardId,
@@ -67,23 +101,6 @@ pub struct InspectResponse {
 
 impl StorageController {
     pub fn from_env(env: &LocalEnv) -> Self {
-        // Makes no sense to construct this if pageservers aren't going to use it: assume
-        // pageservers have control plane API set
-        let listen_url = env.control_plane_api.clone().unwrap();
-
-        let listen = format!(
-            "{}:{}",
-            listen_url.host_str().unwrap(),
-            listen_url.port().unwrap()
-        );
-
-        // Convention: NeonEnv in python tests reserves the next port after the control_plane_api
-        // port, for use by our captive postgres.
-        let postgres_port = listen_url
-            .port()
-            .expect("Control plane API setting should always have a port")
-            + 1;
-
         // Assume all pageservers have symmetric auth configuration: this service
         // expects to use one JWT token to talk to all of them.
         let ps_conf = env
@@ -126,20 +143,28 @@ impl StorageController {
 
         Self {
             env: env.clone(),
-            listen,
             private_key,
             public_key,
-            postgres_port,
             client: reqwest::ClientBuilder::new()
                 .build()
                 .expect("Failed to construct http client"),
             config: env.storage_controller.clone(),
+            listen: OnceLock::default(),
         }
     }
 
-    fn pid_file(&self) -> Utf8PathBuf {
-        Utf8PathBuf::from_path_buf(self.env.base_data_dir.join("storage_controller.pid"))
-            .expect("non-Unicode path")
+    fn storage_controller_instance_dir(&self, instance_id: u8) -> PathBuf {
+        self.env
+            .base_data_dir
+            .join(format!("storage_controller_{}", instance_id))
+    }
+
+    fn pid_file(&self, instance_id: u8) -> Utf8PathBuf {
+        Utf8PathBuf::from_path_buf(
+            self.storage_controller_instance_dir(instance_id)
+                .join("storage_controller.pid"),
+        )
+        .expect("non-Unicode path")
     }
 
     /// PIDFile for the postgres instance used to store storage controller state
@@ -184,23 +209,23 @@ impl StorageController {
     }
 
     /// Readiness check for our postgres process
-    async fn pg_isready(&self, pg_bin_dir: &Utf8Path) -> anyhow::Result<bool> {
+    async fn pg_isready(&self, pg_bin_dir: &Utf8Path, postgres_port: u16) -> anyhow::Result<bool> {
         let bin_path = pg_bin_dir.join("pg_isready");
-        let args = ["-h", "localhost", "-p", &format!("{}", self.postgres_port)];
+        let args = ["-h", "localhost", "-p", &format!("{}", postgres_port)];
         let exitcode = Command::new(bin_path).args(args).spawn()?.wait().await?;
 
         Ok(exitcode.success())
     }
 
-    /// Create our database if it doesn't exist, and run migrations.
+    /// Create our database if it doesn't exist
     ///
     /// This function is equivalent to the `diesel setup` command in the diesel CLI.  We implement
     /// the same steps by hand to avoid imposing a dependency on installing diesel-cli for developers
     /// who just want to run `cargo neon_local` without knowing about diesel.
     ///
     /// Returns the database url
-    pub async fn setup_database(&self) -> anyhow::Result<String> {
-        let database_url = format!("postgresql://localhost:{}/{DB_NAME}", self.postgres_port);
+    pub async fn setup_database(&self, postgres_port: u16) -> anyhow::Result<String> {
+        let database_url = format!("postgresql://localhost:{}/{DB_NAME}", postgres_port);
 
         let pg_bin_dir = self.get_pg_bin_dir().await?;
         let createdb_path = pg_bin_dir.join("createdb");
@@ -209,7 +234,7 @@ impl StorageController {
                 "-h",
                 "localhost",
                 "-p",
-                &format!("{}", self.postgres_port),
+                &format!("{}", postgres_port),
                 DB_NAME,
             ])
             .output()
@@ -230,13 +255,14 @@ impl StorageController {
 
     pub async fn connect_to_database(
         &self,
+        postgres_port: u16,
     ) -> anyhow::Result<(
         tokio_postgres::Client,
         tokio_postgres::Connection<tokio_postgres::Socket, tokio_postgres::tls::NoTlsStream>,
     )> {
         tokio_postgres::Config::new()
             .host("localhost")
-            .port(self.postgres_port)
+            .port(postgres_port)
             // The user is the ambient operating system user name.
             // That is an impurity which we want to fix in => TODO https://github.com/neondatabase/neon/issues/8400
             //
@@ -252,72 +278,114 @@ impl StorageController {
             .map_err(anyhow::Error::new)
     }
 
-    pub async fn start(&self, retry_timeout: &Duration) -> anyhow::Result<()> {
-        // Start a vanilla Postgres process used by the storage controller for persistence.
-        let pg_data_path = Utf8PathBuf::from_path_buf(self.env.base_data_dir.clone())
-            .unwrap()
-            .join("storage_controller_db");
-        let pg_bin_dir = self.get_pg_bin_dir().await?;
-        let pg_lib_dir = self.get_pg_lib_dir().await?;
-        let pg_log_path = pg_data_path.join("postgres.log");
+    pub async fn start(&self, start_args: NeonStorageControllerStartArgs) -> anyhow::Result<()> {
+        let instance_dir = self.storage_controller_instance_dir(start_args.instance_id);
+        if let Err(err) = tokio::fs::create_dir(&instance_dir).await {
+            if err.kind() != std::io::ErrorKind::AlreadyExists {
+                panic!("Failed to create instance dir {instance_dir:?}");
+            }
+        }
 
-        if !tokio::fs::try_exists(&pg_data_path).await? {
-            // Initialize empty database
-            let initdb_path = pg_bin_dir.join("initdb");
-            let mut child = Command::new(&initdb_path)
-                .envs(vec![
-                    ("LD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
-                    ("DYLD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
-                ])
-                .args(["-D", pg_data_path.as_ref()])
-                .spawn()
-                .expect("Failed to spawn initdb");
-            let status = child.wait().await?;
-            if !status.success() {
-                anyhow::bail!("initdb failed with status {status}");
+        let (listen, postgres_port) = {
+            if let Some(base_port) = start_args.base_port {
+                (
+                    format!("127.0.0.1:{base_port}"),
+                    self.config
+                        .database_url
+                        .expect("--base-port requires NeonStorageControllerConf::database_url")
+                        .port(),
+                )
+            } else {
+                let listen_url = self.env.control_plane_api.clone().unwrap();
+
+                let listen = format!(
+                    "{}:{}",
+                    listen_url.host_str().unwrap(),
+                    listen_url.port().unwrap()
+                );
+
+                (listen, listen_url.port().unwrap() + 1)
             }
         };
 
-        // Write a minimal config file:
-        // - Specify the port, since this is chosen dynamically
-        // - Switch off fsync, since we're running on lightweight test environments and when e.g. scale testing
-        //   the storage controller we don't want a slow local disk to interfere with that.
-        //
-        // NB: it's important that we rewrite this file on each start command so we propagate changes
-        // from `LocalEnv`'s config file (`.neon/config`).
-        tokio::fs::write(
-            &pg_data_path.join("postgresql.conf"),
-            format!("port = {}\nfsync=off\n", self.postgres_port),
-        )
-        .await?;
+        let socket_addr = listen
+            .parse()
+            .expect("listen address is a valid socket address");
+        self.listen
+            .set(socket_addr)
+            .expect("StorageController::listen is only set here");
 
-        println!("Starting storage controller database...");
-        let db_start_args = [
-            "-w",
-            "-D",
-            pg_data_path.as_ref(),
-            "-l",
-            pg_log_path.as_ref(),
-            "start",
-        ];
+        // Do we remove the pid file on stop?
+        let pg_started = self.is_postgres_running().await?;
+        let pg_lib_dir = self.get_pg_lib_dir().await?;
 
-        background_process::start_process(
-            "storage_controller_db",
-            &self.env.base_data_dir,
-            pg_bin_dir.join("pg_ctl").as_std_path(),
-            db_start_args,
-            vec![
-                ("LD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
-                ("DYLD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
-            ],
-            background_process::InitialPidFile::Create(self.postgres_pid_file()),
-            retry_timeout,
-            || self.pg_isready(&pg_bin_dir),
-        )
-        .await?;
+        if !pg_started {
+            // Start a vanilla Postgres process used by the storage controller for persistence.
+            let pg_data_path = Utf8PathBuf::from_path_buf(self.env.base_data_dir.clone())
+                .unwrap()
+                .join("storage_controller_db");
+            let pg_bin_dir = self.get_pg_bin_dir().await?;
+            let pg_log_path = pg_data_path.join("postgres.log");
 
-        // Run migrations on every startup, in case something changed.
-        let database_url = self.setup_database().await?;
+            if !tokio::fs::try_exists(&pg_data_path).await? {
+                // Initialize empty database
+                let initdb_path = pg_bin_dir.join("initdb");
+                let mut child = Command::new(&initdb_path)
+                    .envs(vec![
+                        ("LD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
+                        ("DYLD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
+                    ])
+                    .args(["-D", pg_data_path.as_ref()])
+                    .spawn()
+                    .expect("Failed to spawn initdb");
+                let status = child.wait().await?;
+                if !status.success() {
+                    anyhow::bail!("initdb failed with status {status}");
+                }
+            };
+
+            // Write a minimal config file:
+            // - Specify the port, since this is chosen dynamically
+            // - Switch off fsync, since we're running on lightweight test environments and when e.g. scale testing
+            //   the storage controller we don't want a slow local disk to interfere with that.
+            //
+            // NB: it's important that we rewrite this file on each start command so we propagate changes
+            // from `LocalEnv`'s config file (`.neon/config`).
+            tokio::fs::write(
+                &pg_data_path.join("postgresql.conf"),
+                format!("port = {}\nfsync=off\n", postgres_port),
+            )
+            .await?;
+
+            println!("Starting storage controller database...");
+            let db_start_args = [
+                "-w",
+                "-D",
+                pg_data_path.as_ref(),
+                "-l",
+                pg_log_path.as_ref(),
+                "start",
+            ];
+
+            background_process::start_process(
+                "storage_controller_db",
+                &self.env.base_data_dir,
+                pg_bin_dir.join("pg_ctl").as_std_path(),
+                db_start_args,
+                vec![
+                    ("LD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
+                    ("DYLD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
+                ],
+                background_process::InitialPidFile::Create(self.postgres_pid_file()),
+                &start_args.start_timeout,
+                || self.pg_isready(&pg_bin_dir, postgres_port),
+            )
+            .await?;
+
+            self.setup_database(postgres_port).await?;
+        }
+
+        let database_url = format!("postgresql://localhost:{}/{DB_NAME}", postgres_port);
 
         // We support running a startup SQL script to fiddle with the database before we launch storcon.
         // This is used by the test suite.
@@ -339,7 +407,7 @@ impl StorageController {
                 }
             }
         };
-        let (mut client, conn) = self.connect_to_database().await?;
+        let (mut client, conn) = self.connect_to_database(postgres_port).await?;
         let conn = tokio::spawn(conn);
         let tx = client.build_transaction();
         let tx = tx.start().await?;
@@ -348,9 +416,20 @@ impl StorageController {
         drop(client);
         conn.await??;
 
+        let listen = self
+            .listen
+            .get()
+            .expect("cell is set earlier in this function");
+        let address_for_peers = Uri::builder()
+            .scheme("http")
+            .authority(format!("{}:{}", listen.ip(), listen.port()))
+            .path_and_query("")
+            .build()
+            .unwrap();
+
         let mut args = vec![
             "-l",
-            &self.listen,
+            &listen.to_string(),
             "--dev",
             "--database-url",
             &database_url,
@@ -358,15 +437,27 @@ impl StorageController {
             &humantime::Duration::from(self.config.max_offline).to_string(),
             "--max-warming-up-interval",
             &humantime::Duration::from(self.config.max_warming_up).to_string(),
+            "--address-for-peers",
+            &address_for_peers.to_string(),
         ]
         .into_iter()
         .map(|s| s.to_string())
         .collect::<Vec<_>>();
+
+        if self.config.start_as_candidate {
+            args.push("--start-as-candidate".to_string());
+        }
+
         if let Some(private_key) = &self.private_key {
             let claims = Claims::new(None, Scope::PageServerApi);
             let jwt_token =
                 encode_from_key_file(&claims, private_key).expect("failed to generate jwt token");
             args.push(format!("--jwt-token={jwt_token}"));
+
+            let peer_claims = Claims::new(None, Scope::Admin);
+            let peer_jwt_token = encode_from_key_file(&peer_claims, private_key)
+                .expect("failed to generate jwt token");
+            args.push(format!("--peer-jwt-token={peer_jwt_token}"));
         }
 
         if let Some(public_key) = &self.public_key {
@@ -394,15 +485,15 @@ impl StorageController {
 
         background_process::start_process(
             COMMAND,
-            &self.env.base_data_dir,
+            &instance_dir,
             &self.env.storage_controller_bin(),
             args,
             vec![
                 ("LD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
                 ("DYLD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
             ],
-            background_process::InitialPidFile::Create(self.pid_file()),
-            retry_timeout,
+            background_process::InitialPidFile::Create(self.pid_file(start_args.instance_id)),
+            &start_args.start_timeout,
             || async {
                 match self.ready().await {
                     Ok(_) => Ok(true),
@@ -415,8 +506,35 @@ impl StorageController {
         Ok(())
     }
 
-    pub async fn stop(&self, immediate: bool) -> anyhow::Result<()> {
-        background_process::stop_process(immediate, COMMAND, &self.pid_file())?;
+    pub async fn stop(&self, stop_args: NeonStorageControllerStopArgs) -> anyhow::Result<()> {
+        background_process::stop_process(
+            stop_args.immediate,
+            COMMAND,
+            &self.pid_file(stop_args.instance_id),
+        )?;
+
+        let storcon_instances = self.env.storage_controller_instances().await?;
+        for (instance_id, instanced_dir_path) in storcon_instances {
+            if instance_id == stop_args.instance_id {
+                continue;
+            }
+
+            let pid_file = instanced_dir_path.join("storage_controller.pid");
+            let pid = tokio::fs::read_to_string(&pid_file)
+                .await
+                .map_err(|err| {
+                    anyhow::anyhow!("Failed to read storcon pid file at {pid_file:?}: {err}")
+                })?
+                .parse::<i32>()
+                .expect("pid is valid i32");
+
+            let other_proc_alive = !background_process::process_has_stopped(Pid::from_raw(pid))?;
+            if other_proc_alive {
+                // There is another storage controller instance running, so we return
+                // and leave the database running.
+                return Ok(());
+            }
+        }
 
         let pg_data_path = self.env.base_data_dir.join("storage_controller_db");
         let pg_bin_dir = self.get_pg_bin_dir().await?;
@@ -429,27 +547,51 @@ impl StorageController {
             .wait()
             .await?;
         if !stop_status.success() {
-            let pg_status_args = ["-D", &pg_data_path.to_string_lossy(), "status"];
-            let status_exitcode = Command::new(pg_bin_dir.join("pg_ctl"))
-                .args(pg_status_args)
-                .spawn()?
-                .wait()
-                .await?;
-
-            // pg_ctl status returns this exit code if postgres is not running: in this case it is
-            // fine that stop failed.  Otherwise it is an error that stop failed.
-            const PG_STATUS_NOT_RUNNING: i32 = 3;
-            if Some(PG_STATUS_NOT_RUNNING) == status_exitcode.code() {
-                println!("Storage controller database is already stopped");
-                return Ok(());
-            } else {
-                anyhow::bail!("Failed to stop storage controller database: {stop_status}")
+            match self.is_postgres_running().await {
+                Ok(false) => {
+                    println!("Storage controller database is already stopped");
+                    return Ok(());
+                }
+                Ok(true) => {
+                    anyhow::bail!("Failed to stop storage controller database");
+                }
+                Err(err) => {
+                    anyhow::bail!("Failed to stop storage controller database: {err}");
+                }
             }
         }
 
         Ok(())
     }
 
+    async fn is_postgres_running(&self) -> anyhow::Result<bool> {
+        let pg_data_path = self.env.base_data_dir.join("storage_controller_db");
+        let pg_bin_dir = self.get_pg_bin_dir().await?;
+
+        let pg_status_args = ["-D", &pg_data_path.to_string_lossy(), "status"];
+        let status_exitcode = Command::new(pg_bin_dir.join("pg_ctl"))
+            .args(pg_status_args)
+            .spawn()?
+            .wait()
+            .await?;
+
+        // pg_ctl status returns this exit code if postgres is not running: in this case it is
+        // fine that stop failed.  Otherwise it is an error that stop failed.
+        const PG_STATUS_NOT_RUNNING: i32 = 3;
+        const PG_NO_DATA_DIR: i32 = 4;
+        const PG_STATUS_RUNNING: i32 = 0;
+        match status_exitcode.code() {
+            Some(PG_STATUS_NOT_RUNNING) => Ok(false),
+            Some(PG_NO_DATA_DIR) => Ok(false),
+            Some(PG_STATUS_RUNNING) => Ok(true),
+            Some(code) => Err(anyhow::anyhow!(
+                "pg_ctl status returned unexpected status code: {:?}",
+                code
+            )),
+            None => Err(anyhow::anyhow!("pg_ctl status returned no status code")),
+        }
+    }
+
     fn get_claims_for_path(path: &str) -> anyhow::Result<Option<Claims>> {
         let category = match path.find('/') {
             Some(idx) => &path[..idx],
@@ -475,15 +617,31 @@ impl StorageController {
         RQ: Serialize + Sized,
         RS: DeserializeOwned + Sized,
     {
-        // The configured URL has the /upcall path prefix for pageservers to use: we will strip that out
-        // for general purpose API access.
-        let listen_url = self.env.control_plane_api.clone().unwrap();
-        let url = Url::from_str(&format!(
-            "http://{}:{}/{path}",
-            listen_url.host_str().unwrap(),
-            listen_url.port().unwrap()
-        ))
-        .unwrap();
+        // In the special case of the `storage_controller start` subcommand, we wish
+        // to use the API endpoint of the newly started storage controller in order
+        // to pass the readiness check. In this scenario [`Self::listen`] will be set
+        // (see [`Self::start`]).
+        //
+        // Otherwise, we infer the storage controller api endpoint from the configured
+        // control plane API.
+        let url = if let Some(socket_addr) = self.listen.get() {
+            Url::from_str(&format!(
+                "http://{}:{}/{path}",
+                socket_addr.ip().to_canonical(),
+                socket_addr.port()
+            ))
+            .unwrap()
+        } else {
+            // The configured URL has the /upcall path prefix for pageservers to use: we will strip that out
+            // for general purpose API access.
+            let listen_url = self.env.control_plane_api.clone().unwrap();
+            Url::from_str(&format!(
+                "http://{}:{}/{path}",
+                listen_url.host_str().unwrap(),
+                listen_url.port().unwrap()
+            ))
+            .unwrap()
+        };
 
         let mut builder = self.client.request(method, url);
         if let Some(body) = body {

docs/rfcs/033-storage-controller-drain-and-fill.md

@@ -14,7 +14,7 @@ picked tenant (which requested on-demand activation) for around 30 seconds
 during the restart at 2024-04-03 16:37 UTC.
 
 Note that lots of shutdowns on loaded pageservers do not finish within the
-[10 second systemd enforced timeout](https://github.com/neondatabase/aws/blob/0a5280b383e43c063d43cbf87fa026543f6d6ad4/.github/ansible/systemd/pageserver.service#L16). This means we are shutting down without flushing ephemeral layers
+[10 second systemd enforced timeout](https://github.com/neondatabase/infra/blob/0a5280b383e43c063d43cbf87fa026543f6d6ad4/.github/ansible/systemd/pageserver.service#L16). This means we are shutting down without flushing ephemeral layers
 and have to reingest data in order to serve requests after restarting, potentially making first request latencies worse.
 
 This problem is not yet very acutely felt in storage controller managed pageservers since

docs/rfcs/036-physical-replication.md

@@ -0,0 +1,265 @@
+# Physical Replication
+
+This RFC is a bit special in that we have already implemented physical
+replication a long time ago. However, we never properly wrote down all
+the decisions and assumptions, and in the last months when more users
+have started to use the feature, numerous issues have surfaced.
+
+This RFC documents the design decisions that have been made.
+
+## Summary
+
+PostgreSQL has a feature called streaming replication, where a replica
+streams WAL from the primary and continuously applies it. It is also
+known as "physical replication", to distinguish it from logical
+replication.  In PostgreSQL, a replica is initialized by taking a
+physical backup of the primary. In Neon, the replica is initialized
+from a slim "base backup" from the pageserver, just like a primary,
+and the primary and the replicas connect to the same pageserver,
+sharing the storage.
+
+There are two kinds of read-only replicas in Neon:
+- replicas that follow the primary, and
+- "static" replicas that are pinned at a particular LSN.
+
+A static replica is useful e.g. for performing time-travel queries and
+running one-off slow queries without affecting the primary. A replica
+that follows the primary can be used e.g. to scale out read-only
+workloads.
+
+## Motivation
+
+Read-only replicas allow offloading read-only queries. It's useful for
+isolation, if you want to make sure that read-only queries don't
+affect the primary, and it's also an easy way to provide guaranteed
+read-only access to an application, without having to mess with access
+controls.
+
+## Non Goals (if relevant)
+
+This RFC is all about WAL-based *physical* replication. Logical
+replication is a different feature.
+
+Neon also has the capability to launch "static" read-only nodes which
+do not follow the primary, but are pinned to a particular LSN. They
+can be used for long-running one-off queries, or for Point-in-time
+queries. They work similarly to read replicas that follow the primary,
+but some things are simpler: there are no concerns about cache
+invalidation when the data changes on the primary, or worrying about
+transactions that are in-progress on the primary.
+
+## Impacted components (e.g. pageserver, safekeeper, console, etc)
+
+- Control plane launches the replica
+- Replica Postgres instance connects to the safekeepers, to stream the WAL
+- The primary does not know about the standby, except for the hot standby feedback
+- The primary and replicas all connect to the same pageservers
+
+
+# Context
+
+Some useful things to know about hot standby and replicas in
+PostgreSQL.
+
+## PostgreSQL startup sequence
+
+"Running" and "start up" terms are little imprecise. PostgreSQL
+replica startup goes through several stages:
+
+1. First, the process is started up, and various initialization steps
+   are performed, like initializing shared memory. If you try to
+   connect to the server in this stage, you get an error: ERROR: the
+   database system is starting up. This stage happens very quickly, no
+
+2. Then the server reads the checpoint record from the WAL and starts
+   the WAL replay starting from the checkpoint. This works differently
+   in Neon: we start the WAL replay at the basebackup LSN, not from a
+   checkpoint! If you connect to the server in this state, you get an
+   error: ERROR: the database system is not yet accepting
+   connections. We proceed to the next stage, when the WAL replay sees
+   a running-xacts record. Or in Neon, the "CLOG scanning" mechanism
+   can allow us to move directly to next stage, with all the caveats
+   listed in this RFC.
+
+3. When the running-xacts information is established, the server
+   starts to accept connections normally.
+
+From PostgreSQL's point of view, the server is already running in
+stage 2, even though it's not accepting connections yet. Our
+`compute_ctl` does not consider it as running until stage 3. If the
+transition from stage 2 to 3 doesn't happen fast enough, the control
+plane will mark the start operation as failed.
+
+
+## Decisions, Issues
+
+### Cache invalidation in replica
+
+When a read replica follows the primary in PostgreSQL, it needs to
+stream all the WAL from the primary and apply all the records, to keep
+the local copy of the data consistent with the primary. In Neon, the
+replica can fetch the updated page versions from the pageserver, so
+it's not necessary to apply all the WAL. However, it needs to ensure
+that any pages that are currently in the Postgres buffer cache, or the
+Local File Cache, are either updated, or thrown away so that the next
+read of the page will fetch the latest version.
+
+We choose to apply the WAL records for pages that are already in the
+buffer cache, and skip records for other pages. Somewhat arbitrarily,
+we also apply records affecting catalog relations, fetching the old
+page version from the pageserver if necessary first. See
+`neon_redo_read_buffer_filter()` function.
+
+The replica wouldn't necessarily need to see all the WAL records, only
+the records that apply to cached pages. For simplicity, we do stream
+all the WAL to the replica, and the replica simply ignores WAL records
+that require no action.
+
+Like in PostgreSQL, the read replica maintains a "replay LSN", which
+is the LSN up to which the replica has received and replayed the
+WAL. The replica can lag behind the primary, if it cannot quite keep
+up with the primary, or if a long-running query conflicts with changes
+that are about to be applied, or even intentionally if the user wishes
+to see delayed data (see recovery_min_apply_delay). It's important
+that the replica sees a consistent view of the whole cluster at the
+replay LSN, when it's lagging behind.
+
+In Neon, the replica connects to a safekeeper to get the WAL
+stream. That means that the safekeepers must be able to regurgitate
+the original WAL as far back as the replay LSN of any running read
+replica. (A static read-only node that does not follow the primary
+does not require a WAL stream however). The primary does not need to
+be running, and when it is, the replicas don't incur any extra
+overhead to the primary (see hot standby feedback though).
+
+### In-progress transactions
+
+In PostgreSQL, when a hot standby server starts up, it cannot
+immediately open up for queries (see [PostgreSQL startup
+sequence]). It first needs to establish a complete list of in-progress
+transactions, including subtransactions, that are running at the
+primary, at the current replay LSN. Normally that happens quickly,
+when the replica sees a "running-xacts" WAL record, because the
+primary writes a running-xacts WAL record at every checkpoint, and in
+PostgreSQL the replica always starts the WAL replay from a checkpoint
+REDO point. (A shutdown checkpoint WAL record also implies that all
+the non-prepared transactions have ended.) If there are a lot of
+subtransactions in progress, however, the standby might need to wait
+for old transactions to complete before it can open up for queries.
+
+In Neon that problem is worse: a replica can start at any LSN, so
+there's no guarantee that it will see a running-xacts record any time
+soon. In particular, if the primary is not running when the replica is
+started, it might never see a running-xacts record.
+
+To make things worse, we initially missed this issue, and always
+started accepting queries at replica startup, even if it didn't have
+the transaction information. That could lead to incorrect query
+results and data corruption later. However, as we fixed that, we
+introduced a new problem compared to what we had before: previously
+the replica would always start up, but after fixing that bug, it might
+not. In a superficial way, the old behavior was better (but could lead
+to serious issues later!). That made fixing that bug was very hard,
+because as we fixed it, we made things (superficially) worse for
+others.
+
+See https://github.com/neondatabase/neon/pull/7288 which fixed the
+bug, and follow-up PRs https://github.com/neondatabase/neon/pull/8323
+and https://github.com/neondatabase/neon/pull/8484 to try to claw back
+the cases that started to cause trouble as fixing it. As of this
+writing, there are still cases where a replica might not immediately
+start up, causing the control plane operation to fail, the remaining
+issues are tracked in https://github.com/neondatabase/neon/issues/6211.
+
+One long-term fix for this is to switch to using so-called CSN
+snapshots in read replica. That would make it unnecessary to have the
+full in-progress transaction list in the replica at startup time. See
+https://commitfest.postgresql.org/48/4912/ for a work-in-progress
+patch to upstream to implement that.
+
+Another thing we could do is to teach the control plane about that
+distinction between "starting up" and "running but haven't received
+running-xacts information yet", so that we could keep the replica
+waiting longer in that stage, and also give any client connections the
+same `ERROR: the database system is not yet accepting connections`
+error that you get in standalone PostgreSQL in that state.
+
+
+### Recovery conflicts and Hot standby feedback
+
+It's possible that a tuple version is vacuumed away in the primary,
+even though it is still needed by a running transactions in the
+replica. This is called a "recovery conflict", and PostgreSQL provides
+various options for dealing with it. By default, the WAL replay will
+wait up to 30 s for the conflicting query to finish. After that, it
+will kill the running query, so that the WAL replay can proceed.
+
+Another way to avoid the situation is to enable the
+[`hot_standby_feedback`](https://www.postgresql.org/docs/current/runtime-config-replication.html#GUC-HOT-STANDBY-FEEDBACK)
+option. When it is enabled, the primary will refrain from vacuuming
+tuples that are still needed in the primary. That means potentially
+bloating the primary, which violates the usual rule that read replicas
+don't affect the operations on the primary, which is why it's off by
+default. We leave it to users to decide if they want to turn it on,
+same as PostgreSQL.
+
+Neon supports `hot_standby_feedback` by passing the feedback messages
+from the replica to the safekeepers, and from safekeepers to the
+primary.
+
+### Relationship of settings between primary and replica
+
+In order to enter hot standby mode, some configuration options need to
+be set to the same or larger values in the standby, compared to the
+primary.  See [explanation in the PostgreSQL
+docs](https://www.postgresql.org/docs/current/hot-standby.html#HOT-STANDBY-ADMIN)
+
+In Neon, we have this problem too. To prevent customers from hitting
+it, the control plane automatically adjusts the settings of a replica,
+so that they match or exceed the primary's settings (see
+https://github.com/neondatabase/cloud/issues/14903). However, you
+can still hit the issue if the primary is restarted with larger
+settings, while the replica is running.
+
+
+### Interaction with Pageserver GC
+
+The read replica can lag behind the primary. If there are recovery
+conflicts or the replica cannot keep up for some reason, the lag can
+in principle grow indefinitely. The replica will issue all GetPage
+requests to the pageservers at the current replay LSN, and needs to
+see the old page versions.
+
+If the retention period in the pageserver is set to be small, it may
+have already garbage collected away the old page versions. That will
+cause read errors in the compute, and can mean that the replica cannot
+make progress with the replication anymore.
+
+There is a mechanism for replica to pass information about its replay
+LSN to the pageserver, so that the pageserver refrains from GC'ing
+data that is still needed by the standby. It's called
+'standby_horizon' in the pageserver code, see
+https://github.com/neondatabase/neon/pull/7368. A separate "lease"
+mechanism also is in the works, where the replica could hold a lease
+on the old LSN, preventing the pageserver from advancing the GC
+horizon past that point. The difference is that the standby_horizon
+mechanism relies on a feedback message from replica to safekeeper,
+while the least API is exposed directly from the pageserver. A static
+read-only node is not connected to safekeepers, so it cannot use the
+standby_horizon mechanism.
+
+
+### Synchronous replication
+
+We haven't put any effort into synchronous replication yet.
+
+PostgreSQL provides multiple levels of synchronicity. In the weaker
+levels, a transaction is not acknowledged as committed to the client
+in the primary until the WAL has been streamed to a replica or flushed
+to disk there. Those modes don't make senses in Neon, because the
+safekeepers handle durability.
+
+`synchronous_commit=remote_apply` mode would make sense. In that mode,
+the commit is not acknowledged to the client until it has been
+replayed in the replica. That ensures that after commit, you can see
+the commit in the replica too (aka. read-your-write consistency).

libs/remote_storage/src/azure_blob.rs

@@ -383,6 +383,48 @@ impl RemoteStorage for AzureBlobStorage {
         }
     }
 
+    async fn head_object(
+        &self,
+        key: &RemotePath,
+        cancel: &CancellationToken,
+    ) -> Result<ListingObject, DownloadError> {
+        let kind = RequestKind::Head;
+        let _permit = self.permit(kind, cancel).await?;
+
+        let started_at = start_measuring_requests(kind);
+
+        let blob_client = self.client.blob_client(self.relative_path_to_name(key));
+        let properties_future = blob_client.get_properties().into_future();
+
+        let properties_future = tokio::time::timeout(self.timeout, properties_future);
+
+        let res = tokio::select! {
+            res = properties_future => res,
+            _ = cancel.cancelled() => return Err(TimeoutOrCancel::Cancel.into()),
+        };
+
+        if let Ok(inner) = &res {
+            // do not incl. timeouts as errors in metrics but cancellations
+            let started_at = ScopeGuard::into_inner(started_at);
+            crate::metrics::BUCKET_METRICS
+                .req_seconds
+                .observe_elapsed(kind, inner, started_at);
+        }
+
+        let data = match res {
+            Ok(Ok(data)) => Ok(data),
+            Ok(Err(sdk)) => Err(to_download_error(sdk)),
+            Err(_timeout) => Err(DownloadError::Timeout),
+        }?;
+
+        let properties = data.blob.properties;
+        Ok(ListingObject {
+            key: key.to_owned(),
+            last_modified: SystemTime::from(properties.last_modified),
+            size: properties.content_length,
+        })
+    }
+
     async fn upload(
         &self,
         from: impl Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static,

libs/remote_storage/src/lib.rs

@@ -150,7 +150,7 @@ pub enum ListingMode {
     NoDelimiter,
 }
 
-#[derive(PartialEq, Eq, Debug)]
+#[derive(PartialEq, Eq, Debug, Clone)]
 pub struct ListingObject {
     pub key: RemotePath,
     pub last_modified: SystemTime,
@@ -215,6 +215,13 @@ pub trait RemoteStorage: Send + Sync + 'static {
         Ok(combined)
     }
 
+    /// Obtain metadata information about an object.
+    async fn head_object(
+        &self,
+        key: &RemotePath,
+        cancel: &CancellationToken,
+    ) -> Result<ListingObject, DownloadError>;
+
     /// Streams the local file contents into remote into the remote storage entry.
     ///
     /// If the operation fails because of timeout or cancellation, the root cause of the error will be
@@ -363,6 +370,20 @@ impl<Other: RemoteStorage> GenericRemoteStorage<Arc<Other>> {
         }
     }
 
+    // See [`RemoteStorage::head_object`].
+    pub async fn head_object(
+        &self,
+        key: &RemotePath,
+        cancel: &CancellationToken,
+    ) -> Result<ListingObject, DownloadError> {
+        match self {
+            Self::LocalFs(s) => s.head_object(key, cancel).await,
+            Self::AwsS3(s) => s.head_object(key, cancel).await,
+            Self::AzureBlob(s) => s.head_object(key, cancel).await,
+            Self::Unreliable(s) => s.head_object(key, cancel).await,
+        }
+    }
+
     /// See [`RemoteStorage::upload`]
     pub async fn upload(
         &self,
@@ -598,6 +619,7 @@ impl ConcurrencyLimiter {
             RequestKind::Delete => &self.write,
             RequestKind::Copy => &self.write,
             RequestKind::TimeTravel => &self.write,
+            RequestKind::Head => &self.read,
         }
     }
 

libs/remote_storage/src/local_fs.rs

@@ -445,6 +445,20 @@ impl RemoteStorage for LocalFs {
         }
     }
 
+    async fn head_object(
+        &self,
+        key: &RemotePath,
+        _cancel: &CancellationToken,
+    ) -> Result<ListingObject, DownloadError> {
+        let target_file_path = key.with_base(&self.storage_root);
+        let metadata = file_metadata(&target_file_path).await?;
+        Ok(ListingObject {
+            key: key.clone(),
+            last_modified: metadata.modified()?,
+            size: metadata.len(),
+        })
+    }
+
     async fn upload(
         &self,
         data: impl Stream<Item = std::io::Result<Bytes>> + Send + Sync,

libs/remote_storage/src/metrics.rs

@@ -13,6 +13,7 @@ pub(crate) enum RequestKind {
     List = 3,
     Copy = 4,
     TimeTravel = 5,
+    Head = 6,
 }
 
 use scopeguard::ScopeGuard;
@@ -27,6 +28,7 @@ impl RequestKind {
             List => "list_objects",
             Copy => "copy_object",
             TimeTravel => "time_travel_recover",
+            Head => "head_object",
         }
     }
     const fn as_index(&self) -> usize {
@@ -34,7 +36,8 @@ impl RequestKind {
     }
 }
 
-pub(crate) struct RequestTyped<C>([C; 6]);
+const REQUEST_KIND_COUNT: usize = 7;
+pub(crate) struct RequestTyped<C>([C; REQUEST_KIND_COUNT]);
 
 impl<C> RequestTyped<C> {
     pub(crate) fn get(&self, kind: RequestKind) -> &C {
@@ -43,8 +46,8 @@ impl<C> RequestTyped<C> {
 
     fn build_with(mut f: impl FnMut(RequestKind) -> C) -> Self {
         use RequestKind::*;
-        let mut it = [Get, Put, Delete, List, Copy, TimeTravel].into_iter();
-        let arr = std::array::from_fn::<C, 6, _>(|index| {
+        let mut it = [Get, Put, Delete, List, Copy, TimeTravel, Head].into_iter();
+        let arr = std::array::from_fn::<C, REQUEST_KIND_COUNT, _>(|index| {
             let next = it.next().unwrap();
             assert_eq!(index, next.as_index());
             f(next)

libs/remote_storage/src/s3_bucket.rs

@@ -23,7 +23,7 @@ use aws_config::{
 use aws_sdk_s3::{
     config::{AsyncSleep, IdentityCache, Region, SharedAsyncSleep},
     error::SdkError,
-    operation::get_object::GetObjectError,
+    operation::{get_object::GetObjectError, head_object::HeadObjectError},
     types::{Delete, DeleteMarkerEntry, ObjectIdentifier, ObjectVersion, StorageClass},
     Client,
 };
@@ -604,6 +604,78 @@ impl RemoteStorage for S3Bucket {
         }
     }
 
+    async fn head_object(
+        &self,
+        key: &RemotePath,
+        cancel: &CancellationToken,
+    ) -> Result<ListingObject, DownloadError> {
+        let kind = RequestKind::Head;
+        let _permit = self.permit(kind, cancel).await?;
+
+        let started_at = start_measuring_requests(kind);
+
+        let head_future = self
+            .client
+            .head_object()
+            .bucket(self.bucket_name())
+            .key(self.relative_path_to_s3_object(key))
+            .send();
+
+        let head_future = tokio::time::timeout(self.timeout, head_future);
+
+        let res = tokio::select! {
+            res = head_future => res,
+            _ = cancel.cancelled() => return Err(TimeoutOrCancel::Cancel.into()),
+        };
+
+        let res = res.map_err(|_e| DownloadError::Timeout)?;
+
+        // do not incl. timeouts as errors in metrics but cancellations
+        let started_at = ScopeGuard::into_inner(started_at);
+        crate::metrics::BUCKET_METRICS
+            .req_seconds
+            .observe_elapsed(kind, &res, started_at);
+
+        let data = match res {
+            Ok(object_output) => object_output,
+            Err(SdkError::ServiceError(e)) if matches!(e.err(), HeadObjectError::NotFound(_)) => {
+                // Count this in the AttemptOutcome::Ok bucket, because 404 is not
+                // an error: we expect to sometimes fetch an object and find it missing,
+                // e.g. when probing for timeline indices.
+                crate::metrics::BUCKET_METRICS.req_seconds.observe_elapsed(
+                    kind,
+                    AttemptOutcome::Ok,
+                    started_at,
+                );
+                return Err(DownloadError::NotFound);
+            }
+            Err(e) => {
+                crate::metrics::BUCKET_METRICS.req_seconds.observe_elapsed(
+                    kind,
+                    AttemptOutcome::Err,
+                    started_at,
+                );
+
+                return Err(DownloadError::Other(
+                    anyhow::Error::new(e).context("s3 head object"),
+                ));
+            }
+        };
+
+        let (Some(last_modified), Some(size)) = (data.last_modified, data.content_length) else {
+            return Err(DownloadError::Other(anyhow!(
+                "head_object doesn't contain last_modified or content_length"
+            )))?;
+        };
+        Ok(ListingObject {
+            key: key.to_owned(),
+            last_modified: SystemTime::try_from(last_modified).map_err(|e| {
+                DownloadError::Other(anyhow!("can't convert time '{last_modified}': {e}"))
+            })?,
+            size: size as u64,
+        })
+    }
+
     async fn upload(
         &self,
         from: impl Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static,

libs/remote_storage/src/simulate_failures.rs

@@ -30,6 +30,7 @@ pub struct UnreliableWrapper {
 #[derive(Debug, Hash, Eq, PartialEq)]
 enum RemoteOp {
     ListPrefixes(Option<RemotePath>),
+    HeadObject(RemotePath),
     Upload(RemotePath),
     Download(RemotePath),
     Delete(RemotePath),
@@ -137,6 +138,16 @@ impl RemoteStorage for UnreliableWrapper {
         self.inner.list(prefix, mode, max_keys, cancel).await
     }
 
+    async fn head_object(
+        &self,
+        key: &RemotePath,
+        cancel: &CancellationToken,
+    ) -> Result<crate::ListingObject, DownloadError> {
+        self.attempt(RemoteOp::HeadObject(key.clone()))
+            .map_err(DownloadError::Other)?;
+        self.inner.head_object(key, cancel).await
+    }
+
     async fn upload(
         &self,
         data: impl Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static,

pageserver/src/basebackup.rs

@@ -413,6 +413,10 @@ where
             let mut content = Vec::with_capacity(n_origins * 16 + 8);
             content.extend_from_slice(&pg_constants::REPLICATION_STATE_MAGIC.to_le_bytes());
             for (origin_id, origin_lsn) in repl_origins {
+                info!(
+                    "Include origin_id={}, origin_lsn={} in basebackup",
+                    origin_id, origin_lsn
+                );
                 content.extend_from_slice(&origin_id.to_le_bytes());
                 content.extend_from_slice(&[0u8; 6]); // align to 8 bytes
                 content.extend_from_slice(&origin_lsn.0.to_le_bytes());

pageserver/src/config.rs

@@ -50,7 +50,6 @@ pub mod defaults {
         DEFAULT_HTTP_LISTEN_ADDR, DEFAULT_HTTP_LISTEN_PORT, DEFAULT_PG_LISTEN_ADDR,
         DEFAULT_PG_LISTEN_PORT,
     };
-    use pageserver_api::models::ImageCompressionAlgorithm;
     pub use storage_broker::DEFAULT_ENDPOINT as BROKER_DEFAULT_ENDPOINT;
 
     pub const DEFAULT_WAIT_LSN_TIMEOUT: &str = "300 s";
@@ -90,8 +89,7 @@ pub mod defaults {
 
     pub const DEFAULT_MAX_VECTORED_READ_BYTES: usize = 128 * 1024; // 128 KiB
 
-    pub const DEFAULT_IMAGE_COMPRESSION: ImageCompressionAlgorithm =
-        ImageCompressionAlgorithm::Disabled;
+    pub const DEFAULT_IMAGE_COMPRESSION: &str = "zstd(1)";
 
     pub const DEFAULT_VALIDATE_VECTORED_GET: bool = false;
 
@@ -478,7 +476,7 @@ impl PageServerConfigBuilder {
             max_vectored_read_bytes: Set(MaxVectoredReadBytes(
                 NonZeroUsize::new(DEFAULT_MAX_VECTORED_READ_BYTES).unwrap(),
             )),
-            image_compression: Set(DEFAULT_IMAGE_COMPRESSION),
+            image_compression: Set(DEFAULT_IMAGE_COMPRESSION.parse().unwrap()),
             ephemeral_bytes_per_memory_kb: Set(DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB),
             l0_flush: Set(L0FlushConfig::default()),
             compact_level0_phase1_value_access: Set(CompactL0Phase1ValueAccess::default()),
@@ -1065,7 +1063,7 @@ impl PageServerConf {
                 NonZeroUsize::new(defaults::DEFAULT_MAX_VECTORED_READ_BYTES)
                     .expect("Invalid default constant"),
             ),
-            image_compression: defaults::DEFAULT_IMAGE_COMPRESSION,
+            image_compression: defaults::DEFAULT_IMAGE_COMPRESSION.parse().unwrap(),
             ephemeral_bytes_per_memory_kb: defaults::DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB,
             l0_flush: L0FlushConfig::default(),
             compact_level0_phase1_value_access: CompactL0Phase1ValueAccess::default(),
@@ -1305,7 +1303,7 @@ background_task_maximum_delay = '334 s'
                     NonZeroUsize::new(defaults::DEFAULT_MAX_VECTORED_READ_BYTES)
                         .expect("Invalid default constant")
                 ),
-                image_compression: defaults::DEFAULT_IMAGE_COMPRESSION,
+                image_compression: defaults::DEFAULT_IMAGE_COMPRESSION.parse().unwrap(),
                 ephemeral_bytes_per_memory_kb: defaults::DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB,
                 l0_flush: L0FlushConfig::default(),
                 compact_level0_phase1_value_access: CompactL0Phase1ValueAccess::default(),
@@ -1378,7 +1376,7 @@ background_task_maximum_delay = '334 s'
                     NonZeroUsize::new(defaults::DEFAULT_MAX_VECTORED_READ_BYTES)
                         .expect("Invalid default constant")
                 ),
-                image_compression: defaults::DEFAULT_IMAGE_COMPRESSION,
+                image_compression: defaults::DEFAULT_IMAGE_COMPRESSION.parse().unwrap(),
                 ephemeral_bytes_per_memory_kb: defaults::DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB,
                 l0_flush: L0FlushConfig::default(),
                 compact_level0_phase1_value_access: CompactL0Phase1ValueAccess::default(),

pageserver/src/l0_flush.rs

@@ -1,15 +1,10 @@
 use std::{num::NonZeroUsize, sync::Arc};
 
-use crate::tenant::ephemeral_file;
-
 #[derive(Debug, PartialEq, Eq, Clone, serde::Deserialize)]
 #[serde(tag = "mode", rename_all = "kebab-case", deny_unknown_fields)]
 pub enum L0FlushConfig {
-    PageCached,
     #[serde(rename_all = "snake_case")]
-    Direct {
-        max_concurrency: NonZeroUsize,
-    },
+    Direct { max_concurrency: NonZeroUsize },
 }
 
 impl Default for L0FlushConfig {
@@ -25,14 +20,12 @@ impl Default for L0FlushConfig {
 pub struct L0FlushGlobalState(Arc<Inner>);
 
 pub enum Inner {
-    PageCached,
     Direct { semaphore: tokio::sync::Semaphore },
 }
 
 impl L0FlushGlobalState {
     pub fn new(config: L0FlushConfig) -> Self {
         match config {
-            L0FlushConfig::PageCached => Self(Arc::new(Inner::PageCached)),
             L0FlushConfig::Direct { max_concurrency } => {
                 let semaphore = tokio::sync::Semaphore::new(max_concurrency.get());
                 Self(Arc::new(Inner::Direct { semaphore }))
@@ -44,13 +37,3 @@ impl L0FlushGlobalState {
         &self.0
     }
 }
-
-impl L0FlushConfig {
-    pub(crate) fn prewarm_on_write(&self) -> ephemeral_file::PrewarmPageCacheOnWrite {
-        use L0FlushConfig::*;
-        match self {
-            PageCached => ephemeral_file::PrewarmPageCacheOnWrite::Yes,
-            Direct { .. } => ephemeral_file::PrewarmPageCacheOnWrite::No,
-        }
-    }
-}

pageserver/src/lib.rs

@@ -49,7 +49,7 @@ use tracing::{info, info_span};
 /// backwards-compatible changes to the metadata format.
 pub const STORAGE_FORMAT_VERSION: u16 = 3;
 
-pub const DEFAULT_PG_VERSION: u32 = 15;
+pub const DEFAULT_PG_VERSION: u32 = 16;
 
 // Magic constants used to identify different kinds of files
 pub const IMAGE_FILE_MAGIC: u16 = 0x5A60;

pageserver/src/metrics.rs

@@ -1803,6 +1803,15 @@ pub(crate) static SECONDARY_RESIDENT_PHYSICAL_SIZE: Lazy<UIntGaugeVec> = Lazy::n
     .expect("failed to define a metric")
 });
 
+pub(crate) static SECONDARY_HEATMAP_TOTAL_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
+    register_uint_gauge_vec!(
+        "pageserver_secondary_heatmap_total_size",
+        "The total size in bytes of all layers in the most recently downloaded heatmap.",
+        &["tenant_id", "shard_id"]
+    )
+    .expect("failed to define a metric")
+});
+
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
 pub enum RemoteOpKind {
     Upload,

pageserver/src/task_mgr.rs

@@ -393,7 +393,7 @@ struct PageServerTask {
 
     /// Tasks may optionally be launched for a particular tenant/timeline, enabling
     /// later cancelling tasks for that tenant/timeline in [`shutdown_tasks`]
-    tenant_shard_id: Option<TenantShardId>,
+    tenant_shard_id: TenantShardId,
     timeline_id: Option<TimelineId>,
 
     mutable: Mutex<MutableTaskState>,
@@ -405,7 +405,7 @@ struct PageServerTask {
 pub fn spawn<F>(
     runtime: &tokio::runtime::Handle,
     kind: TaskKind,
-    tenant_shard_id: Option<TenantShardId>,
+    tenant_shard_id: TenantShardId,
     timeline_id: Option<TimelineId>,
     name: &str,
     future: F,
@@ -550,7 +550,7 @@ pub async fn shutdown_tasks(
         let tasks = TASKS.lock().unwrap();
         for task in tasks.values() {
             if (kind.is_none() || Some(task.kind) == kind)
-                && (tenant_shard_id.is_none() || task.tenant_shard_id == tenant_shard_id)
+                && (tenant_shard_id.is_none() || Some(task.tenant_shard_id) == tenant_shard_id)
                 && (timeline_id.is_none() || task.timeline_id == timeline_id)
             {
                 task.cancel.cancel();
@@ -573,13 +573,8 @@ pub async fn shutdown_tasks(
         };
         if let Some(mut join_handle) = join_handle {
             if log_all {
-                if tenant_shard_id.is_none() {
-                    // there are quite few of these
-                    info!(name = task.name, kind = ?task_kind, "stopping global task");
-                } else {
-                    // warn to catch these in tests; there shouldn't be any
-                    warn!(name = task.name, tenant_shard_id = ?tenant_shard_id, timeline_id = ?timeline_id, kind = ?task_kind, "stopping left-over");
-                }
+                // warn to catch these in tests; there shouldn't be any
+                warn!(name = task.name, tenant_shard_id = ?tenant_shard_id, timeline_id = ?timeline_id, kind = ?task_kind, "stopping left-over");
             }
             if tokio::time::timeout(std::time::Duration::from_secs(1), &mut join_handle)
                 .await

pageserver/src/tenant.rs

@@ -798,7 +798,7 @@ impl Tenant {
         task_mgr::spawn(
             &tokio::runtime::Handle::current(),
             TaskKind::Attach,
-            Some(tenant_shard_id),
+            tenant_shard_id,
             None,
             "attach tenant",
             async move {

pageserver/src/tenant/ephemeral_file.rs

@@ -21,7 +21,6 @@ pub struct EphemeralFile {
 }
 
 mod page_caching;
-pub(crate) use page_caching::PrewarmOnWrite as PrewarmPageCacheOnWrite;
 mod zero_padded_read_write;
 
 impl EphemeralFile {
@@ -52,12 +51,10 @@ impl EphemeralFile {
         )
         .await?;
 
-        let prewarm = conf.l0_flush.prewarm_on_write();
-
         Ok(EphemeralFile {
             _tenant_shard_id: tenant_shard_id,
             _timeline_id: timeline_id,
-            rw: page_caching::RW::new(file, prewarm, gate_guard),
+            rw: page_caching::RW::new(file, gate_guard),
         })
     }
 

pageserver/src/tenant/ephemeral_file/page_caching.rs

@@ -1,15 +1,15 @@
 //! Wrapper around [`super::zero_padded_read_write::RW`] that uses the
 //! [`crate::page_cache`] to serve reads that need to go to the underlying [`VirtualFile`].
+//!
+//! Subject to removal in <https://github.com/neondatabase/neon/pull/8537>
 
 use crate::context::RequestContext;
 use crate::page_cache::{self, PAGE_SZ};
 use crate::tenant::block_io::BlockLease;
-use crate::virtual_file::owned_buffers_io::io_buf_ext::FullSlice;
+use crate::virtual_file::owned_buffers_io::util::size_tracking_writer;
 use crate::virtual_file::VirtualFile;
 
-use once_cell::sync::Lazy;
-use std::io::{self, ErrorKind};
-use std::ops::{Deref, Range};
+use std::io::{self};
 use tokio_epoll_uring::BoundedBuf;
 use tracing::*;
 
@@ -18,33 +18,17 @@ use super::zero_padded_read_write;
 /// See module-level comment.
 pub struct RW {
     page_cache_file_id: page_cache::FileId,
-    rw: super::zero_padded_read_write::RW<PreWarmingWriter>,
+    rw: super::zero_padded_read_write::RW<size_tracking_writer::Writer<VirtualFile>>,
     /// Gate guard is held on as long as we need to do operations in the path (delete on drop).
     _gate_guard: utils::sync::gate::GateGuard,
 }
 
-/// When we flush a block to the underlying [`crate::virtual_file::VirtualFile`],
-/// should we pre-warm the [`crate::page_cache`] with the contents?
-#[derive(Clone, Copy)]
-pub enum PrewarmOnWrite {
-    Yes,
-    No,
-}
-
 impl RW {
-    pub fn new(
-        file: VirtualFile,
-        prewarm_on_write: PrewarmOnWrite,
-        _gate_guard: utils::sync::gate::GateGuard,
-    ) -> Self {
+    pub fn new(file: VirtualFile, _gate_guard: utils::sync::gate::GateGuard) -> Self {
         let page_cache_file_id = page_cache::next_file_id();
         Self {
             page_cache_file_id,
-            rw: super::zero_padded_read_write::RW::new(PreWarmingWriter::new(
-                page_cache_file_id,
-                file,
-                prewarm_on_write,
-            )),
+            rw: super::zero_padded_read_write::RW::new(size_tracking_writer::Writer::new(file)),
             _gate_guard,
         }
     }
@@ -84,10 +68,10 @@ impl RW {
         let vec = Vec::with_capacity(size);
 
         // read from disk what we've already flushed
-        let writer = self.rw.as_writer();
-        let flushed_range = writer.written_range();
-        let mut vec = writer
-            .file
+        let file_size_tracking_writer = self.rw.as_writer();
+        let flushed_range = 0..usize::try_from(file_size_tracking_writer.bytes_written()).unwrap();
+        let mut vec = file_size_tracking_writer
+            .as_inner()
             .read_exact_at(
                 vec.slice(0..(flushed_range.end - flushed_range.start)),
                 u64::try_from(flushed_range.start).unwrap(),
@@ -122,7 +106,7 @@ impl RW {
                             format!(
                                 "ephemeral file: read immutable page #{}: {}: {:#}",
                                 blknum,
-                                self.rw.as_writer().file.path,
+                                self.rw.as_writer().as_inner().path,
                                 e,
                             ),
                         )
@@ -132,7 +116,7 @@ impl RW {
                     }
                     page_cache::ReadBufResult::NotFound(write_guard) => {
                         let write_guard = writer
-                            .file
+                            .as_inner()
                             .read_exact_at_page(write_guard, blknum as u64 * PAGE_SZ as u64, ctx)
                             .await?;
                         let read_guard = write_guard.mark_valid();
@@ -154,137 +138,16 @@ impl Drop for RW {
 
         // unlink the file
         // we are clear to do this, because we have entered a gate
-        let res = std::fs::remove_file(&self.rw.as_writer().file.path);
+        let path = &self.rw.as_writer().as_inner().path;
+        let res = std::fs::remove_file(path);
         if let Err(e) = res {
             if e.kind() != std::io::ErrorKind::NotFound {
                 // just never log the not found errors, we cannot do anything for them; on detach
                 // the tenant directory is already gone.
                 //
                 // not found files might also be related to https://github.com/neondatabase/neon/issues/2442
-                error!(
-                    "could not remove ephemeral file '{}': {}",
-                    self.rw.as_writer().file.path,
-                    e
-                );
+                error!("could not remove ephemeral file '{path}': {e}");
             }
         }
     }
 }
-
-struct PreWarmingWriter {
-    prewarm_on_write: PrewarmOnWrite,
-    nwritten_blocks: u32,
-    page_cache_file_id: page_cache::FileId,
-    file: VirtualFile,
-}
-
-impl PreWarmingWriter {
-    fn new(
-        page_cache_file_id: page_cache::FileId,
-        file: VirtualFile,
-        prewarm_on_write: PrewarmOnWrite,
-    ) -> Self {
-        Self {
-            prewarm_on_write,
-            nwritten_blocks: 0,
-            page_cache_file_id,
-            file,
-        }
-    }
-
-    /// Return the byte range within `file` that has been written though `write_all`.
-    ///
-    /// The returned range would be invalidated by another `write_all`. To prevent that, we capture `&_`.
-    fn written_range(&self) -> (impl Deref<Target = Range<usize>> + '_) {
-        let nwritten_blocks = usize::try_from(self.nwritten_blocks).unwrap();
-        struct Wrapper(Range<usize>);
-        impl Deref for Wrapper {
-            type Target = Range<usize>;
-            fn deref(&self) -> &Range<usize> {
-                &self.0
-            }
-        }
-        Wrapper(0..nwritten_blocks * PAGE_SZ)
-    }
-}
-
-impl crate::virtual_file::owned_buffers_io::write::OwnedAsyncWriter for PreWarmingWriter {
-    async fn write_all<Buf: tokio_epoll_uring::IoBuf + Send>(
-        &mut self,
-        buf: FullSlice<Buf>,
-        ctx: &RequestContext,
-    ) -> std::io::Result<(usize, FullSlice<Buf>)> {
-        let buflen = buf.len();
-        assert_eq!(
-            buflen % PAGE_SZ,
-            0,
-            "{buflen} ; we know TAIL_SZ is a PAGE_SZ multiple, and write_buffered_borrowed is used"
-        );
-
-        // Do the IO.
-        let buf = match self.file.write_all(buf, ctx).await {
-            (buf, Ok(nwritten)) => {
-                assert_eq!(nwritten, buflen);
-                buf
-            }
-            (_, Err(e)) => {
-                return Err(std::io::Error::new(
-                    ErrorKind::Other,
-                    // order error before path because path is long and error is short
-                    format!(
-                        "ephemeral_file: write_blob: write-back tail self.nwritten_blocks={}, buflen={}, {:#}: {}",
-                        self.nwritten_blocks, buflen, e, self.file.path,
-                    ),
-                ));
-            }
-        };
-
-        let nblocks = buflen / PAGE_SZ;
-        let nblocks32 = u32::try_from(nblocks).unwrap();
-
-        if matches!(self.prewarm_on_write, PrewarmOnWrite::Yes) {
-            // Pre-warm page cache with the contents.
-            // At least in isolated bulk ingest benchmarks (test_bulk_insert.py), the pre-warming
-            // benefits the code that writes InMemoryLayer=>L0 layers.
-
-            let cache = page_cache::get();
-            static CTX: Lazy<RequestContext> = Lazy::new(|| {
-                RequestContext::new(
-                    crate::task_mgr::TaskKind::EphemeralFilePreWarmPageCache,
-                    crate::context::DownloadBehavior::Error,
-                )
-            });
-            for blknum_in_buffer in 0..nblocks {
-                let blk_in_buffer =
-                    &buf[blknum_in_buffer * PAGE_SZ..(blknum_in_buffer + 1) * PAGE_SZ];
-                let blknum = self
-                    .nwritten_blocks
-                    .checked_add(blknum_in_buffer as u32)
-                    .unwrap();
-                match cache
-                    .read_immutable_buf(self.page_cache_file_id, blknum, &CTX)
-                    .await
-                {
-                    Err(e) => {
-                        error!("ephemeral_file write_blob failed to get immutable buf to pre-warm page cache: {e:?}");
-                        // fail gracefully, it's not the end of the world if we can't pre-warm the cache here
-                    }
-                    Ok(v) => match v {
-                        page_cache::ReadBufResult::Found(_guard) => {
-                            // This function takes &mut self, so, it shouldn't be possible to reach this point.
-                            unreachable!("we just wrote block {blknum} to the VirtualFile, which is owned by Self, \
-                                      and this function takes &mut self, so, no concurrent read_blk is possible");
-                        }
-                        page_cache::ReadBufResult::NotFound(mut write_guard) => {
-                            write_guard.copy_from_slice(blk_in_buffer);
-                            let _ = write_guard.mark_valid();
-                        }
-                    },
-                }
-            }
-        }
-
-        self.nwritten_blocks = self.nwritten_blocks.checked_add(nblocks32).unwrap();
-        Ok((buflen, buf))
-    }
-}

pageserver/src/tenant/metadata.rs

@@ -565,7 +565,7 @@ mod tests {
         );
         let expected_bytes = vec![
             /* TimelineMetadataHeader */
-            4, 37, 101, 34, 0, 70, 0, 4, // checksum, size, format_version (4 + 2 + 2)
+            74, 104, 158, 105, 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)
@@ -574,7 +574,7 @@ mod tests {
             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)
+            0, 0, 0, 16, // 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,

pageserver/src/tenant/remote_timeline_client.rs

@@ -1728,7 +1728,7 @@ impl RemoteTimelineClient {
             task_mgr::spawn(
                 &self.runtime,
                 TaskKind::RemoteUploadTask,
-                Some(self.tenant_shard_id),
+                self.tenant_shard_id,
                 Some(self.timeline_id),
                 "remote upload",
                 async move {

pageserver/src/tenant/secondary.rs

@@ -8,6 +8,7 @@ use std::{sync::Arc, time::SystemTime};
 use crate::{
     context::RequestContext,
     disk_usage_eviction_task::DiskUsageEvictionInfo,
+    metrics::SECONDARY_HEATMAP_TOTAL_SIZE,
     task_mgr::{self, TaskKind, BACKGROUND_RUNTIME},
 };
 
@@ -105,6 +106,9 @@ pub(crate) struct SecondaryTenant {
 
     // Sum of layer sizes on local disk
     pub(super) resident_size_metric: UIntGauge,
+
+    // Sum of layer sizes in the most recently downloaded heatmap
+    pub(super) heatmap_total_size_metric: UIntGauge,
 }
 
 impl Drop for SecondaryTenant {
@@ -112,6 +116,7 @@ impl Drop for SecondaryTenant {
         let tenant_id = self.tenant_shard_id.tenant_id.to_string();
         let shard_id = format!("{}", self.tenant_shard_id.shard_slug());
         let _ = SECONDARY_RESIDENT_PHYSICAL_SIZE.remove_label_values(&[&tenant_id, &shard_id]);
+        let _ = SECONDARY_HEATMAP_TOTAL_SIZE.remove_label_values(&[&tenant_id, &shard_id]);
     }
 }
 
@@ -128,6 +133,10 @@ impl SecondaryTenant {
             .get_metric_with_label_values(&[&tenant_id, &shard_id])
             .unwrap();
 
+        let heatmap_total_size_metric = SECONDARY_HEATMAP_TOTAL_SIZE
+            .get_metric_with_label_values(&[&tenant_id, &shard_id])
+            .unwrap();
+
         Arc::new(Self {
             tenant_shard_id,
             // todo: shall we make this a descendent of the
@@ -145,6 +154,7 @@ impl SecondaryTenant {
             progress: std::sync::Mutex::default(),
 
             resident_size_metric,
+            heatmap_total_size_metric,
         })
     }
 

pageserver/src/tenant/secondary/downloader.rs

@@ -829,6 +829,12 @@ impl<'a> TenantDownloader<'a> {
             layers_downloaded: 0,
             bytes_downloaded: 0,
         };
+
+        // Also expose heatmap bytes_total as a metric
+        self.secondary_state
+            .heatmap_total_size_metric
+            .set(heatmap_stats.bytes);
+
         // Accumulate list of things to delete while holding the detail lock, for execution after dropping the lock
         let mut delete_layers = Vec::new();
         let mut delete_timelines = Vec::new();

pageserver/src/tenant/secondary/heatmap.rs

@@ -29,16 +29,16 @@ pub(super) struct HeatMapTenant {
 #[derive(Serialize, Deserialize)]
 pub(crate) struct HeatMapTimeline {
     #[serde_as(as = "DisplayFromStr")]
-    pub(super) timeline_id: TimelineId,
+    pub(crate) timeline_id: TimelineId,
 
-    pub(super) layers: Vec<HeatMapLayer>,
+    pub(crate) layers: Vec<HeatMapLayer>,
 }
 
 #[serde_as]
 #[derive(Serialize, Deserialize)]
 pub(crate) struct HeatMapLayer {
-    pub(super) name: LayerName,
-    pub(super) metadata: LayerFileMetadata,
+    pub(crate) name: LayerName,
+    pub(crate) metadata: LayerFileMetadata,
 
     #[serde_as(as = "TimestampSeconds<i64>")]
     pub(super) access_time: SystemTime,

pageserver/src/tenant/storage_layer/inmemory_layer.rs

@@ -13,7 +13,7 @@ use crate::tenant::ephemeral_file::EphemeralFile;
 use crate::tenant::timeline::GetVectoredError;
 use crate::tenant::PageReconstructError;
 use crate::virtual_file::owned_buffers_io::io_buf_ext::IoBufExt;
-use crate::{l0_flush, page_cache, walrecord};
+use crate::{l0_flush, page_cache};
 use anyhow::{anyhow, Result};
 use camino::Utf8PathBuf;
 use pageserver_api::key::CompactKey;
@@ -249,9 +249,7 @@ impl InMemoryLayer {
     /// debugging function to print out the contents of the layer
     ///
     /// this is likely completly unused
-    pub async fn dump(&self, verbose: bool, ctx: &RequestContext) -> Result<()> {
-        let inner = self.inner.read().await;
-
+    pub async fn dump(&self, _verbose: bool, _ctx: &RequestContext) -> Result<()> {
         let end_str = self.end_lsn_or_max();
 
         println!(
@@ -259,39 +257,6 @@ impl InMemoryLayer {
             self.timeline_id, self.start_lsn, end_str,
         );
 
-        if !verbose {
-            return Ok(());
-        }
-
-        let cursor = inner.file.block_cursor();
-        let mut buf = Vec::new();
-        for (key, vec_map) in inner.index.iter() {
-            for (lsn, pos) in vec_map.as_slice() {
-                let mut desc = String::new();
-                cursor.read_blob_into_buf(*pos, &mut buf, ctx).await?;
-                let val = Value::des(&buf);
-                match val {
-                    Ok(Value::Image(img)) => {
-                        write!(&mut desc, " img {} bytes", img.len())?;
-                    }
-                    Ok(Value::WalRecord(rec)) => {
-                        let wal_desc = walrecord::describe_wal_record(&rec).unwrap();
-                        write!(
-                            &mut desc,
-                            " rec {} bytes will_init: {} {}",
-                            buf.len(),
-                            rec.will_init(),
-                            wal_desc
-                        )?;
-                    }
-                    Err(err) => {
-                        write!(&mut desc, " DESERIALIZATION ERROR: {}", err)?;
-                    }
-                }
-                println!("  key {} at {}: {}", key, lsn, desc);
-            }
-        }
-
         Ok(())
     }
 
@@ -536,7 +501,6 @@ impl InMemoryLayer {
 
         use l0_flush::Inner;
         let _concurrency_permit = match l0_flush_global_state {
-            Inner::PageCached => None,
             Inner::Direct { semaphore, .. } => Some(semaphore.acquire().await),
         };
 
@@ -568,34 +532,6 @@ impl InMemoryLayer {
         .await?;
 
         match l0_flush_global_state {
-            l0_flush::Inner::PageCached => {
-                let ctx = RequestContextBuilder::extend(ctx)
-                    .page_content_kind(PageContentKind::InMemoryLayer)
-                    .build();
-
-                let mut buf = Vec::new();
-
-                let cursor = inner.file.block_cursor();
-
-                for (key, vec_map) in inner.index.iter() {
-                    // Write all page versions
-                    for (lsn, pos) in vec_map.as_slice() {
-                        cursor.read_blob_into_buf(*pos, &mut buf, &ctx).await?;
-                        let will_init = Value::des(&buf)?.will_init();
-                        let (tmp, res) = delta_layer_writer
-                            .put_value_bytes(
-                                Key::from_compact(*key),
-                                *lsn,
-                                buf.slice_len(),
-                                will_init,
-                                &ctx,
-                            )
-                            .await;
-                        res?;
-                        buf = tmp.into_raw_slice().into_inner();
-                    }
-                }
-            }
             l0_flush::Inner::Direct { .. } => {
                 let file_contents: Vec<u8> = inner.file.load_to_vec(ctx).await?;
                 assert_eq!(

pageserver/src/tenant/storage_layer/split_writer.rs

@@ -208,6 +208,8 @@ impl SplitDeltaLayerWriter {
 
 #[cfg(test)]
 mod tests {
+    use rand::{RngCore, SeedableRng};
+
     use crate::{
         tenant::{
             harness::{TenantHarness, TIMELINE_ID},
@@ -229,7 +231,10 @@ mod tests {
     }
 
     fn get_large_img() -> Bytes {
-        vec![0; 8192].into()
+        let mut rng = rand::rngs::SmallRng::seed_from_u64(42);
+        let mut data = vec![0; 8192];
+        rng.fill_bytes(&mut data);
+        data.into()
     }
 
     #[tokio::test]

pageserver/src/tenant/tasks.rs

@@ -98,7 +98,7 @@ pub fn start_background_loops(
     task_mgr::spawn(
         BACKGROUND_RUNTIME.handle(),
         TaskKind::Compaction,
-        Some(tenant_shard_id),
+        tenant_shard_id,
         None,
         &format!("compactor for tenant {tenant_shard_id}"),
         {
@@ -121,7 +121,7 @@ pub fn start_background_loops(
     task_mgr::spawn(
         BACKGROUND_RUNTIME.handle(),
         TaskKind::GarbageCollector,
-        Some(tenant_shard_id),
+        tenant_shard_id,
         None,
         &format!("garbage collector for tenant {tenant_shard_id}"),
         {
@@ -144,7 +144,7 @@ pub fn start_background_loops(
     task_mgr::spawn(
         BACKGROUND_RUNTIME.handle(),
         TaskKind::IngestHousekeeping,
-        Some(tenant_shard_id),
+        tenant_shard_id,
         None,
         &format!("ingest housekeeping for tenant {tenant_shard_id}"),
         {

pageserver/src/tenant/timeline.rs

@@ -1645,6 +1645,20 @@ impl Timeline {
         self.last_record_lsn.shutdown();
 
         if try_freeze_and_flush {
+            if let Some((open, frozen)) = self
+                .layers
+                .read()
+                .await
+                .layer_map()
+                .map(|lm| (lm.open_layer.is_some(), lm.frozen_layers.len()))
+                .ok()
+                .filter(|(open, frozen)| *open || *frozen > 0)
+            {
+                tracing::info!(?open, frozen, "flushing and freezing on shutdown");
+            } else {
+                // this is double-shutdown, ignore it
+            }
+
             // we shut down walreceiver above, so, we won't add anything more
             // to the InMemoryLayer; freeze it and wait for all frozen layers
             // to reach the disk & upload queue, then shut the upload queue and
@@ -2267,7 +2281,7 @@ impl Timeline {
         task_mgr::spawn(
             task_mgr::BACKGROUND_RUNTIME.handle(),
             task_mgr::TaskKind::LayerFlushTask,
-            Some(self.tenant_shard_id),
+            self.tenant_shard_id,
             Some(self.timeline_id),
             "layer flush task",
             async move {
@@ -2621,7 +2635,7 @@ impl Timeline {
         task_mgr::spawn(
             task_mgr::BACKGROUND_RUNTIME.handle(),
             task_mgr::TaskKind::InitialLogicalSizeCalculation,
-            Some(self.tenant_shard_id),
+            self.tenant_shard_id,
             Some(self.timeline_id),
             "initial size calculation",
             // NB: don't log errors here, task_mgr will do that.
@@ -2789,7 +2803,7 @@ impl Timeline {
         task_mgr::spawn(
             task_mgr::BACKGROUND_RUNTIME.handle(),
             task_mgr::TaskKind::OndemandLogicalSizeCalculation,
-            Some(self.tenant_shard_id),
+            self.tenant_shard_id,
             Some(self.timeline_id),
             "ondemand logical size calculation",
             async move {
@@ -2963,11 +2977,7 @@ impl Timeline {
                 LayerVisibilityHint::Visible => {
                     // Layer is visible to one or more read LSNs: elegible for inclusion in layer map
                     let last_activity_ts = layer.latest_activity();
-                    Some(HeatMapLayer::new(
-                        layer.layer_desc().layer_name(),
-                        layer.metadata(),
-                        last_activity_ts,
-                    ))
+                    Some((layer.layer_desc(), layer.metadata(), last_activity_ts))
                 }
                 LayerVisibilityHint::Covered => {
                     // Layer is resident but unlikely to be read: not elegible for inclusion in heatmap.
@@ -2976,7 +2986,23 @@ impl Timeline {
             }
         });
 
-        let layers = resident.collect();
+        let mut layers = resident.collect::<Vec<_>>();
+
+        // Sort layers in order of which to download first.  For a large set of layers to download, we
+        // want to prioritize those layers which are most likely to still be in the resident many minutes
+        // or hours later:
+        // - Download L0s last, because they churn the fastest: L0s on a fast-writing tenant might
+        //   only exist for a few minutes before being compacted into L1s.
+        // - For L1 & image layers, download most recent LSNs first: the older the LSN, the sooner
+        //   the layer is likely to be covered by an image layer during compaction.
+        layers.sort_by_key(|(desc, _meta, _atime)| {
+            std::cmp::Reverse((!LayerMap::is_l0(&desc.key_range), desc.lsn_range.end))
+        });
+
+        let layers = layers
+            .into_iter()
+            .map(|(desc, meta, atime)| HeatMapLayer::new(desc.layer_name(), meta, atime))
+            .collect();
 
         Some(HeatMapTimeline::new(self.timeline_id, layers))
     }
@@ -4502,6 +4528,7 @@ impl DurationRecorder {
 /// the layer descriptor requires the user to provide the ranges, which should cover all
 /// keys specified in the `data` field.
 #[cfg(test)]
+#[derive(Clone)]
 pub struct DeltaLayerTestDesc {
     pub lsn_range: Range<Lsn>,
     pub key_range: Range<Key>,
@@ -4531,6 +4558,13 @@ impl DeltaLayerTestDesc {
             data,
         }
     }
+
+    pub(crate) fn layer_name(&self) -> LayerName {
+        LayerName::Delta(super::storage_layer::DeltaLayerName {
+            key_range: self.key_range.clone(),
+            lsn_range: self.lsn_range.clone(),
+        })
+    }
 }
 
 impl Timeline {
@@ -5128,7 +5162,7 @@ impl Timeline {
         let task_id = task_mgr::spawn(
             task_mgr::BACKGROUND_RUNTIME.handle(),
             task_mgr::TaskKind::DownloadAllRemoteLayers,
-            Some(self.tenant_shard_id),
+            self.tenant_shard_id,
             Some(self.timeline_id),
             "download all remote layers task",
             async move {
@@ -5754,12 +5788,110 @@ fn is_send() {
 
 #[cfg(test)]
 mod tests {
+    use pageserver_api::key::Key;
     use utils::{id::TimelineId, lsn::Lsn};
 
-    use crate::tenant::{
-        harness::TenantHarness, storage_layer::Layer, timeline::EvictionError, Timeline,
+    use crate::{
+        repository::Value,
+        tenant::{
+            harness::{test_img, TenantHarness},
+            layer_map::LayerMap,
+            storage_layer::{Layer, LayerName},
+            timeline::{DeltaLayerTestDesc, EvictionError},
+            Timeline,
+        },
     };
 
+    #[tokio::test]
+    async fn test_heatmap_generation() {
+        let harness = TenantHarness::create("heatmap_generation").await.unwrap();
+
+        let covered_delta = DeltaLayerTestDesc::new_with_inferred_key_range(
+            Lsn(0x10)..Lsn(0x20),
+            vec![(
+                Key::from_hex("620000000033333333444444445500000000").unwrap(),
+                Lsn(0x11),
+                Value::Image(test_img("foo")),
+            )],
+        );
+        let visible_delta = DeltaLayerTestDesc::new_with_inferred_key_range(
+            Lsn(0x10)..Lsn(0x20),
+            vec![(
+                Key::from_hex("720000000033333333444444445500000000").unwrap(),
+                Lsn(0x11),
+                Value::Image(test_img("foo")),
+            )],
+        );
+        let l0_delta = DeltaLayerTestDesc::new(
+            Lsn(0x20)..Lsn(0x30),
+            Key::from_hex("000000000000000000000000000000000000").unwrap()
+                ..Key::from_hex("FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF").unwrap(),
+            vec![(
+                Key::from_hex("720000000033333333444444445500000000").unwrap(),
+                Lsn(0x25),
+                Value::Image(test_img("foo")),
+            )],
+        );
+        let delta_layers = vec![
+            covered_delta.clone(),
+            visible_delta.clone(),
+            l0_delta.clone(),
+        ];
+
+        let image_layer = (
+            Lsn(0x40),
+            vec![(
+                Key::from_hex("620000000033333333444444445500000000").unwrap(),
+                test_img("bar"),
+            )],
+        );
+        let image_layers = vec![image_layer];
+
+        let (tenant, ctx) = harness.load().await;
+        let timeline = tenant
+            .create_test_timeline_with_layers(
+                TimelineId::generate(),
+                Lsn(0x10),
+                14,
+                &ctx,
+                delta_layers,
+                image_layers,
+                Lsn(0x100),
+            )
+            .await
+            .unwrap();
+
+        // Layer visibility is an input to heatmap generation, so refresh it first
+        timeline.update_layer_visibility().await.unwrap();
+
+        let heatmap = timeline
+            .generate_heatmap()
+            .await
+            .expect("Infallible while timeline is not shut down");
+
+        assert_eq!(heatmap.timeline_id, timeline.timeline_id);
+
+        // L0 should come last
+        assert_eq!(heatmap.layers.last().unwrap().name, l0_delta.layer_name());
+
+        let mut last_lsn = Lsn::MAX;
+        for layer in heatmap.layers {
+            // Covered layer should be omitted
+            assert!(layer.name != covered_delta.layer_name());
+
+            let layer_lsn = match &layer.name {
+                LayerName::Delta(d) => d.lsn_range.end,
+                LayerName::Image(i) => i.lsn,
+            };
+
+            // Apart from L0s, newest Layers should come first
+            if !LayerMap::is_l0(layer.name.key_range()) {
+                assert!(layer_lsn <= last_lsn);
+                last_lsn = layer_lsn;
+            }
+        }
+    }
+
     #[tokio::test]
     async fn two_layer_eviction_attempts_at_the_same_time() {
         let harness = TenantHarness::create("two_layer_eviction_attempts_at_the_same_time")

pageserver/src/tenant/timeline/delete.rs

@@ -395,7 +395,7 @@ impl DeleteTimelineFlow {
         task_mgr::spawn(
             task_mgr::BACKGROUND_RUNTIME.handle(),
             TaskKind::TimelineDeletionWorker,
-            Some(tenant_shard_id),
+            tenant_shard_id,
             Some(timeline_id),
             "timeline_delete",
             async move {

pageserver/src/tenant/timeline/eviction_task.rs

@@ -60,7 +60,7 @@ impl Timeline {
         task_mgr::spawn(
             BACKGROUND_RUNTIME.handle(),
             TaskKind::Eviction,
-            Some(self.tenant_shard_id),
+            self.tenant_shard_id,
             Some(self.timeline_id),
             &format!(
                 "layer eviction for {}/{}",

pageserver/src/virtual_file.rs

@@ -756,11 +756,23 @@ impl VirtualFile {
         })
     }
 
+    /// The function aborts the process if the error is fatal.
     async fn write_at<B: IoBuf + Send>(
         &self,
         buf: FullSlice<B>,
         offset: u64,
         _ctx: &RequestContext, /* TODO: use for metrics: https://github.com/neondatabase/neon/issues/6107 */
+    ) -> (FullSlice<B>, Result<usize, Error>) {
+        let (slice, result) = self.write_at_inner(buf, offset, _ctx).await;
+        let result = result.maybe_fatal_err("write_at");
+        (slice, result)
+    }
+
+    async fn write_at_inner<B: IoBuf + Send>(
+        &self,
+        buf: FullSlice<B>,
+        offset: u64,
+        _ctx: &RequestContext, /* TODO: use for metrics: https://github.com/neondatabase/neon/issues/6107 */
     ) -> (FullSlice<B>, Result<usize, Error>) {
         let file_guard = match self.lock_file().await {
             Ok(file_guard) => file_guard,

pageserver/src/walingest.rs

@@ -235,6 +235,7 @@ impl WalIngest {
                         &parsed_xact,
                         info == pg_constants::XLOG_XACT_COMMIT,
                         decoded.origin_id,
+                        lsn,
                         ctx,
                     )
                     .await?;
@@ -248,6 +249,7 @@ impl WalIngest {
                         &parsed_xact,
                         info == pg_constants::XLOG_XACT_COMMIT_PREPARED,
                         decoded.origin_id,
+                        lsn,
                         ctx,
                     )
                     .await?;
@@ -408,11 +410,16 @@ impl WalIngest {
                 let info = decoded.xl_info & pg_constants::XLR_RMGR_INFO_MASK;
                 if info == pg_constants::XLOG_REPLORIGIN_SET {
                     let xlrec = crate::walrecord::XlReploriginSet::decode(&mut buf);
+                    info!(
+                        "Set replication origin_id={}, origin_lsn={}",
+                        xlrec.node_id, xlrec.remote_lsn
+                    );
                     modification
                         .set_replorigin(xlrec.node_id, xlrec.remote_lsn)
                         .await?
                 } else if info == pg_constants::XLOG_REPLORIGIN_DROP {
                     let xlrec = crate::walrecord::XlReploriginDrop::decode(&mut buf);
+                    info!("Drop replication origin_id={}", xlrec.node_id);
                     modification.drop_replorigin(xlrec.node_id).await?
                 }
             }
@@ -1220,6 +1227,7 @@ impl WalIngest {
         parsed: &XlXactParsedRecord,
         is_commit: bool,
         origin_id: u16,
+        lsn: Lsn,
         ctx: &RequestContext,
     ) -> anyhow::Result<()> {
         // Record update of CLOG pages
@@ -1286,6 +1294,10 @@ impl WalIngest {
             }
         }
         if origin_id != 0 {
+            info!(
+                "Commit at {} origin_id={}, origin_lsn={}",
+                lsn, origin_id, parsed.origin_lsn
+            );
             modification
                 .set_replorigin(origin_id, parsed.origin_lsn)
                 .await?;

pgxn/neon/file_cache.c

@@ -41,6 +41,8 @@
 
 #include "hll.h"
 
+#define CriticalAssert(cond) do if (!(cond)) elog(PANIC, "Assertion %s failed at %s:%d: ", #cond, __FILE__, __LINE__); while (0)
+
 /*
  * Local file cache is used to temporary store relations pages in local file system.
  * All blocks of all relations are stored inside one file and addressed using shared hash map.
@@ -51,19 +53,43 @@
  *
  * Cache is always reconstructed at node startup, so we do not need to save mapping somewhere and worry about
  * its consistency.
+
+ *
+ * ## Holes
+ *
+ * The LFC can be resized on the fly, up to a maximum size that's determined
+ * at server startup (neon.max_file_cache_size). After server startup, we
+ * expand the underlying file when needed, until it reaches the soft limit
+ * (neon.file_cache_size_limit). If the soft limit is later reduced, we shrink
+ * the LFC by punching holes in the underlying file with a
+ * fallocate(FALLOC_FL_PUNCH_HOLE) call. The nominal size of the file doesn't
+ * shrink, but the disk space it uses does.
+ *
+ * Each hole is tracked by a dummy FileCacheEntry, which are kept in the
+ * 'holes' linked list. They are entered into the chunk hash table, with a
+ * special key where the blockNumber is used to store the 'offset' of the
+ * hole, and all other fields are zero. Holes are never looked up in the hash
+ * table, we only enter them there to have a FileCacheEntry that we can keep
+ * in the linked list. If the soft limit is raised again, we reuse the holes
+ * before extending the nominal size of the file.
  */
 
 /* Local file storage allocation chunk.
- * Should be power of two and not less than 32. Using larger than page chunks can
+ * Should be power of two. Using larger than page chunks can
  * 1. Reduce hash-map memory footprint: 8TB database contains billion pages
  *    and size of hash entry is 40 bytes, so we need 40Gb just for hash map.
  *    1Mb chunks can reduce hash map size to 320Mb.
  * 2. Improve access locality, subsequent pages will be allocated together improving seqscan speed
  */
 #define BLOCKS_PER_CHUNK	128 /* 1Mb chunk */
+/*
+ * Smaller chunk seems to be better for OLTP workload
+ */
+// #define BLOCKS_PER_CHUNK	8 /* 64kb chunk */
 #define MB					((uint64)1024*1024)
 
 #define SIZE_MB_TO_CHUNKS(size) ((uint32)((size) * MB / BLCKSZ / BLOCKS_PER_CHUNK))
+#define CHUNK_BITMAP_SIZE ((BLOCKS_PER_CHUNK + 31) / 32)
 
 typedef struct FileCacheEntry
 {
@@ -71,8 +97,8 @@ typedef struct FileCacheEntry
 	uint32		hash;
 	uint32		offset;
 	uint32		access_count;
-	uint32		bitmap[BLOCKS_PER_CHUNK / 32];
-	dlist_node	lru_node;		/* LRU list node */
+	uint32		bitmap[CHUNK_BITMAP_SIZE];
+	dlist_node	list_node;		/* LRU/holes list node */
 } FileCacheEntry;
 
 typedef struct FileCacheControl
@@ -87,6 +113,7 @@ typedef struct FileCacheControl
 	uint64		writes;
 	dlist_head	lru;			/* double linked list for LRU replacement
 								 * algorithm */
+	dlist_head  holes;          /* double linked list of punched holes */
 	HyperLogLogState wss_estimation; /* estimation of working set size */
 } FileCacheControl;
 
@@ -135,6 +162,7 @@ lfc_disable(char const *op)
 		lfc_ctl->used = 0;
 		lfc_ctl->limit = 0;
 		dlist_init(&lfc_ctl->lru);
+		dlist_init(&lfc_ctl->holes);
 
 		if (lfc_desc > 0)
 		{
@@ -214,18 +242,18 @@ lfc_shmem_startup(void)
 	if (!found)
 	{
 		int			fd;
-		uint32		lfc_size = SIZE_MB_TO_CHUNKS(lfc_max_size);
+		uint32		n_chunks = SIZE_MB_TO_CHUNKS(lfc_max_size);
 
 		lfc_lock = (LWLockId) GetNamedLWLockTranche("lfc_lock");
 		info.keysize = sizeof(BufferTag);
 		info.entrysize = sizeof(FileCacheEntry);
 
 		/*
-		 * lfc_size+1 because we add new element to hash table before eviction
+		 * n_chunks+1 because we add new element to hash table before eviction
 		 * of victim
 		 */
 		lfc_hash = ShmemInitHash("lfc_hash",
-								 lfc_size + 1, lfc_size + 1,
+								 n_chunks + 1, n_chunks + 1,
 								 &info,
 								 HASH_ELEM | HASH_BLOBS);
 		lfc_ctl->generation = 0;
@@ -235,6 +263,7 @@ lfc_shmem_startup(void)
 		lfc_ctl->misses = 0;
 		lfc_ctl->writes = 0;
 		dlist_init(&lfc_ctl->lru);
+		dlist_init(&lfc_ctl->holes);
 
 		/* Initialize hyper-log-log structure for estimating working set size */
 		initSHLL(&lfc_ctl->wss_estimation);
@@ -310,14 +339,31 @@ lfc_change_limit_hook(int newval, void *extra)
 		 * Shrink cache by throwing away least recently accessed chunks and
 		 * returning their space to file system
 		 */
-		FileCacheEntry *victim = dlist_container(FileCacheEntry, lru_node, dlist_pop_head_node(&lfc_ctl->lru));
+		FileCacheEntry *victim = dlist_container(FileCacheEntry, list_node, dlist_pop_head_node(&lfc_ctl->lru));
+		FileCacheEntry *hole;
+		uint32		offset = victim->offset;
+		uint32		hash;
+		bool		found;
+		BufferTag	holetag;
 
-		Assert(victim->access_count == 0);
+		CriticalAssert(victim->access_count == 0);
 #ifdef FALLOC_FL_PUNCH_HOLE
 		if (fallocate(lfc_desc, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, (off_t) victim->offset * BLOCKS_PER_CHUNK * BLCKSZ, BLOCKS_PER_CHUNK * BLCKSZ) < 0)
 			neon_log(LOG, "Failed to punch hole in file: %m");
 #endif
+		/* We remove the old entry, and re-enter a hole to the hash table */
 		hash_search_with_hash_value(lfc_hash, &victim->key, victim->hash, HASH_REMOVE, NULL);
+
+		memset(&holetag, 0, sizeof(holetag));
+		holetag.blockNum = offset;
+		hash = get_hash_value(lfc_hash, &holetag);
+		hole = hash_search_with_hash_value(lfc_hash, &holetag, hash, HASH_ENTER, &found);
+		hole->hash = hash;
+		hole->offset = offset;
+		hole->access_count = 0;
+		CriticalAssert(!found);
+		dlist_push_tail(&lfc_ctl->holes, &hole->list_node);
+
 		lfc_ctl->used -= 1;
 	}
 	lfc_ctl->limit = new_size;
@@ -409,6 +455,8 @@ lfc_cache_contains(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno)
 	CopyNRelFileInfoToBufTag(tag, rinfo);
 	tag.forkNum = forkNum;
 	tag.blockNum = blkno & ~(BLOCKS_PER_CHUNK - 1);
+
+	CriticalAssert(BufTagGetRelNumber(&tag) != InvalidRelFileNumber);
 	hash = get_hash_value(lfc_hash, &tag);
 
 	LWLockAcquire(lfc_lock, LW_SHARED);
@@ -440,6 +488,7 @@ lfc_evict(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno)
 	tag.forkNum = forkNum;
 	tag.blockNum = (blkno & ~(BLOCKS_PER_CHUNK - 1));
 
+	CriticalAssert(BufTagGetRelNumber(&tag) != InvalidRelFileNumber);
 	hash = get_hash_value(lfc_hash, &tag);
 
 	LWLockAcquire(lfc_lock, LW_EXCLUSIVE);
@@ -470,7 +519,7 @@ lfc_evict(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno)
 	{
 		bool		has_remaining_pages;
 
-		for (int i = 0; i < (BLOCKS_PER_CHUNK / 32); i++)
+		for (int i = 0; i < CHUNK_BITMAP_SIZE; i++)
 		{
 			if (entry->bitmap[i] != 0)
 			{
@@ -485,8 +534,8 @@ lfc_evict(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno)
 		 */
 		if (!has_remaining_pages)
 		{
-			dlist_delete(&entry->lru_node);
-			dlist_push_head(&lfc_ctl->lru, &entry->lru_node);
+			dlist_delete(&entry->list_node);
+			dlist_push_head(&lfc_ctl->lru, &entry->list_node);
 		}
 	}
 
@@ -525,6 +574,8 @@ lfc_read(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
 	CopyNRelFileInfoToBufTag(tag, rinfo);
 	tag.forkNum = forkNum;
 	tag.blockNum = blkno & ~(BLOCKS_PER_CHUNK - 1);
+
+	CriticalAssert(BufTagGetRelNumber(&tag) != InvalidRelFileNumber);
 	hash = get_hash_value(lfc_hash, &tag);
 
 	LWLockAcquire(lfc_lock, LW_EXCLUSIVE);
@@ -551,7 +602,7 @@ lfc_read(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
 	}
 	/* Unlink entry from LRU list to pin it for the duration of IO operation */
 	if (entry->access_count++ == 0)
-		dlist_delete(&entry->lru_node);
+		dlist_delete(&entry->list_node);
 	generation = lfc_ctl->generation;
 	entry_offset = entry->offset;
 
@@ -569,12 +620,12 @@ lfc_read(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
 
 	if (lfc_ctl->generation == generation)
 	{
-		Assert(LFC_ENABLED());
+		CriticalAssert(LFC_ENABLED());
 		lfc_ctl->hits += 1;
 		pgBufferUsage.file_cache.hits += 1;
-		Assert(entry->access_count > 0);
+		CriticalAssert(entry->access_count > 0);
 		if (--entry->access_count == 0)
-			dlist_push_tail(&lfc_ctl->lru, &entry->lru_node);
+			dlist_push_tail(&lfc_ctl->lru, &entry->list_node);
 	}
 	else
 		result = false;
@@ -613,6 +664,8 @@ lfc_write(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno, const void
 	tag.forkNum = forkNum;
 	tag.blockNum = blkno & ~(BLOCKS_PER_CHUNK - 1);
 	CopyNRelFileInfoToBufTag(tag, rinfo);
+
+	CriticalAssert(BufTagGetRelNumber(&tag) != InvalidRelFileNumber);
 	hash = get_hash_value(lfc_hash, &tag);
 
 	LWLockAcquire(lfc_lock, LW_EXCLUSIVE);
@@ -632,7 +685,7 @@ lfc_write(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno, const void
 		 * operation
 		 */
 		if (entry->access_count++ == 0)
-			dlist_delete(&entry->lru_node);
+			dlist_delete(&entry->list_node);
 	}
 	else
 	{
@@ -655,13 +708,26 @@ lfc_write(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno, const void
 		if (lfc_ctl->used >= lfc_ctl->limit && !dlist_is_empty(&lfc_ctl->lru))
 		{
 			/* Cache overflow: evict least recently used chunk */
-			FileCacheEntry *victim = dlist_container(FileCacheEntry, lru_node, dlist_pop_head_node(&lfc_ctl->lru));
+			FileCacheEntry *victim = dlist_container(FileCacheEntry, list_node, dlist_pop_head_node(&lfc_ctl->lru));
 
-			Assert(victim->access_count == 0);
+			CriticalAssert(victim->access_count == 0);
 			entry->offset = victim->offset; /* grab victim's chunk */
 			hash_search_with_hash_value(lfc_hash, &victim->key, victim->hash, HASH_REMOVE, NULL);
 			neon_log(DEBUG2, "Swap file cache page");
 		}
+		else if (!dlist_is_empty(&lfc_ctl->holes))
+		{
+			/* We can reuse a hole that was left behind when the LFC was shrunk previously */
+			FileCacheEntry *hole = dlist_container(FileCacheEntry, list_node, dlist_pop_head_node(&lfc_ctl->holes));
+			uint32		offset = hole->offset;
+			bool		found;
+
+			hash_search_with_hash_value(lfc_hash, &hole->key, hole->hash, HASH_REMOVE, &found);
+			CriticalAssert(found);
+
+			lfc_ctl->used += 1;
+			entry->offset = offset;	/* reuse the hole */
+		}
 		else
 		{
 			lfc_ctl->used += 1;
@@ -689,11 +755,11 @@ lfc_write(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno, const void
 
 		if (lfc_ctl->generation == generation)
 		{
-			Assert(LFC_ENABLED());
+			CriticalAssert(LFC_ENABLED());
 			/* Place entry to the head of LRU list */
-			Assert(entry->access_count > 0);
+			CriticalAssert(entry->access_count > 0);
 			if (--entry->access_count == 0)
-				dlist_push_tail(&lfc_ctl->lru, &entry->lru_node);
+				dlist_push_tail(&lfc_ctl->lru, &entry->list_node);
 
 			entry->bitmap[chunk_offs >> 5] |= (1 << (chunk_offs & 31));
 		}
@@ -708,7 +774,6 @@ typedef struct
 } NeonGetStatsCtx;
 
 #define NUM_NEON_GET_STATS_COLS	2
-#define NUM_NEON_GET_STATS_ROWS	3
 
 PG_FUNCTION_INFO_V1(neon_get_lfc_stats);
 Datum
@@ -744,7 +809,6 @@ neon_get_lfc_stats(PG_FUNCTION_ARGS)
 						   INT8OID, -1, 0);
 
 		fctx->tupdesc = BlessTupleDesc(tupledesc);
-		funcctx->max_calls = NUM_NEON_GET_STATS_ROWS;
 		funcctx->user_fctx = fctx;
 
 		/* Return to original context when allocating transient memory */
@@ -778,6 +842,11 @@ neon_get_lfc_stats(PG_FUNCTION_ARGS)
 			if (lfc_ctl)
 				value = lfc_ctl->writes;
 			break;
+		case 4:
+			key = "file_cache_size";
+			if (lfc_ctl)
+				value = lfc_ctl->size;
+			break;
 		default:
 			SRF_RETURN_DONE(funcctx);
 	}
@@ -901,7 +970,7 @@ local_cache_pages(PG_FUNCTION_ARGS)
 				hash_seq_init(&status, lfc_hash);
 				while ((entry = hash_seq_search(&status)) != NULL)
 				{
-					for (int i = 0; i < BLOCKS_PER_CHUNK / 32; i++)
+					for (int i = 0; i < CHUNK_BITMAP_SIZE; i++)
 						n_pages += pg_popcount32(entry->bitmap[i]);
 				}
 			}

pgxn/neon/neon.c

@@ -192,6 +192,13 @@ LogicalSlotsMonitorMain(Datum main_arg)
 	{
 		XLogRecPtr	cutoff_lsn;
 
+		/* In case of a SIGHUP, just reload the configuration. */
+		if (ConfigReloadPending)
+		{
+			ConfigReloadPending = false;
+			ProcessConfigFile(PGC_SIGHUP);
+		}
+
 		/*
 		 * If there are too many .snap files, just drop all logical slots to
 		 * prevent aux files bloat.

pgxn/neon/neon_pgversioncompat.h

@@ -54,6 +54,10 @@
 
 #define BufTagGetNRelFileInfo(tag) tag.rnode
 
+#define BufTagGetRelNumber(tagp) ((tagp)->rnode.relNode)
+
+#define InvalidRelFileNumber InvalidOid
+
 #define SMgrRelGetRelInfo(reln) \
 	(reln->smgr_rnode.node)
 

proxy/src/auth.rs

@@ -113,38 +113,36 @@ impl<E: Into<AuthErrorImpl>> From<E> for AuthError {
 
 impl UserFacingError for AuthError {
     fn to_string_client(&self) -> String {
-        use AuthErrorImpl::*;
         match self.0.as_ref() {
-            Link(e) => e.to_string_client(),
-            GetAuthInfo(e) => e.to_string_client(),
-            Sasl(e) => e.to_string_client(),
-            AuthFailed(_) => self.to_string(),
-            BadAuthMethod(_) => self.to_string(),
-            MalformedPassword(_) => self.to_string(),
-            MissingEndpointName => self.to_string(),
-            Io(_) => "Internal error".to_string(),
-            IpAddressNotAllowed(_) => self.to_string(),
-            TooManyConnections => self.to_string(),
-            UserTimeout(_) => self.to_string(),
+            AuthErrorImpl::Link(e) => e.to_string_client(),
+            AuthErrorImpl::GetAuthInfo(e) => e.to_string_client(),
+            AuthErrorImpl::Sasl(e) => e.to_string_client(),
+            AuthErrorImpl::AuthFailed(_) => self.to_string(),
+            AuthErrorImpl::BadAuthMethod(_) => self.to_string(),
+            AuthErrorImpl::MalformedPassword(_) => self.to_string(),
+            AuthErrorImpl::MissingEndpointName => self.to_string(),
+            AuthErrorImpl::Io(_) => "Internal error".to_string(),
+            AuthErrorImpl::IpAddressNotAllowed(_) => self.to_string(),
+            AuthErrorImpl::TooManyConnections => self.to_string(),
+            AuthErrorImpl::UserTimeout(_) => self.to_string(),
         }
     }
 }
 
 impl ReportableError for AuthError {
     fn get_error_kind(&self) -> crate::error::ErrorKind {
-        use AuthErrorImpl::*;
         match self.0.as_ref() {
-            Link(e) => e.get_error_kind(),
-            GetAuthInfo(e) => e.get_error_kind(),
-            Sasl(e) => e.get_error_kind(),
-            AuthFailed(_) => crate::error::ErrorKind::User,
-            BadAuthMethod(_) => crate::error::ErrorKind::User,
-            MalformedPassword(_) => crate::error::ErrorKind::User,
-            MissingEndpointName => crate::error::ErrorKind::User,
-            Io(_) => crate::error::ErrorKind::ClientDisconnect,
-            IpAddressNotAllowed(_) => crate::error::ErrorKind::User,
-            TooManyConnections => crate::error::ErrorKind::RateLimit,
-            UserTimeout(_) => crate::error::ErrorKind::User,
+            AuthErrorImpl::Link(e) => e.get_error_kind(),
+            AuthErrorImpl::GetAuthInfo(e) => e.get_error_kind(),
+            AuthErrorImpl::Sasl(e) => e.get_error_kind(),
+            AuthErrorImpl::AuthFailed(_) => crate::error::ErrorKind::User,
+            AuthErrorImpl::BadAuthMethod(_) => crate::error::ErrorKind::User,
+            AuthErrorImpl::MalformedPassword(_) => crate::error::ErrorKind::User,
+            AuthErrorImpl::MissingEndpointName => crate::error::ErrorKind::User,
+            AuthErrorImpl::Io(_) => crate::error::ErrorKind::ClientDisconnect,
+            AuthErrorImpl::IpAddressNotAllowed(_) => crate::error::ErrorKind::User,
+            AuthErrorImpl::TooManyConnections => crate::error::ErrorKind::RateLimit,
+            AuthErrorImpl::UserTimeout(_) => crate::error::ErrorKind::User,
         }
     }
 }

proxy/src/auth/backend.rs

@@ -80,9 +80,8 @@ pub trait TestBackend: Send + Sync + 'static {
 
 impl std::fmt::Display for BackendType<'_, (), ()> {
     fn fmt(&self, fmt: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        use BackendType::*;
         match self {
-            Console(api, _) => match &**api {
+            Self::Console(api, _) => match &**api {
                 ConsoleBackend::Console(endpoint) => {
                     fmt.debug_tuple("Console").field(&endpoint.url()).finish()
                 }
@@ -93,7 +92,7 @@ impl std::fmt::Display for BackendType<'_, (), ()> {
                 #[cfg(test)]
                 ConsoleBackend::Test(_) => fmt.debug_tuple("Test").finish(),
             },
-            Link(url, _) => fmt.debug_tuple("Link").field(&url.as_str()).finish(),
+            Self::Link(url, _) => fmt.debug_tuple("Link").field(&url.as_str()).finish(),
         }
     }
 }
@@ -102,10 +101,9 @@ impl<T, D> BackendType<'_, T, D> {
     /// Very similar to [`std::option::Option::as_ref`].
     /// This helps us pass structured config to async tasks.
     pub fn as_ref(&self) -> BackendType<'_, &T, &D> {
-        use BackendType::*;
         match self {
-            Console(c, x) => Console(MaybeOwned::Borrowed(c), x),
-            Link(c, x) => Link(MaybeOwned::Borrowed(c), x),
+            Self::Console(c, x) => BackendType::Console(MaybeOwned::Borrowed(c), x),
+            Self::Link(c, x) => BackendType::Link(MaybeOwned::Borrowed(c), x),
         }
     }
 }
@@ -115,10 +113,9 @@ impl<'a, T, D> BackendType<'a, T, D> {
     /// Maps [`BackendType<T>`] to [`BackendType<R>`] by applying
     /// a function to a contained value.
     pub fn map<R>(self, f: impl FnOnce(T) -> R) -> BackendType<'a, R, D> {
-        use BackendType::*;
         match self {
-            Console(c, x) => Console(c, f(x)),
-            Link(c, x) => Link(c, x),
+            Self::Console(c, x) => BackendType::Console(c, f(x)),
+            Self::Link(c, x) => BackendType::Link(c, x),
         }
     }
 }
@@ -126,10 +123,9 @@ impl<'a, T, D, E> BackendType<'a, Result<T, E>, D> {
     /// Very similar to [`std::option::Option::transpose`].
     /// This is most useful for error handling.
     pub fn transpose(self) -> Result<BackendType<'a, T, D>, E> {
-        use BackendType::*;
         match self {
-            Console(c, x) => x.map(|x| Console(c, x)),
-            Link(c, x) => Ok(Link(c, x)),
+            Self::Console(c, x) => x.map(|x| BackendType::Console(c, x)),
+            Self::Link(c, x) => Ok(BackendType::Link(c, x)),
         }
     }
 }
@@ -293,7 +289,9 @@ async fn auth_quirks(
             ctx.set_endpoint_id(res.info.endpoint.clone());
             let password = match res.keys {
                 ComputeCredentialKeys::Password(p) => p,
-                _ => unreachable!("password hack should return a password"),
+                ComputeCredentialKeys::AuthKeys(_) => {
+                    unreachable!("password hack should return a password")
+                }
             };
             (res.info, Some(password))
         }
@@ -400,21 +398,17 @@ async fn authenticate_with_secret(
 impl<'a> BackendType<'a, ComputeUserInfoMaybeEndpoint, &()> {
     /// Get compute endpoint name from the credentials.
     pub fn get_endpoint(&self) -> Option<EndpointId> {
-        use BackendType::*;
-
         match self {
-            Console(_, user_info) => user_info.endpoint_id.clone(),
-            Link(_, _) => Some("link".into()),
+            Self::Console(_, user_info) => user_info.endpoint_id.clone(),
+            Self::Link(_, _) => Some("link".into()),
         }
     }
 
     /// Get username from the credentials.
     pub fn get_user(&self) -> &str {
-        use BackendType::*;
-
         match self {
-            Console(_, user_info) => &user_info.user,
-            Link(_, _) => "link",
+            Self::Console(_, user_info) => &user_info.user,
+            Self::Link(_, _) => "link",
         }
     }
 
@@ -428,10 +422,8 @@ impl<'a> BackendType<'a, ComputeUserInfoMaybeEndpoint, &()> {
         config: &'static AuthenticationConfig,
         endpoint_rate_limiter: Arc<EndpointRateLimiter>,
     ) -> auth::Result<BackendType<'a, ComputeCredentials, NodeInfo>> {
-        use BackendType::*;
-
         let res = match self {
-            Console(api, user_info) => {
+            Self::Console(api, user_info) => {
                 info!(
                     user = &*user_info.user,
                     project = user_info.endpoint(),
@@ -451,7 +443,7 @@ impl<'a> BackendType<'a, ComputeUserInfoMaybeEndpoint, &()> {
                 BackendType::Console(api, credentials)
             }
             // NOTE: this auth backend doesn't use client credentials.
-            Link(url, _) => {
+            Self::Link(url, _) => {
                 info!("performing link authentication");
 
                 let info = link::authenticate(ctx, &url, client).await?;
@@ -470,10 +462,9 @@ impl BackendType<'_, ComputeUserInfo, &()> {
         &self,
         ctx: &RequestMonitoring,
     ) -> Result<CachedRoleSecret, GetAuthInfoError> {
-        use BackendType::*;
         match self {
-            Console(api, user_info) => api.get_role_secret(ctx, user_info).await,
-            Link(_, _) => Ok(Cached::new_uncached(None)),
+            Self::Console(api, user_info) => api.get_role_secret(ctx, user_info).await,
+            Self::Link(_, _) => Ok(Cached::new_uncached(None)),
         }
     }
 
@@ -481,10 +472,9 @@ impl BackendType<'_, ComputeUserInfo, &()> {
         &self,
         ctx: &RequestMonitoring,
     ) -> Result<(CachedAllowedIps, Option<CachedRoleSecret>), GetAuthInfoError> {
-        use BackendType::*;
         match self {
-            Console(api, user_info) => api.get_allowed_ips_and_secret(ctx, user_info).await,
-            Link(_, _) => Ok((Cached::new_uncached(Arc::new(vec![])), None)),
+            Self::Console(api, user_info) => api.get_allowed_ips_and_secret(ctx, user_info).await,
+            Self::Link(_, _) => Ok((Cached::new_uncached(Arc::new(vec![])), None)),
         }
     }
 }
@@ -495,18 +485,16 @@ impl ComputeConnectBackend for BackendType<'_, ComputeCredentials, NodeInfo> {
         &self,
         ctx: &RequestMonitoring,
     ) -> Result<CachedNodeInfo, console::errors::WakeComputeError> {
-        use BackendType::*;
-
         match self {
-            Console(api, creds) => api.wake_compute(ctx, &creds.info).await,
-            Link(_, info) => Ok(Cached::new_uncached(info.clone())),
+            Self::Console(api, creds) => api.wake_compute(ctx, &creds.info).await,
+            Self::Link(_, info) => Ok(Cached::new_uncached(info.clone())),
         }
     }
 
     fn get_keys(&self) -> Option<&ComputeCredentialKeys> {
         match self {
-            BackendType::Console(_, creds) => Some(&creds.keys),
-            BackendType::Link(_, _) => None,
+            Self::Console(_, creds) => Some(&creds.keys),
+            Self::Link(_, _) => None,
         }
     }
 }
@@ -517,18 +505,16 @@ impl ComputeConnectBackend for BackendType<'_, ComputeCredentials, &()> {
         &self,
         ctx: &RequestMonitoring,
     ) -> Result<CachedNodeInfo, console::errors::WakeComputeError> {
-        use BackendType::*;
-
         match self {
-            Console(api, creds) => api.wake_compute(ctx, &creds.info).await,
-            Link(_, _) => unreachable!("link auth flow doesn't support waking the compute"),
+            Self::Console(api, creds) => api.wake_compute(ctx, &creds.info).await,
+            Self::Link(_, _) => unreachable!("link auth flow doesn't support waking the compute"),
         }
     }
 
     fn get_keys(&self) -> Option<&ComputeCredentialKeys> {
         match self {
-            BackendType::Console(_, creds) => Some(&creds.keys),
-            BackendType::Link(_, _) => None,
+            Self::Console(_, creds) => Some(&creds.keys),
+            Self::Link(_, _) => None,
         }
     }
 }

proxy/src/auth/backend/jwt.rs

@@ -195,7 +195,7 @@ impl JwkCacheEntryLock {
 
         let header = base64::decode_config(header, base64::URL_SAFE_NO_PAD)
             .context("Provided authentication token is not a valid JWT encoding")?;
-        let header = serde_json::from_slice::<JWTHeader>(&header)
+        let header = serde_json::from_slice::<JWTHeader<'_>>(&header)
             .context("Provided authentication token is not a valid JWT encoding")?;
 
         let sig = base64::decode_config(signature, base64::URL_SAFE_NO_PAD)
@@ -340,7 +340,7 @@ impl JwkRenewalPermit<'_> {
         }
     }
 
-    async fn acquire_permit(from: &Arc<JwkCacheEntryLock>) -> JwkRenewalPermit {
+    async fn acquire_permit(from: &Arc<JwkCacheEntryLock>) -> JwkRenewalPermit<'_> {
         match from.lookup.acquire().await {
             Ok(permit) => {
                 permit.forget();
@@ -352,7 +352,7 @@ impl JwkRenewalPermit<'_> {
         }
     }
 
-    fn try_acquire_permit(from: &Arc<JwkCacheEntryLock>) -> Option<JwkRenewalPermit> {
+    fn try_acquire_permit(from: &Arc<JwkCacheEntryLock>) -> Option<JwkRenewalPermit<'_>> {
         match from.lookup.try_acquire() {
             Ok(permit) => {
                 permit.forget();

proxy/src/auth/credentials.rs

@@ -89,10 +89,12 @@ impl ComputeUserInfoMaybeEndpoint {
         sni: Option<&str>,
         common_names: Option<&HashSet<String>>,
     ) -> Result<Self, ComputeUserInfoParseError> {
-        use ComputeUserInfoParseError::*;
-
         // Some parameters are stored in the startup message.
-        let get_param = |key| params.get(key).ok_or(MissingKey(key));
+        let get_param = |key| {
+            params
+                .get(key)
+                .ok_or(ComputeUserInfoParseError::MissingKey(key))
+        };
         let user: RoleName = get_param("user")?.into();
 
         // Project name might be passed via PG's command-line options.
@@ -122,11 +124,14 @@ impl ComputeUserInfoMaybeEndpoint {
         let endpoint = match (endpoint_option, endpoint_from_domain) {
             // Invariant: if we have both project name variants, they should match.
             (Some(option), Some(domain)) if option != domain => {
-                Some(Err(InconsistentProjectNames { domain, option }))
+                Some(Err(ComputeUserInfoParseError::InconsistentProjectNames {
+                    domain,
+                    option,
+                }))
             }
             // Invariant: project name may not contain certain characters.
             (a, b) => a.or(b).map(|name| match project_name_valid(name.as_ref()) {
-                false => Err(MalformedProjectName(name)),
+                false => Err(ComputeUserInfoParseError::MalformedProjectName(name)),
                 true => Ok(name),
             }),
         }
@@ -186,7 +191,7 @@ impl<'de> serde::de::Deserialize<'de> for IpPattern {
         impl<'de> serde::de::Visitor<'de> for StrVisitor {
             type Value = IpPattern;
 
-            fn expecting(&self, formatter: &mut std::fmt::Formatter) -> std::fmt::Result {
+            fn expecting(&self, formatter: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
                 write!(formatter, "comma separated list with ip address, ip address range, or ip address subnet mask")
             }
 

proxy/src/bin/proxy.rs

@@ -173,9 +173,6 @@ struct ProxyCliArgs {
     /// cache for `role_secret` (use `size=0` to disable)
     #[clap(long, default_value = config::CacheOptions::CACHE_DEFAULT_OPTIONS)]
     role_secret_cache: String,
-    /// disable ip check for http requests. If it is too time consuming, it could be turned off.
-    #[clap(long, default_value_t = false, value_parser = clap::builder::BoolishValueParser::new(), action = clap::ArgAction::Set)]
-    disable_ip_check_for_http: bool,
     /// redis url for notifications (if empty, redis_host:port will be used for both notifications and streaming connections)
     #[clap(long)]
     redis_notifications: Option<String>,
@@ -661,6 +658,7 @@ fn build_config(args: &ProxyCliArgs) -> anyhow::Result<&'static ProxyConfig> {
     )?;
 
     let http_config = HttpConfig {
+        accept_websockets: true,
         pool_options: GlobalConnPoolOptions {
             max_conns_per_endpoint: args.sql_over_http.sql_over_http_pool_max_conns_per_endpoint,
             gc_epoch: args.sql_over_http.sql_over_http_pool_gc_epoch,

proxy/src/cache/common.rs

@@ -24,7 +24,7 @@ impl<C: Cache> Cache for &C {
     type LookupInfo<Key> = C::LookupInfo<Key>;
 
     fn invalidate(&self, info: &Self::LookupInfo<Self::Key>) {
-        C::invalidate(self, info)
+        C::invalidate(self, info);
     }
 }
 

proxy/src/cache/timed_lru.rs

@@ -58,7 +58,7 @@ impl<K: Hash + Eq, V> Cache for TimedLru<K, V> {
     type LookupInfo<Key> = LookupInfo<Key>;
 
     fn invalidate(&self, info: &Self::LookupInfo<K>) {
-        self.invalidate_raw(info)
+        self.invalidate_raw(info);
     }
 }
 

proxy/src/compute.rs

@@ -44,11 +44,10 @@ pub enum ConnectionError {
 
 impl UserFacingError for ConnectionError {
     fn to_string_client(&self) -> String {
-        use ConnectionError::*;
         match self {
             // This helps us drop irrelevant library-specific prefixes.
             // TODO: propagate severity level and other parameters.
-            Postgres(err) => match err.as_db_error() {
+            ConnectionError::Postgres(err) => match err.as_db_error() {
                 Some(err) => {
                     let msg = err.message();
 
@@ -62,8 +61,8 @@ impl UserFacingError for ConnectionError {
                 }
                 None => err.to_string(),
             },
-            WakeComputeError(err) => err.to_string_client(),
-            TooManyConnectionAttempts(_) => {
+            ConnectionError::WakeComputeError(err) => err.to_string_client(),
+            ConnectionError::TooManyConnectionAttempts(_) => {
                 "Failed to acquire permit to connect to the database. Too many database connection attempts are currently ongoing.".to_owned()
             }
             _ => COULD_NOT_CONNECT.to_owned(),
@@ -366,16 +365,16 @@ static TLS_ROOTS: OnceCell<Arc<rustls::RootCertStore>> = OnceCell::new();
 struct AcceptEverythingVerifier;
 impl ServerCertVerifier for AcceptEverythingVerifier {
     fn supported_verify_schemes(&self) -> Vec<rustls::SignatureScheme> {
-        use rustls::SignatureScheme::*;
+        use rustls::SignatureScheme;
         // The schemes for which `SignatureScheme::supported_in_tls13` returns true.
         vec![
-            ECDSA_NISTP521_SHA512,
-            ECDSA_NISTP384_SHA384,
-            ECDSA_NISTP256_SHA256,
-            RSA_PSS_SHA512,
-            RSA_PSS_SHA384,
-            RSA_PSS_SHA256,
-            ED25519,
+            SignatureScheme::ECDSA_NISTP521_SHA512,
+            SignatureScheme::ECDSA_NISTP384_SHA384,
+            SignatureScheme::ECDSA_NISTP256_SHA256,
+            SignatureScheme::RSA_PSS_SHA512,
+            SignatureScheme::RSA_PSS_SHA384,
+            SignatureScheme::RSA_PSS_SHA256,
+            SignatureScheme::ED25519,
         ]
     }
     fn verify_server_cert(

proxy/src/config.rs

@@ -52,6 +52,7 @@ pub struct TlsConfig {
 }
 
 pub struct HttpConfig {
+    pub accept_websockets: bool,
     pub pool_options: GlobalConnPoolOptions,
     pub cancel_set: CancelSet,
     pub client_conn_threshold: u64,
@@ -155,7 +156,7 @@ pub enum TlsServerEndPoint {
 }
 
 impl TlsServerEndPoint {
-    pub fn new(cert: &CertificateDer) -> anyhow::Result<Self> {
+    pub fn new(cert: &CertificateDer<'_>) -> anyhow::Result<Self> {
         let sha256_oids = [
             // I'm explicitly not adding MD5 or SHA1 here... They're bad.
             oid_registry::OID_SIG_ECDSA_WITH_SHA256,
@@ -278,7 +279,7 @@ impl CertResolver {
 impl rustls::server::ResolvesServerCert for CertResolver {
     fn resolve(
         &self,
-        client_hello: rustls::server::ClientHello,
+        client_hello: rustls::server::ClientHello<'_>,
     ) -> Option<Arc<rustls::sign::CertifiedKey>> {
         self.resolve(client_hello.server_name()).map(|x| x.0)
     }
@@ -559,7 +560,7 @@ impl RetryConfig {
             match key {
                 "num_retries" => num_retries = Some(value.parse()?),
                 "base_retry_wait_duration" => {
-                    base_retry_wait_duration = Some(humantime::parse_duration(value)?)
+                    base_retry_wait_duration = Some(humantime::parse_duration(value)?);
                 }
                 "retry_wait_exponent_base" => retry_wait_exponent_base = Some(value.parse()?),
                 unknown => bail!("unknown key: {unknown}"),

proxy/src/console/messages.rs

@@ -22,16 +22,15 @@ impl ConsoleError {
         self.status
             .as_ref()
             .and_then(|s| s.details.error_info.as_ref())
-            .map(|e| e.reason)
-            .unwrap_or(Reason::Unknown)
+            .map_or(Reason::Unknown, |e| e.reason)
     }
+
     pub fn get_user_facing_message(&self) -> String {
         use super::provider::errors::REQUEST_FAILED;
         self.status
             .as_ref()
             .and_then(|s| s.details.user_facing_message.as_ref())
-            .map(|m| m.message.clone().into())
-            .unwrap_or_else(|| {
+            .map_or_else(|| {
                 // Ask @neondatabase/control-plane for review before adding more.
                 match self.http_status_code {
                     http::StatusCode::NOT_FOUND => {
@@ -48,19 +47,18 @@ impl ConsoleError {
                     }
                     _ => REQUEST_FAILED.to_owned(),
                 }
-            })
+            }, |m| m.message.clone().into())
     }
 }
 
 impl Display for ConsoleError {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        let msg = self
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        let msg: &str = self
             .status
             .as_ref()
             .and_then(|s| s.details.user_facing_message.as_ref())
-            .map(|m| m.message.as_ref())
-            .unwrap_or_else(|| &self.error);
-        write!(f, "{}", msg)
+            .map_or_else(|| self.error.as_ref(), |m| m.message.as_ref());
+        write!(f, "{msg}")
     }
 }
 
@@ -286,7 +284,7 @@ pub struct DatabaseInfo {
 
 // Manually implement debug to omit sensitive info.
 impl fmt::Debug for DatabaseInfo {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         f.debug_struct("DatabaseInfo")
             .field("host", &self.host)
             .field("port", &self.port)
@@ -373,7 +371,7 @@ mod tests {
                 }
             }
         });
-        let _: KickSession = serde_json::from_str(&json.to_string())?;
+        let _: KickSession<'_> = serde_json::from_str(&json.to_string())?;
 
         Ok(())
     }

proxy/src/console/mgmt.rs

@@ -93,7 +93,8 @@ impl postgres_backend::Handler<tokio::net::TcpStream> for MgmtHandler {
 }
 
 fn try_process_query(pgb: &mut PostgresBackendTCP, query: &str) -> Result<(), QueryError> {
-    let resp: KickSession = serde_json::from_str(query).context("Failed to parse query as json")?;
+    let resp: KickSession<'_> =
+        serde_json::from_str(query).context("Failed to parse query as json")?;
 
     let span = info_span!("event", session_id = resp.session_id);
     let _enter = span.enter();

proxy/src/console/provider.rs

@@ -26,7 +26,7 @@ use tracing::info;
 pub mod errors {
     use crate::{
         console::messages::{self, ConsoleError, Reason},
-        error::{io_error, ReportableError, UserFacingError},
+        error::{io_error, ErrorKind, ReportableError, UserFacingError},
         proxy::retry::CouldRetry,
     };
     use thiserror::Error;
@@ -51,21 +51,19 @@ pub mod errors {
     impl ApiError {
         /// Returns HTTP status code if it's the reason for failure.
         pub fn get_reason(&self) -> messages::Reason {
-            use ApiError::*;
             match self {
-                Console(e) => e.get_reason(),
-                _ => messages::Reason::Unknown,
+                ApiError::Console(e) => e.get_reason(),
+                ApiError::Transport(_) => messages::Reason::Unknown,
             }
         }
     }
 
     impl UserFacingError for ApiError {
         fn to_string_client(&self) -> String {
-            use ApiError::*;
             match self {
                 // To minimize risks, only select errors are forwarded to users.
-                Console(c) => c.get_user_facing_message(),
-                _ => REQUEST_FAILED.to_owned(),
+                ApiError::Console(c) => c.get_user_facing_message(),
+                ApiError::Transport(_) => REQUEST_FAILED.to_owned(),
             }
         }
     }
@@ -73,57 +71,53 @@ pub mod errors {
     impl ReportableError for ApiError {
         fn get_error_kind(&self) -> crate::error::ErrorKind {
             match self {
-                ApiError::Console(e) => {
-                    use crate::error::ErrorKind::*;
-                    match e.get_reason() {
-                        Reason::RoleProtected => User,
-                        Reason::ResourceNotFound => User,
-                        Reason::ProjectNotFound => User,
-                        Reason::EndpointNotFound => User,
-                        Reason::BranchNotFound => User,
-                        Reason::RateLimitExceeded => ServiceRateLimit,
-                        Reason::NonDefaultBranchComputeTimeExceeded => User,
-                        Reason::ActiveTimeQuotaExceeded => User,
-                        Reason::ComputeTimeQuotaExceeded => User,
-                        Reason::WrittenDataQuotaExceeded => User,
-                        Reason::DataTransferQuotaExceeded => User,
-                        Reason::LogicalSizeQuotaExceeded => User,
-                        Reason::ConcurrencyLimitReached => ControlPlane,
-                        Reason::LockAlreadyTaken => ControlPlane,
-                        Reason::RunningOperations => ControlPlane,
-                        Reason::Unknown => match &e {
-                            ConsoleError {
-                                http_status_code:
-                                    http::StatusCode::NOT_FOUND | http::StatusCode::NOT_ACCEPTABLE,
-                                ..
-                            } => crate::error::ErrorKind::User,
-                            ConsoleError {
-                                http_status_code: http::StatusCode::UNPROCESSABLE_ENTITY,
-                                error,
-                                ..
-                            } if error.contains(
-                                "compute time quota of non-primary branches is exceeded",
-                            ) =>
-                            {
-                                crate::error::ErrorKind::User
-                            }
-                            ConsoleError {
-                                http_status_code: http::StatusCode::LOCKED,
-                                error,
-                                ..
-                            } if error.contains("quota exceeded")
-                                || error.contains("the limit for current plan reached") =>
-                            {
-                                crate::error::ErrorKind::User
-                            }
-                            ConsoleError {
-                                http_status_code: http::StatusCode::TOO_MANY_REQUESTS,
-                                ..
-                            } => crate::error::ErrorKind::ServiceRateLimit,
-                            ConsoleError { .. } => crate::error::ErrorKind::ControlPlane,
-                        },
-                    }
-                }
+                ApiError::Console(e) => match e.get_reason() {
+                    Reason::RoleProtected => ErrorKind::User,
+                    Reason::ResourceNotFound => ErrorKind::User,
+                    Reason::ProjectNotFound => ErrorKind::User,
+                    Reason::EndpointNotFound => ErrorKind::User,
+                    Reason::BranchNotFound => ErrorKind::User,
+                    Reason::RateLimitExceeded => ErrorKind::ServiceRateLimit,
+                    Reason::NonDefaultBranchComputeTimeExceeded => ErrorKind::User,
+                    Reason::ActiveTimeQuotaExceeded => ErrorKind::User,
+                    Reason::ComputeTimeQuotaExceeded => ErrorKind::User,
+                    Reason::WrittenDataQuotaExceeded => ErrorKind::User,
+                    Reason::DataTransferQuotaExceeded => ErrorKind::User,
+                    Reason::LogicalSizeQuotaExceeded => ErrorKind::User,
+                    Reason::ConcurrencyLimitReached => ErrorKind::ControlPlane,
+                    Reason::LockAlreadyTaken => ErrorKind::ControlPlane,
+                    Reason::RunningOperations => ErrorKind::ControlPlane,
+                    Reason::Unknown => match &e {
+                        ConsoleError {
+                            http_status_code:
+                                http::StatusCode::NOT_FOUND | http::StatusCode::NOT_ACCEPTABLE,
+                            ..
+                        } => crate::error::ErrorKind::User,
+                        ConsoleError {
+                            http_status_code: http::StatusCode::UNPROCESSABLE_ENTITY,
+                            error,
+                            ..
+                        } if error
+                            .contains("compute time quota of non-primary branches is exceeded") =>
+                        {
+                            crate::error::ErrorKind::User
+                        }
+                        ConsoleError {
+                            http_status_code: http::StatusCode::LOCKED,
+                            error,
+                            ..
+                        } if error.contains("quota exceeded")
+                            || error.contains("the limit for current plan reached") =>
+                        {
+                            crate::error::ErrorKind::User
+                        }
+                        ConsoleError {
+                            http_status_code: http::StatusCode::TOO_MANY_REQUESTS,
+                            ..
+                        } => crate::error::ErrorKind::ServiceRateLimit,
+                        ConsoleError { .. } => crate::error::ErrorKind::ControlPlane,
+                    },
+                },
                 ApiError::Transport(_) => crate::error::ErrorKind::ControlPlane,
             }
         }
@@ -170,12 +164,11 @@ pub mod errors {
 
     impl UserFacingError for GetAuthInfoError {
         fn to_string_client(&self) -> String {
-            use GetAuthInfoError::*;
             match self {
                 // We absolutely should not leak any secrets!
-                BadSecret => REQUEST_FAILED.to_owned(),
+                Self::BadSecret => REQUEST_FAILED.to_owned(),
                 // However, API might return a meaningful error.
-                ApiError(e) => e.to_string_client(),
+                Self::ApiError(e) => e.to_string_client(),
             }
         }
     }
@@ -183,8 +176,8 @@ pub mod errors {
     impl ReportableError for GetAuthInfoError {
         fn get_error_kind(&self) -> crate::error::ErrorKind {
             match self {
-                GetAuthInfoError::BadSecret => crate::error::ErrorKind::ControlPlane,
-                GetAuthInfoError::ApiError(_) => crate::error::ErrorKind::ControlPlane,
+                Self::BadSecret => crate::error::ErrorKind::ControlPlane,
+                Self::ApiError(_) => crate::error::ErrorKind::ControlPlane,
             }
         }
     }
@@ -213,17 +206,16 @@ pub mod errors {
 
     impl UserFacingError for WakeComputeError {
         fn to_string_client(&self) -> String {
-            use WakeComputeError::*;
             match self {
                 // We shouldn't show user the address even if it's broken.
                 // Besides, user is unlikely to care about this detail.
-                BadComputeAddress(_) => REQUEST_FAILED.to_owned(),
+                Self::BadComputeAddress(_) => REQUEST_FAILED.to_owned(),
                 // However, API might return a meaningful error.
-                ApiError(e) => e.to_string_client(),
+                Self::ApiError(e) => e.to_string_client(),
 
-                TooManyConnections => self.to_string(),
+                Self::TooManyConnections => self.to_string(),
 
-                TooManyConnectionAttempts(_) => {
+                Self::TooManyConnectionAttempts(_) => {
                     "Failed to acquire permit to connect to the database. Too many database connection attempts are currently ongoing.".to_owned()
                 }
             }
@@ -233,10 +225,10 @@ pub mod errors {
     impl ReportableError for WakeComputeError {
         fn get_error_kind(&self) -> crate::error::ErrorKind {
             match self {
-                WakeComputeError::BadComputeAddress(_) => crate::error::ErrorKind::ControlPlane,
-                WakeComputeError::ApiError(e) => e.get_error_kind(),
-                WakeComputeError::TooManyConnections => crate::error::ErrorKind::RateLimit,
-                WakeComputeError::TooManyConnectionAttempts(e) => e.get_error_kind(),
+                Self::BadComputeAddress(_) => crate::error::ErrorKind::ControlPlane,
+                Self::ApiError(e) => e.get_error_kind(),
+                Self::TooManyConnections => crate::error::ErrorKind::RateLimit,
+                Self::TooManyConnectionAttempts(e) => e.get_error_kind(),
             }
         }
     }
@@ -244,10 +236,10 @@ pub mod errors {
     impl CouldRetry for WakeComputeError {
         fn could_retry(&self) -> bool {
             match self {
-                WakeComputeError::BadComputeAddress(_) => false,
-                WakeComputeError::ApiError(e) => e.could_retry(),
-                WakeComputeError::TooManyConnections => false,
-                WakeComputeError::TooManyConnectionAttempts(_) => false,
+                Self::BadComputeAddress(_) => false,
+                Self::ApiError(e) => e.could_retry(),
+                Self::TooManyConnections => false,
+                Self::TooManyConnectionAttempts(_) => false,
             }
         }
     }
@@ -366,13 +358,14 @@ impl Api for ConsoleBackend {
         ctx: &RequestMonitoring,
         user_info: &ComputeUserInfo,
     ) -> Result<CachedRoleSecret, errors::GetAuthInfoError> {
-        use ConsoleBackend::*;
         match self {
-            Console(api) => api.get_role_secret(ctx, user_info).await,
+            Self::Console(api) => api.get_role_secret(ctx, user_info).await,
             #[cfg(any(test, feature = "testing"))]
-            Postgres(api) => api.get_role_secret(ctx, user_info).await,
+            Self::Postgres(api) => api.get_role_secret(ctx, user_info).await,
             #[cfg(test)]
-            Test(_) => unreachable!("this function should never be called in the test backend"),
+            Self::Test(_) => {
+                unreachable!("this function should never be called in the test backend")
+            }
         }
     }
 
@@ -381,13 +374,12 @@ impl Api for ConsoleBackend {
         ctx: &RequestMonitoring,
         user_info: &ComputeUserInfo,
     ) -> Result<(CachedAllowedIps, Option<CachedRoleSecret>), errors::GetAuthInfoError> {
-        use ConsoleBackend::*;
         match self {
-            Console(api) => api.get_allowed_ips_and_secret(ctx, user_info).await,
+            Self::Console(api) => api.get_allowed_ips_and_secret(ctx, user_info).await,
             #[cfg(any(test, feature = "testing"))]
-            Postgres(api) => api.get_allowed_ips_and_secret(ctx, user_info).await,
+            Self::Postgres(api) => api.get_allowed_ips_and_secret(ctx, user_info).await,
             #[cfg(test)]
-            Test(api) => api.get_allowed_ips_and_secret(),
+            Self::Test(api) => api.get_allowed_ips_and_secret(),
         }
     }
 
@@ -396,14 +388,12 @@ impl Api for ConsoleBackend {
         ctx: &RequestMonitoring,
         user_info: &ComputeUserInfo,
     ) -> Result<CachedNodeInfo, errors::WakeComputeError> {
-        use ConsoleBackend::*;
-
         match self {
-            Console(api) => api.wake_compute(ctx, user_info).await,
+            Self::Console(api) => api.wake_compute(ctx, user_info).await,
             #[cfg(any(test, feature = "testing"))]
-            Postgres(api) => api.wake_compute(ctx, user_info).await,
+            Self::Postgres(api) => api.wake_compute(ctx, user_info).await,
             #[cfg(test)]
-            Test(api) => api.wake_compute(),
+            Self::Test(api) => api.wake_compute(),
         }
     }
 }
@@ -549,7 +539,7 @@ impl WakeComputePermit {
         !self.permit.is_disabled()
     }
     pub fn release(self, outcome: Outcome) {
-        self.permit.release(outcome)
+        self.permit.release(outcome);
     }
     pub fn release_result<T, E>(self, res: Result<T, E>) -> Result<T, E> {
         match res {

proxy/src/context.rs

@@ -166,7 +166,7 @@ impl RequestMonitoring {
     pub fn set_project(&self, x: MetricsAuxInfo) {
         let mut this = self.0.try_lock().expect("should not deadlock");
         if this.endpoint_id.is_none() {
-            this.set_endpoint_id(x.endpoint_id.as_str().into())
+            this.set_endpoint_id(x.endpoint_id.as_str().into());
         }
         this.branch = Some(x.branch_id);
         this.project = Some(x.project_id);
@@ -260,7 +260,7 @@ impl RequestMonitoring {
             .cold_start_info
     }
 
-    pub fn latency_timer_pause(&self, waiting_for: Waiting) -> LatencyTimerPause {
+    pub fn latency_timer_pause(&self, waiting_for: Waiting) -> LatencyTimerPause<'_> {
         LatencyTimerPause {
             ctx: self,
             start: tokio::time::Instant::now(),
@@ -273,7 +273,7 @@ impl RequestMonitoring {
             .try_lock()
             .expect("should not deadlock")
             .latency_timer
-            .success()
+            .success();
     }
 }
 
@@ -328,7 +328,7 @@ impl RequestMonitoringInner {
     fn has_private_peer_addr(&self) -> bool {
         match self.peer_addr {
             IpAddr::V4(ip) => ip.is_private(),
-            _ => false,
+            IpAddr::V6(_) => false,
         }
     }
 

proxy/src/context/parquet.rs

@@ -736,7 +736,7 @@ mod tests {
                 while let Some(r) = s.next().await {
                     tx.send(r).unwrap();
                 }
-                time::sleep(time::Duration::from_secs(70)).await
+                time::sleep(time::Duration::from_secs(70)).await;
             }
         });
 

proxy/src/intern.rs

@@ -56,7 +56,7 @@ impl<'de, Id: InternId> serde::de::Deserialize<'de> for InternedString<Id> {
         impl<'de, Id: InternId> serde::de::Visitor<'de> for Visitor<Id> {
             type Value = InternedString<Id>;
 
-            fn expecting(&self, formatter: &mut std::fmt::Formatter) -> std::fmt::Result {
+            fn expecting(&self, formatter: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
                 formatter.write_str("a string")
             }
 

proxy/src/metrics.rs

@@ -252,7 +252,7 @@ impl Drop for HttpEndpointPoolsGuard<'_> {
 }
 
 impl HttpEndpointPools {
-    pub fn guard(&self) -> HttpEndpointPoolsGuard {
+    pub fn guard(&self) -> HttpEndpointPoolsGuard<'_> {
         self.http_pool_endpoints_registered_total.inc();
         HttpEndpointPoolsGuard {
             dec: &self.http_pool_endpoints_unregistered_total,

proxy/src/proxy/copy_bidirectional.rs

@@ -184,7 +184,7 @@ impl CopyBuffer {
                 }
                 Poll::Pending
             }
-            res => res.map_err(ErrorDirection::Write),
+            res @ Poll::Ready(_) => res.map_err(ErrorDirection::Write),
         }
     }
 

proxy/src/proxy/handshake.rs

@@ -82,9 +82,8 @@ pub async fn handshake<S: AsyncRead + AsyncWrite + Unpin>(
     let mut stream = PqStream::new(Stream::from_raw(stream));
     loop {
         let msg = stream.read_startup_packet().await?;
-        use FeStartupPacket::*;
         match msg {
-            SslRequest { direct } => match stream.get_ref() {
+            FeStartupPacket::SslRequest { direct } => match stream.get_ref() {
                 Stream::Raw { .. } if !tried_ssl => {
                     tried_ssl = true;
 
@@ -139,7 +138,7 @@ pub async fn handshake<S: AsyncRead + AsyncWrite + Unpin>(
 
                         let tls_stream = accept.await.inspect_err(|_| {
                             if record_handshake_error {
-                                Metrics::get().proxy.tls_handshake_failures.inc()
+                                Metrics::get().proxy.tls_handshake_failures.inc();
                             }
                         })?;
 
@@ -182,7 +181,7 @@ pub async fn handshake<S: AsyncRead + AsyncWrite + Unpin>(
                 }
                 _ => return Err(HandshakeError::ProtocolViolation),
             },
-            GssEncRequest => match stream.get_ref() {
+            FeStartupPacket::GssEncRequest => match stream.get_ref() {
                 Stream::Raw { .. } if !tried_gss => {
                     tried_gss = true;
 
@@ -191,7 +190,7 @@ pub async fn handshake<S: AsyncRead + AsyncWrite + Unpin>(
                 }
                 _ => return Err(HandshakeError::ProtocolViolation),
             },
-            StartupMessage { params, version }
+            FeStartupPacket::StartupMessage { params, version }
                 if PG_PROTOCOL_EARLIEST <= version && version <= PG_PROTOCOL_LATEST =>
             {
                 // Check that the config has been consumed during upgrade
@@ -211,7 +210,7 @@ pub async fn handshake<S: AsyncRead + AsyncWrite + Unpin>(
                 break Ok(HandshakeData::Startup(stream, params));
             }
             // downgrade protocol version
-            StartupMessage { params, version }
+            FeStartupPacket::StartupMessage { params, version }
                 if version.major() == 3 && version > PG_PROTOCOL_LATEST =>
             {
                 warn!(?version, "unsupported minor version");
@@ -241,7 +240,7 @@ pub async fn handshake<S: AsyncRead + AsyncWrite + Unpin>(
                 );
                 break Ok(HandshakeData::Startup(stream, params));
             }
-            StartupMessage { version, .. } => {
+            FeStartupPacket::StartupMessage { version, .. } => {
                 warn!(
                     ?version,
                     session_type = "normal",
@@ -249,7 +248,7 @@ pub async fn handshake<S: AsyncRead + AsyncWrite + Unpin>(
                 );
                 return Err(HandshakeError::ProtocolViolation);
             }
-            CancelRequest(cancel_key_data) => {
+            FeStartupPacket::CancelRequest(cancel_key_data) => {
                 info!(session_type = "cancellation", "successful handshake");
                 break Ok(HandshakeData::Cancel(cancel_key_data));
             }

proxy/src/proxy/tests/mitm.rs

@@ -68,7 +68,7 @@ async fn proxy_mitm(
                                 end_client.send(Bytes::from_static(b"R\0\0\0\x17\0\0\0\x0aSCRAM-SHA-256\0\0")).await.unwrap();
                                 continue;
                             }
-                            end_client.send(message).await.unwrap()
+                            end_client.send(message).await.unwrap();
                         }
                         _ => break,
                     }
@@ -88,7 +88,7 @@ async fn proxy_mitm(
                                 end_server.send(buf.freeze()).await.unwrap();
                                 continue;
                             }
-                            end_server.send(message).await.unwrap()
+                            end_server.send(message).await.unwrap();
                         }
                         _ => break,
                     }

proxy/src/rate_limiter/limit_algorithm.rs

@@ -237,7 +237,7 @@ impl Token {
     }
 
     pub fn release(mut self, outcome: Outcome) {
-        self.release_mut(Some(outcome))
+        self.release_mut(Some(outcome));
     }
 
     pub fn release_mut(&mut self, outcome: Option<Outcome>) {
@@ -249,7 +249,7 @@ impl Token {
 
 impl Drop for Token {
     fn drop(&mut self) {
-        self.release_mut(None)
+        self.release_mut(None);
     }
 }
 

proxy/src/rate_limiter/limit_algorithm/aimd.rs

@@ -25,9 +25,8 @@ pub struct Aimd {
 
 impl LimitAlgorithm for Aimd {
     fn update(&self, old_limit: usize, sample: Sample) -> usize {
-        use Outcome::*;
         match sample.outcome {
-            Success => {
+            Outcome::Success => {
                 let utilisation = sample.in_flight as f32 / old_limit as f32;
 
                 if utilisation > self.utilisation {
@@ -42,7 +41,7 @@ impl LimitAlgorithm for Aimd {
                     old_limit
                 }
             }
-            Overload => {
+            Outcome::Overload => {
                 let limit = old_limit as f32 * self.dec;
 
                 // Floor instead of round, so the limit reduces even with small numbers.

proxy/src/redis/connection_with_credentials_provider.rs

@@ -98,7 +98,7 @@ impl ConnectionWithCredentialsProvider {
         info!("Establishing a new connection...");
         self.con = None;
         if let Some(f) = self.refresh_token_task.take() {
-            f.abort()
+            f.abort();
         }
         let mut con = self
             .get_client()

proxy/src/redis/notifications.rs

@@ -108,7 +108,6 @@ impl<C: ProjectInfoCache + Send + Sync + 'static> MessageHandler<C> {
     }
     #[tracing::instrument(skip(self, msg), fields(session_id = tracing::field::Empty))]
     async fn handle_message(&self, msg: redis::Msg) -> anyhow::Result<()> {
-        use Notification::*;
         let payload: String = msg.get_payload()?;
         tracing::debug!(?payload, "received a message payload");
 
@@ -124,7 +123,7 @@ impl<C: ProjectInfoCache + Send + Sync + 'static> MessageHandler<C> {
         };
         tracing::debug!(?msg, "received a message");
         match msg {
-            Cancel(cancel_session) => {
+            Notification::Cancel(cancel_session) => {
                 tracing::Span::current().record(
                     "session_id",
                     tracing::field::display(cancel_session.session_id),
@@ -153,12 +152,12 @@ impl<C: ProjectInfoCache + Send + Sync + 'static> MessageHandler<C> {
             }
             _ => {
                 invalidate_cache(self.cache.clone(), msg.clone());
-                if matches!(msg, AllowedIpsUpdate { .. }) {
+                if matches!(msg, Notification::AllowedIpsUpdate { .. }) {
                     Metrics::get()
                         .proxy
                         .redis_events_count
                         .inc(RedisEventsCount::AllowedIpsUpdate);
-                } else if matches!(msg, PasswordUpdate { .. }) {
+                } else if matches!(msg, Notification::PasswordUpdate { .. }) {
                     Metrics::get()
                         .proxy
                         .redis_events_count
@@ -180,16 +179,16 @@ impl<C: ProjectInfoCache + Send + Sync + 'static> MessageHandler<C> {
 }
 
 fn invalidate_cache<C: ProjectInfoCache>(cache: Arc<C>, msg: Notification) {
-    use Notification::*;
     match msg {
-        AllowedIpsUpdate { allowed_ips_update } => {
-            cache.invalidate_allowed_ips_for_project(allowed_ips_update.project_id)
+        Notification::AllowedIpsUpdate { allowed_ips_update } => {
+            cache.invalidate_allowed_ips_for_project(allowed_ips_update.project_id);
         }
-        PasswordUpdate { password_update } => cache.invalidate_role_secret_for_project(
-            password_update.project_id,
-            password_update.role_name,
-        ),
-        Cancel(_) => unreachable!("cancel message should be handled separately"),
+        Notification::PasswordUpdate { password_update } => cache
+            .invalidate_role_secret_for_project(
+                password_update.project_id,
+                password_update.role_name,
+            ),
+        Notification::Cancel(_) => unreachable!("cancel message should be handled separately"),
     }
 }
 

proxy/src/sasl.rs

@@ -42,10 +42,9 @@ pub enum Error {
 
 impl UserFacingError for Error {
     fn to_string_client(&self) -> String {
-        use Error::*;
         match self {
-            ChannelBindingFailed(m) => m.to_string(),
-            ChannelBindingBadMethod(m) => format!("unsupported channel binding method {m}"),
+            Self::ChannelBindingFailed(m) => (*m).to_string(),
+            Self::ChannelBindingBadMethod(m) => format!("unsupported channel binding method {m}"),
             _ => "authentication protocol violation".to_string(),
         }
     }

proxy/src/sasl/channel_binding.rs

@@ -13,11 +13,10 @@ pub enum ChannelBinding<T> {
 
 impl<T> ChannelBinding<T> {
     pub fn and_then<R, E>(self, f: impl FnOnce(T) -> Result<R, E>) -> Result<ChannelBinding<R>, E> {
-        use ChannelBinding::*;
         Ok(match self {
-            NotSupportedClient => NotSupportedClient,
-            NotSupportedServer => NotSupportedServer,
-            Required(x) => Required(f(x)?),
+            Self::NotSupportedClient => ChannelBinding::NotSupportedClient,
+            Self::NotSupportedServer => ChannelBinding::NotSupportedServer,
+            Self::Required(x) => ChannelBinding::Required(f(x)?),
         })
     }
 }
@@ -25,11 +24,10 @@ impl<T> ChannelBinding<T> {
 impl<'a> ChannelBinding<&'a str> {
     // NB: FromStr doesn't work with lifetimes
     pub fn parse(input: &'a str) -> Option<Self> {
-        use ChannelBinding::*;
         Some(match input {
-            "n" => NotSupportedClient,
-            "y" => NotSupportedServer,
-            other => Required(other.strip_prefix("p=")?),
+            "n" => Self::NotSupportedClient,
+            "y" => Self::NotSupportedServer,
+            other => Self::Required(other.strip_prefix("p=")?),
         })
     }
 }
@@ -40,17 +38,16 @@ impl<T: std::fmt::Display> ChannelBinding<T> {
         &self,
         get_cbind_data: impl FnOnce(&T) -> Result<&'a [u8], E>,
     ) -> Result<std::borrow::Cow<'static, str>, E> {
-        use ChannelBinding::*;
         Ok(match self {
-            NotSupportedClient => {
+            Self::NotSupportedClient => {
                 // base64::encode("n,,")
                 "biws".into()
             }
-            NotSupportedServer => {
+            Self::NotSupportedServer => {
                 // base64::encode("y,,")
                 "eSws".into()
             }
-            Required(mode) => {
+            Self::Required(mode) => {
                 use std::io::Write;
                 let mut cbind_input = vec![];
                 write!(&mut cbind_input, "p={mode},,",).unwrap();

proxy/src/sasl/messages.rs

@@ -42,10 +42,9 @@ pub(super) enum ServerMessage<T> {
 
 impl<'a> ServerMessage<&'a str> {
     pub(super) fn to_reply(&self) -> BeMessage<'a> {
-        use BeAuthenticationSaslMessage::*;
         BeMessage::AuthenticationSasl(match self {
-            ServerMessage::Continue(s) => Continue(s.as_bytes()),
-            ServerMessage::Final(s) => Final(s.as_bytes()),
+            ServerMessage::Continue(s) => BeAuthenticationSaslMessage::Continue(s.as_bytes()),
+            ServerMessage::Final(s) => BeAuthenticationSaslMessage::Final(s.as_bytes()),
         })
     }
 }

proxy/src/scram.rs

@@ -137,12 +137,12 @@ mod tests {
 
     #[tokio::test]
     async fn round_trip() {
-        run_round_trip_test("pencil", "pencil").await
+        run_round_trip_test("pencil", "pencil").await;
     }
 
     #[tokio::test]
     #[should_panic(expected = "password doesn't match")]
     async fn failure() {
-        run_round_trip_test("pencil", "eraser").await
+        run_round_trip_test("pencil", "eraser").await;
     }
 }

proxy/src/scram/countmin.rs

@@ -98,8 +98,6 @@ mod tests {
         // q% of counts will be within p of the actual value
         let mut sketch = CountMinSketch::with_params(p / N as f64, 1.0 - q);
 
-        dbg!(sketch.buckets.len());
-
         // insert a bunch of entries in a random order
         let mut ids2 = ids.clone();
         while !ids2.is_empty() {

proxy/src/scram/exchange.rs

@@ -210,23 +210,23 @@ impl sasl::Mechanism for Exchange<'_> {
     type Output = super::ScramKey;
 
     fn exchange(mut self, input: &str) -> sasl::Result<sasl::Step<Self, Self::Output>> {
-        use {sasl::Step::*, ExchangeState::*};
+        use {sasl::Step, ExchangeState};
         match &self.state {
-            Initial(init) => {
+            ExchangeState::Initial(init) => {
                 match init.transition(self.secret, &self.tls_server_end_point, input)? {
-                    Continue(sent, msg) => {
-                        self.state = SaltSent(sent);
-                        Ok(Continue(self, msg))
+                    Step::Continue(sent, msg) => {
+                        self.state = ExchangeState::SaltSent(sent);
+                        Ok(Step::Continue(self, msg))
                     }
-                    Success(x, _) => match x {},
-                    Failure(msg) => Ok(Failure(msg)),
+                    Step::Success(x, _) => match x {},
+                    Step::Failure(msg) => Ok(Step::Failure(msg)),
                 }
             }
-            SaltSent(sent) => {
+            ExchangeState::SaltSent(sent) => {
                 match sent.transition(self.secret, &self.tls_server_end_point, input)? {
-                    Success(keys, msg) => Ok(Success(keys, msg)),
-                    Continue(x, _) => match x {},
-                    Failure(msg) => Ok(Failure(msg)),
+                    Step::Success(keys, msg) => Ok(Step::Success(keys, msg)),
+                    Step::Continue(x, _) => match x {},
+                    Step::Failure(msg) => Ok(Step::Failure(msg)),
                 }
             }
         }

proxy/src/scram/messages.rs

@@ -59,7 +59,7 @@ impl<'a> ClientFirstMessage<'a> {
 
         // https://github.com/postgres/postgres/blob/f83908798f78c4cafda217ca875602c88ea2ae28/src/backend/libpq/auth-scram.c#L13-L14
         if !username.is_empty() {
-            tracing::warn!(username, "scram username provided, but is not expected")
+            tracing::warn!(username, "scram username provided, but is not expected");
             // TODO(conrad):
             // return None;
         }
@@ -137,7 +137,7 @@ impl<'a> ClientFinalMessage<'a> {
     /// Build a response to [`ClientFinalMessage`].
     pub fn build_server_final_message(
         &self,
-        signature_builder: SignatureBuilder,
+        signature_builder: SignatureBuilder<'_>,
         server_key: &ScramKey,
     ) -> String {
         let mut buf = String::from("v=");
@@ -212,7 +212,7 @@ mod tests {
 
     #[test]
     fn parse_client_first_message_with_invalid_gs2_authz() {
-        assert!(ClientFirstMessage::parse("n,authzid,n=,r=nonce").is_none())
+        assert!(ClientFirstMessage::parse("n,authzid,n=,r=nonce").is_none());
     }
 
     #[test]

proxy/src/scram/pbkdf2.rs

@@ -84,6 +84,6 @@ mod tests {
         };
 
         let expected = pbkdf2_hmac_array::<Sha256, 32>(pass, salt, 600000);
-        assert_eq!(hash, expected)
+        assert_eq!(hash, expected);
     }
 }

proxy/src/scram/threadpool.rs

@@ -270,7 +270,7 @@ fn thread_rt(pool: Arc<ThreadPool>, worker: Worker<JobSpec>, index: usize) {
                         .inc(ThreadPoolWorkerId(index));
 
                     // skip for now
-                    worker.push(job)
+                    worker.push(job);
                 }
             }
 
@@ -316,6 +316,6 @@ mod tests {
             10, 114, 73, 188, 140, 222, 196, 156, 214, 184, 79, 157, 119, 242, 16, 31, 53, 242,
             178, 43, 95, 8, 225, 182, 122, 40, 219, 21, 89, 147, 64, 140,
         ];
-        assert_eq!(actual, expected)
+        assert_eq!(actual, expected);
     }
 }

proxy/src/serverless.rs

@@ -10,6 +10,7 @@ mod json;
 mod sql_over_http;
 mod websocket;
 
+use async_trait::async_trait;
 use atomic_take::AtomicTake;
 use bytes::Bytes;
 pub use conn_pool::GlobalConnPoolOptions;
@@ -26,8 +27,9 @@ use rand::rngs::StdRng;
 use rand::SeedableRng;
 pub use reqwest_middleware::{ClientWithMiddleware, Error};
 pub use reqwest_retry::{policies::ExponentialBackoff, RetryTransientMiddleware};
+use tokio::io::{AsyncRead, AsyncWrite};
 use tokio::time::timeout;
-use tokio_rustls::{server::TlsStream, TlsAcceptor};
+use tokio_rustls::TlsAcceptor;
 use tokio_util::task::TaskTracker;
 
 use crate::cancellation::CancellationHandlerMain;
@@ -41,7 +43,7 @@ use crate::serverless::backend::PoolingBackend;
 use crate::serverless::http_util::{api_error_into_response, json_response};
 
 use std::net::{IpAddr, SocketAddr};
-use std::pin::pin;
+use std::pin::{pin, Pin};
 use std::sync::Arc;
 use tokio::net::{TcpListener, TcpStream};
 use tokio_util::sync::CancellationToken;
@@ -86,18 +88,18 @@ pub async fn task_main(
         config,
         endpoint_rate_limiter: Arc::clone(&endpoint_rate_limiter),
     });
-
-    let tls_config = match config.tls_config.as_ref() {
-        Some(config) => config,
+    let tls_acceptor: Arc<dyn MaybeTlsAcceptor> = match config.tls_config.as_ref() {
+        Some(config) => {
+            let mut tls_server_config = rustls::ServerConfig::clone(&config.to_server_config());
+            // prefer http2, but support http/1.1
+            tls_server_config.alpn_protocols = vec![b"h2".to_vec(), b"http/1.1".to_vec()];
+            Arc::new(tls_server_config) as Arc<_>
+        }
         None => {
-            warn!("TLS config is missing, WebSocket Secure server will not be started");
-            return Ok(());
+            warn!("TLS config is missing");
+            Arc::new(NoTls) as Arc<_>
         }
     };
-    let mut tls_server_config = rustls::ServerConfig::clone(&tls_config.to_server_config());
-    // prefer http2, but support http/1.1
-    tls_server_config.alpn_protocols = vec![b"h2".to_vec(), b"http/1.1".to_vec()];
-    let tls_acceptor: tokio_rustls::TlsAcceptor = Arc::new(tls_server_config).into();
 
     let connections = tokio_util::task::task_tracker::TaskTracker::new();
     connections.close(); // allows `connections.wait to complete`
@@ -120,7 +122,7 @@ pub async fn task_main(
             tracing::trace!("attempting to cancel a random connection");
             if let Some(token) = config.http_config.cancel_set.take() {
                 tracing::debug!("cancelling a random connection");
-                token.cancel()
+                token.cancel();
             }
         }
 
@@ -176,16 +178,41 @@ pub async fn task_main(
     Ok(())
 }
 
+pub trait AsyncReadWrite: AsyncRead + AsyncWrite + Send + 'static {}
+impl<T: AsyncRead + AsyncWrite + Send + 'static> AsyncReadWrite for T {}
+pub type AsyncRW = Pin<Box<dyn AsyncReadWrite>>;
+
+#[async_trait]
+trait MaybeTlsAcceptor: Send + Sync + 'static {
+    async fn accept(self: Arc<Self>, conn: ChainRW<TcpStream>) -> std::io::Result<AsyncRW>;
+}
+
+#[async_trait]
+impl MaybeTlsAcceptor for rustls::ServerConfig {
+    async fn accept(self: Arc<Self>, conn: ChainRW<TcpStream>) -> std::io::Result<AsyncRW> {
+        Ok(Box::pin(TlsAcceptor::from(self).accept(conn).await?))
+    }
+}
+
+struct NoTls;
+
+#[async_trait]
+impl MaybeTlsAcceptor for NoTls {
+    async fn accept(self: Arc<Self>, conn: ChainRW<TcpStream>) -> std::io::Result<AsyncRW> {
+        Ok(Box::pin(conn))
+    }
+}
+
 /// Handles the TCP startup lifecycle.
 /// 1. Parses PROXY protocol V2
 /// 2. Handles TLS handshake
 async fn connection_startup(
     config: &ProxyConfig,
-    tls_acceptor: TlsAcceptor,
+    tls_acceptor: Arc<dyn MaybeTlsAcceptor>,
     session_id: uuid::Uuid,
     conn: TcpStream,
     peer_addr: SocketAddr,
-) -> Option<(TlsStream<ChainRW<TcpStream>>, IpAddr)> {
+) -> Option<(AsyncRW, IpAddr)> {
     // handle PROXY protocol
     let (conn, peer) = match read_proxy_protocol(conn).await {
         Ok(c) => c,
@@ -198,7 +225,7 @@ async fn connection_startup(
     let peer_addr = peer.unwrap_or(peer_addr).ip();
     let has_private_peer_addr = match peer_addr {
         IpAddr::V4(ip) => ip.is_private(),
-        _ => false,
+        IpAddr::V6(_) => false,
     };
     info!(?session_id, %peer_addr, "accepted new TCP connection");
 
@@ -241,7 +268,7 @@ async fn connection_handler(
     cancellation_handler: Arc<CancellationHandlerMain>,
     endpoint_rate_limiter: Arc<EndpointRateLimiter>,
     cancellation_token: CancellationToken,
-    conn: TlsStream<ChainRW<TcpStream>>,
+    conn: AsyncRW,
     peer_addr: IpAddr,
     session_id: uuid::Uuid,
 ) {
@@ -326,7 +353,9 @@ async fn request_handler(
         .map(|s| s.to_string());
 
     // Check if the request is a websocket upgrade request.
-    if framed_websockets::upgrade::is_upgrade_request(&request) {
+    if config.http_config.accept_websockets
+        && framed_websockets::upgrade::is_upgrade_request(&request)
+    {
         let ctx = RequestMonitoring::new(
             session_id,
             peer_addr,

proxy/src/serverless/conn_pool.rs

@@ -390,7 +390,7 @@ impl<C: ClientInnerExt> GlobalConnPool<C> {
             .write()
             .get_conn_entry(conn_info.db_and_user())
         {
-            client = Some(entry.conn)
+            client = Some(entry.conn);
         }
         let endpoint_pool = Arc::downgrade(&endpoint_pool);
 
@@ -662,13 +662,13 @@ impl<C: ClientInnerExt> Discard<'_, C> {
     pub fn check_idle(&mut self, status: ReadyForQueryStatus) {
         let conn_info = &self.conn_info;
         if status != ReadyForQueryStatus::Idle && std::mem::take(self.pool).strong_count() > 0 {
-            info!("pool: throwing away connection '{conn_info}' because connection is not idle")
+            info!("pool: throwing away connection '{conn_info}' because connection is not idle");
         }
     }
     pub fn discard(&mut self) {
         let conn_info = &self.conn_info;
         if std::mem::take(self.pool).strong_count() > 0 {
-            info!("pool: throwing away connection '{conn_info}' because connection is potentially in a broken state")
+            info!("pool: throwing away connection '{conn_info}' because connection is potentially in a broken state");
         }
     }
 }
@@ -758,6 +758,7 @@ mod tests {
     async fn test_pool() {
         let _ = env_logger::try_init();
         let config = Box::leak(Box::new(crate::config::HttpConfig {
+            accept_websockets: false,
             pool_options: GlobalConnPoolOptions {
                 max_conns_per_endpoint: 2,
                 gc_epoch: Duration::from_secs(1),

proxy/src/serverless/sql_over_http.rs

@@ -147,7 +147,7 @@ impl UserFacingError for ConnInfoError {
 fn get_conn_info(
     ctx: &RequestMonitoring,
     headers: &HeaderMap,
-    tls: &TlsConfig,
+    tls: Option<&TlsConfig>,
 ) -> Result<ConnInfo, ConnInfoError> {
     // HTTP only uses cleartext (for now and likely always)
     ctx.set_auth_method(crate::context::AuthMethod::Cleartext);
@@ -184,12 +184,22 @@ fn get_conn_info(
         .ok_or(ConnInfoError::MissingPassword)?;
     let password = urlencoding::decode_binary(password.as_bytes());
 
-    let hostname = connection_url
-        .host_str()
-        .ok_or(ConnInfoError::MissingHostname)?;
-
-    let endpoint =
-        endpoint_sni(hostname, &tls.common_names)?.ok_or(ConnInfoError::MalformedEndpoint)?;
+    let endpoint = match connection_url.host() {
+        Some(url::Host::Domain(hostname)) => {
+            if let Some(tls) = tls {
+                endpoint_sni(hostname, &tls.common_names)?
+                    .ok_or(ConnInfoError::MalformedEndpoint)?
+            } else {
+                hostname
+                    .split_once(".")
+                    .map_or(hostname, |(prefix, _)| prefix)
+                    .into()
+            }
+        }
+        Some(url::Host::Ipv4(_)) | Some(url::Host::Ipv6(_)) | None => {
+            return Err(ConnInfoError::MissingHostname)
+        }
+    };
     ctx.set_endpoint_id(endpoint.clone());
 
     let pairs = connection_url.query_pairs();
@@ -502,7 +512,7 @@ async fn handle_inner(
     let headers = request.headers();
 
     // TLS config should be there.
-    let conn_info = get_conn_info(ctx, headers, config.tls_config.as_ref().unwrap())?;
+    let conn_info = get_conn_info(ctx, headers, config.tls_config.as_ref())?;
     info!(user = conn_info.user_info.user.as_str(), "credentials");
 
     // Allow connection pooling only if explicitly requested

proxy/src/stream.rs

@@ -234,7 +234,7 @@ impl<S: AsyncRead + AsyncWrite + Unpin> Stream<S> {
                 .await
                 .inspect_err(|_| {
                     if record_handshake_error {
-                        Metrics::get().proxy.tls_handshake_failures.inc()
+                        Metrics::get().proxy.tls_handshake_failures.inc();
                     }
                 })?),
             Stream::Tls { .. } => Err(StreamUpgradeError::AlreadyTls),

proxy/src/url.rs

@@ -12,7 +12,7 @@ impl ApiUrl {
     }
 
     /// See [`url::Url::path_segments_mut`].
-    pub fn path_segments_mut(&mut self) -> url::PathSegmentsMut {
+    pub fn path_segments_mut(&mut self) -> url::PathSegmentsMut<'_> {
         // We've already verified that it works during construction.
         self.0.path_segments_mut().expect("bad API url")
     }

proxy/src/waiters.rs

@@ -36,7 +36,7 @@ impl<T> Default for Waiters<T> {
 }
 
 impl<T> Waiters<T> {
-    pub fn register(&self, key: String) -> Result<Waiter<T>, RegisterError> {
+    pub fn register(&self, key: String) -> Result<Waiter<'_, T>, RegisterError> {
         let (tx, rx) = oneshot::channel();
 
         self.0

safekeeper/src/safekeeper.rs

@@ -92,7 +92,7 @@ impl TermHistory {
     }
 
     /// Find point of divergence between leader (walproposer) term history and
-    /// safekeeper. Arguments are not symmetrics as proposer history ends at
+    /// safekeeper. Arguments are not symmetric as proposer history ends at
     /// +infinity while safekeeper at flush_lsn.
     /// C version is at walproposer SendProposerElected.
     pub fn find_highest_common_point(
@@ -701,7 +701,13 @@ where
             .with_label_values(&["handle_elected"])
             .start_timer();
 
-        info!("received ProposerElected {:?}", msg);
+        info!(
+            "received ProposerElected {:?}, term={}, last_log_term={}, flush_lsn={}",
+            msg,
+            self.state.acceptor_state.term,
+            self.get_last_log_term(),
+            self.flush_lsn()
+        );
         if self.state.acceptor_state.term < msg.term {
             let mut state = self.state.start_change();
             state.acceptor_state.term = msg.term;
@@ -713,22 +719,43 @@ where
             return Ok(None);
         }
 
-        // This might happen in a rare race when another (old) connection from
-        // the same walproposer writes + flushes WAL after this connection
-        // already sent flush_lsn in VoteRequest. It is generally safe to
-        // proceed, but to prevent commit_lsn surprisingly going down we should
-        // either refuse the session (simpler) or skip the part we already have
-        // from the stream (can be implemented).
-        if msg.term == self.get_last_log_term() && self.flush_lsn() > msg.start_streaming_at {
-            bail!("refusing ProposerElected which is going to overwrite correct WAL: term={}, flush_lsn={}, start_streaming_at={}; restarting the handshake should help",
-                   msg.term, self.flush_lsn(), msg.start_streaming_at)
+        // Before truncating WAL check-cross the check divergence point received
+        // from the walproposer.
+        let sk_th = self.get_term_history();
+        let last_common_point = match TermHistory::find_highest_common_point(
+            &msg.term_history,
+            &sk_th,
+            self.flush_lsn(),
+        ) {
+            // No common point. Expect streaming from the beginning of the
+            // history like walproposer while we don't have proper init.
+            None => *msg.term_history.0.first().ok_or(anyhow::anyhow!(
+                "empty walproposer term history {:?}",
+                msg.term_history
+            ))?,
+            Some(lcp) => lcp,
+        };
+        // This is expected to happen in a rare race when another connection
+        // from the same walproposer writes + flushes WAL after this connection
+        // sent flush_lsn in VoteRequest; for instance, very late
+        // ProposerElected message delivery after another connection was
+        // established and wrote WAL. In such cases error is transient;
+        // reconnection makes safekeeper send newest term history and flush_lsn
+        // and walproposer recalculates the streaming point. OTOH repeating
+        // error indicates a serious bug.
+        if last_common_point.lsn != msg.start_streaming_at {
+            bail!("refusing ProposerElected with unexpected truncation point: lcp={:?} start_streaming_at={}, term={}, sk_th={:?} flush_lsn={}, wp_th={:?}",
+                    last_common_point, msg.start_streaming_at,
+                    self.state.acceptor_state.term, sk_th, self.flush_lsn(), msg.term_history,
+            );
         }
-        // Otherwise we must never attempt to truncate committed data.
+
+        // We are also expected to never attempt to truncate committed data.
         assert!(
             msg.start_streaming_at >= self.state.inmem.commit_lsn,
-            "attempt to truncate committed data: start_streaming_at={}, commit_lsn={}",
-            msg.start_streaming_at,
-            self.state.inmem.commit_lsn
+            "attempt to truncate committed data: start_streaming_at={}, commit_lsn={}, term={}, sk_th={:?} flush_lsn={}, wp_th={:?}",
+            msg.start_streaming_at, self.state.inmem.commit_lsn,
+            self.state.acceptor_state.term, sk_th, self.flush_lsn(), msg.term_history,
         );
 
         // Before first WAL write initialize its segment. It makes first segment
@@ -743,9 +770,6 @@ where
                 .await?;
         }
 
-        // TODO: cross check divergence point, check if msg.start_streaming_at corresponds to
-        // intersection of our history and history from msg
-
         // truncate wal, update the LSNs
         self.wal_store.truncate_wal(msg.start_streaming_at).await?;
 
@@ -1069,7 +1093,7 @@ mod tests {
 
         let pem = ProposerElected {
             term: 1,
-            start_streaming_at: Lsn(1),
+            start_streaming_at: Lsn(3),
             term_history: TermHistory(vec![TermLsn {
                 term: 1,
                 lsn: Lsn(3),

scripts/ps_ec2_setup_instance_store

@@ -44,7 +44,7 @@ run the following commands from the top of the neon.git checkout
 
     # test suite run
     export TEST_OUTPUT="$TEST_OUTPUT"
-    DEFAULT_PG_VERSION=15 BUILD_TYPE=release ./scripts/pytest test_runner/performance/test_latency.py
+    DEFAULT_PG_VERSION=16 BUILD_TYPE=release ./scripts/pytest test_runner/performance/test_latency.py
 
     # for interactive use
     export NEON_REPO_DIR="$NEON_REPO_DIR"

storage_controller/src/http.rs

@@ -520,6 +520,19 @@ async fn handle_node_status(req: Request<Body>) -> Result<Response<Body>, ApiErr
     json_response(StatusCode::OK, node_status)
 }
 
+async fn handle_get_leader(req: Request<Body>) -> Result<Response<Body>, ApiError> {
+    check_permissions(&req, Scope::Admin)?;
+
+    let state = get_state(&req);
+    let leader = state.service.get_leader().await.map_err(|err| {
+        ApiError::InternalServerError(anyhow::anyhow!(
+            "Failed to read leader from database: {err}"
+        ))
+    })?;
+
+    json_response(StatusCode::OK, leader)
+}
+
 async fn handle_node_drain(req: Request<Body>) -> Result<Response<Body>, ApiError> {
     check_permissions(&req, Scope::Admin)?;
 
@@ -1016,6 +1029,9 @@ pub fn make_router(
         .get("/control/v1/node/:node_id", |r| {
             named_request_span(r, handle_node_status, RequestName("control_v1_node_status"))
         })
+        .get("/control/v1/leader", |r| {
+            named_request_span(r, handle_get_leader, RequestName("control_v1_get_leader"))
+        })
         .put("/control/v1/node/:node_id/drain", |r| {
             named_request_span(r, handle_node_drain, RequestName("control_v1_node_drain"))
         })

storage_controller/src/leadership.rs

@@ -0,0 +1,135 @@
+use std::sync::Arc;
+
+use hyper::Uri;
+use tokio_util::sync::CancellationToken;
+
+use crate::{
+    peer_client::{GlobalObservedState, PeerClient},
+    persistence::{ControllerPersistence, DatabaseError, DatabaseResult, Persistence},
+    service::Config,
+};
+
+/// Helper for storage controller leadership acquisition
+pub(crate) struct Leadership {
+    persistence: Arc<Persistence>,
+    config: Config,
+    cancel: CancellationToken,
+}
+
+#[derive(thiserror::Error, Debug)]
+pub(crate) enum Error {
+    #[error(transparent)]
+    Database(#[from] DatabaseError),
+}
+
+pub(crate) type Result<T> = std::result::Result<T, Error>;
+
+impl Leadership {
+    pub(crate) fn new(
+        persistence: Arc<Persistence>,
+        config: Config,
+        cancel: CancellationToken,
+    ) -> Self {
+        Self {
+            persistence,
+            config,
+            cancel,
+        }
+    }
+
+    /// Find the current leader in the database and request it to step down if required.
+    /// Should be called early on in within the start-up sequence.
+    ///
+    /// Returns a tuple of two optionals: the current leader and its observed state
+    pub(crate) async fn step_down_current_leader(
+        &self,
+    ) -> Result<(Option<ControllerPersistence>, Option<GlobalObservedState>)> {
+        let leader = self.current_leader().await?;
+        let leader_step_down_state = if let Some(ref leader) = leader {
+            if self.config.start_as_candidate {
+                self.request_step_down(leader).await
+            } else {
+                None
+            }
+        } else {
+            tracing::info!("No leader found to request step down from. Will build observed state.");
+            None
+        };
+
+        Ok((leader, leader_step_down_state))
+    }
+
+    /// Mark the current storage controller instance as the leader in the database
+    pub(crate) async fn become_leader(
+        &self,
+        current_leader: Option<ControllerPersistence>,
+    ) -> Result<()> {
+        if let Some(address_for_peers) = &self.config.address_for_peers {
+            // TODO: `address-for-peers` can become a mandatory cli arg
+            // after we update the k8s setup
+            let proposed_leader = ControllerPersistence {
+                address: address_for_peers.to_string(),
+                started_at: chrono::Utc::now(),
+            };
+
+            self.persistence
+                .update_leader(current_leader, proposed_leader)
+                .await
+                .map_err(Error::Database)
+        } else {
+            tracing::info!("No address-for-peers provided. Skipping leader persistence.");
+            Ok(())
+        }
+    }
+
+    async fn current_leader(&self) -> DatabaseResult<Option<ControllerPersistence>> {
+        let res = self.persistence.get_leader().await;
+        if let Err(DatabaseError::Query(diesel::result::Error::DatabaseError(_kind, ref err))) = res
+        {
+            const REL_NOT_FOUND_MSG: &str = "relation \"controllers\" does not exist";
+            if err.message().trim() == REL_NOT_FOUND_MSG {
+                // Special case: if this is a brand new storage controller, migrations will not
+                // have run at this point yet, and, hence, the controllers table does not exist.
+                // Detect this case via the error string (diesel doesn't type it) and allow it.
+                tracing::info!("Detected first storage controller start-up. Allowing missing controllers table ...");
+                return Ok(None);
+            }
+        }
+
+        res
+    }
+
+    /// Request step down from the currently registered leader in the database
+    ///
+    /// If such an entry is persisted, the success path returns the observed
+    /// state and details of the leader. Otherwise, None is returned indicating
+    /// there is no leader currently.
+    async fn request_step_down(
+        &self,
+        leader: &ControllerPersistence,
+    ) -> Option<GlobalObservedState> {
+        tracing::info!("Sending step down request to {leader:?}");
+
+        let client = PeerClient::new(
+            Uri::try_from(leader.address.as_str()).expect("Failed to build leader URI"),
+            self.config.peer_jwt_token.clone(),
+        );
+        let state = client.step_down(&self.cancel).await;
+        match state {
+            Ok(state) => Some(state),
+            Err(err) => {
+                // TODO: Make leaders periodically update a timestamp field in the
+                // database and, if the leader is not reachable from the current instance,
+                // but inferred as alive from the timestamp, abort start-up. This avoids
+                // a potential scenario in which we have two controllers acting as leaders.
+                tracing::error!(
+                    "Leader ({}) did not respond to step-down request: {}",
+                    leader.address,
+                    err
+                );
+
+                None
+            }
+        }
+    }
+}

storage_controller/src/lib.rs

@@ -8,6 +8,7 @@ mod drain_utils;
 mod heartbeater;
 pub mod http;
 mod id_lock_map;
+mod leadership;
 pub mod metrics;
 mod node;
 mod pageserver_client;

storage_controller/src/main.rs

@@ -1,6 +1,5 @@
 use anyhow::{anyhow, Context};
 use clap::Parser;
-use diesel::Connection;
 use hyper::Uri;
 use metrics::launch_timestamp::LaunchTimestamp;
 use metrics::BuildInfo;
@@ -27,9 +26,6 @@ use utils::{project_build_tag, project_git_version, tcp_listener};
 project_git_version!(GIT_VERSION);
 project_build_tag!(BUILD_TAG);
 
-use diesel_migrations::{embed_migrations, EmbeddedMigrations};
-pub const MIGRATIONS: EmbeddedMigrations = embed_migrations!("./migrations");
-
 #[derive(Parser)]
 #[command(author, version, about, long_about = None)]
 #[command(arg_required_else_help(true))]
@@ -51,6 +47,9 @@ struct Cli {
     #[arg(long)]
     control_plane_jwt_token: Option<String>,
 
+    #[arg(long)]
+    peer_jwt_token: Option<String>,
+
     /// URL to control plane compute notification endpoint
     #[arg(long)]
     compute_hook_url: Option<String>,
@@ -130,28 +129,28 @@ struct Secrets {
     public_key: Option<JwtAuth>,
     jwt_token: Option<String>,
     control_plane_jwt_token: Option<String>,
+    peer_jwt_token: Option<String>,
 }
 
 impl Secrets {
     const DATABASE_URL_ENV: &'static str = "DATABASE_URL";
     const PAGESERVER_JWT_TOKEN_ENV: &'static str = "PAGESERVER_JWT_TOKEN";
     const CONTROL_PLANE_JWT_TOKEN_ENV: &'static str = "CONTROL_PLANE_JWT_TOKEN";
+    const PEER_JWT_TOKEN_ENV: &'static str = "PEER_JWT_TOKEN";
     const PUBLIC_KEY_ENV: &'static str = "PUBLIC_KEY";
 
     /// Load secrets from, in order of preference:
     /// - CLI args if database URL is provided on the CLI
     /// - Environment variables if DATABASE_URL is set.
-    /// - AWS Secrets Manager secrets
     async fn load(args: &Cli) -> anyhow::Result<Self> {
-        let Some(database_url) =
-            Self::load_secret(&args.database_url, Self::DATABASE_URL_ENV).await
+        let Some(database_url) = Self::load_secret(&args.database_url, Self::DATABASE_URL_ENV)
         else {
             anyhow::bail!(
                 "Database URL is not set (set `--database-url`, or `DATABASE_URL` environment)"
             )
         };
 
-        let public_key = match Self::load_secret(&args.public_key, Self::PUBLIC_KEY_ENV).await {
+        let public_key = match Self::load_secret(&args.public_key, Self::PUBLIC_KEY_ENV) {
             Some(v) => Some(JwtAuth::from_key(v).context("Loading public key")?),
             None => None,
         };
@@ -159,18 +158,18 @@ impl Secrets {
         let this = Self {
             database_url,
             public_key,
-            jwt_token: Self::load_secret(&args.jwt_token, Self::PAGESERVER_JWT_TOKEN_ENV).await,
+            jwt_token: Self::load_secret(&args.jwt_token, Self::PAGESERVER_JWT_TOKEN_ENV),
             control_plane_jwt_token: Self::load_secret(
                 &args.control_plane_jwt_token,
                 Self::CONTROL_PLANE_JWT_TOKEN_ENV,
-            )
-            .await,
+            ),
+            peer_jwt_token: Self::load_secret(&args.peer_jwt_token, Self::PEER_JWT_TOKEN_ENV),
         };
 
         Ok(this)
     }
 
-    async fn load_secret(cli: &Option<String>, env_name: &str) -> Option<String> {
+    fn load_secret(cli: &Option<String>, env_name: &str) -> Option<String> {
         if let Some(v) = cli {
             Some(v.clone())
         } else if let Ok(v) = std::env::var(env_name) {
@@ -181,29 +180,27 @@ impl Secrets {
     }
 }
 
-/// Execute the diesel migrations that are built into this binary
-async fn migration_run(database_url: &str) -> anyhow::Result<()> {
-    use diesel::PgConnection;
-    use diesel_migrations::{HarnessWithOutput, MigrationHarness};
-    let mut conn = PgConnection::establish(database_url)?;
-
-    HarnessWithOutput::write_to_stdout(&mut conn)
-        .run_pending_migrations(MIGRATIONS)
-        .map(|_| ())
-        .map_err(|e| anyhow::anyhow!(e))?;
-
-    Ok(())
-}
-
 fn main() -> anyhow::Result<()> {
-    let default_panic = std::panic::take_hook();
-    std::panic::set_hook(Box::new(move |info| {
-        default_panic(info);
-        std::process::exit(1);
-    }));
+    logging::init(
+        LogFormat::Plain,
+        logging::TracingErrorLayerEnablement::Disabled,
+        logging::Output::Stdout,
+    )?;
+
+    // log using tracing so we don't get confused output by default hook writing to stderr
+    utils::logging::replace_panic_hook_with_tracing_panic_hook().forget();
 
     let _sentry_guard = init_sentry(Some(GIT_VERSION.into()), &[]);
 
+    let hook = std::panic::take_hook();
+    std::panic::set_hook(Box::new(move |info| {
+        // let sentry send a message (and flush)
+        // and trace the error
+        hook(info);
+
+        std::process::exit(1);
+    }));
+
     tokio::runtime::Builder::new_current_thread()
         // We use spawn_blocking for database operations, so require approximately
         // as many blocking threads as we will open database connections.
@@ -217,12 +214,6 @@ fn main() -> anyhow::Result<()> {
 async fn async_main() -> anyhow::Result<()> {
     let launch_ts = Box::leak(Box::new(LaunchTimestamp::generate()));
 
-    logging::init(
-        LogFormat::Plain,
-        logging::TracingErrorLayerEnablement::Disabled,
-        logging::Output::Stdout,
-    )?;
-
     preinitialize_metrics();
 
     let args = Cli::parse();
@@ -278,6 +269,7 @@ async fn async_main() -> anyhow::Result<()> {
     let config = Config {
         jwt_token: secrets.jwt_token,
         control_plane_jwt_token: secrets.control_plane_jwt_token,
+        peer_jwt_token: secrets.peer_jwt_token,
         compute_hook_url: args.compute_hook_url,
         max_offline_interval: args
             .max_offline_interval
@@ -298,13 +290,9 @@ async fn async_main() -> anyhow::Result<()> {
         http_service_port: args.listen.port() as i32,
     };
 
-    // After loading secrets & config, but before starting anything else, apply database migrations
+    // Validate that we can connect to the database
     Persistence::await_connection(&secrets.database_url, args.db_connect_timeout.into()).await?;
 
-    migration_run(&secrets.database_url)
-        .await
-        .context("Running database migrations")?;
-
     let persistence = Arc::new(Persistence::new(secrets.database_url));
 
     let service = Service::spawn(config, persistence.clone()).await?;

storage_controller/src/metrics.rs

@@ -230,6 +230,7 @@ pub(crate) enum DatabaseErrorLabel {
     Connection,
     ConnectionPool,
     Logical,
+    Migration,
 }
 
 impl DatabaseError {
@@ -239,6 +240,7 @@ impl DatabaseError {
             Self::Connection(_) => DatabaseErrorLabel::Connection,
             Self::ConnectionPool(_) => DatabaseErrorLabel::ConnectionPool,
             Self::Logical(_) => DatabaseErrorLabel::Logical,
+            Self::Migration(_) => DatabaseErrorLabel::Migration,
         }
     }
 }

storage_controller/src/peer_client.rs

@@ -1,7 +1,7 @@
 use crate::tenant_shard::ObservedState;
 use pageserver_api::shard::TenantShardId;
 use serde::{Deserialize, Serialize};
-use std::collections::HashMap;
+use std::{collections::HashMap, time::Duration};
 use tokio_util::sync::CancellationToken;
 
 use hyper::Uri;
@@ -69,6 +69,8 @@ impl PeerClient {
             req
         };
 
+        let req = req.timeout(Duration::from_secs(2));
+
         let res = req
             .send()
             .await

storage_controller/src/persistence.rs

@@ -25,6 +25,9 @@ use crate::metrics::{
 };
 use crate::node::Node;
 
+use diesel_migrations::{embed_migrations, EmbeddedMigrations};
+const MIGRATIONS: EmbeddedMigrations = embed_migrations!("./migrations");
+
 /// ## What do we store?
 ///
 /// The storage controller service does not store most of its state durably.
@@ -72,6 +75,8 @@ pub(crate) enum DatabaseError {
     ConnectionPool(#[from] r2d2::Error),
     #[error("Logical error: {0}")]
     Logical(String),
+    #[error("Migration error: {0}")]
+    Migration(String),
 }
 
 #[derive(measured::FixedCardinalityLabel, Copy, Clone)]
@@ -167,6 +172,19 @@ impl Persistence {
         }
     }
 
+    /// Execute the diesel migrations that are built into this binary
+    pub(crate) async fn migration_run(&self) -> DatabaseResult<()> {
+        use diesel_migrations::{HarnessWithOutput, MigrationHarness};
+
+        self.with_conn(move |conn| -> DatabaseResult<()> {
+            HarnessWithOutput::write_to_stdout(conn)
+                .run_pending_migrations(MIGRATIONS)
+                .map(|_| ())
+                .map_err(|e| DatabaseError::Migration(e.to_string()))
+        })
+        .await
+    }
+
     /// Wraps `with_conn` in order to collect latency and error metrics
     async fn with_measured_conn<F, R>(&self, op: DatabaseOperation, func: F) -> DatabaseResult<R>
     where

storage_controller/src/service.rs

@@ -17,10 +17,12 @@ use crate::{
     compute_hook::NotifyError,
     drain_utils::{self, TenantShardDrain, TenantShardIterator},
     id_lock_map::{trace_exclusive_lock, trace_shared_lock, IdLockMap, TracingExclusiveGuard},
+    leadership::Leadership,
     metrics,
-    peer_client::{GlobalObservedState, PeerClient},
+    peer_client::GlobalObservedState,
     persistence::{
-        AbortShardSplitStatus, ControllerPersistence, MetadataHealthPersistence, TenantFilter,
+        AbortShardSplitStatus, ControllerPersistence, DatabaseResult, MetadataHealthPersistence,
+        TenantFilter,
     },
     reconciler::{ReconcileError, ReconcileUnits, ReconcilerConfig, ReconcilerConfigBuilder},
     scheduler::{MaySchedule, ScheduleContext, ScheduleMode},
@@ -286,6 +288,9 @@ pub struct Config {
     // This JWT token will be used to authenticate this service to the control plane.
     pub control_plane_jwt_token: Option<String>,
 
+    // This JWT token will be used to authenticate with other storage controller instances
+    pub peer_jwt_token: Option<String>,
+
     /// Where the compute hook should send notifications of pageserver attachment locations
     /// (this URL points to the control plane in prod). If this is None, the compute hook will
     /// assume it is running in a test environment and try to update neon_local.
@@ -332,7 +337,7 @@ impl From<DatabaseError> for ApiError {
             DatabaseError::Connection(_) | DatabaseError::ConnectionPool(_) => {
                 ApiError::ShuttingDown
             }
-            DatabaseError::Logical(reason) => {
+            DatabaseError::Logical(reason) | DatabaseError::Migration(reason) => {
                 ApiError::InternalServerError(anyhow::anyhow!(reason))
             }
         }
@@ -489,11 +494,6 @@ pub(crate) enum ReconcileResultRequest {
     Stop,
 }
 
-struct LeaderStepDownState {
-    observed: GlobalObservedState,
-    leader: ControllerPersistence,
-}
-
 impl Service {
     pub fn get_config(&self) -> &Config {
         &self.config
@@ -504,7 +504,8 @@ impl Service {
     #[instrument(skip_all)]
     async fn startup_reconcile(
         self: &Arc<Service>,
-        leader_step_down_state: Option<LeaderStepDownState>,
+        current_leader: Option<ControllerPersistence>,
+        leader_step_down_state: Option<GlobalObservedState>,
         bg_compute_notify_result_tx: tokio::sync::mpsc::Sender<
             Result<(), (TenantShardId, NotifyError)>,
         >,
@@ -522,17 +523,15 @@ impl Service {
             .checked_add(STARTUP_RECONCILE_TIMEOUT / 2)
             .expect("Reconcile timeout is a modest constant");
 
-        let (observed, current_leader) = if let Some(state) = leader_step_down_state {
+        let observed = if let Some(state) = leader_step_down_state {
             tracing::info!(
                 "Using observed state received from leader at {}",
-                state.leader.address,
+                current_leader.as_ref().unwrap().address
             );
-            (state.observed, Some(state.leader))
+
+            state
         } else {
-            (
-                self.build_global_observed_state(node_scan_deadline).await,
-                None,
-            )
+            self.build_global_observed_state(node_scan_deadline).await
         };
 
         // Accumulate a list of any tenant locations that ought to be detached
@@ -611,22 +610,15 @@ impl Service {
 
         // Before making any obeservable changes to the cluster, persist self
         // as leader in database and memory.
-        if let Some(address_for_peers) = &self.config.address_for_peers {
-            // TODO: `address-for-peers` can become a mandatory cli arg
-            // after we update the k8s setup
-            let proposed_leader = ControllerPersistence {
-                address: address_for_peers.to_string(),
-                started_at: chrono::Utc::now(),
-            };
+        let leadership = Leadership::new(
+            self.persistence.clone(),
+            self.config.clone(),
+            self.cancel.child_token(),
+        );
 
-            if let Err(err) = self
-                .persistence
-                .update_leader(current_leader, proposed_leader)
-                .await
-            {
-                tracing::error!("Failed to persist self as leader: {err}. Aborting start-up ...");
-                std::process::exit(1);
-            }
+        if let Err(e) = leadership.become_leader(current_leader).await {
+            tracing::error!("Failed to persist self as leader: {e}. Aborting start-up ...");
+            std::process::exit(1);
         }
 
         self.inner.write().unwrap().become_leader();
@@ -1164,6 +1156,16 @@ impl Service {
         let (result_tx, result_rx) = tokio::sync::mpsc::unbounded_channel();
         let (abort_tx, abort_rx) = tokio::sync::mpsc::unbounded_channel();
 
+        let leadership_cancel = CancellationToken::new();
+        let leadership = Leadership::new(persistence.clone(), config.clone(), leadership_cancel);
+        let (leader, leader_step_down_state) = leadership.step_down_current_leader().await?;
+
+        // Apply the migrations **after** the current leader has stepped down
+        // (or we've given up waiting for it), but **before** reading from the
+        // database. The only exception is reading the current leader before
+        // migrating.
+        persistence.migration_run().await?;
+
         tracing::info!("Loading nodes from database...");
         let nodes = persistence
             .list_nodes()
@@ -1381,14 +1383,7 @@ impl Service {
                     return;
                 };
 
-                let leadership_status = this.inner.read().unwrap().get_leadership_status();
-                let peer_observed_state = match leadership_status {
-                    LeadershipStatus::Candidate => this.request_step_down().await,
-                    LeadershipStatus::Leader => None,
-                    LeadershipStatus::SteppedDown => unreachable!(),
-                };
-
-                this.startup_reconcile(peer_observed_state, bg_compute_notify_result_tx)
+                this.startup_reconcile(leader, leader_step_down_state, bg_compute_notify_result_tx)
                     .await;
 
                 drop(startup_completion);
@@ -4650,6 +4645,10 @@ impl Service {
             ))
     }
 
+    pub(crate) async fn get_leader(&self) -> DatabaseResult<Option<ControllerPersistence>> {
+        self.persistence.get_leader().await
+    }
+
     pub(crate) async fn node_register(
         &self,
         register_req: NodeRegisterRequest,
@@ -6342,6 +6341,7 @@ impl Service {
 
     pub(crate) async fn step_down(&self) -> GlobalObservedState {
         tracing::info!("Received step down request from peer");
+        failpoint_support::sleep_millis_async!("sleep-on-step-down-handling");
 
         self.inner.write().unwrap().step_down();
         // TODO: would it make sense to have a time-out for this?
@@ -6358,61 +6358,4 @@ impl Service {
 
         global_observed
     }
-
-    /// Request step down from the currently registered leader in the database
-    ///
-    /// If such an entry is persisted, the success path returns the observed
-    /// state and details of the leader. Otherwise, None is returned indicating
-    /// there is no leader currently.
-    ///
-    /// On failures to query the database or step down error responses the process is killed
-    /// and we rely on k8s to retry.
-    async fn request_step_down(&self) -> Option<LeaderStepDownState> {
-        let leader = match self.persistence.get_leader().await {
-            Ok(leader) => leader,
-            Err(err) => {
-                tracing::error!(
-                    "Failed to query database for current leader: {err}. Aborting start-up ..."
-                );
-                std::process::exit(1);
-            }
-        };
-
-        match leader {
-            Some(leader) => {
-                tracing::info!("Sending step down request to {leader:?}");
-
-                // TODO: jwt token
-                let client = PeerClient::new(
-                    Uri::try_from(leader.address.as_str()).expect("Failed to build leader URI"),
-                    self.config.jwt_token.clone(),
-                );
-                let state = client.step_down(&self.cancel).await;
-                match state {
-                    Ok(state) => Some(LeaderStepDownState {
-                        observed: state,
-                        leader: leader.clone(),
-                    }),
-                    Err(err) => {
-                        // TODO: Make leaders periodically update a timestamp field in the
-                        // database and, if the leader is not reachable from the current instance,
-                        // but inferred as alive from the timestamp, abort start-up. This avoids
-                        // a potential scenario in which we have two controllers acting as leaders.
-                        tracing::error!(
-                            "Leader ({}) did not respond to step-down request: {}",
-                            leader.address,
-                            err
-                        );
-                        None
-                    }
-                }
-            }
-            None => {
-                tracing::info!(
-                    "No leader found to request step down from. Will build observed state."
-                );
-                None
-            }
-        }
-    }
 }

storage_scrubber/src/checks.rs

@@ -1,10 +1,10 @@
 use std::collections::{HashMap, HashSet};
 
 use anyhow::Context;
-use aws_sdk_s3::Client;
 use pageserver::tenant::layer_map::LayerMap;
 use pageserver::tenant::remote_timeline_client::index::LayerFileMetadata;
 use pageserver_api::shard::ShardIndex;
+use tokio_util::sync::CancellationToken;
 use tracing::{error, info, warn};
 use utils::generation::Generation;
 use utils::id::TimelineId;
@@ -16,7 +16,7 @@ use futures_util::StreamExt;
 use pageserver::tenant::remote_timeline_client::{parse_remote_index_path, remote_layer_path};
 use pageserver::tenant::storage_layer::LayerName;
 use pageserver::tenant::IndexPart;
-use remote_storage::RemotePath;
+use remote_storage::{GenericRemoteStorage, ListingObject, RemotePath};
 
 pub(crate) struct TimelineAnalysis {
     /// Anomalies detected
@@ -48,13 +48,12 @@ impl TimelineAnalysis {
 }
 
 pub(crate) async fn branch_cleanup_and_check_errors(
-    s3_client: &Client,
-    target: &RootTarget,
+    remote_client: &GenericRemoteStorage,
     id: &TenantShardTimelineId,
     tenant_objects: &mut TenantObjectListing,
     s3_active_branch: Option<&BranchData>,
     console_branch: Option<BranchData>,
-    s3_data: Option<S3TimelineBlobData>,
+    s3_data: Option<RemoteTimelineBlobData>,
 ) -> TimelineAnalysis {
     let mut result = TimelineAnalysis::new();
 
@@ -78,7 +77,9 @@ pub(crate) async fn branch_cleanup_and_check_errors(
 
     match s3_data {
         Some(s3_data) => {
-            result.garbage_keys.extend(s3_data.unknown_keys);
+            result
+                .garbage_keys
+                .extend(s3_data.unknown_keys.into_iter().map(|k| k.key.to_string()));
 
             match s3_data.blob_data {
                 BlobDataParseResult::Parsed {
@@ -143,11 +144,8 @@ pub(crate) async fn branch_cleanup_and_check_errors(
 
                             // HEAD request used here to address a race condition  when an index was uploaded concurrently
                             // with our scan. We check if the object is uploaded to S3 after taking the listing snapshot.
-                            let response = s3_client
-                                .head_object()
-                                .bucket(target.bucket_name())
-                                .key(path.get_path().as_str())
-                                .send()
+                            let response = remote_client
+                                .head_object(&path, &CancellationToken::new())
                                 .await;
 
                             if response.is_err() {
@@ -284,14 +282,14 @@ impl TenantObjectListing {
 }
 
 #[derive(Debug)]
-pub(crate) struct S3TimelineBlobData {
+pub(crate) struct RemoteTimelineBlobData {
     pub(crate) blob_data: BlobDataParseResult,
 
     // Index objects that were not used when loading `blob_data`, e.g. those from old generations
-    pub(crate) unused_index_keys: Vec<String>,
+    pub(crate) unused_index_keys: Vec<ListingObject>,
 
     // Objects whose keys were not recognized at all, i.e. not layer files, not indices
-    pub(crate) unknown_keys: Vec<String>,
+    pub(crate) unknown_keys: Vec<ListingObject>,
 }
 
 #[derive(Debug)]
@@ -323,31 +321,37 @@ pub(crate) fn parse_layer_object_name(name: &str) -> Result<(LayerName, Generati
 }
 
 pub(crate) async fn list_timeline_blobs(
-    s3_client: &Client,
+    remote_client: &GenericRemoteStorage,
     id: TenantShardTimelineId,
-    s3_root: &RootTarget,
-) -> anyhow::Result<S3TimelineBlobData> {
+    root_target: &RootTarget,
+) -> anyhow::Result<RemoteTimelineBlobData> {
     let mut s3_layers = HashSet::new();
 
     let mut errors = Vec::new();
     let mut unknown_keys = Vec::new();
 
-    let mut timeline_dir_target = s3_root.timeline_root(&id);
+    let mut timeline_dir_target = root_target.timeline_root(&id);
     timeline_dir_target.delimiter = String::new();
 
-    let mut index_part_keys: Vec<String> = Vec::new();
+    let mut index_part_keys: Vec<ListingObject> = Vec::new();
     let mut initdb_archive: bool = false;
 
-    let mut stream = std::pin::pin!(stream_listing(s3_client, &timeline_dir_target));
-    while let Some(obj) = stream.next().await {
-        let obj = obj?;
-        let key = obj.key();
+    let prefix_str = &timeline_dir_target
+        .prefix_in_bucket
+        .strip_prefix("/")
+        .unwrap_or(&timeline_dir_target.prefix_in_bucket);
 
-        let blob_name = key.strip_prefix(&timeline_dir_target.prefix_in_bucket);
+    let mut stream = std::pin::pin!(stream_listing(remote_client, &timeline_dir_target));
+    while let Some(obj) = stream.next().await {
+        let (key, Some(obj)) = obj? else {
+            panic!("ListingObject not specified");
+        };
+
+        let blob_name = key.get_path().as_str().strip_prefix(prefix_str);
         match blob_name {
             Some(name) if name.starts_with("index_part.json") => {
                 tracing::debug!("Index key {key}");
-                index_part_keys.push(key.to_owned())
+                index_part_keys.push(obj)
             }
             Some("initdb.tar.zst") => {
                 tracing::debug!("initdb archive {key}");
@@ -358,7 +362,7 @@ pub(crate) async fn list_timeline_blobs(
             }
             Some(maybe_layer_name) => match parse_layer_object_name(maybe_layer_name) {
                 Ok((new_layer, gen)) => {
-                    tracing::debug!("Parsed layer key: {} {:?}", new_layer, gen);
+                    tracing::debug!("Parsed layer key: {new_layer} {gen:?}");
                     s3_layers.insert((new_layer, gen));
                 }
                 Err(e) => {
@@ -366,13 +370,13 @@ pub(crate) async fn list_timeline_blobs(
                     errors.push(
                         format!("S3 list response got an object with key {key} that is not a layer name: {e}"),
                     );
-                    unknown_keys.push(key.to_string());
+                    unknown_keys.push(obj);
                 }
             },
             None => {
-                tracing::warn!("Unknown key {}", key);
+                tracing::warn!("Unknown key {key}");
                 errors.push(format!("S3 list response got an object with odd key {key}"));
-                unknown_keys.push(key.to_string());
+                unknown_keys.push(obj);
             }
         }
     }
@@ -381,7 +385,7 @@ pub(crate) async fn list_timeline_blobs(
         tracing::debug!(
             "Timeline is empty apart from initdb archive: expected post-deletion state."
         );
-        return Ok(S3TimelineBlobData {
+        return Ok(RemoteTimelineBlobData {
             blob_data: BlobDataParseResult::Relic,
             unused_index_keys: index_part_keys,
             unknown_keys: Vec::new(),
@@ -395,13 +399,13 @@ pub(crate) async fn list_timeline_blobs(
             // Stripping the index key to the last part, because RemotePath doesn't
             // like absolute paths, and depending on prefix_in_bucket it's possible
             // for the keys we read back to start with a slash.
-            let basename = key.rsplit_once('/').unwrap().1;
+            let basename = key.key.get_path().as_str().rsplit_once('/').unwrap().1;
             parse_remote_index_path(RemotePath::from_string(basename).unwrap()).map(|g| (key, g))
         })
         .max_by_key(|i| i.1)
         .map(|(k, g)| (k.clone(), g))
     {
-        Some((key, gen)) => (Some(key), gen),
+        Some((key, gen)) => (Some::<ListingObject>(key.to_owned()), gen),
         None => {
             // Legacy/missing case: one or zero index parts, which did not have a generation
             (index_part_keys.pop(), Generation::none())
@@ -416,17 +420,14 @@ pub(crate) async fn list_timeline_blobs(
     }
 
     if let Some(index_part_object_key) = index_part_object.as_ref() {
-        let index_part_bytes = download_object_with_retries(
-            s3_client,
-            &timeline_dir_target.bucket_name,
-            index_part_object_key,
-        )
-        .await
-        .context("index_part.json download")?;
+        let index_part_bytes =
+            download_object_with_retries(remote_client, &index_part_object_key.key)
+                .await
+                .context("index_part.json download")?;
 
         match serde_json::from_slice(&index_part_bytes) {
             Ok(index_part) => {
-                return Ok(S3TimelineBlobData {
+                return Ok(RemoteTimelineBlobData {
                     blob_data: BlobDataParseResult::Parsed {
                         index_part: Box::new(index_part),
                         index_part_generation,
@@ -448,7 +449,7 @@ pub(crate) async fn list_timeline_blobs(
         );
     }
 
-    Ok(S3TimelineBlobData {
+    Ok(RemoteTimelineBlobData {
         blob_data: BlobDataParseResult::Incorrect { errors, s3_layers },
         unused_index_keys: index_part_keys,
         unknown_keys,

storage_scrubber/src/find_large_objects.rs

@@ -6,7 +6,7 @@ use remote_storage::ListingMode;
 use serde::{Deserialize, Serialize};
 
 use crate::{
-    checks::parse_layer_object_name, init_remote_generic, metadata_stream::stream_tenants_generic,
+    checks::parse_layer_object_name, init_remote, metadata_stream::stream_tenants,
     stream_objects_with_retries, BucketConfig, NodeKind,
 };
 
@@ -50,9 +50,8 @@ pub async fn find_large_objects(
     ignore_deltas: bool,
     concurrency: usize,
 ) -> anyhow::Result<LargeObjectListing> {
-    let (remote_client, target) =
-        init_remote_generic(bucket_config.clone(), NodeKind::Pageserver).await?;
-    let tenants = pin!(stream_tenants_generic(&remote_client, &target));
+    let (remote_client, target) = init_remote(bucket_config.clone(), NodeKind::Pageserver).await?;
+    let tenants = pin!(stream_tenants(&remote_client, &target));
 
     let objects_stream = tenants.map_ok(|tenant_shard_id| {
         let mut tenant_root = target.tenant_root(&tenant_shard_id);

storage_scrubber/src/garbage.rs

@@ -19,8 +19,8 @@ use utils::id::TenantId;
 
 use crate::{
     cloud_admin_api::{CloudAdminApiClient, MaybeDeleted, ProjectData},
-    init_remote_generic, list_objects_with_retries_generic,
-    metadata_stream::{stream_tenant_timelines_generic, stream_tenants_generic},
+    init_remote, list_objects_with_retries,
+    metadata_stream::{stream_tenant_timelines, stream_tenants},
     BucketConfig, ConsoleConfig, NodeKind, TenantShardTimelineId, TraversingDepth,
 };
 
@@ -153,7 +153,7 @@ async fn find_garbage_inner(
     node_kind: NodeKind,
 ) -> anyhow::Result<GarbageList> {
     // Construct clients for S3 and for Console API
-    let (remote_client, target) = init_remote_generic(bucket_config.clone(), node_kind).await?;
+    let (remote_client, target) = init_remote(bucket_config.clone(), node_kind).await?;
     let cloud_admin_api_client = Arc::new(CloudAdminApiClient::new(console_config));
 
     // Build a set of console-known tenants, for quickly eliminating known-active tenants without having
@@ -179,7 +179,7 @@ async fn find_garbage_inner(
 
     // Enumerate Tenants in S3, and check if each one exists in Console
     tracing::info!("Finding all tenants in bucket {}...", bucket_config.bucket);
-    let tenants = stream_tenants_generic(&remote_client, &target);
+    let tenants = stream_tenants(&remote_client, &target);
     let tenants_checked = tenants.map_ok(|t| {
         let api_client = cloud_admin_api_client.clone();
         let console_cache = console_cache.clone();
@@ -237,14 +237,13 @@ async fn find_garbage_inner(
         // Special case: If it's missing in console, check for known bugs that would enable us to conclusively
         // identify it as purge-able anyway
         if console_result.is_none() {
-            let timelines =
-                stream_tenant_timelines_generic(&remote_client, &target, tenant_shard_id)
-                    .await?
-                    .collect::<Vec<_>>()
-                    .await;
+            let timelines = stream_tenant_timelines(&remote_client, &target, tenant_shard_id)
+                .await?
+                .collect::<Vec<_>>()
+                .await;
             if timelines.is_empty() {
                 // No timelines, but a heatmap: the deletion bug where we deleted everything but heatmaps
-                let tenant_objects = list_objects_with_retries_generic(
+                let tenant_objects = list_objects_with_retries(
                     &remote_client,
                     ListingMode::WithDelimiter,
                     &target.tenant_root(&tenant_shard_id),
@@ -265,7 +264,7 @@ async fn find_garbage_inner(
 
                 for timeline_r in timelines {
                     let timeline = timeline_r?;
-                    let timeline_objects = list_objects_with_retries_generic(
+                    let timeline_objects = list_objects_with_retries(
                         &remote_client,
                         ListingMode::WithDelimiter,
                         &target.timeline_root(&timeline),
@@ -331,8 +330,7 @@ async fn find_garbage_inner(
 
     // Construct a stream of all timelines within active tenants
     let active_tenants = tokio_stream::iter(active_tenants.iter().map(Ok));
-    let timelines =
-        active_tenants.map_ok(|t| stream_tenant_timelines_generic(&remote_client, &target, *t));
+    let timelines = active_tenants.map_ok(|t| stream_tenant_timelines(&remote_client, &target, *t));
     let timelines = timelines.try_buffer_unordered(S3_CONCURRENCY);
     let timelines = timelines.try_flatten();
 
@@ -507,7 +505,7 @@ pub async fn purge_garbage(
     );
 
     let (remote_client, _target) =
-        init_remote_generic(garbage_list.bucket_config.clone(), garbage_list.node_kind).await?;
+        init_remote(garbage_list.bucket_config.clone(), garbage_list.node_kind).await?;
 
     assert_eq!(
         &garbage_list.bucket_config.bucket,

storage_scrubber/src/lib.rs

@@ -15,7 +15,7 @@ use std::fmt::Display;
 use std::sync::Arc;
 use std::time::Duration;
 
-use anyhow::{anyhow, Context};
+use anyhow::Context;
 use aws_config::retry::{RetryConfigBuilder, RetryMode};
 use aws_sdk_s3::config::Region;
 use aws_sdk_s3::error::DisplayErrorContext;
@@ -352,7 +352,7 @@ fn make_root_target(
     }
 }
 
-async fn init_remote(
+async fn init_remote_s3(
     bucket_config: BucketConfig,
     node_kind: NodeKind,
 ) -> anyhow::Result<(Arc<Client>, RootTarget)> {
@@ -369,7 +369,7 @@ async fn init_remote(
     Ok((s3_client, s3_root))
 }
 
-async fn init_remote_generic(
+async fn init_remote(
     bucket_config: BucketConfig,
     node_kind: NodeKind,
 ) -> anyhow::Result<(GenericRemoteStorage, RootTarget)> {
@@ -394,45 +394,10 @@ async fn init_remote_generic(
 
     // We already pass the prefix to the remote client above
     let prefix_in_root_target = String::new();
-    let s3_root = make_root_target(bucket_config.bucket, prefix_in_root_target, node_kind);
+    let root_target = make_root_target(bucket_config.bucket, prefix_in_root_target, node_kind);
 
     let client = GenericRemoteStorage::from_config(&storage_config).await?;
-    Ok((client, s3_root))
-}
-
-async fn list_objects_with_retries(
-    s3_client: &Client,
-    s3_target: &S3Target,
-    continuation_token: Option<String>,
-) -> anyhow::Result<aws_sdk_s3::operation::list_objects_v2::ListObjectsV2Output> {
-    for trial in 0..MAX_RETRIES {
-        match s3_client
-            .list_objects_v2()
-            .bucket(&s3_target.bucket_name)
-            .prefix(&s3_target.prefix_in_bucket)
-            .delimiter(&s3_target.delimiter)
-            .set_continuation_token(continuation_token.clone())
-            .send()
-            .await
-        {
-            Ok(response) => return Ok(response),
-            Err(e) => {
-                if trial == MAX_RETRIES - 1 {
-                    return Err(e)
-                        .with_context(|| format!("Failed to list objects {MAX_RETRIES} times"));
-                }
-                error!(
-                    "list_objects_v2 query failed: bucket_name={}, prefix={}, delimiter={}, error={}",
-                    s3_target.bucket_name,
-                    s3_target.prefix_in_bucket,
-                    s3_target.delimiter,
-                    DisplayErrorContext(e),
-                );
-                tokio::time::sleep(Duration::from_secs(1)).await;
-            }
-        }
-    }
-    Err(anyhow!("unreachable unless MAX_RETRIES==0"))
+    Ok((client, root_target))
 }
 
 /// Listing possibly large amounts of keys in a streaming fashion.
@@ -452,23 +417,26 @@ fn stream_objects_with_retries<'a>(
         let mut list_stream =
             storage_client.list_streaming(Some(&prefix), listing_mode, None, &cancel);
         while let Some(res) = list_stream.next().await {
-            if let Err(err) = res {
-                let yield_err = if err.is_permanent() {
-                    true
-                } else {
-                    let backoff_time = 1 << trial.max(5);
-                    tokio::time::sleep(Duration::from_secs(backoff_time)).await;
-                    trial += 1;
-                    trial == MAX_RETRIES - 1
-                };
-                if yield_err {
-                    yield Err(err)
-                        .with_context(|| format!("Failed to list objects {MAX_RETRIES} times"));
-                    break;
+            match res {
+                Err(err) => {
+                    let yield_err = if err.is_permanent() {
+                        true
+                    } else {
+                        let backoff_time = 1 << trial.max(5);
+                        tokio::time::sleep(Duration::from_secs(backoff_time)).await;
+                        trial += 1;
+                        trial == MAX_RETRIES - 1
+                    };
+                    if yield_err {
+                        yield Err(err)
+                            .with_context(|| format!("Failed to list objects {MAX_RETRIES} times"));
+                        break;
+                    }
+                }
+                Ok(res) => {
+                    trial = 0;
+                    yield Ok(res);
                 }
-            } else {
-                trial = 0;
-                yield res.map_err(anyhow::Error::from);
             }
         }
     }
@@ -476,7 +444,7 @@ fn stream_objects_with_retries<'a>(
 
 /// If you want to list a bounded amount of prefixes or keys. For larger numbers of keys/prefixes,
 /// use [`stream_objects_with_retries`] instead.
-async fn list_objects_with_retries_generic(
+async fn list_objects_with_retries(
     remote_client: &GenericRemoteStorage,
     listing_mode: ListingMode,
     s3_target: &S3Target,
@@ -514,40 +482,34 @@ async fn list_objects_with_retries_generic(
 }
 
 async fn download_object_with_retries(
-    s3_client: &Client,
-    bucket_name: &str,
-    key: &str,
+    remote_client: &GenericRemoteStorage,
+    key: &RemotePath,
 ) -> anyhow::Result<Vec<u8>> {
-    for _ in 0..MAX_RETRIES {
-        let mut body_buf = Vec::new();
-        let response_stream = match s3_client
-            .get_object()
-            .bucket(bucket_name)
-            .key(key)
-            .send()
-            .await
-        {
+    let cancel = CancellationToken::new();
+    for trial in 0..MAX_RETRIES {
+        let mut buf = Vec::new();
+        let download = match remote_client.download(key, &cancel).await {
             Ok(response) => response,
             Err(e) => {
                 error!("Failed to download object for key {key}: {e}");
-                tokio::time::sleep(Duration::from_secs(1)).await;
+                let backoff_time = 1 << trial.max(5);
+                tokio::time::sleep(Duration::from_secs(backoff_time)).await;
                 continue;
             }
         };
 
-        match response_stream
-            .body
-            .into_async_read()
-            .read_to_end(&mut body_buf)
+        match tokio_util::io::StreamReader::new(download.download_stream)
+            .read_to_end(&mut buf)
             .await
         {
             Ok(bytes_read) => {
                 tracing::debug!("Downloaded {bytes_read} bytes for object {key}");
-                return Ok(body_buf);
+                return Ok(buf);
             }
             Err(e) => {
                 error!("Failed to stream object body for key {key}: {e}");
-                tokio::time::sleep(Duration::from_secs(1)).await;
+                let backoff_time = 1 << trial.max(5);
+                tokio::time::sleep(Duration::from_secs(backoff_time)).await;
             }
         }
     }
@@ -555,7 +517,7 @@ async fn download_object_with_retries(
     anyhow::bail!("Failed to download objects with key {key} {MAX_RETRIES} times")
 }
 
-async fn download_object_to_file(
+async fn download_object_to_file_s3(
     s3_client: &Client,
     bucket_name: &str,
     key: &str,

storage_scrubber/src/main.rs

@@ -3,9 +3,10 @@ use camino::Utf8PathBuf;
 use pageserver_api::controller_api::{MetadataHealthUpdateRequest, MetadataHealthUpdateResponse};
 use pageserver_api::shard::TenantShardId;
 use reqwest::{Method, Url};
+use storage_controller_client::control_api;
 use storage_scrubber::garbage::{find_garbage, purge_garbage, PurgeMode};
 use storage_scrubber::pageserver_physical_gc::GcMode;
-use storage_scrubber::scan_pageserver_metadata::scan_metadata;
+use storage_scrubber::scan_pageserver_metadata::scan_pageserver_metadata;
 use storage_scrubber::tenant_snapshot::SnapshotDownloader;
 use storage_scrubber::{find_large_objects, ControllerClientConfig};
 use storage_scrubber::{
@@ -68,7 +69,7 @@ enum Command {
         #[arg(long = "tenant-id", num_args = 0..)]
         tenant_ids: Vec<TenantShardId>,
         #[arg(long = "post", default_value_t = false)]
-        post_to_storage_controller: bool,
+        post_to_storcon: bool,
         #[arg(long, default_value = None)]
         /// For safekeeper node_kind only, points to db with debug dump
         dump_db_connstr: Option<String>,
@@ -100,6 +101,16 @@ enum Command {
         #[arg(long = "concurrency", short = 'j', default_value_t = 64)]
         concurrency: usize,
     },
+    CronJob {
+        // PageserverPhysicalGc
+        #[arg(long = "min-age")]
+        gc_min_age: humantime::Duration,
+        #[arg(short, long, default_value_t = GcMode::IndicesOnly)]
+        gc_mode: GcMode,
+        // ScanMetadata
+        #[arg(long = "post", default_value_t = false)]
+        post_to_storcon: bool,
+    },
 }
 
 #[tokio::main]
@@ -117,6 +128,7 @@ async fn main() -> anyhow::Result<()> {
         Command::TenantSnapshot { .. } => "tenant-snapshot",
         Command::PageserverPhysicalGc { .. } => "pageserver-physical-gc",
         Command::FindLargeObjects { .. } => "find-large-objects",
+        Command::CronJob { .. } => "cron-job",
     };
     let _guard = init_logging(&format!(
         "{}_{}_{}_{}.log",
@@ -126,12 +138,13 @@ async fn main() -> anyhow::Result<()> {
         chrono::Utc::now().format("%Y_%m_%d__%H_%M_%S")
     ));
 
-    let controller_client_conf = cli.controller_api.map(|controller_api| {
+    let controller_client = cli.controller_api.map(|controller_api| {
         ControllerClientConfig {
             controller_api,
             // Default to no key: this is a convenience when working in a development environment
             controller_jwt: cli.controller_jwt.unwrap_or("".to_owned()),
         }
+        .build_client()
     });
 
     match cli.command {
@@ -139,7 +152,7 @@ async fn main() -> anyhow::Result<()> {
             json,
             tenant_ids,
             node_kind,
-            post_to_storage_controller,
+            post_to_storcon,
             dump_db_connstr,
             dump_db_table,
         } => {
@@ -178,53 +191,14 @@ async fn main() -> anyhow::Result<()> {
                 }
                 Ok(())
             } else {
-                if controller_client_conf.is_none() && post_to_storage_controller {
-                    return Err(anyhow!("Posting pageserver scan health status to storage controller requires `--controller-api` and `--controller-jwt` to run"));
-                }
-                match scan_metadata(bucket_config.clone(), tenant_ids).await {
-                    Err(e) => {
-                        tracing::error!("Failed: {e}");
-                        Err(e)
-                    }
-                    Ok(summary) => {
-                        if json {
-                            println!("{}", serde_json::to_string(&summary).unwrap())
-                        } else {
-                            println!("{}", summary.summary_string());
-                        }
-
-                        if post_to_storage_controller {
-                            if let Some(conf) = controller_client_conf {
-                                let controller_client = conf.build_client();
-                                let body = summary.build_health_update_request();
-                                controller_client
-                                    .dispatch::<MetadataHealthUpdateRequest, MetadataHealthUpdateResponse>(
-                                        Method::POST,
-                                        "control/v1/metadata_health/update".to_string(),
-                                        Some(body),
-                                    )
-                                    .await?;
-                            }
-                        }
-
-                        if summary.is_fatal() {
-                            tracing::error!("Fatal scrub errors detected");
-                        } else if summary.is_empty() {
-                            // Strictly speaking an empty bucket is a valid bucket, but if someone ran the
-                            // scrubber they were likely expecting to scan something, and if we see no timelines
-                            // at all then it's likely due to some configuration issues like a bad prefix
-                            tracing::error!(
-                                "No timelines found in bucket {} prefix {}",
-                                bucket_config.bucket,
-                                bucket_config
-                                    .prefix_in_bucket
-                                    .unwrap_or("<none>".to_string())
-                            );
-                        }
-
-                        Ok(())
-                    }
-                }
+                scan_pageserver_metadata_cmd(
+                    bucket_config,
+                    controller_client.as_ref(),
+                    tenant_ids,
+                    json,
+                    post_to_storcon,
+                )
+                .await
             }
         }
         Command::FindGarbage {
@@ -254,31 +228,14 @@ async fn main() -> anyhow::Result<()> {
             min_age,
             mode,
         } => {
-            match (&controller_client_conf, mode) {
-                (Some(_), _) => {
-                    // Any mode may run when controller API is set
-                }
-                (None, GcMode::Full) => {
-                    // The part of physical GC where we erase ancestor layers cannot be done safely without
-                    // confirming the most recent complete shard split with the controller.  Refuse to run, rather
-                    // than doing it unsafely.
-                    return Err(anyhow!("Full physical GC requires `--controller-api` and `--controller-jwt` to run"));
-                }
-                (None, GcMode::DryRun | GcMode::IndicesOnly) => {
-                    // These GcModes do not require the controller to run.
-                }
-            }
-
-            let summary = pageserver_physical_gc(
-                bucket_config,
-                controller_client_conf,
+            pageserver_physical_gc_cmd(
+                &bucket_config,
+                controller_client.as_ref(),
                 tenant_ids,
-                min_age.into(),
+                min_age,
                 mode,
             )
-            .await?;
-            println!("{}", serde_json::to_string(&summary).unwrap());
-            Ok(())
+            .await
         }
         Command::FindLargeObjects {
             min_size,
@@ -295,5 +252,142 @@ async fn main() -> anyhow::Result<()> {
             println!("{}", serde_json::to_string(&summary).unwrap());
             Ok(())
         }
+        Command::CronJob {
+            gc_min_age,
+            gc_mode,
+            post_to_storcon,
+        } => {
+            run_cron_job(
+                bucket_config,
+                controller_client.as_ref(),
+                gc_min_age,
+                gc_mode,
+                post_to_storcon,
+            )
+            .await
+        }
+    }
+}
+
+/// Runs the scrubber cron job.
+/// 1. Do pageserver physical gc
+/// 2. Scan pageserver metadata
+pub async fn run_cron_job(
+    bucket_config: BucketConfig,
+    controller_client: Option<&control_api::Client>,
+    gc_min_age: humantime::Duration,
+    gc_mode: GcMode,
+    post_to_storcon: bool,
+) -> anyhow::Result<()> {
+    tracing::info!(%gc_min_age, %gc_mode, "Running pageserver-physical-gc");
+    pageserver_physical_gc_cmd(
+        &bucket_config,
+        controller_client,
+        Vec::new(),
+        gc_min_age,
+        gc_mode,
+    )
+    .await?;
+    tracing::info!(%post_to_storcon, node_kind = %NodeKind::Pageserver, "Running scan-metadata");
+    scan_pageserver_metadata_cmd(
+        bucket_config,
+        controller_client,
+        Vec::new(),
+        true,
+        post_to_storcon,
+    )
+    .await?;
+
+    Ok(())
+}
+
+pub async fn pageserver_physical_gc_cmd(
+    bucket_config: &BucketConfig,
+    controller_client: Option<&control_api::Client>,
+    tenant_shard_ids: Vec<TenantShardId>,
+    min_age: humantime::Duration,
+    mode: GcMode,
+) -> anyhow::Result<()> {
+    match (controller_client, mode) {
+        (Some(_), _) => {
+            // Any mode may run when controller API is set
+        }
+        (None, GcMode::Full) => {
+            // The part of physical GC where we erase ancestor layers cannot be done safely without
+            // confirming the most recent complete shard split with the controller.  Refuse to run, rather
+            // than doing it unsafely.
+            return Err(anyhow!(
+                "Full physical GC requires `--controller-api` and `--controller-jwt` to run"
+            ));
+        }
+        (None, GcMode::DryRun | GcMode::IndicesOnly) => {
+            // These GcModes do not require the controller to run.
+        }
+    }
+
+    let summary = pageserver_physical_gc(
+        bucket_config,
+        controller_client,
+        tenant_shard_ids,
+        min_age.into(),
+        mode,
+    )
+    .await?;
+    println!("{}", serde_json::to_string(&summary).unwrap());
+    Ok(())
+}
+
+pub async fn scan_pageserver_metadata_cmd(
+    bucket_config: BucketConfig,
+    controller_client: Option<&control_api::Client>,
+    tenant_shard_ids: Vec<TenantShardId>,
+    json: bool,
+    post_to_storcon: bool,
+) -> anyhow::Result<()> {
+    if controller_client.is_none() && post_to_storcon {
+        return Err(anyhow!("Posting pageserver scan health status to storage controller requires `--controller-api` and `--controller-jwt` to run"));
+    }
+    match scan_pageserver_metadata(bucket_config.clone(), tenant_shard_ids).await {
+        Err(e) => {
+            tracing::error!("Failed: {e}");
+            Err(e)
+        }
+        Ok(summary) => {
+            if json {
+                println!("{}", serde_json::to_string(&summary).unwrap())
+            } else {
+                println!("{}", summary.summary_string());
+            }
+
+            if post_to_storcon {
+                if let Some(client) = controller_client {
+                    let body = summary.build_health_update_request();
+                    client
+                        .dispatch::<MetadataHealthUpdateRequest, MetadataHealthUpdateResponse>(
+                            Method::POST,
+                            "control/v1/metadata_health/update".to_string(),
+                            Some(body),
+                        )
+                        .await?;
+                }
+            }
+
+            if summary.is_fatal() {
+                tracing::error!("Fatal scrub errors detected");
+            } else if summary.is_empty() {
+                // Strictly speaking an empty bucket is a valid bucket, but if someone ran the
+                // scrubber they were likely expecting to scan something, and if we see no timelines
+                // at all then it's likely due to some configuration issues like a bad prefix
+                tracing::error!(
+                    "No timelines found in bucket {} prefix {}",
+                    bucket_config.bucket,
+                    bucket_config
+                        .prefix_in_bucket
+                        .unwrap_or("<none>".to_string())
+                );
+            }
+
+            Ok(())
+        }
     }
 }

storage_scrubber/src/metadata_stream.rs

@@ -2,7 +2,6 @@ use std::str::FromStr;
 
 use anyhow::{anyhow, Context};
 use async_stream::{stream, try_stream};
-use aws_sdk_s3::{types::ObjectIdentifier, Client};
 use futures::StreamExt;
 use remote_storage::{GenericRemoteStorage, ListingMode, ListingObject, RemotePath};
 use tokio_stream::Stream;
@@ -15,7 +14,7 @@ use pageserver_api::shard::TenantShardId;
 use utils::id::{TenantId, TimelineId};
 
 /// Given a remote storage and a target, output a stream of TenantIds discovered via listing prefixes
-pub fn stream_tenants_generic<'a>(
+pub fn stream_tenants<'a>(
     remote_client: &'a GenericRemoteStorage,
     target: &'a RootTarget,
 ) -> impl Stream<Item = anyhow::Result<TenantShardId>> + 'a {
@@ -36,92 +35,36 @@ pub fn stream_tenants_generic<'a>(
     }
 }
 
-/// Given an S3 bucket, output a stream of TenantIds discovered via ListObjectsv2
-pub fn stream_tenants<'a>(
-    s3_client: &'a Client,
-    target: &'a RootTarget,
-) -> impl Stream<Item = anyhow::Result<TenantShardId>> + 'a {
-    try_stream! {
-        let mut continuation_token = None;
-        let tenants_target = target.tenants_root();
-        loop {
-            let fetch_response =
-                list_objects_with_retries(s3_client, &tenants_target, continuation_token.clone()).await?;
-
-            let new_entry_ids = fetch_response
-                .common_prefixes()
-                .iter()
-                .filter_map(|prefix| prefix.prefix())
-                .filter_map(|prefix| -> Option<&str> {
-                    prefix
-                        .strip_prefix(&tenants_target.prefix_in_bucket)?
-                        .strip_suffix('/')
-                }).map(|entry_id_str| {
-                entry_id_str
-                    .parse()
-                    .with_context(|| format!("Incorrect entry id str: {entry_id_str}"))
-            });
-
-            for i in new_entry_ids {
-                yield i?;
-            }
-
-            match fetch_response.next_continuation_token {
-                Some(new_token) => continuation_token = Some(new_token),
-                None => break,
-            }
-        }
-    }
-}
-
 pub async fn stream_tenant_shards<'a>(
-    s3_client: &'a Client,
+    remote_client: &'a GenericRemoteStorage,
     target: &'a RootTarget,
     tenant_id: TenantId,
 ) -> anyhow::Result<impl Stream<Item = Result<TenantShardId, anyhow::Error>> + 'a> {
-    let mut tenant_shard_ids: Vec<Result<TenantShardId, anyhow::Error>> = Vec::new();
-    let mut continuation_token = None;
     let shards_target = target.tenant_shards_prefix(&tenant_id);
 
-    loop {
-        tracing::info!("Listing in {}", shards_target.prefix_in_bucket);
-        let fetch_response =
-            list_objects_with_retries(s3_client, &shards_target, continuation_token.clone()).await;
-        let fetch_response = match fetch_response {
-            Err(e) => {
-                tenant_shard_ids.push(Err(e));
-                break;
-            }
-            Ok(r) => r,
-        };
+    let strip_prefix = target.tenants_root().prefix_in_bucket;
+    let prefix_str = &strip_prefix.strip_prefix("/").unwrap_or(&strip_prefix);
 
-        let new_entry_ids = fetch_response
-            .common_prefixes()
-            .iter()
-            .filter_map(|prefix| prefix.prefix())
-            .filter_map(|prefix| -> Option<&str> {
-                prefix
-                    .strip_prefix(&target.tenants_root().prefix_in_bucket)?
-                    .strip_suffix('/')
-            })
-            .map(|entry_id_str| {
-                let first_part = entry_id_str.split('/').next().unwrap();
+    tracing::info!("Listing shards in {}", shards_target.prefix_in_bucket);
+    let listing =
+        list_objects_with_retries(remote_client, ListingMode::WithDelimiter, &shards_target)
+            .await?;
 
-                first_part
-                    .parse::<TenantShardId>()
-                    .with_context(|| format!("Incorrect entry id str: {first_part}"))
-            });
+    let tenant_shard_ids = listing
+        .prefixes
+        .iter()
+        .map(|prefix| prefix.get_path().as_str())
+        .filter_map(|prefix| -> Option<&str> { prefix.strip_prefix(prefix_str) })
+        .map(|entry_id_str| {
+            let first_part = entry_id_str.split('/').next().unwrap();
 
-        for i in new_entry_ids {
-            tenant_shard_ids.push(i);
-        }
-
-        match fetch_response.next_continuation_token {
-            Some(new_token) => continuation_token = Some(new_token),
-            None => break,
-        }
-    }
+            first_part
+                .parse::<TenantShardId>()
+                .with_context(|| format!("Incorrect entry id str: {first_part}"))
+        })
+        .collect::<Vec<_>>();
 
+    tracing::debug!("Yielding {} shards for {tenant_id}", tenant_shard_ids.len());
     Ok(stream! {
         for i in tenant_shard_ids {
             let id = i?;
@@ -130,69 +73,10 @@ pub async fn stream_tenant_shards<'a>(
     })
 }
 
-/// Given a TenantShardId, output a stream of the timelines within that tenant, discovered
-/// using ListObjectsv2.  The listing is done before the stream is built, so that this
-/// function can be used to generate concurrency on a stream using buffer_unordered.
-pub async fn stream_tenant_timelines<'a>(
-    s3_client: &'a Client,
-    target: &'a RootTarget,
-    tenant: TenantShardId,
-) -> anyhow::Result<impl Stream<Item = Result<TenantShardTimelineId, anyhow::Error>> + 'a> {
-    let mut timeline_ids: Vec<Result<TimelineId, anyhow::Error>> = Vec::new();
-    let mut continuation_token = None;
-    let timelines_target = target.timelines_root(&tenant);
-
-    loop {
-        tracing::debug!("Listing in {}", tenant);
-        let fetch_response =
-            list_objects_with_retries(s3_client, &timelines_target, continuation_token.clone())
-                .await;
-        let fetch_response = match fetch_response {
-            Err(e) => {
-                timeline_ids.push(Err(e));
-                break;
-            }
-            Ok(r) => r,
-        };
-
-        let new_entry_ids = fetch_response
-            .common_prefixes()
-            .iter()
-            .filter_map(|prefix| prefix.prefix())
-            .filter_map(|prefix| -> Option<&str> {
-                prefix
-                    .strip_prefix(&timelines_target.prefix_in_bucket)?
-                    .strip_suffix('/')
-            })
-            .map(|entry_id_str| {
-                entry_id_str
-                    .parse::<TimelineId>()
-                    .with_context(|| format!("Incorrect entry id str: {entry_id_str}"))
-            });
-
-        for i in new_entry_ids {
-            timeline_ids.push(i);
-        }
-
-        match fetch_response.next_continuation_token {
-            Some(new_token) => continuation_token = Some(new_token),
-            None => break,
-        }
-    }
-
-    tracing::debug!("Yielding for {}", tenant);
-    Ok(stream! {
-        for i in timeline_ids {
-            let id = i?;
-            yield Ok(TenantShardTimelineId::new(tenant, id));
-        }
-    })
-}
-
 /// Given a `TenantShardId`, output a stream of the timelines within that tenant, discovered
 /// using a listing. The listing is done before the stream is built, so that this
 /// function can be used to generate concurrency on a stream using buffer_unordered.
-pub async fn stream_tenant_timelines_generic<'a>(
+pub async fn stream_tenant_timelines<'a>(
     remote_client: &'a GenericRemoteStorage,
     target: &'a RootTarget,
     tenant: TenantShardId,
@@ -200,6 +84,11 @@ pub async fn stream_tenant_timelines_generic<'a>(
     let mut timeline_ids: Vec<Result<TimelineId, anyhow::Error>> = Vec::new();
     let timelines_target = target.timelines_root(&tenant);
 
+    let prefix_str = &timelines_target
+        .prefix_in_bucket
+        .strip_prefix("/")
+        .unwrap_or(&timelines_target.prefix_in_bucket);
+
     let mut objects_stream = std::pin::pin!(stream_objects_with_retries(
         remote_client,
         ListingMode::WithDelimiter,
@@ -220,11 +109,7 @@ pub async fn stream_tenant_timelines_generic<'a>(
             .prefixes
             .iter()
             .filter_map(|prefix| -> Option<&str> {
-                prefix
-                    .get_path()
-                    .as_str()
-                    .strip_prefix(&timelines_target.prefix_in_bucket)?
-                    .strip_suffix('/')
+                prefix.get_path().as_str().strip_prefix(prefix_str)
             })
             .map(|entry_id_str| {
                 entry_id_str
@@ -237,7 +122,7 @@ pub async fn stream_tenant_timelines_generic<'a>(
         }
     }
 
-    tracing::debug!("Yielding for {}", tenant);
+    tracing::debug!("Yielding {} timelines for {}", timeline_ids.len(), tenant);
     Ok(stream! {
         for i in timeline_ids {
             let id = i?;
@@ -247,37 +132,6 @@ pub async fn stream_tenant_timelines_generic<'a>(
 }
 
 pub(crate) fn stream_listing<'a>(
-    s3_client: &'a Client,
-    target: &'a S3Target,
-) -> impl Stream<Item = anyhow::Result<ObjectIdentifier>> + 'a {
-    try_stream! {
-        let mut continuation_token = None;
-        loop {
-            let fetch_response =
-                list_objects_with_retries(s3_client, target, continuation_token.clone()).await?;
-
-            if target.delimiter.is_empty() {
-                for object_key in fetch_response.contents().iter().filter_map(|object| object.key())
-                {
-                    let object_id = ObjectIdentifier::builder().key(object_key).build()?;
-                    yield object_id;
-                }
-            } else {
-                for prefix in fetch_response.common_prefixes().iter().filter_map(|p| p.prefix()) {
-                    let object_id = ObjectIdentifier::builder().key(prefix).build()?;
-                    yield object_id;
-                }
-            }
-
-            match fetch_response.next_continuation_token {
-                Some(new_token) => continuation_token = Some(new_token),
-                None => break,
-            }
-        }
-    }
-}
-
-pub(crate) fn stream_listing_generic<'a>(
     remote_client: &'a GenericRemoteStorage,
     target: &'a S3Target,
 ) -> impl Stream<Item = anyhow::Result<(RemotePath, Option<ListingObject>)>> + 'a {

storage_scrubber/src/pageserver_physical_gc.rs

@@ -1,13 +1,10 @@
 use std::collections::{BTreeMap, BTreeSet, HashMap};
 use std::sync::Arc;
-use std::time::{Duration, SystemTime};
+use std::time::Duration;
 
 use crate::checks::{list_timeline_blobs, BlobDataParseResult};
 use crate::metadata_stream::{stream_tenant_timelines, stream_tenants};
-use crate::{
-    init_remote, BucketConfig, ControllerClientConfig, NodeKind, RootTarget, TenantShardTimelineId,
-};
-use aws_sdk_s3::Client;
+use crate::{init_remote, BucketConfig, NodeKind, RootTarget, TenantShardTimelineId};
 use futures_util::{StreamExt, TryStreamExt};
 use pageserver::tenant::remote_timeline_client::index::LayerFileMetadata;
 use pageserver::tenant::remote_timeline_client::{parse_remote_index_path, remote_layer_path};
@@ -15,10 +12,11 @@ use pageserver::tenant::storage_layer::LayerName;
 use pageserver::tenant::IndexPart;
 use pageserver_api::controller_api::TenantDescribeResponse;
 use pageserver_api::shard::{ShardIndex, TenantShardId};
-use remote_storage::RemotePath;
+use remote_storage::{GenericRemoteStorage, ListingObject, RemotePath};
 use reqwest::Method;
 use serde::Serialize;
 use storage_controller_client::control_api;
+use tokio_util::sync::CancellationToken;
 use tracing::{info_span, Instrument};
 use utils::generation::Generation;
 use utils::id::{TenantId, TenantTimelineId};
@@ -242,38 +240,13 @@ impl TenantRefAccumulator {
     }
 }
 
-async fn is_old_enough(
-    s3_client: &Client,
-    bucket_config: &BucketConfig,
-    min_age: &Duration,
-    key: &str,
-    summary: &mut GcSummary,
-) -> bool {
+fn is_old_enough(min_age: &Duration, key: &ListingObject, summary: &mut GcSummary) -> bool {
     // Validation: we will only GC indices & layers after a time threshold (e.g. one week) so that during an incident
     // it is easier to read old data for analysis, and easier to roll back shard splits without having to un-delete any objects.
-    let age: Duration = match s3_client
-        .head_object()
-        .bucket(&bucket_config.bucket)
-        .key(key)
-        .send()
-        .await
-    {
-        Ok(response) => match response.last_modified {
-            None => {
-                tracing::warn!("Missing last_modified");
-                summary.remote_storage_errors += 1;
-                return false;
-            }
-            Some(last_modified) => match SystemTime::try_from(last_modified).map(|t| t.elapsed()) {
-                Ok(Ok(e)) => e,
-                Err(_) | Ok(Err(_)) => {
-                    tracing::warn!("Bad last_modified time: {last_modified:?}");
-                    return false;
-                }
-            },
-        },
-        Err(e) => {
-            tracing::warn!("Failed to HEAD {key}: {e}");
+    let age = match key.last_modified.elapsed() {
+        Ok(e) => e,
+        Err(_) => {
+            tracing::warn!("Bad last_modified time: {:?}", key.last_modified);
             summary.remote_storage_errors += 1;
             return false;
         }
@@ -291,17 +264,30 @@ async fn is_old_enough(
     old_enough
 }
 
+/// Same as [`is_old_enough`], but doesn't require a [`ListingObject`] passed to it.
+async fn check_is_old_enough(
+    remote_client: &GenericRemoteStorage,
+    key: &RemotePath,
+    min_age: &Duration,
+    summary: &mut GcSummary,
+) -> Option<bool> {
+    let listing_object = remote_client
+        .head_object(key, &CancellationToken::new())
+        .await
+        .ok()?;
+    Some(is_old_enough(min_age, &listing_object, summary))
+}
+
 async fn maybe_delete_index(
-    s3_client: &Client,
-    bucket_config: &BucketConfig,
+    remote_client: &GenericRemoteStorage,
     min_age: &Duration,
     latest_gen: Generation,
-    key: &str,
+    obj: &ListingObject,
     mode: GcMode,
     summary: &mut GcSummary,
 ) {
     // Validation: we will only delete things that parse cleanly
-    let basename = key.rsplit_once('/').unwrap().1;
+    let basename = obj.key.get_path().file_name().unwrap();
     let candidate_generation =
         match parse_remote_index_path(RemotePath::from_string(basename).unwrap()) {
             Some(g) => g,
@@ -330,7 +316,7 @@ async fn maybe_delete_index(
         return;
     }
 
-    if !is_old_enough(s3_client, bucket_config, min_age, key, summary).await {
+    if !is_old_enough(min_age, obj, summary) {
         return;
     }
 
@@ -340,11 +326,8 @@ async fn maybe_delete_index(
     }
 
     // All validations passed: erase the object
-    match s3_client
-        .delete_object()
-        .bucket(&bucket_config.bucket)
-        .key(key)
-        .send()
+    match remote_client
+        .delete(&obj.key, &CancellationToken::new())
         .await
     {
         Ok(_) => {
@@ -360,8 +343,7 @@ async fn maybe_delete_index(
 
 #[allow(clippy::too_many_arguments)]
 async fn gc_ancestor(
-    s3_client: &Client,
-    bucket_config: &BucketConfig,
+    remote_client: &GenericRemoteStorage,
     root_target: &RootTarget,
     min_age: &Duration,
     ancestor: TenantShardId,
@@ -370,7 +352,7 @@ async fn gc_ancestor(
     summary: &mut GcSummary,
 ) -> anyhow::Result<()> {
     // Scan timelines in the ancestor
-    let timelines = stream_tenant_timelines(s3_client, root_target, ancestor).await?;
+    let timelines = stream_tenant_timelines(remote_client, root_target, ancestor).await?;
     let mut timelines = std::pin::pin!(timelines);
 
     // Build a list of keys to retain
@@ -378,7 +360,7 @@ async fn gc_ancestor(
     while let Some(ttid) = timelines.next().await {
         let ttid = ttid?;
 
-        let data = list_timeline_blobs(s3_client, ttid, root_target).await?;
+        let data = list_timeline_blobs(remote_client, ttid, root_target).await?;
 
         let s3_layers = match data.blob_data {
             BlobDataParseResult::Parsed {
@@ -429,7 +411,8 @@ async fn gc_ancestor(
 
             // We apply a time threshold to GCing objects that are un-referenced: this preserves our ability
             // to roll back a shard split if we have to, by avoiding deleting ancestor layers right away
-            if !is_old_enough(s3_client, bucket_config, min_age, &key, summary).await {
+            let path = RemotePath::from_string(key.strip_prefix("/").unwrap_or(&key)).unwrap();
+            if check_is_old_enough(remote_client, &path, min_age, summary).await != Some(true) {
                 continue;
             }
 
@@ -439,13 +422,7 @@ async fn gc_ancestor(
             }
 
             // All validations passed: erase the object
-            match s3_client
-                .delete_object()
-                .bucket(&bucket_config.bucket)
-                .key(&key)
-                .send()
-                .await
-            {
+            match remote_client.delete(&path, &CancellationToken::new()).await {
                 Ok(_) => {
                     tracing::info!("Successfully deleted unreferenced ancestor layer {key}");
                     summary.ancestor_layers_deleted += 1;
@@ -473,16 +450,16 @@ async fn gc_ancestor(
 /// This type of GC is not necessary for correctness: rather it serves to reduce wasted storage capacity, and
 /// make sure that object listings don't get slowed down by large numbers of garbage objects.
 pub async fn pageserver_physical_gc(
-    bucket_config: BucketConfig,
-    controller_client_conf: Option<ControllerClientConfig>,
+    bucket_config: &BucketConfig,
+    controller_client: Option<&control_api::Client>,
     tenant_shard_ids: Vec<TenantShardId>,
     min_age: Duration,
     mode: GcMode,
 ) -> anyhow::Result<GcSummary> {
-    let (s3_client, target) = init_remote(bucket_config.clone(), NodeKind::Pageserver).await?;
+    let (remote_client, target) = init_remote(bucket_config.clone(), NodeKind::Pageserver).await?;
 
     let tenants = if tenant_shard_ids.is_empty() {
-        futures::future::Either::Left(stream_tenants(&s3_client, &target))
+        futures::future::Either::Left(stream_tenants(&remote_client, &target))
     } else {
         futures::future::Either::Right(futures::stream::iter(tenant_shard_ids.into_iter().map(Ok)))
     };
@@ -495,14 +472,13 @@ pub async fn pageserver_physical_gc(
     let accumulator = Arc::new(std::sync::Mutex::new(TenantRefAccumulator::default()));
 
     // Generate a stream of TenantTimelineId
-    let timelines = tenants.map_ok(|t| stream_tenant_timelines(&s3_client, &target, t));
+    let timelines = tenants.map_ok(|t| stream_tenant_timelines(&remote_client, &target, t));
     let timelines = timelines.try_buffered(CONCURRENCY);
     let timelines = timelines.try_flatten();
 
     // Generate a stream of S3TimelineBlobData
     async fn gc_timeline(
-        s3_client: &Client,
-        bucket_config: &BucketConfig,
+        remote_client: &GenericRemoteStorage,
         min_age: &Duration,
         target: &RootTarget,
         mode: GcMode,
@@ -510,7 +486,7 @@ pub async fn pageserver_physical_gc(
         accumulator: &Arc<std::sync::Mutex<TenantRefAccumulator>>,
     ) -> anyhow::Result<GcSummary> {
         let mut summary = GcSummary::default();
-        let data = list_timeline_blobs(s3_client, ttid, target).await?;
+        let data = list_timeline_blobs(remote_client, ttid, target).await?;
 
         let (index_part, latest_gen, candidates) = match &data.blob_data {
             BlobDataParseResult::Parsed {
@@ -535,17 +511,9 @@ pub async fn pageserver_physical_gc(
         accumulator.lock().unwrap().update(ttid, index_part);
 
         for key in candidates {
-            maybe_delete_index(
-                s3_client,
-                bucket_config,
-                min_age,
-                latest_gen,
-                &key,
-                mode,
-                &mut summary,
-            )
-            .instrument(info_span!("maybe_delete_index", %ttid, ?latest_gen, key))
-            .await;
+            maybe_delete_index(remote_client, min_age, latest_gen, &key, mode, &mut summary)
+                .instrument(info_span!("maybe_delete_index", %ttid, ?latest_gen, %key.key))
+                .await;
         }
 
         Ok(summary)
@@ -556,15 +524,7 @@ pub async fn pageserver_physical_gc(
     // Drain futures for per-shard GC, populating accumulator as a side effect
     {
         let timelines = timelines.map_ok(|ttid| {
-            gc_timeline(
-                &s3_client,
-                &bucket_config,
-                &min_age,
-                &target,
-                mode,
-                ttid,
-                &accumulator,
-            )
+            gc_timeline(&remote_client, &min_age, &target, mode, ttid, &accumulator)
         });
         let mut timelines = std::pin::pin!(timelines.try_buffered(CONCURRENCY));
 
@@ -574,7 +534,7 @@ pub async fn pageserver_physical_gc(
     }
 
     // Execute cross-shard GC, using the accumulator's full view of all the shards built in the per-shard GC
-    let Some(controller_client) = controller_client_conf.map(|c| c.build_client()) else {
+    let Some(client) = controller_client else {
         tracing::info!("Skipping ancestor layer GC, because no `--controller-api` was specified");
         return Ok(summary);
     };
@@ -583,13 +543,12 @@ pub async fn pageserver_physical_gc(
         .unwrap()
         .into_inner()
         .unwrap()
-        .into_gc_ancestors(&controller_client, &mut summary)
+        .into_gc_ancestors(client, &mut summary)
         .await;
 
     for ancestor_shard in ancestor_shards {
         gc_ancestor(
-            &s3_client,
-            &bucket_config,
+            &remote_client,
             &target,
             &min_age,
             ancestor_shard,

storage_scrubber/src/scan_pageserver_metadata.rs

@@ -1,16 +1,16 @@
 use std::collections::{HashMap, HashSet};
 
 use crate::checks::{
-    branch_cleanup_and_check_errors, list_timeline_blobs, BlobDataParseResult, S3TimelineBlobData,
-    TenantObjectListing, TimelineAnalysis,
+    branch_cleanup_and_check_errors, list_timeline_blobs, BlobDataParseResult,
+    RemoteTimelineBlobData, TenantObjectListing, TimelineAnalysis,
 };
 use crate::metadata_stream::{stream_tenant_timelines, stream_tenants};
 use crate::{init_remote, BucketConfig, NodeKind, RootTarget, TenantShardTimelineId};
-use aws_sdk_s3::Client;
 use futures_util::{StreamExt, TryStreamExt};
 use pageserver::tenant::remote_timeline_client::remote_layer_path;
 use pageserver_api::controller_api::MetadataHealthUpdateRequest;
 use pageserver_api::shard::TenantShardId;
+use remote_storage::GenericRemoteStorage;
 use serde::Serialize;
 use utils::id::TenantId;
 use utils::shard::ShardCount;
@@ -36,7 +36,7 @@ impl MetadataSummary {
         Self::default()
     }
 
-    fn update_data(&mut self, data: &S3TimelineBlobData) {
+    fn update_data(&mut self, data: &RemoteTimelineBlobData) {
         self.timeline_shard_count += 1;
         if let BlobDataParseResult::Parsed {
             index_part,
@@ -116,14 +116,14 @@ Index versions: {version_summary}
 }
 
 /// Scan the pageserver metadata in an S3 bucket, reporting errors and statistics.
-pub async fn scan_metadata(
+pub async fn scan_pageserver_metadata(
     bucket_config: BucketConfig,
     tenant_ids: Vec<TenantShardId>,
 ) -> anyhow::Result<MetadataSummary> {
-    let (s3_client, target) = init_remote(bucket_config, NodeKind::Pageserver).await?;
+    let (remote_client, target) = init_remote(bucket_config, NodeKind::Pageserver).await?;
 
     let tenants = if tenant_ids.is_empty() {
-        futures::future::Either::Left(stream_tenants(&s3_client, &target))
+        futures::future::Either::Left(stream_tenants(&remote_client, &target))
     } else {
         futures::future::Either::Right(futures::stream::iter(tenant_ids.into_iter().map(Ok)))
     };
@@ -133,20 +133,20 @@ pub async fn scan_metadata(
     const CONCURRENCY: usize = 32;
 
     // Generate a stream of TenantTimelineId
-    let timelines = tenants.map_ok(|t| stream_tenant_timelines(&s3_client, &target, t));
+    let timelines = tenants.map_ok(|t| stream_tenant_timelines(&remote_client, &target, t));
     let timelines = timelines.try_buffered(CONCURRENCY);
     let timelines = timelines.try_flatten();
 
     // Generate a stream of S3TimelineBlobData
     async fn report_on_timeline(
-        s3_client: &Client,
+        remote_client: &GenericRemoteStorage,
         target: &RootTarget,
         ttid: TenantShardTimelineId,
-    ) -> anyhow::Result<(TenantShardTimelineId, S3TimelineBlobData)> {
-        let data = list_timeline_blobs(s3_client, ttid, target).await?;
+    ) -> anyhow::Result<(TenantShardTimelineId, RemoteTimelineBlobData)> {
+        let data = list_timeline_blobs(remote_client, ttid, target).await?;
         Ok((ttid, data))
     }
-    let timelines = timelines.map_ok(|ttid| report_on_timeline(&s3_client, &target, ttid));
+    let timelines = timelines.map_ok(|ttid| report_on_timeline(&remote_client, &target, ttid));
     let mut timelines = std::pin::pin!(timelines.try_buffered(CONCURRENCY));
 
     // We must gather all the TenantShardTimelineId->S3TimelineBlobData for each tenant, because different
@@ -157,12 +157,11 @@ pub async fn scan_metadata(
     let mut tenant_timeline_results = Vec::new();
 
     async fn analyze_tenant(
-        s3_client: &Client,
-        target: &RootTarget,
+        remote_client: &GenericRemoteStorage,
         tenant_id: TenantId,
         summary: &mut MetadataSummary,
         mut tenant_objects: TenantObjectListing,
-        timelines: Vec<(TenantShardTimelineId, S3TimelineBlobData)>,
+        timelines: Vec<(TenantShardTimelineId, RemoteTimelineBlobData)>,
         highest_shard_count: ShardCount,
     ) {
         summary.tenant_count += 1;
@@ -191,8 +190,7 @@ pub async fn scan_metadata(
                 // Apply checks to this timeline shard's metadata, and in the process update `tenant_objects`
                 // reference counts for layers across the tenant.
                 let analysis = branch_cleanup_and_check_errors(
-                    s3_client,
-                    target,
+                    remote_client,
                     &ttid,
                     &mut tenant_objects,
                     None,
@@ -273,8 +271,7 @@ pub async fn scan_metadata(
                     let tenant_objects = std::mem::take(&mut tenant_objects);
                     let timelines = std::mem::take(&mut tenant_timeline_results);
                     analyze_tenant(
-                        &s3_client,
-                        &target,
+                        &remote_client,
                         prev_tenant_id,
                         &mut summary,
                         tenant_objects,
@@ -311,8 +308,7 @@ pub async fn scan_metadata(
 
     if !tenant_timeline_results.is_empty() {
         analyze_tenant(
-            &s3_client,
-            &target,
+            &remote_client,
             tenant_id.expect("Must be set if results are present"),
             &mut summary,
             tenant_objects,

storage_scrubber/src/scan_safekeeper_metadata.rs

@@ -14,9 +14,8 @@ use utils::{
 };
 
 use crate::{
-    cloud_admin_api::CloudAdminApiClient, init_remote_generic,
-    metadata_stream::stream_listing_generic, BucketConfig, ConsoleConfig, NodeKind, RootTarget,
-    TenantShardTimelineId,
+    cloud_admin_api::CloudAdminApiClient, init_remote, metadata_stream::stream_listing,
+    BucketConfig, ConsoleConfig, NodeKind, RootTarget, TenantShardTimelineId,
 };
 
 /// Generally we should ask safekeepers, but so far we use everywhere default 16MB.
@@ -107,7 +106,7 @@ pub async fn scan_safekeeper_metadata(
     let timelines = client.query(&query, &[]).await?;
     info!("loaded {} timelines", timelines.len());
 
-    let (remote_client, target) = init_remote_generic(bucket_config, NodeKind::Safekeeper).await?;
+    let (remote_client, target) = init_remote(bucket_config, NodeKind::Safekeeper).await?;
     let console_config = ConsoleConfig::from_env()?;
     let cloud_admin_api_client = CloudAdminApiClient::new(console_config);
 
@@ -188,14 +187,19 @@ async fn check_timeline(
     // we need files, so unset it.
     timeline_dir_target.delimiter = String::new();
 
-    let mut stream = std::pin::pin!(stream_listing_generic(remote_client, &timeline_dir_target));
+    let prefix_str = &timeline_dir_target
+        .prefix_in_bucket
+        .strip_prefix("/")
+        .unwrap_or(&timeline_dir_target.prefix_in_bucket);
+
+    let mut stream = std::pin::pin!(stream_listing(remote_client, &timeline_dir_target));
     while let Some(obj) = stream.next().await {
         let (key, _obj) = obj?;
 
         let seg_name = key
             .get_path()
             .as_str()
-            .strip_prefix(&timeline_dir_target.prefix_in_bucket)
+            .strip_prefix(prefix_str)
             .expect("failed to extract segment name");
         expected_segfiles.remove(seg_name);
     }

storage_scrubber/src/tenant_snapshot.rs

@@ -1,10 +1,11 @@
 use std::collections::HashMap;
 use std::sync::Arc;
 
-use crate::checks::{list_timeline_blobs, BlobDataParseResult, S3TimelineBlobData};
+use crate::checks::{list_timeline_blobs, BlobDataParseResult, RemoteTimelineBlobData};
 use crate::metadata_stream::{stream_tenant_shards, stream_tenant_timelines};
 use crate::{
-    download_object_to_file, init_remote, BucketConfig, NodeKind, RootTarget, TenantShardTimelineId,
+    download_object_to_file_s3, init_remote, init_remote_s3, BucketConfig, NodeKind, RootTarget,
+    TenantShardTimelineId,
 };
 use anyhow::Context;
 use async_stream::stream;
@@ -15,6 +16,7 @@ use pageserver::tenant::remote_timeline_client::index::LayerFileMetadata;
 use pageserver::tenant::storage_layer::LayerName;
 use pageserver::tenant::IndexPart;
 use pageserver_api::shard::TenantShardId;
+use remote_storage::GenericRemoteStorage;
 use utils::generation::Generation;
 use utils::id::TenantId;
 
@@ -34,7 +36,8 @@ impl SnapshotDownloader {
         output_path: Utf8PathBuf,
         concurrency: usize,
     ) -> anyhow::Result<Self> {
-        let (s3_client, s3_root) = init_remote(bucket_config.clone(), NodeKind::Pageserver).await?;
+        let (s3_client, s3_root) =
+            init_remote_s3(bucket_config.clone(), NodeKind::Pageserver).await?;
         Ok(Self {
             s3_client,
             s3_root,
@@ -91,7 +94,7 @@ impl SnapshotDownloader {
             let Some(version) = versions.versions.as_ref().and_then(|v| v.first()) else {
                 return Err(anyhow::anyhow!("No versions found for {remote_layer_path}"));
             };
-            download_object_to_file(
+            download_object_to_file_s3(
                 &self.s3_client,
                 &self.bucket_config.bucket,
                 &remote_layer_path,
@@ -215,11 +218,11 @@ impl SnapshotDownloader {
     }
 
     pub async fn download(&self) -> anyhow::Result<()> {
-        let (s3_client, target) =
+        let (remote_client, target) =
             init_remote(self.bucket_config.clone(), NodeKind::Pageserver).await?;
 
         // Generate a stream of TenantShardId
-        let shards = stream_tenant_shards(&s3_client, &target, self.tenant_id).await?;
+        let shards = stream_tenant_shards(&remote_client, &target, self.tenant_id).await?;
         let shards: Vec<TenantShardId> = shards.try_collect().await?;
 
         // Only read from shards that have the highest count: avoids redundantly downloading
@@ -237,18 +240,19 @@ impl SnapshotDownloader {
 
         for shard in shards.into_iter().filter(|s| s.shard_count == shard_count) {
             // Generate a stream of TenantTimelineId
-            let timelines = stream_tenant_timelines(&s3_client, &self.s3_root, shard).await?;
+            let timelines = stream_tenant_timelines(&remote_client, &target, shard).await?;
 
             // Generate a stream of S3TimelineBlobData
             async fn load_timeline_index(
-                s3_client: &Client,
+                remote_client: &GenericRemoteStorage,
                 target: &RootTarget,
                 ttid: TenantShardTimelineId,
-            ) -> anyhow::Result<(TenantShardTimelineId, S3TimelineBlobData)> {
-                let data = list_timeline_blobs(s3_client, ttid, target).await?;
+            ) -> anyhow::Result<(TenantShardTimelineId, RemoteTimelineBlobData)> {
+                let data = list_timeline_blobs(remote_client, ttid, target).await?;
                 Ok((ttid, data))
             }
-            let timelines = timelines.map_ok(|ttid| load_timeline_index(&s3_client, &target, ttid));
+            let timelines =
+                timelines.map_ok(|ttid| load_timeline_index(&remote_client, &target, ttid));
             let mut timelines = std::pin::pin!(timelines.try_buffered(8));
 
             while let Some(i) = timelines.next().await {
@@ -278,7 +282,7 @@ impl SnapshotDownloader {
 
         for (ttid, layers) in ancestor_layers.into_iter() {
             tracing::info!(
-                "Downloading {} layers from ancvestor timeline {ttid}...",
+                "Downloading {} layers from ancestor timeline {ttid}...",
                 layers.len()
             );