starvation prevention while allowing concurrent find_victims

This reverts commit 74601238ee.

.github/actions/allure-report-generate/action.yml

@@ -76,8 +76,8 @@ runs:
           rm -f ${ALLURE_ZIP}
         fi
       env:
-        ALLURE_VERSION: 2.24.0
-        ALLURE_ZIP_SHA256: 60b1d6ce65d9ef24b23cf9c2c19fd736a123487c38e54759f1ed1a7a77353c90
+        ALLURE_VERSION: 2.23.1
+        ALLURE_ZIP_SHA256: 11141bfe727504b3fd80c0f9801eb317407fd0ac983ebb57e671f14bac4bcd86
 
     # Potentially we could have several running build for the same key (for example, for the main branch), so we use improvised lock for this
     - name: Acquire lock

.github/workflows/build_and_test.yml

@@ -1092,10 +1092,8 @@ jobs:
         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
-
-            # TODO: move deployPreprodRegion to release (`"$GITHUB_REF_NAME" == "release"` block), once Staging support different compute tag prefixes for different regions
-            gh workflow --repo neondatabase/aws run deploy-dev.yml --ref main -f branch=main -f dockerTag=${{needs.tag.outputs.build-tag}} -f deployPreprodRegion=true
           elif [[ "$GITHUB_REF_NAME" == "release" ]]; then
+            gh workflow --repo neondatabase/aws run deploy-dev.yml --ref main -f branch=main -f dockerTag=${{needs.tag.outputs.build-tag}} -f deployPreprodRegion=true
             gh workflow --repo neondatabase/aws run deploy-prod.yml --ref main -f branch=main -f dockerTag=${{needs.tag.outputs.build-tag}} -f disclamerAcknowledged=true
           else
             echo "GITHUB_REF_NAME (value '$GITHUB_REF_NAME') is not set to either 'main' or 'release'"

Cargo.toml

@@ -36,9 +36,6 @@ license = "Apache-2.0"
 [workspace.dependencies]
 anyhow = { version = "1.0", features = ["backtrace"] }
 async-compression = { version = "0.4.0", features = ["tokio", "gzip"] }
-azure_core = "0.16.0"
-azure_storage = "0.16"
-azure_storage_blobs = "0.16.0"
 flate2 = "1.0.26"
 async-stream = "0.3"
 async-trait = "0.1"
@@ -54,7 +51,6 @@ bindgen = "0.65"
 bstr = "1.0"
 byteorder = "1.4"
 bytes = "1.0"
-camino = "1.1.6"
 cfg-if = "1.0.0"
 chrono = { version = "0.4", default-features = false, features = ["clock"] }
 clap = { version = "4.0", features = ["derive"] }
@@ -79,7 +75,6 @@ hex = "0.4"
 hex-literal = "0.4"
 hmac = "0.12.1"
 hostname = "0.3.1"
-http-types = "2"
 humantime = "2.1"
 humantime-serde = "1.1.1"
 hyper = "0.14"
@@ -192,7 +187,7 @@ workspace_hack = { version = "0.1", path = "./workspace_hack/" }
 criterion = "0.5.1"
 rcgen = "0.11"
 rstest = "0.18"
-camino-tempfile = "1.0.2"
+tempfile = "3.4"
 tonic-build = "0.9"
 
 [patch.crates-io]

Dockerfile.compute-node

@@ -224,8 +224,8 @@ RUN wget https://github.com/df7cb/postgresql-unit/archive/refs/tags/7.7.tar.gz -
 FROM build-deps AS vector-pg-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
-RUN wget https://github.com/pgvector/pgvector/archive/refs/tags/v0.5.1.tar.gz -O pgvector.tar.gz && \
-    echo "cc7a8e034a96e30a819911ac79d32f6bc47bdd1aa2de4d7d4904e26b83209dc8 pgvector.tar.gz" | sha256sum --check && \
+RUN wget https://github.com/pgvector/pgvector/archive/refs/tags/v0.5.0.tar.gz -O pgvector.tar.gz && \
+    echo "d8aa3504b215467ca528525a6de12c3f85f9891b091ce0e5864dd8a9b757f77b pgvector.tar.gz" | sha256sum --check && \
     mkdir pgvector-src && cd pgvector-src && tar xvzf ../pgvector.tar.gz --strip-components=1 -C . && \
     make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
     make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
@@ -368,8 +368,8 @@ RUN wget https://github.com/citusdata/postgresql-hll/archive/refs/tags/v2.18.tar
 FROM build-deps AS plpgsql-check-pg-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
-RUN wget https://github.com/okbob/plpgsql_check/archive/refs/tags/v2.5.3.tar.gz -O plpgsql_check.tar.gz && \
-    echo "6631ec3e7fb3769eaaf56e3dfedb829aa761abf163d13dba354b4c218508e1c0 plpgsql_check.tar.gz" | sha256sum --check && \
+RUN wget https://github.com/okbob/plpgsql_check/archive/refs/tags/v2.4.0.tar.gz -O plpgsql_check.tar.gz && \
+    echo "9ba58387a279b35a3bfa39ee611e5684e6cddb2ba046ddb2c5190b3bd2ca254a plpgsql_check.tar.gz" | sha256sum --check && \
     mkdir plpgsql_check-src && cd plpgsql_check-src && tar xvzf ../plpgsql_check.tar.gz --strip-components=1 -C . && \
     make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config USE_PGXS=1 && \
     make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config USE_PGXS=1 && \
@@ -612,13 +612,51 @@ RUN wget https://gitlab.com/dalibo/postgresql_anonymizer/-/archive/1.1.0/postgre
     sort -o /before.txt /before.txt && sort -o /after.txt /after.txt && \
     comm -13 /before.txt /after.txt | tar --directory=/usr/local/pgsql --zstd -cf /extensions/anon.tar.zst -T -
 
+#########################################################################################
+#
+# Layer "rust extensions" for older extension which hasn't been updated to `pgrx` yet
+# This layer is used to build `pgx` deps
+#
+#########################################################################################
+FROM build-deps AS rust-extensions-build-pgx
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+RUN apt-get update && \
+    apt-get install -y curl libclang-dev cmake && \
+    useradd -ms /bin/bash nonroot -b /home
+
+ENV HOME=/home/nonroot
+ENV PATH="/home/nonroot/.cargo/bin:/usr/local/pgsql/bin/:$PATH"
+USER nonroot
+WORKDIR /home/nonroot
+ARG PG_VERSION
+
+RUN case "${PG_VERSION}" in \
+      "v14" | "v15") \
+        ;; \
+      "v16") \
+        echo "TODO: Not yet supported for PostgreSQL 16. Need to update pgrx dependencies" && exit 0 \
+	;; \
+      *) \
+        echo "unexpected PostgreSQL version ${PG_VERSION}" && exit 1 \
+        ;; \
+    esac && \
+    curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux-gnu/rustup-init && \
+    chmod +x rustup-init && \
+    ./rustup-init -y --no-modify-path --profile minimal --default-toolchain stable && \
+    rm rustup-init && \
+    cargo install --locked --version 0.7.3 cargo-pgx && \
+    /bin/bash -c 'cargo pgx init --pg${PG_VERSION:1}=/usr/local/pgsql/bin/pg_config'
+
+USER root
+
 #########################################################################################
 #
 # Layer "rust extensions"
 # This layer is used to build `pgrx` deps
 #
 #########################################################################################
-FROM build-deps AS rust-extensions-build
+FROM build-deps AS rust-extensions-build-pgrx
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 RUN apt-get update && \
@@ -647,14 +685,26 @@ USER root
 #
 #########################################################################################
 
-FROM rust-extensions-build AS pg-jsonschema-pg-build
+FROM rust-extensions-build-pgx AS pg-jsonschema-pg-build
 ARG PG_VERSION
 
-RUN wget https://github.com/supabase/pg_jsonschema/archive/refs/tags/v0.2.0.tar.gz -O pg_jsonschema.tar.gz && \
-    echo "9118fc508a6e231e7a39acaa6f066fcd79af17a5db757b47d2eefbe14f7794f0 pg_jsonschema.tar.gz" | sha256sum --check && \
+# caeab60d70b2fd3ae421ec66466a3abbb37b7ee6 made on 06/03/2023
+# there is no release tag yet, but we need it due to the superuser fix in the control file, switch to git tag after release >= 0.1.5
+RUN case "${PG_VERSION}" in \
+      "v14" | "v15") \
+        ;; \
+      "v16") \
+        echo "TODO: Not yet supported for PostgreSQL 16. Need to update pgrx dependencies" && exit 0 \
+	;; \
+      *) \
+        echo "unexpected PostgreSQL version \"${PG_VERSION}\"" && exit 1 \
+        ;; \
+    esac && \
+    wget https://github.com/supabase/pg_jsonschema/archive/caeab60d70b2fd3ae421ec66466a3abbb37b7ee6.tar.gz -O pg_jsonschema.tar.gz && \
+    echo "54129ce2e7ee7a585648dbb4cef6d73f795d94fe72f248ac01119992518469a4 pg_jsonschema.tar.gz" | sha256sum --check && \
     mkdir pg_jsonschema-src && cd pg_jsonschema-src && tar xvzf ../pg_jsonschema.tar.gz --strip-components=1 -C . && \
-    sed -i 's/pgrx = "0.10.2"/pgrx = { version = "0.10.2", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
-    cargo pgrx install --release && \
+    sed -i 's/pgx = "0.7.1"/pgx = { version = "0.7.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    cargo pgx install --release && \
     echo "trusted = true" >> /usr/local/pgsql/share/extension/pg_jsonschema.control
 
 #########################################################################################
@@ -664,14 +714,29 @@ RUN wget https://github.com/supabase/pg_jsonschema/archive/refs/tags/v0.2.0.tar.
 #
 #########################################################################################
 
-FROM rust-extensions-build AS pg-graphql-pg-build
+FROM rust-extensions-build-pgx AS pg-graphql-pg-build
 ARG PG_VERSION
 
-RUN wget https://github.com/supabase/pg_graphql/archive/refs/tags/v1.4.0.tar.gz -O pg_graphql.tar.gz && \
-    echo "bd8dc7230282b3efa9ae5baf053a54151ed0e66881c7c53750e2d0c765776edc pg_graphql.tar.gz" | sha256sum --check && \
+# b4988843647450a153439be367168ed09971af85 made on 22/02/2023 (from remove-pgx-contrib-spiext branch)
+# Currently pgx version bump to >= 0.7.2  causes "call to unsafe function" compliation errors in
+# pgx-contrib-spiext. There is a branch that removes that dependency, so use it. It is on the
+# same 1.1 version we've used before.
+RUN case "${PG_VERSION}" in \
+      "v14" | "v15") \
+        ;; \
+      "v16") \
+        echo "TODO: Not yet supported for PostgreSQL 16. Need to update pgrx dependencies" && exit 0 \
+	;; \
+      *) \
+        echo "unexpected PostgreSQL version" && exit 1 \
+        ;; \
+    esac && \
+    wget https://github.com/yrashk/pg_graphql/archive/b4988843647450a153439be367168ed09971af85.tar.gz -O pg_graphql.tar.gz && \
+    echo "0c7b0e746441b2ec24187d0e03555faf935c2159e2839bddd14df6dafbc8c9bd pg_graphql.tar.gz" | sha256sum --check && \
     mkdir pg_graphql-src && cd pg_graphql-src && tar xvzf ../pg_graphql.tar.gz --strip-components=1 -C . && \
-    sed -i 's/pgrx = "=0.10.2"/pgrx = { version = "0.10.2", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
-    cargo pgrx install --release && \
+    sed -i 's/pgx = "~0.7.1"/pgx = { version = "0.7.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    sed -i 's/pgx-tests = "~0.7.1"/pgx-tests = "0.7.3"/g' Cargo.toml && \
+    cargo pgx install --release && \
     # it's needed to enable extension because it uses untrusted C language
     sed -i 's/superuser = false/superuser = true/g' /usr/local/pgsql/share/extension/pg_graphql.control && \
     echo "trusted = true" >> /usr/local/pgsql/share/extension/pg_graphql.control
@@ -683,7 +748,7 @@ RUN wget https://github.com/supabase/pg_graphql/archive/refs/tags/v1.4.0.tar.gz
 #
 #########################################################################################
 
-FROM rust-extensions-build AS pg-tiktoken-pg-build
+FROM rust-extensions-build-pgrx AS pg-tiktoken-pg-build
 ARG PG_VERSION
 
 # 26806147b17b60763039c6a6878884c41a262318 made on 26/09/2023
@@ -700,7 +765,7 @@ RUN wget https://github.com/kelvich/pg_tiktoken/archive/26806147b17b60763039c6a6
 #
 #########################################################################################
 
-FROM rust-extensions-build AS pg-pgx-ulid-build
+FROM rust-extensions-build-pgrx AS pg-pgx-ulid-build
 ARG PG_VERSION
 
 RUN wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.1.3.tar.gz -O pgx_ulid.tar.gz && \

compute_tools/src/compute.rs

@@ -692,11 +692,10 @@ impl ComputeNode {
         // Proceed with post-startup configuration. Note, that order of operations is important.
         let spec = &compute_state.pspec.as_ref().expect("spec must be set").spec;
         create_neon_superuser(spec, &mut client)?;
-        cleanup_instance(&mut client)?;
         handle_roles(spec, &mut client)?;
         handle_databases(spec, &mut client)?;
         handle_role_deletions(spec, self.connstr.as_str(), &mut client)?;
-        handle_grants(spec, &mut client, self.connstr.as_str())?;
+        handle_grants(spec, self.connstr.as_str())?;
         handle_extensions(spec, &mut client)?;
         create_availability_check_data(&mut client)?;
 
@@ -732,11 +731,10 @@ impl ComputeNode {
         // Disable DDL forwarding because control plane already knows about these roles/databases.
         if spec.mode == ComputeMode::Primary {
             client.simple_query("SET neon.forward_ddl = false")?;
-            cleanup_instance(&mut client)?;
             handle_roles(&spec, &mut client)?;
             handle_databases(&spec, &mut client)?;
             handle_role_deletions(&spec, self.connstr.as_str(), &mut client)?;
-            handle_grants(&spec, &mut client, self.connstr.as_str())?;
+            handle_grants(&spec, self.connstr.as_str())?;
             handle_extensions(&spec, &mut client)?;
         }
 
@@ -1041,7 +1039,7 @@ LIMIT 100",
         let remote_extensions = spec
             .remote_extensions
             .as_ref()
-            .ok_or(anyhow::anyhow!("Remote extensions are not configured"))?;
+            .ok_or(anyhow::anyhow!("Remote extensions are not configured",))?;
 
         info!("parse shared_preload_libraries from spec.cluster.settings");
         let mut libs_vec = Vec::new();

compute_tools/src/monitor.rs

@@ -1,5 +1,5 @@
 use std::sync::Arc;
-use std::{thread, time::Duration};
+use std::{thread, time};
 
 use chrono::{DateTime, Utc};
 use postgres::{Client, NoTls};
@@ -7,7 +7,7 @@ use tracing::{debug, info};
 
 use crate::compute::ComputeNode;
 
-const MONITOR_CHECK_INTERVAL: Duration = Duration::from_millis(500);
+const MONITOR_CHECK_INTERVAL: u64 = 500; // milliseconds
 
 // Spin in a loop and figure out the last activity time in the Postgres.
 // Then update it in the shared state. This function never errors out.
@@ -17,12 +17,13 @@ fn watch_compute_activity(compute: &ComputeNode) {
     let connstr = compute.connstr.as_str();
     // Define `client` outside of the loop to reuse existing connection if it's active.
     let mut client = Client::connect(connstr, NoTls);
+    let timeout = time::Duration::from_millis(MONITOR_CHECK_INTERVAL);
 
     info!("watching Postgres activity at {}", connstr);
 
     loop {
         // Should be outside of the write lock to allow others to read while we sleep.
-        thread::sleep(MONITOR_CHECK_INTERVAL);
+        thread::sleep(timeout);
 
         match &mut client {
             Ok(cli) => {

compute_tools/src/pg_helpers.rs

@@ -1,4 +1,3 @@
-use std::collections::HashMap;
 use std::fmt::Write;
 use std::fs;
 use std::fs::File;
@@ -206,37 +205,22 @@ pub fn get_existing_roles(xact: &mut Transaction<'_>) -> Result<Vec<Role>> {
 }
 
 /// Build a list of existing Postgres databases
-pub fn get_existing_dbs(client: &mut Client) -> Result<HashMap<String, Database>> {
-    // `pg_database.datconnlimit = -2` means that the database is in the
-    // invalid state. See:
-    //   https://github.com/postgres/postgres/commit/a4b4cc1d60f7e8ccfcc8ff8cb80c28ee411ad9a9
-    let postgres_dbs: Vec<Database> = client
+pub fn get_existing_dbs(client: &mut Client) -> Result<Vec<Database>> {
+    let postgres_dbs = client
         .query(
-            "SELECT
-                datname AS name,
-                datdba::regrole::text AS owner,
-                NOT datallowconn AS restrict_conn,
-                datconnlimit = - 2 AS invalid
-            FROM
-                pg_catalog.pg_database;",
+            "SELECT datname, datdba::regrole::text as owner
+               FROM pg_catalog.pg_database;",
             &[],
         )?
         .iter()
         .map(|row| Database {
-            name: row.get("name"),
+            name: row.get("datname"),
             owner: row.get("owner"),
-            restrict_conn: row.get("restrict_conn"),
-            invalid: row.get("invalid"),
             options: None,
         })
         .collect();
 
-    let dbs_map = postgres_dbs
-        .iter()
-        .map(|db| (db.name.clone(), db.clone()))
-        .collect::<HashMap<_, _>>();
-
-    Ok(dbs_map)
+    Ok(postgres_dbs)
 }
 
 /// Wait for Postgres to become ready to accept connections. It's ready to

compute_tools/src/spec.rs

@@ -13,7 +13,7 @@ use crate::params::PG_HBA_ALL_MD5;
 use crate::pg_helpers::*;
 
 use compute_api::responses::{ControlPlaneComputeStatus, ControlPlaneSpecResponse};
-use compute_api::spec::{ComputeSpec, PgIdent, Role};
+use compute_api::spec::{ComputeSpec, Database, PgIdent, Role};
 
 // Do control plane request and return response if any. In case of error it
 // returns a bool flag indicating whether it makes sense to retry the request
@@ -161,38 +161,6 @@ pub fn add_standby_signal(pgdata_path: &Path) -> Result<()> {
     Ok(())
 }
 
-/// Compute could be unexpectedly shut down, for example, during the
-/// database dropping. This leaves the database in the invalid state,
-/// which prevents new db creation with the same name. This function
-/// will clean it up before proceeding with catalog updates. All
-/// possible future cleanup operations may go here too.
-#[instrument(skip_all)]
-pub fn cleanup_instance(client: &mut Client) -> Result<()> {
-    let existing_dbs = get_existing_dbs(client)?;
-
-    for (_, db) in existing_dbs {
-        if db.invalid {
-            // After recent commit in Postgres, interrupted DROP DATABASE
-            // leaves the database in the invalid state. According to the
-            // commit message, the only option for user is to drop it again.
-            // See:
-            //   https://github.com/postgres/postgres/commit/a4b4cc1d60f7e8ccfcc8ff8cb80c28ee411ad9a9
-            //
-            // Postgres Neon extension is done the way, that db is de-registered
-            // in the control plane metadata only after it is dropped. So there is
-            // a chance that it still thinks that db should exist. This means
-            // that it will be re-created by `handle_databases()`. Yet, it's fine
-            // as user can just repeat drop (in vanilla Postgres they would need
-            // to do the same, btw).
-            let query = format!("DROP DATABASE IF EXISTS {}", db.name.pg_quote());
-            info!("dropping invalid database {}", db.name);
-            client.execute(query.as_str(), &[])?;
-        }
-    }
-
-    Ok(())
-}
-
 /// Given a cluster spec json and open transaction it handles roles creation,
 /// deletion and update.
 #[instrument(skip_all)]
@@ -411,13 +379,13 @@ fn reassign_owned_objects(spec: &ComputeSpec, connstr: &str, role_name: &PgIdent
 /// which together provide us idempotency.
 #[instrument(skip_all)]
 pub fn handle_databases(spec: &ComputeSpec, client: &mut Client) -> Result<()> {
-    let existing_dbs = get_existing_dbs(client)?;
+    let existing_dbs: Vec<Database> = get_existing_dbs(client)?;
 
     // Print a list of existing Postgres databases (only in debug mode)
     if span_enabled!(Level::INFO) {
         info!("postgres databases:");
-        for (dbname, db) in &existing_dbs {
-            info!("    {}:{}", dbname, db.owner);
+        for r in &existing_dbs {
+            info!("    {}:{}", r.name, r.owner);
         }
     }
 
@@ -471,7 +439,8 @@ pub fn handle_databases(spec: &ComputeSpec, client: &mut Client) -> Result<()> {
                 "rename_db" => {
                     let new_name = op.new_name.as_ref().unwrap();
 
-                    if existing_dbs.get(&op.name).is_some() {
+                    // XXX: with a limited number of roles it is fine, but consider making it a HashMap
+                    if existing_dbs.iter().any(|r| r.name == op.name) {
                         let query: String = format!(
                             "ALTER DATABASE {} RENAME TO {}",
                             op.name.pg_quote(),
@@ -488,12 +457,14 @@ pub fn handle_databases(spec: &ComputeSpec, client: &mut Client) -> Result<()> {
     }
 
     // Refresh Postgres databases info to handle possible renames
-    let existing_dbs = get_existing_dbs(client)?;
+    let existing_dbs: Vec<Database> = get_existing_dbs(client)?;
 
     info!("cluster spec databases:");
     for db in &spec.cluster.databases {
         let name = &db.name;
-        let pg_db = existing_dbs.get(name);
+
+        // XXX: with a limited number of databases it is fine, but consider making it a HashMap
+        let pg_db = existing_dbs.iter().find(|r| r.name == *name);
 
         enum DatabaseAction {
             None,
@@ -559,32 +530,13 @@ pub fn handle_databases(spec: &ComputeSpec, client: &mut Client) -> Result<()> {
 /// Grant CREATE ON DATABASE to the database owner and do some other alters and grants
 /// to allow users creating trusted extensions and re-creating `public` schema, for example.
 #[instrument(skip_all)]
-pub fn handle_grants(spec: &ComputeSpec, client: &mut Client, connstr: &str) -> Result<()> {
-    info!("modifying database permissions");
-    let existing_dbs = get_existing_dbs(client)?;
+pub fn handle_grants(spec: &ComputeSpec, connstr: &str) -> Result<()> {
+    info!("cluster spec grants:");
 
     // Do some per-database access adjustments. We'd better do this at db creation time,
     // but CREATE DATABASE isn't transactional. So we cannot create db + do some grants
     // atomically.
     for db in &spec.cluster.databases {
-        match existing_dbs.get(&db.name) {
-            Some(pg_db) => {
-                if pg_db.restrict_conn || pg_db.invalid {
-                    info!(
-                        "skipping grants for db {} (invalid: {}, connections not allowed: {})",
-                        db.name, pg_db.invalid, pg_db.restrict_conn
-                    );
-                    continue;
-                }
-            }
-            None => {
-                bail!(
-                    "database {} doesn't exist in Postgres after handle_databases()",
-                    db.name
-                );
-            }
-        }
-
         let mut conf = Config::from_str(connstr)?;
         conf.dbname(&db.name);
 
@@ -623,11 +575,6 @@ pub fn handle_grants(spec: &ComputeSpec, client: &mut Client, connstr: &str) ->
 
         // Explicitly grant CREATE ON SCHEMA PUBLIC to the web_access user.
         // This is needed because since postgres 15 this privilege is removed by default.
-        // TODO: web_access isn't created for almost 1 year. It could be that we have
-        // active users of 1 year old projects, but hopefully not, so check it and
-        // remove this code if possible. The worst thing that could happen is that
-        // user won't be able to use public schema in NEW databases created in the
-        // very OLD project.
         let grant_query = "DO $$\n\
                 BEGIN\n\
                     IF EXISTS(\n\

control_plane/Cargo.toml

@@ -6,7 +6,6 @@ license.workspace = true
 
 [dependencies]
 anyhow.workspace = true
-camino.workspace = true
 clap.workspace = true
 comfy-table.workspace = true
 git-version.workspace = true

control_plane/src/attachment_service.rs

@@ -1,6 +1,5 @@
 use crate::{background_process, local_env::LocalEnv};
 use anyhow::anyhow;
-use camino::Utf8PathBuf;
 use serde::{Deserialize, Serialize};
 use serde_with::{serde_as, DisplayFromStr};
 use std::{path::PathBuf, process::Child};
@@ -48,9 +47,8 @@ impl AttachmentService {
         }
     }
 
-    fn pid_file(&self) -> Utf8PathBuf {
-        Utf8PathBuf::from_path_buf(self.env.base_data_dir.join("attachment_service.pid"))
-            .expect("non-Unicode path")
+    fn pid_file(&self) -> PathBuf {
+        self.env.base_data_dir.join("attachment_service.pid")
     }
 
     pub fn start(&self) -> anyhow::Result<Child> {

control_plane/src/background_process.rs

@@ -16,13 +16,12 @@ use std::ffi::OsStr;
 use std::io::Write;
 use std::os::unix::prelude::AsRawFd;
 use std::os::unix::process::CommandExt;
-use std::path::Path;
+use std::path::{Path, PathBuf};
 use std::process::{Child, Command};
 use std::time::Duration;
 use std::{fs, io, thread};
 
 use anyhow::Context;
-use camino::{Utf8Path, Utf8PathBuf};
 use nix::errno::Errno;
 use nix::fcntl::{FcntlArg, FdFlag};
 use nix::sys::signal::{kill, Signal};
@@ -46,9 +45,9 @@ const NOTICE_AFTER_RETRIES: u64 = 50;
 /// it itself.
 pub enum InitialPidFile<'t> {
     /// Create a pidfile, to allow future CLI invocations to manipulate the process.
-    Create(&'t Utf8Path),
+    Create(&'t Path),
     /// The process will create the pidfile itself, need to wait for that event.
-    Expect(&'t Utf8Path),
+    Expect(&'t Path),
 }
 
 /// Start a background child process using the parameters given.
@@ -86,7 +85,7 @@ where
         .stdout(process_log_file)
         .stderr(same_file_for_stderr)
         .args(args);
-    let filled_cmd = fill_remote_storage_secrets_vars(fill_rust_env_vars(background_command));
+    let filled_cmd = fill_aws_secrets_vars(fill_rust_env_vars(background_command));
     filled_cmd.envs(envs);
 
     let pid_file_to_check = match initial_pid_file {
@@ -138,11 +137,7 @@ where
 }
 
 /// Stops the process, using the pid file given. Returns Ok also if the process is already not running.
-pub fn stop_process(
-    immediate: bool,
-    process_name: &str,
-    pid_file: &Utf8Path,
-) -> anyhow::Result<()> {
+pub fn stop_process(immediate: bool, process_name: &str, pid_file: &Path) -> anyhow::Result<()> {
     let pid = match pid_file::read(pid_file)
         .with_context(|| format!("read pid_file {pid_file:?}"))?
     {
@@ -238,13 +233,11 @@ fn fill_rust_env_vars(cmd: &mut Command) -> &mut Command {
     filled_cmd
 }
 
-fn fill_remote_storage_secrets_vars(mut cmd: &mut Command) -> &mut Command {
+fn fill_aws_secrets_vars(mut cmd: &mut Command) -> &mut Command {
     for env_key in [
         "AWS_ACCESS_KEY_ID",
         "AWS_SECRET_ACCESS_KEY",
         "AWS_SESSION_TOKEN",
-        "AZURE_STORAGE_ACCOUNT",
-        "AZURE_STORAGE_ACCESS_KEY",
     ] {
         if let Ok(value) = std::env::var(env_key) {
             cmd = cmd.env(env_key, value);
@@ -259,9 +252,9 @@ fn fill_remote_storage_secrets_vars(mut cmd: &mut Command) -> &mut Command {
 ///    will remain held until the cmd exits.
 fn pre_exec_create_pidfile<P>(cmd: &mut Command, path: P) -> &mut Command
 where
-    P: Into<Utf8PathBuf>,
+    P: Into<PathBuf>,
 {
-    let path: Utf8PathBuf = path.into();
+    let path: PathBuf = path.into();
     // SAFETY
     // pre_exec is marked unsafe because it runs between fork and exec.
     // Why is that dangerous in various ways?
@@ -318,7 +311,7 @@ where
 
 fn process_started<F>(
     pid: Pid,
-    pid_file_to_check: Option<&Utf8Path>,
+    pid_file_to_check: Option<&Path>,
     status_check: &F,
 ) -> anyhow::Result<bool>
 where

control_plane/src/bin/neon_local.rs

@@ -116,7 +116,6 @@ fn main() -> Result<()> {
             "attachment_service" => handle_attachment_service(sub_args, &env),
             "safekeeper" => handle_safekeeper(sub_args, &env),
             "endpoint" => handle_endpoint(sub_args, &env),
-            "mappings" => handle_mappings(sub_args, &mut env),
             "pg" => bail!("'pg' subcommand has been renamed to 'endpoint'"),
             _ => bail!("unexpected subcommand {sub_name}"),
         };
@@ -817,38 +816,6 @@ fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<(
     Ok(())
 }
 
-fn handle_mappings(sub_match: &ArgMatches, env: &mut local_env::LocalEnv) -> Result<()> {
-    let (sub_name, sub_args) = match sub_match.subcommand() {
-        Some(ep_subcommand_data) => ep_subcommand_data,
-        None => bail!("no mappings subcommand provided"),
-    };
-
-    match sub_name {
-        "map" => {
-            let branch_name = sub_args
-                .get_one::<String>("branch-name")
-                .expect("branch-name argument missing");
-
-            let tenant_id = sub_args
-                .get_one::<String>("tenant-id")
-                .map(|x| TenantId::from_str(x))
-                .expect("tenant-id argument missing")
-                .expect("malformed tenant-id arg");
-
-            let timeline_id = sub_args
-                .get_one::<String>("timeline-id")
-                .map(|x| TimelineId::from_str(x))
-                .expect("timeline-id argument missing")
-                .expect("malformed timeline-id arg");
-
-            env.register_branch_mapping(branch_name.to_owned(), tenant_id, timeline_id)?;
-
-            Ok(())
-        }
-        other => unimplemented!("mappings subcommand {other}"),
-    }
-}
-
 fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
     fn get_pageserver(env: &local_env::LocalEnv, args: &ArgMatches) -> Result<PageServerNode> {
         let node_id = if let Some(id_str) = args.get_one::<String>("pageserver-id") {
@@ -1117,7 +1084,6 @@ fn cli() -> Command {
     // --id, when using a pageserver command
     let pageserver_id_arg = Arg::new("pageserver-id")
         .long("id")
-        .global(true)
         .help("pageserver id")
         .required(false);
     // --pageserver-id when using a non-pageserver command
@@ -1288,20 +1254,17 @@ fn cli() -> Command {
             Command::new("pageserver")
                 .arg_required_else_help(true)
                 .about("Manage pageserver")
-                .arg(pageserver_id_arg)
                 .subcommand(Command::new("status"))
-                .subcommand(Command::new("start")
-                    .about("Start local pageserver")
-                    .arg(pageserver_config_args.clone())
-                )
-                .subcommand(Command::new("stop")
-                    .about("Stop local pageserver")
-                    .arg(stop_mode_arg.clone())
-                )
-                .subcommand(Command::new("restart")
-                    .about("Restart local pageserver")
-                    .arg(pageserver_config_args.clone())
-                )
+                .arg(pageserver_id_arg.clone())
+                .subcommand(Command::new("start").about("Start local pageserver")
+                .arg(pageserver_id_arg.clone())
+                .arg(pageserver_config_args.clone()))
+                .subcommand(Command::new("stop").about("Stop local pageserver")
+                .arg(pageserver_id_arg.clone())
+                            .arg(stop_mode_arg.clone()))
+                .subcommand(Command::new("restart").about("Restart local pageserver")
+                .arg(pageserver_id_arg.clone())
+                .arg(pageserver_config_args.clone()))
         )
         .subcommand(
             Command::new("attachment_service")
@@ -1358,8 +1321,8 @@ fn cli() -> Command {
                     .about("Start postgres.\n If the endpoint doesn't exist yet, it is created.")
                     .arg(endpoint_id_arg.clone())
                     .arg(tenant_id_arg.clone())
-                    .arg(branch_name_arg.clone())
-                    .arg(timeline_id_arg.clone())
+                    .arg(branch_name_arg)
+                    .arg(timeline_id_arg)
                     .arg(lsn_arg)
                     .arg(pg_port_arg)
                     .arg(http_port_arg)
@@ -1372,7 +1335,7 @@ fn cli() -> Command {
                 .subcommand(
                     Command::new("stop")
                     .arg(endpoint_id_arg)
-                    .arg(tenant_id_arg.clone())
+                    .arg(tenant_id_arg)
                     .arg(
                         Arg::new("destroy")
                             .help("Also delete data directory (now optional, should be default in future)")
@@ -1383,18 +1346,6 @@ fn cli() -> Command {
                 )
 
         )
-        .subcommand(
-            Command::new("mappings")
-                .arg_required_else_help(true)
-                .about("Manage neon_local branch name mappings")
-                .subcommand(
-                    Command::new("map")
-                        .about("Create new mapping which cannot exist already")
-                        .arg(branch_name_arg.clone())
-                        .arg(tenant_id_arg.clone())
-                        .arg(timeline_id_arg.clone())
-                )
-        )
         // Obsolete old name for 'endpoint'. We now just print an error if it's used.
         .subcommand(
             Command::new("pg")

control_plane/src/broker.rs

@@ -7,7 +7,7 @@
 //! ```
 use anyhow::Context;
 
-use camino::Utf8PathBuf;
+use std::path::PathBuf;
 
 use crate::{background_process, local_env};
 
@@ -30,7 +30,7 @@ pub fn start_broker_process(env: &local_env::LocalEnv) -> anyhow::Result<()> {
         || {
             let url = broker.client_url();
             let status_url = url.join("status").with_context(|| {
-                format!("Failed to append /status path to broker endpoint {url}")
+                format!("Failed to append /status path to broker endpoint {url}",)
             })?;
             let request = client
                 .get(status_url)
@@ -50,7 +50,6 @@ pub fn stop_broker_process(env: &local_env::LocalEnv) -> anyhow::Result<()> {
     background_process::stop_process(true, "storage_broker", &storage_broker_pid_file_path(env))
 }
 
-fn storage_broker_pid_file_path(env: &local_env::LocalEnv) -> Utf8PathBuf {
-    Utf8PathBuf::from_path_buf(env.base_data_dir.join("storage_broker.pid"))
-        .expect("non-Unicode path")
+fn storage_broker_pid_file_path(env: &local_env::LocalEnv) -> PathBuf {
+    env.base_data_dir.join("storage_broker.pid")
 }

control_plane/src/pageserver.rs

@@ -14,7 +14,6 @@ use std::process::{Child, Command};
 use std::{io, result};
 
 use anyhow::{bail, Context};
-use camino::Utf8PathBuf;
 use pageserver_api::models::{self, TenantInfo, TimelineInfo};
 use postgres_backend::AuthType;
 use postgres_connection::{parse_host_port, PgConnectionConfig};
@@ -145,7 +144,7 @@ impl PageServerNode {
     pub fn initialize(&self, config_overrides: &[&str]) -> anyhow::Result<()> {
         // First, run `pageserver --init` and wait for it to write a config into FS and exit.
         self.pageserver_init(config_overrides)
-            .with_context(|| format!("Failed to run init for pageserver node {}", self.conf.id))
+            .with_context(|| format!("Failed to run init for pageserver node {}", self.conf.id,))
     }
 
     pub fn repo_path(&self) -> PathBuf {
@@ -155,9 +154,8 @@ impl PageServerNode {
     /// The pid file is created by the pageserver process, with its pid stored inside.
     /// Other pageservers cannot lock the same file and overwrite it for as long as the current
     /// pageserver runs. (Unless someone removes the file manually; never do that!)
-    fn pid_file(&self) -> Utf8PathBuf {
-        Utf8PathBuf::from_path_buf(self.repo_path().join("pageserver.pid"))
-            .expect("non-Unicode path")
+    fn pid_file(&self) -> PathBuf {
+        self.repo_path().join("pageserver.pid")
     }
 
     pub fn start(&self, config_overrides: &[&str]) -> anyhow::Result<Child> {

control_plane/src/safekeeper.rs

@@ -11,7 +11,6 @@ use std::process::Child;
 use std::{io, result};
 
 use anyhow::Context;
-use camino::Utf8PathBuf;
 use postgres_connection::PgConnectionConfig;
 use reqwest::blocking::{Client, RequestBuilder, Response};
 use reqwest::{IntoUrl, Method};
@@ -98,9 +97,8 @@ impl SafekeeperNode {
         SafekeeperNode::datadir_path_by_id(&self.env, self.id)
     }
 
-    pub fn pid_file(&self) -> Utf8PathBuf {
-        Utf8PathBuf::from_path_buf(self.datadir_path().join("safekeeper.pid"))
-            .expect("non-Unicode path")
+    pub fn pid_file(&self) -> PathBuf {
+        self.datadir_path().join("safekeeper.pid")
     }
 
     pub fn start(&self, extra_opts: Vec<String>) -> anyhow::Result<Child> {

docs/pageserver-services.md

@@ -96,16 +96,6 @@ prefix_in_bucket = '/test_prefix/'
 
 `AWS_SECRET_ACCESS_KEY` and `AWS_ACCESS_KEY_ID` env variables can be used to specify the S3 credentials if needed.
 
-or
-
-```toml
-[remote_storage]
-container_name = 'some-container-name'
-container_region = 'us-east'
-prefix_in_container = '/test-prefix/'
-```
-
-`AZURE_STORAGE_ACCOUNT` and `AZURE_STORAGE_ACCESS_KEY` env variables can be used to specify the azure credentials if needed.
 
 ## Repository background tasks
 

libs/compute_api/src/spec.rs

@@ -200,12 +200,6 @@ pub struct Database {
     pub name: PgIdent,
     pub owner: PgIdent,
     pub options: GenericOptions,
-    // These are derived flags, not present in the spec file.
-    // They are never set by the control plane.
-    #[serde(skip_deserializing, default)]
-    pub restrict_conn: bool,
-    #[serde(skip_deserializing, default)]
-    pub invalid: bool,
 }
 
 /// Common type representing both SQL statement params with or without value,

libs/pageserver_api/src/models.rs

@@ -10,7 +10,6 @@ use serde_with::{serde_as, DisplayFromStr};
 use strum_macros;
 use utils::{
     completion,
-    generation::Generation,
     history_buffer::HistoryBufferWithDropCounter,
     id::{NodeId, TenantId, TimelineId},
     lsn::Lsn,
@@ -219,8 +218,6 @@ impl std::ops::Deref for TenantCreateRequest {
     }
 }
 
-/// An alternative representation of `pageserver::tenant::TenantConf` with
-/// simpler types.
 #[derive(Serialize, Deserialize, Debug, Default)]
 pub struct TenantConfig {
     pub checkpoint_distance: Option<u64>,
@@ -246,39 +243,6 @@ pub struct TenantConfig {
     pub gc_feedback: Option<bool>,
 }
 
-/// A flattened analog of a `pagesever::tenant::LocationMode`, which
-/// lists out all possible states (and the virtual "Detached" state)
-/// in a flat form rather than using rust-style enums.
-#[derive(Serialize, Deserialize, Debug)]
-pub enum LocationConfigMode {
-    AttachedSingle,
-    AttachedMulti,
-    AttachedStale,
-    Secondary,
-    Detached,
-}
-
-#[derive(Serialize, Deserialize, Debug)]
-pub struct LocationConfigSecondary {
-    pub warm: bool,
-}
-
-/// An alternative representation of `pageserver::tenant::LocationConf`,
-/// for use in external-facing APIs.
-#[derive(Serialize, Deserialize, Debug)]
-pub struct LocationConfig {
-    pub mode: LocationConfigMode,
-    /// If attaching, in what generation?
-    #[serde(default)]
-    pub generation: Option<Generation>,
-    #[serde(default)]
-    pub secondary_conf: Option<LocationConfigSecondary>,
-
-    // If requesting mode `Secondary`, configuration for that.
-    // Custom storage configuration for the tenant, if any
-    pub tenant_conf: TenantConfig,
-}
-
 #[serde_as]
 #[derive(Serialize, Deserialize)]
 #[serde(transparent)]
@@ -289,16 +253,6 @@ pub struct StatusResponse {
     pub id: NodeId,
 }
 
-#[serde_as]
-#[derive(Serialize, Deserialize, Debug)]
-#[serde(deny_unknown_fields)]
-pub struct TenantLocationConfigRequest {
-    #[serde_as(as = "DisplayFromStr")]
-    pub tenant_id: TenantId,
-    #[serde(flatten)]
-    pub config: LocationConfig, // as we have a flattened field, we should reject all unknown fields in it
-}
-
 #[serde_as]
 #[derive(Serialize, Deserialize, Debug)]
 #[serde(deny_unknown_fields)]

libs/postgres_backend/src/lib.rs

@@ -442,20 +442,10 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> PostgresBackend<IO> {
             trace!("got message {:?}", msg);
 
             let result = self.process_message(handler, msg, &mut query_string).await;
-            tokio::select!(
-                biased;
-                _ = shutdown_watcher() => {
-                    // We were requested to shut down.
-                    tracing::info!("shutdown request received during response flush");
-                    return Ok(())
-                },
-                flush_r = self.flush() => {
-                    flush_r?;
-                }
-            );
-
+            self.flush().await?;
             match result? {
                 ProcessMsgResult::Continue => {
+                    self.flush().await?;
                     continue;
                 }
                 ProcessMsgResult::Break => break,

libs/postgres_ffi/wal_craft/Cargo.toml

@@ -12,7 +12,7 @@ log.workspace = true
 once_cell.workspace = true
 postgres.workspace = true
 postgres_ffi.workspace = true
-camino-tempfile.workspace = true
+tempfile.workspace = true
 
 workspace_hack.workspace = true
 

libs/postgres_ffi/wal_craft/src/lib.rs

@@ -1,5 +1,4 @@
 use anyhow::{bail, ensure};
-use camino_tempfile::{tempdir, Utf8TempDir};
 use log::*;
 use postgres::types::PgLsn;
 use postgres::Client;
@@ -9,6 +8,7 @@ use std::cmp::Ordering;
 use std::path::{Path, PathBuf};
 use std::process::Command;
 use std::time::{Duration, Instant};
+use tempfile::{tempdir, TempDir};
 
 macro_rules! xlog_utils_test {
     ($version:ident) => {
@@ -33,7 +33,7 @@ pub struct Conf {
 
 pub struct PostgresServer {
     process: std::process::Child,
-    _unix_socket_dir: Utf8TempDir,
+    _unix_socket_dir: TempDir,
     client_config: postgres::Config,
 }
 

libs/remote_storage/Cargo.toml

@@ -13,8 +13,6 @@ aws-types.workspace = true
 aws-config.workspace = true
 aws-sdk-s3.workspace = true
 aws-credential-types.workspace = true
-bytes.workspace = true
-camino.workspace = true
 hyper = { workspace = true, features = ["stream"] }
 serde.workspace = true
 serde_json.workspace = true
@@ -27,14 +25,8 @@ metrics.workspace = true
 utils.workspace = true
 pin-project-lite.workspace = true
 workspace_hack.workspace = true
-azure_core.workspace = true
-azure_storage.workspace = true
-azure_storage_blobs.workspace = true
-futures-util.workspace = true
-http-types.workspace = true
-itertools.workspace = true
 
 [dev-dependencies]
-camino-tempfile.workspace = true
+tempfile.workspace = true
 test-context.workspace = true
 rand.workspace = true

libs/remote_storage/src/azure_blob.rs

@@ -1,381 +0,0 @@
-//! Azure Blob Storage wrapper
-
-use std::num::NonZeroU32;
-use std::{borrow::Cow, collections::HashMap, io::Cursor};
-
-use super::REMOTE_STORAGE_PREFIX_SEPARATOR;
-use anyhow::Result;
-use azure_core::request_options::{MaxResults, Metadata, Range};
-use azure_core::Header;
-use azure_storage::StorageCredentials;
-use azure_storage_blobs::prelude::ClientBuilder;
-use azure_storage_blobs::{
-    blob::operations::GetBlobBuilder,
-    prelude::{BlobClient, ContainerClient},
-};
-use futures_util::StreamExt;
-use http_types::StatusCode;
-use tokio::io::AsyncRead;
-use tracing::debug;
-
-use crate::s3_bucket::RequestKind;
-use crate::{
-    AzureConfig, ConcurrencyLimiter, Download, DownloadError, RemotePath, RemoteStorage,
-    StorageMetadata,
-};
-
-pub struct AzureBlobStorage {
-    client: ContainerClient,
-    prefix_in_container: Option<String>,
-    max_keys_per_list_response: Option<NonZeroU32>,
-    concurrency_limiter: ConcurrencyLimiter,
-}
-
-impl AzureBlobStorage {
-    pub fn new(azure_config: &AzureConfig) -> Result<Self> {
-        debug!(
-            "Creating azure remote storage for azure container {}",
-            azure_config.container_name
-        );
-
-        let account =
-            std::env::var("AZURE_STORAGE_ACCOUNT").expect("missing AZURE_STORAGE_ACCOUNT");
-        let access_key =
-            std::env::var("AZURE_STORAGE_ACCESS_KEY").expect("missing AZURE_STORAGE_ACCESS_KEY");
-
-        let credentials = StorageCredentials::access_key(account.clone(), access_key);
-
-        let builder = ClientBuilder::new(account, credentials);
-
-        let client = builder.container_client(azure_config.container_name.to_owned());
-
-        let max_keys_per_list_response =
-            if let Some(limit) = azure_config.max_keys_per_list_response {
-                Some(
-                    NonZeroU32::new(limit as u32)
-                        .ok_or_else(|| anyhow::anyhow!("max_keys_per_list_response can't be 0"))?,
-                )
-            } else {
-                None
-            };
-
-        Ok(AzureBlobStorage {
-            client,
-            prefix_in_container: azure_config.prefix_in_container.to_owned(),
-            max_keys_per_list_response,
-            concurrency_limiter: ConcurrencyLimiter::new(azure_config.concurrency_limit.get()),
-        })
-    }
-
-    pub fn relative_path_to_name(&self, path: &RemotePath) -> String {
-        assert_eq!(std::path::MAIN_SEPARATOR, REMOTE_STORAGE_PREFIX_SEPARATOR);
-        let path_string = path
-            .get_path()
-            .as_str()
-            .trim_end_matches(REMOTE_STORAGE_PREFIX_SEPARATOR);
-        match &self.prefix_in_container {
-            Some(prefix) => {
-                if prefix.ends_with(REMOTE_STORAGE_PREFIX_SEPARATOR) {
-                    prefix.clone() + path_string
-                } else {
-                    format!("{prefix}{REMOTE_STORAGE_PREFIX_SEPARATOR}{path_string}")
-                }
-            }
-            None => path_string.to_string(),
-        }
-    }
-
-    fn name_to_relative_path(&self, key: &str) -> RemotePath {
-        let relative_path =
-            match key.strip_prefix(self.prefix_in_container.as_deref().unwrap_or_default()) {
-                Some(stripped) => stripped,
-                // we rely on Azure to return properly prefixed paths
-                // for requests with a certain prefix
-                None => panic!(
-                    "Key {key} does not start with container prefix {:?}",
-                    self.prefix_in_container
-                ),
-            };
-        RemotePath(
-            relative_path
-                .split(REMOTE_STORAGE_PREFIX_SEPARATOR)
-                .collect(),
-        )
-    }
-
-    async fn download_for_builder(
-        &self,
-        metadata: StorageMetadata,
-        builder: GetBlobBuilder,
-    ) -> Result<Download, DownloadError> {
-        let mut response = builder.into_stream();
-
-        // TODO give proper streaming response instead of buffering into RAM
-        // https://github.com/neondatabase/neon/issues/5563
-        let mut buf = Vec::new();
-        while let Some(part) = response.next().await {
-            let part = match part {
-                Ok(l) => l,
-                Err(e) => {
-                    return Err(if let Some(http_err) = e.as_http_error() {
-                        match http_err.status() {
-                            StatusCode::NotFound => DownloadError::NotFound,
-                            StatusCode::BadRequest => {
-                                DownloadError::BadInput(anyhow::Error::new(e))
-                            }
-                            _ => DownloadError::Other(anyhow::Error::new(e)),
-                        }
-                    } else {
-                        DownloadError::Other(e.into())
-                    });
-                }
-            };
-            let data = part
-                .data
-                .collect()
-                .await
-                .map_err(|e| DownloadError::Other(e.into()))?;
-            buf.extend_from_slice(&data.slice(..));
-        }
-        Ok(Download {
-            download_stream: Box::pin(Cursor::new(buf)),
-            metadata: Some(metadata),
-        })
-    }
-    // TODO get rid of this function once we have metadata included in the response
-    // https://github.com/Azure/azure-sdk-for-rust/issues/1439
-    async fn get_metadata(
-        &self,
-        blob_client: &BlobClient,
-    ) -> Result<StorageMetadata, DownloadError> {
-        let builder = blob_client.get_metadata();
-
-        match builder.into_future().await {
-            Ok(r) => {
-                let mut map = HashMap::new();
-
-                for md in r.metadata.iter() {
-                    map.insert(
-                        md.name().as_str().to_string(),
-                        md.value().as_str().to_string(),
-                    );
-                }
-                Ok(StorageMetadata(map))
-            }
-            Err(e) => {
-                return Err(if let Some(http_err) = e.as_http_error() {
-                    match http_err.status() {
-                        StatusCode::NotFound => DownloadError::NotFound,
-                        StatusCode::BadRequest => DownloadError::BadInput(anyhow::Error::new(e)),
-                        _ => DownloadError::Other(anyhow::Error::new(e)),
-                    }
-                } else {
-                    DownloadError::Other(e.into())
-                });
-            }
-        }
-    }
-
-    async fn permit(&self, kind: RequestKind) -> tokio::sync::SemaphorePermit<'_> {
-        self.concurrency_limiter
-            .acquire(kind)
-            .await
-            .expect("semaphore is never closed")
-    }
-}
-
-fn to_azure_metadata(metadata: StorageMetadata) -> Metadata {
-    let mut res = Metadata::new();
-    for (k, v) in metadata.0.into_iter() {
-        res.insert(k, v);
-    }
-    res
-}
-
-#[async_trait::async_trait]
-impl RemoteStorage for AzureBlobStorage {
-    async fn list_prefixes(
-        &self,
-        prefix: Option<&RemotePath>,
-    ) -> Result<Vec<RemotePath>, DownloadError> {
-        // get the passed prefix or if it is not set use prefix_in_bucket value
-        let list_prefix = prefix
-            .map(|p| self.relative_path_to_name(p))
-            .or_else(|| self.prefix_in_container.clone())
-            .map(|mut p| {
-                // required to end with a separator
-                // otherwise request will return only the entry of a prefix
-                if !p.ends_with(REMOTE_STORAGE_PREFIX_SEPARATOR) {
-                    p.push(REMOTE_STORAGE_PREFIX_SEPARATOR);
-                }
-                p
-            });
-
-        let mut builder = self
-            .client
-            .list_blobs()
-            .delimiter(REMOTE_STORAGE_PREFIX_SEPARATOR.to_string());
-
-        if let Some(prefix) = list_prefix {
-            builder = builder.prefix(Cow::from(prefix.to_owned()));
-        }
-
-        if let Some(limit) = self.max_keys_per_list_response {
-            builder = builder.max_results(MaxResults::new(limit));
-        }
-
-        let mut response = builder.into_stream();
-        let mut res = Vec::new();
-        while let Some(l) = response.next().await {
-            let entry = match l {
-                Ok(l) => l,
-                Err(e) => {
-                    return Err(if let Some(http_err) = e.as_http_error() {
-                        match http_err.status() {
-                            StatusCode::NotFound => DownloadError::NotFound,
-                            StatusCode::BadRequest => {
-                                DownloadError::BadInput(anyhow::Error::new(e))
-                            }
-                            _ => DownloadError::Other(anyhow::Error::new(e)),
-                        }
-                    } else {
-                        DownloadError::Other(e.into())
-                    });
-                }
-            };
-            let name_iter = entry
-                .blobs
-                .prefixes()
-                .map(|prefix| self.name_to_relative_path(&prefix.name));
-            res.extend(name_iter);
-        }
-        Ok(res)
-    }
-
-    async fn list_files(&self, folder: Option<&RemotePath>) -> anyhow::Result<Vec<RemotePath>> {
-        let folder_name = folder
-            .map(|p| self.relative_path_to_name(p))
-            .or_else(|| self.prefix_in_container.clone());
-
-        let mut builder = self.client.list_blobs();
-
-        if let Some(folder_name) = folder_name {
-            builder = builder.prefix(Cow::from(folder_name.to_owned()));
-        }
-
-        if let Some(limit) = self.max_keys_per_list_response {
-            builder = builder.max_results(MaxResults::new(limit));
-        }
-
-        let mut response = builder.into_stream();
-        let mut res = Vec::new();
-        while let Some(l) = response.next().await {
-            let entry = l.map_err(anyhow::Error::new)?;
-            let name_iter = entry
-                .blobs
-                .blobs()
-                .map(|bl| self.name_to_relative_path(&bl.name));
-            res.extend(name_iter);
-        }
-        Ok(res)
-    }
-
-    async fn upload(
-        &self,
-        mut from: impl AsyncRead + Unpin + Send + Sync + 'static,
-        data_size_bytes: usize,
-        to: &RemotePath,
-        metadata: Option<StorageMetadata>,
-    ) -> anyhow::Result<()> {
-        let _permit = self.permit(RequestKind::Put).await;
-        let blob_client = self.client.blob_client(self.relative_path_to_name(to));
-
-        // TODO FIX THIS UGLY HACK and don't buffer the entire object
-        // into RAM here, but use the streaming interface. For that,
-        // we'd have to change the interface though...
-        // https://github.com/neondatabase/neon/issues/5563
-        let mut buf = Vec::with_capacity(data_size_bytes);
-        tokio::io::copy(&mut from, &mut buf).await?;
-        let body = azure_core::Body::Bytes(buf.into());
-
-        let mut builder = blob_client.put_block_blob(body);
-
-        if let Some(metadata) = metadata {
-            builder = builder.metadata(to_azure_metadata(metadata));
-        }
-
-        let _response = builder.into_future().await?;
-
-        Ok(())
-    }
-
-    async fn download(&self, from: &RemotePath) -> Result<Download, DownloadError> {
-        let _permit = self.permit(RequestKind::Get).await;
-        let blob_client = self.client.blob_client(self.relative_path_to_name(from));
-
-        let metadata = self.get_metadata(&blob_client).await?;
-
-        let builder = blob_client.get();
-
-        self.download_for_builder(metadata, builder).await
-    }
-
-    async fn download_byte_range(
-        &self,
-        from: &RemotePath,
-        start_inclusive: u64,
-        end_exclusive: Option<u64>,
-    ) -> Result<Download, DownloadError> {
-        let _permit = self.permit(RequestKind::Get).await;
-        let blob_client = self.client.blob_client(self.relative_path_to_name(from));
-
-        let metadata = self.get_metadata(&blob_client).await?;
-
-        let mut builder = blob_client.get();
-
-        if let Some(end_exclusive) = end_exclusive {
-            builder = builder.range(Range::new(start_inclusive, end_exclusive));
-        } else {
-            // Open ranges are not supported by the SDK so we work around
-            // by setting the upper limit extremely high (but high enough
-            // to still be representable by signed 64 bit integers).
-            // TODO remove workaround once the SDK adds open range support
-            // https://github.com/Azure/azure-sdk-for-rust/issues/1438
-            let end_exclusive = u64::MAX / 4;
-            builder = builder.range(Range::new(start_inclusive, end_exclusive));
-        }
-
-        self.download_for_builder(metadata, builder).await
-    }
-
-    async fn delete(&self, path: &RemotePath) -> anyhow::Result<()> {
-        let _permit = self.permit(RequestKind::Delete).await;
-        let blob_client = self.client.blob_client(self.relative_path_to_name(path));
-
-        let builder = blob_client.delete();
-
-        match builder.into_future().await {
-            Ok(_response) => Ok(()),
-            Err(e) => {
-                if let Some(http_err) = e.as_http_error() {
-                    if http_err.status() == StatusCode::NotFound {
-                        return Ok(());
-                    }
-                }
-                Err(anyhow::Error::new(e))
-            }
-        }
-    }
-
-    async fn delete_objects<'a>(&self, paths: &'a [RemotePath]) -> anyhow::Result<()> {
-        // Permit is already obtained by inner delete function
-
-        // TODO batch requests are also not supported by the SDK
-        // https://github.com/Azure/azure-sdk-for-rust/issues/1068
-        // https://github.com/Azure/azure-sdk-for-rust/issues/1249
-        for path in paths {
-            self.delete(path).await?;
-        }
-        Ok(())
-    }
-}

libs/remote_storage/src/lib.rs

@@ -4,10 +4,7 @@
 //! [`RemoteStorage`] trait a CRUD-like generic abstraction to use for adapting external storages with a few implementations:
 //!   * [`local_fs`] allows to use local file system as an external storage
 //!   * [`s3_bucket`] uses AWS S3 bucket as an external storage
-//!   * [`azure_blob`] allows to use Azure Blob storage as an external storage
 //!
-
-mod azure_blob;
 mod local_fs;
 mod s3_bucket;
 mod simulate_failures;
@@ -16,23 +13,19 @@ use std::{
     collections::HashMap,
     fmt::Debug,
     num::{NonZeroU32, NonZeroUsize},
+    path::{Path, PathBuf},
     pin::Pin,
     sync::Arc,
 };
 
 use anyhow::{bail, Context};
-use camino::{Utf8Path, Utf8PathBuf};
 
 use serde::{Deserialize, Serialize};
-use tokio::{io, sync::Semaphore};
+use tokio::io;
 use toml_edit::Item;
 use tracing::info;
 
-pub use self::{
-    azure_blob::AzureBlobStorage, local_fs::LocalFs, s3_bucket::S3Bucket,
-    simulate_failures::UnreliableWrapper,
-};
-use s3_bucket::RequestKind;
+pub use self::{local_fs::LocalFs, s3_bucket::S3Bucket, simulate_failures::UnreliableWrapper};
 
 /// How many different timelines can be processed simultaneously when synchronizing layers with the remote storage.
 /// During regular work, pageserver produces one layer file per timeline checkpoint, with bursts of concurrency
@@ -46,11 +39,6 @@ pub const DEFAULT_REMOTE_STORAGE_MAX_SYNC_ERRORS: u32 = 10;
 /// ~3500 PUT/COPY/POST/DELETE or 5500 GET/HEAD S3 requests
 /// <https://aws.amazon.com/premiumsupport/knowledge-center/s3-request-limit-avoid-throttling/>
 pub const DEFAULT_REMOTE_STORAGE_S3_CONCURRENCY_LIMIT: usize = 100;
-/// We set this a little bit low as we currently buffer the entire file into RAM
-///
-/// Here, a limit of max 20k concurrent connections was noted.
-/// <https://learn.microsoft.com/en-us/answers/questions/1301863/is-there-any-limitation-to-concurrent-connections>
-pub const DEFAULT_REMOTE_STORAGE_AZURE_CONCURRENCY_LIMIT: usize = 30;
 /// No limits on the client side, which currenltly means 1000 for AWS S3.
 /// <https://docs.aws.amazon.com/AmazonS3/latest/API/API_ListObjectsV2.html#API_ListObjectsV2_RequestSyntax>
 pub const DEFAULT_MAX_KEYS_PER_LIST_RESPONSE: Option<i32> = None;
@@ -64,7 +52,7 @@ const REMOTE_STORAGE_PREFIX_SEPARATOR: char = '/';
 /// The prefix is an implementation detail, that allows representing local paths
 /// as the remote ones, stripping the local storage prefix away.
 #[derive(Debug, Clone, PartialEq, Eq, PartialOrd, Ord, Hash)]
-pub struct RemotePath(Utf8PathBuf);
+pub struct RemotePath(PathBuf);
 
 impl Serialize for RemotePath {
     fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
@@ -81,18 +69,18 @@ impl<'de> Deserialize<'de> for RemotePath {
         D: serde::Deserializer<'de>,
     {
         let str = String::deserialize(deserializer)?;
-        Ok(Self(Utf8PathBuf::from(&str)))
+        Ok(Self(PathBuf::from(&str)))
     }
 }
 
 impl std::fmt::Display for RemotePath {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        std::fmt::Display::fmt(&self.0, f)
+        write!(f, "{}", self.0.display())
     }
 }
 
 impl RemotePath {
-    pub fn new(relative_path: &Utf8Path) -> anyhow::Result<Self> {
+    pub fn new(relative_path: &Path) -> anyhow::Result<Self> {
         anyhow::ensure!(
             relative_path.is_relative(),
             "Path {relative_path:?} is not relative"
@@ -101,30 +89,30 @@ impl RemotePath {
     }
 
     pub fn from_string(relative_path: &str) -> anyhow::Result<Self> {
-        Self::new(Utf8Path::new(relative_path))
+        Self::new(Path::new(relative_path))
     }
 
-    pub fn with_base(&self, base_path: &Utf8Path) -> Utf8PathBuf {
+    pub fn with_base(&self, base_path: &Path) -> PathBuf {
         base_path.join(&self.0)
     }
 
     pub fn object_name(&self) -> Option<&str> {
-        self.0.file_name()
+        self.0.file_name().and_then(|os_str| os_str.to_str())
     }
 
-    pub fn join(&self, segment: &Utf8Path) -> Self {
+    pub fn join(&self, segment: &Path) -> Self {
         Self(self.0.join(segment))
     }
 
-    pub fn get_path(&self) -> &Utf8PathBuf {
+    pub fn get_path(&self) -> &PathBuf {
         &self.0
     }
 
     pub fn extension(&self) -> Option<&str> {
-        self.0.extension()
+        self.0.extension()?.to_str()
     }
 
-    pub fn strip_prefix(&self, p: &RemotePath) -> Result<&Utf8Path, std::path::StripPrefixError> {
+    pub fn strip_prefix(&self, p: &RemotePath) -> Result<&Path, std::path::StripPrefixError> {
         self.0.strip_prefix(&p.0)
     }
 }
@@ -229,7 +217,6 @@ impl std::error::Error for DownloadError {}
 pub enum GenericRemoteStorage {
     LocalFs(LocalFs),
     AwsS3(Arc<S3Bucket>),
-    AzureBlob(Arc<AzureBlobStorage>),
     Unreliable(Arc<UnreliableWrapper>),
 }
 
@@ -241,7 +228,6 @@ impl GenericRemoteStorage {
         match self {
             Self::LocalFs(s) => s.list_files(folder).await,
             Self::AwsS3(s) => s.list_files(folder).await,
-            Self::AzureBlob(s) => s.list_files(folder).await,
             Self::Unreliable(s) => s.list_files(folder).await,
         }
     }
@@ -256,7 +242,6 @@ impl GenericRemoteStorage {
         match self {
             Self::LocalFs(s) => s.list_prefixes(prefix).await,
             Self::AwsS3(s) => s.list_prefixes(prefix).await,
-            Self::AzureBlob(s) => s.list_prefixes(prefix).await,
             Self::Unreliable(s) => s.list_prefixes(prefix).await,
         }
     }
@@ -271,7 +256,6 @@ impl GenericRemoteStorage {
         match self {
             Self::LocalFs(s) => s.upload(from, data_size_bytes, to, metadata).await,
             Self::AwsS3(s) => s.upload(from, data_size_bytes, to, metadata).await,
-            Self::AzureBlob(s) => s.upload(from, data_size_bytes, to, metadata).await,
             Self::Unreliable(s) => s.upload(from, data_size_bytes, to, metadata).await,
         }
     }
@@ -280,7 +264,6 @@ impl GenericRemoteStorage {
         match self {
             Self::LocalFs(s) => s.download(from).await,
             Self::AwsS3(s) => s.download(from).await,
-            Self::AzureBlob(s) => s.download(from).await,
             Self::Unreliable(s) => s.download(from).await,
         }
     }
@@ -300,10 +283,6 @@ impl GenericRemoteStorage {
                 s.download_byte_range(from, start_inclusive, end_exclusive)
                     .await
             }
-            Self::AzureBlob(s) => {
-                s.download_byte_range(from, start_inclusive, end_exclusive)
-                    .await
-            }
             Self::Unreliable(s) => {
                 s.download_byte_range(from, start_inclusive, end_exclusive)
                     .await
@@ -315,7 +294,6 @@ impl GenericRemoteStorage {
         match self {
             Self::LocalFs(s) => s.delete(path).await,
             Self::AwsS3(s) => s.delete(path).await,
-            Self::AzureBlob(s) => s.delete(path).await,
             Self::Unreliable(s) => s.delete(path).await,
         }
     }
@@ -324,7 +302,6 @@ impl GenericRemoteStorage {
         match self {
             Self::LocalFs(s) => s.delete_objects(paths).await,
             Self::AwsS3(s) => s.delete_objects(paths).await,
-            Self::AzureBlob(s) => s.delete_objects(paths).await,
             Self::Unreliable(s) => s.delete_objects(paths).await,
         }
     }
@@ -334,7 +311,7 @@ impl GenericRemoteStorage {
     pub fn from_config(storage_config: &RemoteStorageConfig) -> anyhow::Result<Self> {
         Ok(match &storage_config.storage {
             RemoteStorageKind::LocalFs(root) => {
-                info!("Using fs root '{root}' as a remote storage");
+                info!("Using fs root '{}' as a remote storage", root.display());
                 Self::LocalFs(LocalFs::new(root.clone())?)
             }
             RemoteStorageKind::AwsS3(s3_config) => {
@@ -342,11 +319,6 @@ impl GenericRemoteStorage {
                       s3_config.bucket_name, s3_config.bucket_region, s3_config.prefix_in_bucket, s3_config.endpoint);
                 Self::AwsS3(Arc::new(S3Bucket::new(s3_config)?))
             }
-            RemoteStorageKind::AzureContainer(azure_config) => {
-                info!("Using azure container '{}' in region '{}' as a remote storage, prefix in container: '{:?}'",
-                      azure_config.container_name, azure_config.container_region, azure_config.prefix_in_container);
-                Self::AzureBlob(Arc::new(AzureBlobStorage::new(azure_config)?))
-            }
         })
     }
 
@@ -407,13 +379,10 @@ pub struct RemoteStorageConfig {
 pub enum RemoteStorageKind {
     /// Storage based on local file system.
     /// Specify a root folder to place all stored files into.
-    LocalFs(Utf8PathBuf),
+    LocalFs(PathBuf),
     /// AWS S3 based storage, storing all files in the S3 bucket
     /// specified by the config
     AwsS3(S3Config),
-    /// Azure Blob based storage, storing all files in the container
-    /// specified by the config
-    AzureContainer(AzureConfig),
 }
 
 /// AWS S3 bucket coordinates and access credentials to manage the bucket contents (read and write).
@@ -453,45 +422,11 @@ impl Debug for S3Config {
     }
 }
 
-/// Azure  bucket coordinates and access credentials to manage the bucket contents (read and write).
-#[derive(Clone, PartialEq, Eq)]
-pub struct AzureConfig {
-    /// Name of the container to connect to.
-    pub container_name: String,
-    /// The region where the bucket is located at.
-    pub container_region: String,
-    /// A "subfolder" in the container, to use the same container separately by multiple remote storage users at once.
-    pub prefix_in_container: Option<String>,
-    /// Azure has various limits on its API calls, we need not to exceed those.
-    /// See [`DEFAULT_REMOTE_STORAGE_AZURE_CONCURRENCY_LIMIT`] for more details.
-    pub concurrency_limit: NonZeroUsize,
-    pub max_keys_per_list_response: Option<i32>,
-}
-
-impl Debug for AzureConfig {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        f.debug_struct("AzureConfig")
-            .field("bucket_name", &self.container_name)
-            .field("bucket_region", &self.container_region)
-            .field("prefix_in_bucket", &self.prefix_in_container)
-            .field("concurrency_limit", &self.concurrency_limit)
-            .field(
-                "max_keys_per_list_response",
-                &self.max_keys_per_list_response,
-            )
-            .finish()
-    }
-}
-
 impl RemoteStorageConfig {
     pub fn from_toml(toml: &toml_edit::Item) -> anyhow::Result<Option<RemoteStorageConfig>> {
         let local_path = toml.get("local_path");
         let bucket_name = toml.get("bucket_name");
         let bucket_region = toml.get("bucket_region");
-        let container_name = toml.get("container_name");
-        let container_region = toml.get("container_region");
-
-        let use_azure = container_name.is_some() && container_region.is_some();
 
         let max_concurrent_syncs = NonZeroUsize::new(
             parse_optional_integer("max_concurrent_syncs", toml)?
@@ -505,13 +440,9 @@ impl RemoteStorageConfig {
         )
         .context("Failed to parse 'max_sync_errors' as a positive integer")?;
 
-        let default_concurrency_limit = if use_azure {
-            DEFAULT_REMOTE_STORAGE_AZURE_CONCURRENCY_LIMIT
-        } else {
-            DEFAULT_REMOTE_STORAGE_S3_CONCURRENCY_LIMIT
-        };
         let concurrency_limit = NonZeroUsize::new(
-            parse_optional_integer("concurrency_limit", toml)?.unwrap_or(default_concurrency_limit),
+            parse_optional_integer("concurrency_limit", toml)?
+                .unwrap_or(DEFAULT_REMOTE_STORAGE_S3_CONCURRENCY_LIMIT),
         )
         .context("Failed to parse 'concurrency_limit' as a positive integer")?;
 
@@ -520,70 +451,33 @@ impl RemoteStorageConfig {
                 .context("Failed to parse 'max_keys_per_list_response' as a positive integer")?
                 .or(DEFAULT_MAX_KEYS_PER_LIST_RESPONSE);
 
-        let endpoint = toml
-            .get("endpoint")
-            .map(|endpoint| parse_toml_string("endpoint", endpoint))
-            .transpose()?;
-
-        let storage = match (
-            local_path,
-            bucket_name,
-            bucket_region,
-            container_name,
-            container_region,
-        ) {
+        let storage = match (local_path, bucket_name, bucket_region) {
             // no 'local_path' nor 'bucket_name' options are provided, consider this remote storage disabled
-            (None, None, None, None, None) => return Ok(None),
-            (_, Some(_), None, ..) => {
+            (None, None, None) => return Ok(None),
+            (_, Some(_), None) => {
                 bail!("'bucket_region' option is mandatory if 'bucket_name' is given ")
             }
-            (_, None, Some(_), ..) => {
+            (_, None, Some(_)) => {
                 bail!("'bucket_name' option is mandatory if 'bucket_region' is given ")
             }
-            (None, Some(bucket_name), Some(bucket_region), ..) => {
-                RemoteStorageKind::AwsS3(S3Config {
-                    bucket_name: parse_toml_string("bucket_name", bucket_name)?,
-                    bucket_region: parse_toml_string("bucket_region", bucket_region)?,
-                    prefix_in_bucket: toml
-                        .get("prefix_in_bucket")
-                        .map(|prefix_in_bucket| {
-                            parse_toml_string("prefix_in_bucket", prefix_in_bucket)
-                        })
-                        .transpose()?,
-                    endpoint,
-                    concurrency_limit,
-                    max_keys_per_list_response,
-                })
-            }
-            (_, _, _, Some(_), None) => {
-                bail!("'container_name' option is mandatory if 'container_region' is given ")
-            }
-            (_, _, _, None, Some(_)) => {
-                bail!("'container_name' option is mandatory if 'container_region' is given ")
-            }
-            (None, None, None, Some(container_name), Some(container_region)) => {
-                RemoteStorageKind::AzureContainer(AzureConfig {
-                    container_name: parse_toml_string("container_name", container_name)?,
-                    container_region: parse_toml_string("container_region", container_region)?,
-                    prefix_in_container: toml
-                        .get("prefix_in_container")
-                        .map(|prefix_in_container| {
-                            parse_toml_string("prefix_in_container", prefix_in_container)
-                        })
-                        .transpose()?,
-                    concurrency_limit,
-                    max_keys_per_list_response,
-                })
-            }
-            (Some(local_path), None, None, None, None) => RemoteStorageKind::LocalFs(
-                Utf8PathBuf::from(parse_toml_string("local_path", local_path)?),
-            ),
-            (Some(_), Some(_), ..) => {
-                bail!("'local_path' and 'bucket_name' are mutually exclusive")
-            }
-            (Some(_), _, _, Some(_), Some(_)) => {
-                bail!("local_path and 'container_name' are mutually exclusive")
-            }
+            (None, Some(bucket_name), Some(bucket_region)) => RemoteStorageKind::AwsS3(S3Config {
+                bucket_name: parse_toml_string("bucket_name", bucket_name)?,
+                bucket_region: parse_toml_string("bucket_region", bucket_region)?,
+                prefix_in_bucket: toml
+                    .get("prefix_in_bucket")
+                    .map(|prefix_in_bucket| parse_toml_string("prefix_in_bucket", prefix_in_bucket))
+                    .transpose()?,
+                endpoint: toml
+                    .get("endpoint")
+                    .map(|endpoint| parse_toml_string("endpoint", endpoint))
+                    .transpose()?,
+                concurrency_limit,
+                max_keys_per_list_response,
+            }),
+            (Some(local_path), None, None) => RemoteStorageKind::LocalFs(PathBuf::from(
+                parse_toml_string("local_path", local_path)?,
+            )),
+            (Some(_), Some(_), _) => bail!("local_path and bucket_name are mutually exclusive"),
         };
 
         Ok(Some(RemoteStorageConfig {
@@ -619,69 +513,29 @@ fn parse_toml_string(name: &str, item: &Item) -> anyhow::Result<String> {
     Ok(s.to_string())
 }
 
-struct ConcurrencyLimiter {
-    // Every request to S3 can be throttled or cancelled, if a certain number of requests per second is exceeded.
-    // Same goes to IAM, which is queried before every S3 request, if enabled. IAM has even lower RPS threshold.
-    // The helps to ensure we don't exceed the thresholds.
-    write: Arc<Semaphore>,
-    read: Arc<Semaphore>,
-}
-
-impl ConcurrencyLimiter {
-    fn for_kind(&self, kind: RequestKind) -> &Arc<Semaphore> {
-        match kind {
-            RequestKind::Get => &self.read,
-            RequestKind::Put => &self.write,
-            RequestKind::List => &self.read,
-            RequestKind::Delete => &self.write,
-        }
-    }
-
-    async fn acquire(
-        &self,
-        kind: RequestKind,
-    ) -> Result<tokio::sync::SemaphorePermit<'_>, tokio::sync::AcquireError> {
-        self.for_kind(kind).acquire().await
-    }
-
-    async fn acquire_owned(
-        &self,
-        kind: RequestKind,
-    ) -> Result<tokio::sync::OwnedSemaphorePermit, tokio::sync::AcquireError> {
-        Arc::clone(self.for_kind(kind)).acquire_owned().await
-    }
-
-    fn new(limit: usize) -> ConcurrencyLimiter {
-        Self {
-            read: Arc::new(Semaphore::new(limit)),
-            write: Arc::new(Semaphore::new(limit)),
-        }
-    }
-}
-
 #[cfg(test)]
 mod tests {
     use super::*;
 
     #[test]
     fn test_object_name() {
-        let k = RemotePath::new(Utf8Path::new("a/b/c")).unwrap();
+        let k = RemotePath::new(Path::new("a/b/c")).unwrap();
         assert_eq!(k.object_name(), Some("c"));
 
-        let k = RemotePath::new(Utf8Path::new("a/b/c/")).unwrap();
+        let k = RemotePath::new(Path::new("a/b/c/")).unwrap();
         assert_eq!(k.object_name(), Some("c"));
 
-        let k = RemotePath::new(Utf8Path::new("a/")).unwrap();
+        let k = RemotePath::new(Path::new("a/")).unwrap();
         assert_eq!(k.object_name(), Some("a"));
 
         // XXX is it impossible to have an empty key?
-        let k = RemotePath::new(Utf8Path::new("")).unwrap();
+        let k = RemotePath::new(Path::new("")).unwrap();
         assert_eq!(k.object_name(), None);
     }
 
     #[test]
     fn rempte_path_cannot_be_created_from_absolute_ones() {
-        let err = RemotePath::new(Utf8Path::new("/")).expect_err("Should fail on absolute paths");
+        let err = RemotePath::new(Path::new("/")).expect_err("Should fail on absolute paths");
         assert_eq!(err.to_string(), "Path \"/\" is not relative");
     }
 }

libs/remote_storage/src/local_fs.rs

@@ -4,10 +4,15 @@
 //! This storage used in tests, but can also be used in cases when a certain persistent
 //! volume is mounted to the local FS.
 
-use std::{borrow::Cow, future::Future, io::ErrorKind, pin::Pin};
+use std::{
+    borrow::Cow,
+    future::Future,
+    io::ErrorKind,
+    path::{Path, PathBuf},
+    pin::Pin,
+};
 
 use anyhow::{bail, ensure, Context};
-use camino::{Utf8Path, Utf8PathBuf};
 use tokio::{
     fs,
     io::{self, AsyncReadExt, AsyncSeekExt, AsyncWriteExt},
@@ -23,20 +28,20 @@ const LOCAL_FS_TEMP_FILE_SUFFIX: &str = "___temp";
 
 #[derive(Debug, Clone)]
 pub struct LocalFs {
-    storage_root: Utf8PathBuf,
+    storage_root: PathBuf,
 }
 
 impl LocalFs {
     /// Attempts to create local FS storage, along with its root directory.
     /// Storage root will be created (if does not exist) and transformed into an absolute path (if passed as relative).
-    pub fn new(mut storage_root: Utf8PathBuf) -> anyhow::Result<Self> {
+    pub fn new(mut storage_root: PathBuf) -> anyhow::Result<Self> {
         if !storage_root.exists() {
             std::fs::create_dir_all(&storage_root).with_context(|| {
                 format!("Failed to create all directories in the given root path {storage_root:?}")
             })?;
         }
         if !storage_root.is_absolute() {
-            storage_root = storage_root.canonicalize_utf8().with_context(|| {
+            storage_root = storage_root.canonicalize().with_context(|| {
                 format!("Failed to represent path {storage_root:?} as an absolute path")
             })?;
         }
@@ -45,7 +50,7 @@ impl LocalFs {
     }
 
     // mirrors S3Bucket::s3_object_to_relative_path
-    fn local_file_to_relative_path(&self, key: Utf8PathBuf) -> RemotePath {
+    fn local_file_to_relative_path(&self, key: PathBuf) -> RemotePath {
         let relative_path = key
             .strip_prefix(&self.storage_root)
             .expect("relative path must contain storage_root as prefix");
@@ -54,18 +59,22 @@ impl LocalFs {
 
     async fn read_storage_metadata(
         &self,
-        file_path: &Utf8Path,
+        file_path: &Path,
     ) -> anyhow::Result<Option<StorageMetadata>> {
         let metadata_path = storage_metadata_path(file_path);
         if metadata_path.exists() && metadata_path.is_file() {
             let metadata_string = fs::read_to_string(&metadata_path).await.with_context(|| {
-                format!("Failed to read metadata from the local storage at '{metadata_path}'")
+                format!(
+                    "Failed to read metadata from the local storage at '{}'",
+                    metadata_path.display()
+                )
             })?;
 
             serde_json::from_str(&metadata_string)
                 .with_context(|| {
                     format!(
-                        "Failed to deserialize metadata from the local storage at '{metadata_path}'",
+                        "Failed to deserialize metadata from the local storage at '{}'",
+                        metadata_path.display()
                     )
                 })
                 .map(|metadata| Some(StorageMetadata(metadata)))
@@ -162,21 +171,25 @@ impl RemoteStorage for LocalFs {
             }
         }
 
-        // Note that Utf8PathBuf starts_with only considers full path segments, but
+        // Note that PathBuf starts_with only considers full path segments, but
         // object prefixes are arbitrary strings, so we need the strings for doing
         // starts_with later.
-        let prefix = full_path.as_str();
+        let prefix = full_path.to_string_lossy();
 
         let mut files = vec![];
-        let mut directory_queue = vec![initial_dir];
+        let mut directory_queue = vec![initial_dir.clone()];
         while let Some(cur_folder) = directory_queue.pop() {
-            let mut entries = cur_folder.read_dir_utf8()?;
-            while let Some(Ok(entry)) = entries.next() {
-                let file_name = entry.file_name();
-                let full_file_name = cur_folder.join(file_name);
-                if full_file_name.as_str().starts_with(prefix) {
+            let mut entries = fs::read_dir(cur_folder.clone()).await?;
+            while let Some(entry) = entries.next_entry().await? {
+                let file_name: PathBuf = entry.file_name().into();
+                let full_file_name = cur_folder.clone().join(&file_name);
+                if full_file_name
+                    .to_str()
+                    .map(|s| s.starts_with(prefix.as_ref()))
+                    .unwrap_or(false)
+                {
                     let file_remote_path = self.local_file_to_relative_path(full_file_name.clone());
-                    files.push(file_remote_path);
+                    files.push(file_remote_path.clone());
                     if full_file_name.is_dir() {
                         directory_queue.push(full_file_name);
                     }
@@ -217,7 +230,10 @@ impl RemoteStorage for LocalFs {
                 .open(&temp_file_path)
                 .await
                 .with_context(|| {
-                    format!("Failed to open target fs destination at '{target_file_path}'")
+                    format!(
+                        "Failed to open target fs destination at '{}'",
+                        target_file_path.display()
+                    )
                 })?,
         );
 
@@ -228,7 +244,8 @@ impl RemoteStorage for LocalFs {
             .await
             .with_context(|| {
                 format!(
-                    "Failed to upload file (write temp) to the local storage at '{temp_file_path}'",
+                    "Failed to upload file (write temp) to the local storage at '{}'",
+                    temp_file_path.display()
                 )
             })?;
 
@@ -245,7 +262,8 @@ impl RemoteStorage for LocalFs {
 
         destination.flush().await.with_context(|| {
             format!(
-                "Failed to upload (flush temp) file to the local storage at '{temp_file_path}'",
+                "Failed to upload (flush temp) file to the local storage at '{}'",
+                temp_file_path.display()
             )
         })?;
 
@@ -253,7 +271,8 @@ impl RemoteStorage for LocalFs {
             .await
             .with_context(|| {
                 format!(
-                    "Failed to upload (rename) file to the local storage at '{target_file_path}'",
+                    "Failed to upload (rename) file to the local storage at '{}'",
+                    target_file_path.display()
                 )
             })?;
 
@@ -267,7 +286,8 @@ impl RemoteStorage for LocalFs {
             .await
             .with_context(|| {
                 format!(
-                    "Failed to write metadata to the local storage at '{storage_metadata_path}'",
+                    "Failed to write metadata to the local storage at '{}'",
+                    storage_metadata_path.display()
                 )
             })?;
         }
@@ -373,16 +393,16 @@ impl RemoteStorage for LocalFs {
     }
 }
 
-fn storage_metadata_path(original_path: &Utf8Path) -> Utf8PathBuf {
+fn storage_metadata_path(original_path: &Path) -> PathBuf {
     path_with_suffix_extension(original_path, "metadata")
 }
 
 fn get_all_files<'a, P>(
     directory_path: P,
     recursive: bool,
-) -> Pin<Box<dyn Future<Output = anyhow::Result<Vec<Utf8PathBuf>>> + Send + Sync + 'a>>
+) -> Pin<Box<dyn Future<Output = anyhow::Result<Vec<PathBuf>>> + Send + Sync + 'a>>
 where
-    P: AsRef<Utf8Path> + Send + Sync + 'a,
+    P: AsRef<Path> + Send + Sync + 'a,
 {
     Box::pin(async move {
         let directory_path = directory_path.as_ref();
@@ -392,13 +412,7 @@ where
                 let mut dir_contents = fs::read_dir(directory_path).await?;
                 while let Some(dir_entry) = dir_contents.next_entry().await? {
                     let file_type = dir_entry.file_type().await?;
-                    let entry_path =
-                        Utf8PathBuf::from_path_buf(dir_entry.path()).map_err(|pb| {
-                            anyhow::Error::msg(format!(
-                                "non-Unicode path: {}",
-                                pb.to_string_lossy()
-                            ))
-                        })?;
+                    let entry_path = dir_entry.path();
                     if file_type.is_symlink() {
                         debug!("{entry_path:?} is a symlink, skipping")
                     } else if file_type.is_dir() {
@@ -421,10 +435,13 @@ where
     })
 }
 
-async fn create_target_directory(target_file_path: &Utf8Path) -> anyhow::Result<()> {
+async fn create_target_directory(target_file_path: &Path) -> anyhow::Result<()> {
     let target_dir = match target_file_path.parent() {
         Some(parent_dir) => parent_dir,
-        None => bail!("File path '{target_file_path}' has no parent directory"),
+        None => bail!(
+            "File path '{}' has no parent directory",
+            target_file_path.display()
+        ),
     };
     if !target_dir.exists() {
         fs::create_dir_all(target_dir).await?;
@@ -432,9 +449,13 @@ async fn create_target_directory(target_file_path: &Utf8Path) -> anyhow::Result<
     Ok(())
 }
 
-fn file_exists(file_path: &Utf8Path) -> anyhow::Result<bool> {
+fn file_exists(file_path: &Path) -> anyhow::Result<bool> {
     if file_path.exists() {
-        ensure!(file_path.is_file(), "file path '{file_path}' is not a file");
+        ensure!(
+            file_path.is_file(),
+            "file path '{}' is not a file",
+            file_path.display()
+        );
         Ok(true)
     } else {
         Ok(false)
@@ -445,13 +466,13 @@ fn file_exists(file_path: &Utf8Path) -> anyhow::Result<bool> {
 mod fs_tests {
     use super::*;
 
-    use camino_tempfile::tempdir;
     use std::{collections::HashMap, io::Write};
+    use tempfile::tempdir;
 
     async fn read_and_assert_remote_file_contents(
         storage: &LocalFs,
         #[allow(clippy::ptr_arg)]
-        // have to use &Utf8PathBuf due to `storage.local_path` parameter requirements
+        // have to use &PathBuf due to `storage.local_path` parameter requirements
         remote_storage_path: &RemotePath,
         expected_metadata: Option<&StorageMetadata>,
     ) -> anyhow::Result<String> {
@@ -498,7 +519,7 @@ mod fs_tests {
     async fn upload_file_negatives() -> anyhow::Result<()> {
         let storage = create_storage()?;
 
-        let id = RemotePath::new(Utf8Path::new("dummy"))?;
+        let id = RemotePath::new(Path::new("dummy"))?;
         let content = std::io::Cursor::new(b"12345");
 
         // Check that you get an error if the size parameter doesn't match the actual
@@ -523,8 +544,7 @@ mod fs_tests {
     }
 
     fn create_storage() -> anyhow::Result<LocalFs> {
-        let storage_root = tempdir()?.path().to_path_buf();
-        LocalFs::new(storage_root)
+        LocalFs::new(tempdir()?.path().to_owned())
     }
 
     #[tokio::test]
@@ -541,7 +561,7 @@ mod fs_tests {
         );
 
         let non_existing_path = "somewhere/else";
-        match storage.download(&RemotePath::new(Utf8Path::new(non_existing_path))?).await {
+        match storage.download(&RemotePath::new(Path::new(non_existing_path))?).await {
             Err(DownloadError::NotFound) => {} // Should get NotFound for non existing keys
             other => panic!("Should get a NotFound error when downloading non-existing storage files, but got: {other:?}"),
         }
@@ -755,7 +775,7 @@ mod fs_tests {
     }
 
     async fn create_file_for_upload(
-        path: &Utf8Path,
+        path: &Path,
         contents: &str,
     ) -> anyhow::Result<(io::BufReader<fs::File>, usize)> {
         std::fs::create_dir_all(path.parent().unwrap())?;

libs/remote_storage/src/s3_bucket.rs

@@ -4,7 +4,7 @@
 //! allowing multiple api users to independently work with the same S3 bucket, if
 //! their bucket prefixes are both specified and different.
 
-use std::borrow::Cow;
+use std::sync::Arc;
 
 use anyhow::Context;
 use aws_config::{
@@ -24,20 +24,22 @@ use aws_sdk_s3::{
 use aws_smithy_http::body::SdkBody;
 use hyper::Body;
 use scopeguard::ScopeGuard;
-use tokio::io::{self, AsyncRead};
+use tokio::{
+    io::{self, AsyncRead},
+    sync::Semaphore,
+};
 use tokio_util::io::ReaderStream;
 use tracing::debug;
 
 use super::StorageMetadata;
 use crate::{
-    ConcurrencyLimiter, Download, DownloadError, RemotePath, RemoteStorage, S3Config,
-    MAX_KEYS_PER_DELETE, REMOTE_STORAGE_PREFIX_SEPARATOR,
+    Download, DownloadError, RemotePath, RemoteStorage, S3Config, MAX_KEYS_PER_DELETE,
+    REMOTE_STORAGE_PREFIX_SEPARATOR,
 };
 
 pub(super) mod metrics;
 
-use self::metrics::AttemptOutcome;
-pub(super) use self::metrics::RequestKind;
+use self::metrics::{AttemptOutcome, RequestKind};
 
 /// AWS S3 storage.
 pub struct S3Bucket {
@@ -45,7 +47,10 @@ pub struct S3Bucket {
     bucket_name: String,
     prefix_in_bucket: Option<String>,
     max_keys_per_list_response: Option<i32>,
-    concurrency_limiter: ConcurrencyLimiter,
+    // Every request to S3 can be throttled or cancelled, if a certain number of requests per second is exceeded.
+    // Same goes to IAM, which is queried before every S3 request, if enabled. IAM has even lower RPS threshold.
+    // The helps to ensure we don't exceed the thresholds.
+    concurrency_limiter: Arc<Semaphore>,
 }
 
 #[derive(Default)]
@@ -112,7 +117,7 @@ impl S3Bucket {
             bucket_name: aws_config.bucket_name.clone(),
             max_keys_per_list_response: aws_config.max_keys_per_list_response,
             prefix_in_bucket,
-            concurrency_limiter: ConcurrencyLimiter::new(aws_config.concurrency_limit.get()),
+            concurrency_limiter: Arc::new(Semaphore::new(aws_config.concurrency_limit.get())),
         })
     }
 
@@ -138,11 +143,12 @@ impl S3Bucket {
         assert_eq!(std::path::MAIN_SEPARATOR, REMOTE_STORAGE_PREFIX_SEPARATOR);
         let path_string = path
             .get_path()
-            .as_str()
-            .trim_end_matches(REMOTE_STORAGE_PREFIX_SEPARATOR);
+            .to_string_lossy()
+            .trim_end_matches(REMOTE_STORAGE_PREFIX_SEPARATOR)
+            .to_string();
         match &self.prefix_in_bucket {
-            Some(prefix) => prefix.clone() + "/" + path_string,
-            None => path_string.to_string(),
+            Some(prefix) => prefix.clone() + "/" + &path_string,
+            None => path_string,
         }
     }
 
@@ -150,7 +156,7 @@ impl S3Bucket {
         let started_at = start_counting_cancelled_wait(kind);
         let permit = self
             .concurrency_limiter
-            .acquire(kind)
+            .acquire()
             .await
             .expect("semaphore is never closed");
 
@@ -166,7 +172,8 @@ impl S3Bucket {
         let started_at = start_counting_cancelled_wait(kind);
         let permit = self
             .concurrency_limiter
-            .acquire_owned(kind)
+            .clone()
+            .acquire_owned()
             .await
             .expect("semaphore is never closed");
 
@@ -514,20 +521,6 @@ impl RemoteStorage for S3Bucket {
                         .deleted_objects_total
                         .inc_by(chunk.len() as u64);
                     if let Some(errors) = resp.errors {
-                        // Log a bounded number of the errors within the response:
-                        // these requests can carry 1000 keys so logging each one
-                        // would be too verbose, especially as errors may lead us
-                        // to retry repeatedly.
-                        const LOG_UP_TO_N_ERRORS: usize = 10;
-                        for e in errors.iter().take(LOG_UP_TO_N_ERRORS) {
-                            tracing::warn!(
-                                "DeleteObjects key {} failed: {}: {}",
-                                e.key.as_ref().map(Cow::from).unwrap_or("".into()),
-                                e.code.as_ref().map(Cow::from).unwrap_or("".into()),
-                                e.message.as_ref().map(Cow::from).unwrap_or("".into())
-                            );
-                        }
-
                         return Err(anyhow::format_err!(
                             "Failed to delete {} objects",
                             errors.len()
@@ -572,8 +565,8 @@ fn start_measuring_requests(
 
 #[cfg(test)]
 mod tests {
-    use camino::Utf8Path;
     use std::num::NonZeroUsize;
+    use std::path::Path;
 
     use crate::{RemotePath, S3Bucket, S3Config};
 
@@ -582,7 +575,7 @@ mod tests {
         let all_paths = ["", "some/path", "some/path/"];
         let all_paths: Vec<RemotePath> = all_paths
             .iter()
-            .map(|x| RemotePath::new(Utf8Path::new(x)).expect("bad path"))
+            .map(|x| RemotePath::new(Path::new(x)).expect("bad path"))
             .collect();
         let prefixes = [
             None,

libs/remote_storage/src/s3_bucket/metrics.rs

@@ -6,7 +6,7 @@ use once_cell::sync::Lazy;
 pub(super) static BUCKET_METRICS: Lazy<BucketMetrics> = Lazy::new(Default::default);
 
 #[derive(Clone, Copy, Debug)]
-pub(crate) enum RequestKind {
+pub(super) enum RequestKind {
     Get = 0,
     Put = 1,
     Delete = 2,

libs/remote_storage/tests/test_real_azure.rs

@@ -1,619 +0,0 @@
-use std::collections::HashSet;
-use std::env;
-use std::num::{NonZeroU32, NonZeroUsize};
-use std::ops::ControlFlow;
-use std::path::PathBuf;
-use std::sync::Arc;
-use std::time::UNIX_EPOCH;
-
-use anyhow::Context;
-use camino::Utf8Path;
-use once_cell::sync::OnceCell;
-use remote_storage::{
-    AzureConfig, Download, GenericRemoteStorage, RemotePath, RemoteStorageConfig, RemoteStorageKind,
-};
-use test_context::{test_context, AsyncTestContext};
-use tokio::task::JoinSet;
-use tracing::{debug, error, info};
-
-static LOGGING_DONE: OnceCell<()> = OnceCell::new();
-
-const ENABLE_REAL_AZURE_REMOTE_STORAGE_ENV_VAR_NAME: &str = "ENABLE_REAL_AZURE_REMOTE_STORAGE";
-
-const BASE_PREFIX: &str = "test";
-
-/// Tests that the Azure client can list all prefixes, even if the response comes paginated and requires multiple HTTP queries.
-/// Uses real Azure and requires [`ENABLE_REAL_AZURE_REMOTE_STORAGE_ENV_VAR_NAME`] and related Azure cred env vars specified.
-/// See the client creation in [`create_azure_client`] for details on the required env vars.
-/// If real Azure tests are disabled, the test passes, skipping any real test run: currently, there's no way to mark the test ignored in runtime with the
-/// deafult test framework, see https://github.com/rust-lang/rust/issues/68007 for details.
-///
-/// First, the test creates a set of Azure blobs with keys `/${random_prefix_part}/${base_prefix_str}/sub_prefix_${i}/blob_${i}` in [`upload_azure_data`]
-/// where
-/// * `random_prefix_part` is set for the entire Azure client during the Azure client creation in [`create_azure_client`], to avoid multiple test runs interference
-/// * `base_prefix_str` is a common prefix to use in the client requests: we would want to ensure that the client is able to list nested prefixes inside the bucket
-///
-/// Then, verifies that the client does return correct prefixes when queried:
-/// * with no prefix, it lists everything after its `${random_prefix_part}/` — that should be `${base_prefix_str}` value only
-/// * with `${base_prefix_str}/` prefix, it lists every `sub_prefix_${i}`
-///
-/// With the real Azure enabled and `#[cfg(test)]` Rust configuration used, the Azure client test adds a `max-keys` param to limit the response keys.
-/// This way, we are able to test the pagination implicitly, by ensuring all results are returned from the remote storage and avoid uploading too many blobs to Azure.
-///
-/// Lastly, the test attempts to clean up and remove all uploaded Azure files.
-/// If any errors appear during the clean up, they get logged, but the test is not failed or stopped until clean up is finished.
-#[test_context(MaybeEnabledAzureWithTestBlobs)]
-#[tokio::test]
-async fn azure_pagination_should_work(
-    ctx: &mut MaybeEnabledAzureWithTestBlobs,
-) -> anyhow::Result<()> {
-    let ctx = match ctx {
-        MaybeEnabledAzureWithTestBlobs::Enabled(ctx) => ctx,
-        MaybeEnabledAzureWithTestBlobs::Disabled => return Ok(()),
-        MaybeEnabledAzureWithTestBlobs::UploadsFailed(e, _) => {
-            anyhow::bail!("Azure init failed: {e:?}")
-        }
-    };
-
-    let test_client = Arc::clone(&ctx.enabled.client);
-    let expected_remote_prefixes = ctx.remote_prefixes.clone();
-
-    let base_prefix = RemotePath::new(Utf8Path::new(ctx.enabled.base_prefix))
-        .context("common_prefix construction")?;
-    let root_remote_prefixes = test_client
-        .list_prefixes(None)
-        .await
-        .context("client list root prefixes failure")?
-        .into_iter()
-        .collect::<HashSet<_>>();
-    assert_eq!(
-        root_remote_prefixes, HashSet::from([base_prefix.clone()]),
-        "remote storage root prefixes list mismatches with the uploads. Returned prefixes: {root_remote_prefixes:?}"
-    );
-
-    let nested_remote_prefixes = test_client
-        .list_prefixes(Some(&base_prefix))
-        .await
-        .context("client list nested prefixes failure")?
-        .into_iter()
-        .collect::<HashSet<_>>();
-    let remote_only_prefixes = nested_remote_prefixes
-        .difference(&expected_remote_prefixes)
-        .collect::<HashSet<_>>();
-    let missing_uploaded_prefixes = expected_remote_prefixes
-        .difference(&nested_remote_prefixes)
-        .collect::<HashSet<_>>();
-    assert_eq!(
-        remote_only_prefixes.len() + missing_uploaded_prefixes.len(), 0,
-        "remote storage nested prefixes list mismatches with the uploads. Remote only prefixes: {remote_only_prefixes:?}, missing uploaded prefixes: {missing_uploaded_prefixes:?}",
-    );
-
-    Ok(())
-}
-
-/// Tests that Azure client can list all files in a folder, even if the response comes paginated and requirees multiple Azure queries.
-/// Uses real Azure and requires [`ENABLE_REAL_AZURE_REMOTE_STORAGE_ENV_VAR_NAME`] and related Azure cred env vars specified. Test will skip real code and pass if env vars not set.
-/// See `Azure_pagination_should_work` for more information.
-///
-/// First, create a set of Azure objects with keys `random_prefix/folder{j}/blob_{i}.txt` in [`upload_azure_data`]
-/// Then performs the following queries:
-///    1. `list_files(None)`. This should return all files `random_prefix/folder{j}/blob_{i}.txt`
-///    2. `list_files("folder1")`.  This  should return all files `random_prefix/folder1/blob_{i}.txt`
-#[test_context(MaybeEnabledAzureWithSimpleTestBlobs)]
-#[tokio::test]
-async fn azure_list_files_works(
-    ctx: &mut MaybeEnabledAzureWithSimpleTestBlobs,
-) -> anyhow::Result<()> {
-    let ctx = match ctx {
-        MaybeEnabledAzureWithSimpleTestBlobs::Enabled(ctx) => ctx,
-        MaybeEnabledAzureWithSimpleTestBlobs::Disabled => return Ok(()),
-        MaybeEnabledAzureWithSimpleTestBlobs::UploadsFailed(e, _) => {
-            anyhow::bail!("Azure init failed: {e:?}")
-        }
-    };
-    let test_client = Arc::clone(&ctx.enabled.client);
-    let base_prefix =
-        RemotePath::new(Utf8Path::new("folder1")).context("common_prefix construction")?;
-    let root_files = test_client
-        .list_files(None)
-        .await
-        .context("client list root files failure")?
-        .into_iter()
-        .collect::<HashSet<_>>();
-    assert_eq!(
-        root_files,
-        ctx.remote_blobs.clone(),
-        "remote storage list_files on root mismatches with the uploads."
-    );
-    let nested_remote_files = test_client
-        .list_files(Some(&base_prefix))
-        .await
-        .context("client list nested files failure")?
-        .into_iter()
-        .collect::<HashSet<_>>();
-    let trim_remote_blobs: HashSet<_> = ctx
-        .remote_blobs
-        .iter()
-        .map(|x| x.get_path())
-        .filter(|x| x.starts_with("folder1"))
-        .map(|x| RemotePath::new(x).expect("must be valid path"))
-        .collect();
-    assert_eq!(
-        nested_remote_files, trim_remote_blobs,
-        "remote storage list_files on subdirrectory mismatches with the uploads."
-    );
-    Ok(())
-}
-
-#[test_context(MaybeEnabledAzure)]
-#[tokio::test]
-async fn azure_delete_non_exising_works(ctx: &mut MaybeEnabledAzure) -> anyhow::Result<()> {
-    let ctx = match ctx {
-        MaybeEnabledAzure::Enabled(ctx) => ctx,
-        MaybeEnabledAzure::Disabled => return Ok(()),
-    };
-
-    let path = RemotePath::new(Utf8Path::new(
-        format!("{}/for_sure_there_is_nothing_there_really", ctx.base_prefix).as_str(),
-    ))
-    .with_context(|| "RemotePath conversion")?;
-
-    ctx.client.delete(&path).await.expect("should succeed");
-
-    Ok(())
-}
-
-#[test_context(MaybeEnabledAzure)]
-#[tokio::test]
-async fn azure_delete_objects_works(ctx: &mut MaybeEnabledAzure) -> anyhow::Result<()> {
-    let ctx = match ctx {
-        MaybeEnabledAzure::Enabled(ctx) => ctx,
-        MaybeEnabledAzure::Disabled => return Ok(()),
-    };
-
-    let path1 = RemotePath::new(Utf8Path::new(format!("{}/path1", ctx.base_prefix).as_str()))
-        .with_context(|| "RemotePath conversion")?;
-
-    let path2 = RemotePath::new(Utf8Path::new(format!("{}/path2", ctx.base_prefix).as_str()))
-        .with_context(|| "RemotePath conversion")?;
-
-    let path3 = RemotePath::new(Utf8Path::new(format!("{}/path3", ctx.base_prefix).as_str()))
-        .with_context(|| "RemotePath conversion")?;
-
-    let data1 = "remote blob data1".as_bytes();
-    let data1_len = data1.len();
-    let data2 = "remote blob data2".as_bytes();
-    let data2_len = data2.len();
-    let data3 = "remote blob data3".as_bytes();
-    let data3_len = data3.len();
-    ctx.client
-        .upload(std::io::Cursor::new(data1), data1_len, &path1, None)
-        .await?;
-
-    ctx.client
-        .upload(std::io::Cursor::new(data2), data2_len, &path2, None)
-        .await?;
-
-    ctx.client
-        .upload(std::io::Cursor::new(data3), data3_len, &path3, None)
-        .await?;
-
-    ctx.client.delete_objects(&[path1, path2]).await?;
-
-    let prefixes = ctx.client.list_prefixes(None).await?;
-
-    assert_eq!(prefixes.len(), 1);
-
-    ctx.client.delete_objects(&[path3]).await?;
-
-    Ok(())
-}
-
-#[test_context(MaybeEnabledAzure)]
-#[tokio::test]
-async fn azure_upload_download_works(ctx: &mut MaybeEnabledAzure) -> anyhow::Result<()> {
-    let MaybeEnabledAzure::Enabled(ctx) = ctx else {
-        return Ok(());
-    };
-
-    let path = RemotePath::new(Utf8Path::new(format!("{}/file", ctx.base_prefix).as_str()))
-        .with_context(|| "RemotePath conversion")?;
-
-    let data = "remote blob data here".as_bytes();
-    let data_len = data.len() as u64;
-
-    ctx.client
-        .upload(std::io::Cursor::new(data), data.len(), &path, None)
-        .await?;
-
-    async fn download_and_compare(mut dl: Download) -> anyhow::Result<Vec<u8>> {
-        let mut buf = Vec::new();
-        tokio::io::copy(&mut dl.download_stream, &mut buf).await?;
-        Ok(buf)
-    }
-    // Normal download request
-    let dl = ctx.client.download(&path).await?;
-    let buf = download_and_compare(dl).await?;
-    assert_eq!(buf, data);
-
-    // Full range (end specified)
-    let dl = ctx
-        .client
-        .download_byte_range(&path, 0, Some(data_len))
-        .await?;
-    let buf = download_and_compare(dl).await?;
-    assert_eq!(buf, data);
-
-    // partial range (end specified)
-    let dl = ctx.client.download_byte_range(&path, 4, Some(10)).await?;
-    let buf = download_and_compare(dl).await?;
-    assert_eq!(buf, data[4..10]);
-
-    // partial range (end beyond real end)
-    let dl = ctx
-        .client
-        .download_byte_range(&path, 8, Some(data_len * 100))
-        .await?;
-    let buf = download_and_compare(dl).await?;
-    assert_eq!(buf, data[8..]);
-
-    // Partial range (end unspecified)
-    let dl = ctx.client.download_byte_range(&path, 4, None).await?;
-    let buf = download_and_compare(dl).await?;
-    assert_eq!(buf, data[4..]);
-
-    // Full range (end unspecified)
-    let dl = ctx.client.download_byte_range(&path, 0, None).await?;
-    let buf = download_and_compare(dl).await?;
-    assert_eq!(buf, data);
-
-    Ok(())
-}
-
-fn ensure_logging_ready() {
-    LOGGING_DONE.get_or_init(|| {
-        utils::logging::init(
-            utils::logging::LogFormat::Test,
-            utils::logging::TracingErrorLayerEnablement::Disabled,
-        )
-        .expect("logging init failed");
-    });
-}
-
-struct EnabledAzure {
-    client: Arc<GenericRemoteStorage>,
-    base_prefix: &'static str,
-}
-
-impl EnabledAzure {
-    async fn setup(max_keys_in_list_response: Option<i32>) -> Self {
-        let client = create_azure_client(max_keys_in_list_response)
-            .context("Azure client creation")
-            .expect("Azure client creation failed");
-
-        EnabledAzure {
-            client,
-            base_prefix: BASE_PREFIX,
-        }
-    }
-}
-
-enum MaybeEnabledAzure {
-    Enabled(EnabledAzure),
-    Disabled,
-}
-
-#[async_trait::async_trait]
-impl AsyncTestContext for MaybeEnabledAzure {
-    async fn setup() -> Self {
-        ensure_logging_ready();
-
-        if env::var(ENABLE_REAL_AZURE_REMOTE_STORAGE_ENV_VAR_NAME).is_err() {
-            info!(
-                "`{}` env variable is not set, skipping the test",
-                ENABLE_REAL_AZURE_REMOTE_STORAGE_ENV_VAR_NAME
-            );
-            return Self::Disabled;
-        }
-
-        Self::Enabled(EnabledAzure::setup(None).await)
-    }
-}
-
-enum MaybeEnabledAzureWithTestBlobs {
-    Enabled(AzureWithTestBlobs),
-    Disabled,
-    UploadsFailed(anyhow::Error, AzureWithTestBlobs),
-}
-
-struct AzureWithTestBlobs {
-    enabled: EnabledAzure,
-    remote_prefixes: HashSet<RemotePath>,
-    remote_blobs: HashSet<RemotePath>,
-}
-
-#[async_trait::async_trait]
-impl AsyncTestContext for MaybeEnabledAzureWithTestBlobs {
-    async fn setup() -> Self {
-        ensure_logging_ready();
-        if env::var(ENABLE_REAL_AZURE_REMOTE_STORAGE_ENV_VAR_NAME).is_err() {
-            info!(
-                "`{}` env variable is not set, skipping the test",
-                ENABLE_REAL_AZURE_REMOTE_STORAGE_ENV_VAR_NAME
-            );
-            return Self::Disabled;
-        }
-
-        let max_keys_in_list_response = 10;
-        let upload_tasks_count = 1 + (2 * usize::try_from(max_keys_in_list_response).unwrap());
-
-        let enabled = EnabledAzure::setup(Some(max_keys_in_list_response)).await;
-
-        match upload_azure_data(&enabled.client, enabled.base_prefix, upload_tasks_count).await {
-            ControlFlow::Continue(uploads) => {
-                info!("Remote objects created successfully");
-
-                Self::Enabled(AzureWithTestBlobs {
-                    enabled,
-                    remote_prefixes: uploads.prefixes,
-                    remote_blobs: uploads.blobs,
-                })
-            }
-            ControlFlow::Break(uploads) => Self::UploadsFailed(
-                anyhow::anyhow!("One or multiple blobs failed to upload to Azure"),
-                AzureWithTestBlobs {
-                    enabled,
-                    remote_prefixes: uploads.prefixes,
-                    remote_blobs: uploads.blobs,
-                },
-            ),
-        }
-    }
-
-    async fn teardown(self) {
-        match self {
-            Self::Disabled => {}
-            Self::Enabled(ctx) | Self::UploadsFailed(_, ctx) => {
-                cleanup(&ctx.enabled.client, ctx.remote_blobs).await;
-            }
-        }
-    }
-}
-
-// NOTE: the setups for the list_prefixes test and the list_files test are very similar
-// However, they are not idential. The list_prefixes function is concerned with listing prefixes,
-// whereas the list_files function is concerned with listing files.
-// See `RemoteStorage::list_files` documentation for more details
-enum MaybeEnabledAzureWithSimpleTestBlobs {
-    Enabled(AzureWithSimpleTestBlobs),
-    Disabled,
-    UploadsFailed(anyhow::Error, AzureWithSimpleTestBlobs),
-}
-struct AzureWithSimpleTestBlobs {
-    enabled: EnabledAzure,
-    remote_blobs: HashSet<RemotePath>,
-}
-
-#[async_trait::async_trait]
-impl AsyncTestContext for MaybeEnabledAzureWithSimpleTestBlobs {
-    async fn setup() -> Self {
-        ensure_logging_ready();
-        if env::var(ENABLE_REAL_AZURE_REMOTE_STORAGE_ENV_VAR_NAME).is_err() {
-            info!(
-                "`{}` env variable is not set, skipping the test",
-                ENABLE_REAL_AZURE_REMOTE_STORAGE_ENV_VAR_NAME
-            );
-            return Self::Disabled;
-        }
-
-        let max_keys_in_list_response = 10;
-        let upload_tasks_count = 1 + (2 * usize::try_from(max_keys_in_list_response).unwrap());
-
-        let enabled = EnabledAzure::setup(Some(max_keys_in_list_response)).await;
-
-        match upload_simple_azure_data(&enabled.client, upload_tasks_count).await {
-            ControlFlow::Continue(uploads) => {
-                info!("Remote objects created successfully");
-
-                Self::Enabled(AzureWithSimpleTestBlobs {
-                    enabled,
-                    remote_blobs: uploads,
-                })
-            }
-            ControlFlow::Break(uploads) => Self::UploadsFailed(
-                anyhow::anyhow!("One or multiple blobs failed to upload to Azure"),
-                AzureWithSimpleTestBlobs {
-                    enabled,
-                    remote_blobs: uploads,
-                },
-            ),
-        }
-    }
-
-    async fn teardown(self) {
-        match self {
-            Self::Disabled => {}
-            Self::Enabled(ctx) | Self::UploadsFailed(_, ctx) => {
-                cleanup(&ctx.enabled.client, ctx.remote_blobs).await;
-            }
-        }
-    }
-}
-
-fn create_azure_client(
-    max_keys_per_list_response: Option<i32>,
-) -> anyhow::Result<Arc<GenericRemoteStorage>> {
-    use rand::Rng;
-
-    let remote_storage_azure_container = env::var("REMOTE_STORAGE_AZURE_CONTAINER").context(
-        "`REMOTE_STORAGE_AZURE_CONTAINER` env var is not set, but real Azure tests are enabled",
-    )?;
-    let remote_storage_azure_region = env::var("REMOTE_STORAGE_AZURE_REGION").context(
-        "`REMOTE_STORAGE_AZURE_REGION` env var is not set, but real Azure tests are enabled",
-    )?;
-
-    // due to how time works, we've had test runners use the same nanos as bucket prefixes.
-    // millis is just a debugging aid for easier finding the prefix later.
-    let millis = std::time::SystemTime::now()
-        .duration_since(UNIX_EPOCH)
-        .context("random Azure test prefix part calculation")?
-        .as_millis();
-
-    // because nanos can be the same for two threads so can millis, add randomness
-    let random = rand::thread_rng().gen::<u32>();
-
-    let remote_storage_config = RemoteStorageConfig {
-        max_concurrent_syncs: NonZeroUsize::new(100).unwrap(),
-        max_sync_errors: NonZeroU32::new(5).unwrap(),
-        storage: RemoteStorageKind::AzureContainer(AzureConfig {
-            container_name: remote_storage_azure_container,
-            container_region: remote_storage_azure_region,
-            prefix_in_container: Some(format!("test_{millis}_{random:08x}/")),
-            concurrency_limit: NonZeroUsize::new(100).unwrap(),
-            max_keys_per_list_response,
-        }),
-    };
-    Ok(Arc::new(
-        GenericRemoteStorage::from_config(&remote_storage_config).context("remote storage init")?,
-    ))
-}
-
-struct Uploads {
-    prefixes: HashSet<RemotePath>,
-    blobs: HashSet<RemotePath>,
-}
-
-async fn upload_azure_data(
-    client: &Arc<GenericRemoteStorage>,
-    base_prefix_str: &'static str,
-    upload_tasks_count: usize,
-) -> ControlFlow<Uploads, Uploads> {
-    info!("Creating {upload_tasks_count} Azure files");
-    let mut upload_tasks = JoinSet::new();
-    for i in 1..upload_tasks_count + 1 {
-        let task_client = Arc::clone(client);
-        upload_tasks.spawn(async move {
-            let prefix = format!("{base_prefix_str}/sub_prefix_{i}/");
-            let blob_prefix = RemotePath::new(Utf8Path::new(&prefix))
-                .with_context(|| format!("{prefix:?} to RemotePath conversion"))?;
-            let blob_path = blob_prefix.join(Utf8Path::new(&format!("blob_{i}")));
-            debug!("Creating remote item {i} at path {blob_path:?}");
-
-            let data = format!("remote blob data {i}").into_bytes();
-            let data_len = data.len();
-            task_client
-                .upload(std::io::Cursor::new(data), data_len, &blob_path, None)
-                .await?;
-
-            Ok::<_, anyhow::Error>((blob_prefix, blob_path))
-        });
-    }
-
-    let mut upload_tasks_failed = false;
-    let mut uploaded_prefixes = HashSet::with_capacity(upload_tasks_count);
-    let mut uploaded_blobs = HashSet::with_capacity(upload_tasks_count);
-    while let Some(task_run_result) = upload_tasks.join_next().await {
-        match task_run_result
-            .context("task join failed")
-            .and_then(|task_result| task_result.context("upload task failed"))
-        {
-            Ok((upload_prefix, upload_path)) => {
-                uploaded_prefixes.insert(upload_prefix);
-                uploaded_blobs.insert(upload_path);
-            }
-            Err(e) => {
-                error!("Upload task failed: {e:?}");
-                upload_tasks_failed = true;
-            }
-        }
-    }
-
-    let uploads = Uploads {
-        prefixes: uploaded_prefixes,
-        blobs: uploaded_blobs,
-    };
-    if upload_tasks_failed {
-        ControlFlow::Break(uploads)
-    } else {
-        ControlFlow::Continue(uploads)
-    }
-}
-
-async fn cleanup(client: &Arc<GenericRemoteStorage>, objects_to_delete: HashSet<RemotePath>) {
-    info!(
-        "Removing {} objects from the remote storage during cleanup",
-        objects_to_delete.len()
-    );
-    let mut delete_tasks = JoinSet::new();
-    for object_to_delete in objects_to_delete {
-        let task_client = Arc::clone(client);
-        delete_tasks.spawn(async move {
-            debug!("Deleting remote item at path {object_to_delete:?}");
-            task_client
-                .delete(&object_to_delete)
-                .await
-                .with_context(|| format!("{object_to_delete:?} removal"))
-        });
-    }
-
-    while let Some(task_run_result) = delete_tasks.join_next().await {
-        match task_run_result {
-            Ok(task_result) => match task_result {
-                Ok(()) => {}
-                Err(e) => error!("Delete task failed: {e:?}"),
-            },
-            Err(join_err) => error!("Delete task did not finish correctly: {join_err}"),
-        }
-    }
-}
-
-// Uploads files `folder{j}/blob{i}.txt`. See test description for more details.
-async fn upload_simple_azure_data(
-    client: &Arc<GenericRemoteStorage>,
-    upload_tasks_count: usize,
-) -> ControlFlow<HashSet<RemotePath>, HashSet<RemotePath>> {
-    info!("Creating {upload_tasks_count} Azure files");
-    let mut upload_tasks = JoinSet::new();
-    for i in 1..upload_tasks_count + 1 {
-        let task_client = Arc::clone(client);
-        upload_tasks.spawn(async move {
-            let blob_path = PathBuf::from(format!("folder{}/blob_{}.txt", i / 7, i));
-            let blob_path = RemotePath::new(
-                Utf8Path::from_path(blob_path.as_path()).expect("must be valid blob path"),
-            )
-            .with_context(|| format!("{blob_path:?} to RemotePath conversion"))?;
-            debug!("Creating remote item {i} at path {blob_path:?}");
-
-            let data = format!("remote blob data {i}").into_bytes();
-            let data_len = data.len();
-            task_client
-                .upload(std::io::Cursor::new(data), data_len, &blob_path, None)
-                .await?;
-
-            Ok::<_, anyhow::Error>(blob_path)
-        });
-    }
-
-    let mut upload_tasks_failed = false;
-    let mut uploaded_blobs = HashSet::with_capacity(upload_tasks_count);
-    while let Some(task_run_result) = upload_tasks.join_next().await {
-        match task_run_result
-            .context("task join failed")
-            .and_then(|task_result| task_result.context("upload task failed"))
-        {
-            Ok(upload_path) => {
-                uploaded_blobs.insert(upload_path);
-            }
-            Err(e) => {
-                error!("Upload task failed: {e:?}");
-                upload_tasks_failed = true;
-            }
-        }
-    }
-
-    if upload_tasks_failed {
-        ControlFlow::Break(uploaded_blobs)
-    } else {
-        ControlFlow::Continue(uploaded_blobs)
-    }
-}

libs/remote_storage/tests/test_real_s3.rs

@@ -2,12 +2,11 @@ use std::collections::HashSet;
 use std::env;
 use std::num::{NonZeroU32, NonZeroUsize};
 use std::ops::ControlFlow;
-use std::path::PathBuf;
+use std::path::{Path, PathBuf};
 use std::sync::Arc;
 use std::time::UNIX_EPOCH;
 
 use anyhow::Context;
-use camino::Utf8Path;
 use once_cell::sync::OnceCell;
 use remote_storage::{
     GenericRemoteStorage, RemotePath, RemoteStorageConfig, RemoteStorageKind, S3Config,
@@ -56,7 +55,7 @@ async fn s3_pagination_should_work(ctx: &mut MaybeEnabledS3WithTestBlobs) -> any
     let test_client = Arc::clone(&ctx.enabled.client);
     let expected_remote_prefixes = ctx.remote_prefixes.clone();
 
-    let base_prefix = RemotePath::new(Utf8Path::new(ctx.enabled.base_prefix))
+    let base_prefix = RemotePath::new(Path::new(ctx.enabled.base_prefix))
         .context("common_prefix construction")?;
     let root_remote_prefixes = test_client
         .list_prefixes(None)
@@ -109,7 +108,7 @@ async fn s3_list_files_works(ctx: &mut MaybeEnabledS3WithSimpleTestBlobs) -> any
     };
     let test_client = Arc::clone(&ctx.enabled.client);
     let base_prefix =
-        RemotePath::new(Utf8Path::new("folder1")).context("common_prefix construction")?;
+        RemotePath::new(Path::new("folder1")).context("common_prefix construction")?;
     let root_files = test_client
         .list_files(None)
         .await
@@ -130,9 +129,9 @@ async fn s3_list_files_works(ctx: &mut MaybeEnabledS3WithSimpleTestBlobs) -> any
     let trim_remote_blobs: HashSet<_> = ctx
         .remote_blobs
         .iter()
-        .map(|x| x.get_path())
+        .map(|x| x.get_path().to_str().expect("must be valid name"))
         .filter(|x| x.starts_with("folder1"))
-        .map(|x| RemotePath::new(x).expect("must be valid path"))
+        .map(|x| RemotePath::new(Path::new(x)).expect("must be valid name"))
         .collect();
     assert_eq!(
         nested_remote_files, trim_remote_blobs,
@@ -149,9 +148,10 @@ async fn s3_delete_non_exising_works(ctx: &mut MaybeEnabledS3) -> anyhow::Result
         MaybeEnabledS3::Disabled => return Ok(()),
     };
 
-    let path = RemotePath::new(Utf8Path::new(
-        format!("{}/for_sure_there_is_nothing_there_really", ctx.base_prefix).as_str(),
-    ))
+    let path = RemotePath::new(&PathBuf::from(format!(
+        "{}/for_sure_there_is_nothing_there_really",
+        ctx.base_prefix,
+    )))
     .with_context(|| "RemotePath conversion")?;
 
     ctx.client.delete(&path).await.expect("should succeed");
@@ -167,13 +167,13 @@ async fn s3_delete_objects_works(ctx: &mut MaybeEnabledS3) -> anyhow::Result<()>
         MaybeEnabledS3::Disabled => return Ok(()),
     };
 
-    let path1 = RemotePath::new(Utf8Path::new(format!("{}/path1", ctx.base_prefix).as_str()))
+    let path1 = RemotePath::new(&PathBuf::from(format!("{}/path1", ctx.base_prefix,)))
         .with_context(|| "RemotePath conversion")?;
 
-    let path2 = RemotePath::new(Utf8Path::new(format!("{}/path2", ctx.base_prefix).as_str()))
+    let path2 = RemotePath::new(&PathBuf::from(format!("{}/path2", ctx.base_prefix,)))
         .with_context(|| "RemotePath conversion")?;
 
-    let path3 = RemotePath::new(Utf8Path::new(format!("{}/path3", ctx.base_prefix).as_str()))
+    let path3 = RemotePath::new(&PathBuf::from(format!("{}/path3", ctx.base_prefix,)))
         .with_context(|| "RemotePath conversion")?;
 
     let data1 = "remote blob data1".as_bytes();
@@ -427,10 +427,10 @@ async fn upload_s3_data(
     for i in 1..upload_tasks_count + 1 {
         let task_client = Arc::clone(client);
         upload_tasks.spawn(async move {
-            let prefix = format!("{base_prefix_str}/sub_prefix_{i}/");
-            let blob_prefix = RemotePath::new(Utf8Path::new(&prefix))
+            let prefix = PathBuf::from(format!("{base_prefix_str}/sub_prefix_{i}/"));
+            let blob_prefix = RemotePath::new(&prefix)
                 .with_context(|| format!("{prefix:?} to RemotePath conversion"))?;
-            let blob_path = blob_prefix.join(Utf8Path::new(&format!("blob_{i}")));
+            let blob_path = blob_prefix.join(Path::new(&format!("blob_{i}")));
             debug!("Creating remote item {i} at path {blob_path:?}");
 
             let data = format!("remote blob data {i}").into_bytes();
@@ -512,10 +512,8 @@ async fn upload_simple_s3_data(
         let task_client = Arc::clone(client);
         upload_tasks.spawn(async move {
             let blob_path = PathBuf::from(format!("folder{}/blob_{}.txt", i / 7, i));
-            let blob_path = RemotePath::new(
-                Utf8Path::from_path(blob_path.as_path()).expect("must be valid blob path"),
-            )
-            .with_context(|| format!("{blob_path:?} to RemotePath conversion"))?;
+            let blob_path = RemotePath::new(&blob_path)
+                .with_context(|| format!("{blob_path:?} to RemotePath conversion"))?;
             debug!("Creating remote item {i} at path {blob_path:?}");
 
             let data = format!("remote blob data {i}").into_bytes();

libs/utils/Cargo.toml

@@ -10,7 +10,6 @@ async-trait.workspace = true
 anyhow.workspace = true
 bincode.workspace = true
 bytes.workspace = true
-camino.workspace = true
 chrono.workspace = true
 heapless.workspace = true
 hex = { workspace = true, features = ["serde"] }
@@ -54,7 +53,7 @@ byteorder.workspace = true
 bytes.workspace = true
 criterion.workspace = true
 hex-literal.workspace = true
-camino-tempfile.workspace = true
+tempfile.workspace = true
 
 [[bench]]
 name = "benchmarks"

libs/utils/src/auth.rs

@@ -2,9 +2,9 @@
 
 use serde;
 use std::fs;
+use std::path::Path;
 
 use anyhow::Result;
-use camino::Utf8Path;
 use jsonwebtoken::{
     decode, encode, Algorithm, DecodingKey, EncodingKey, Header, TokenData, Validation,
 };
@@ -65,7 +65,7 @@ impl JwtAuth {
         }
     }
 
-    pub fn from_key_path(key_path: &Utf8Path) -> Result<Self> {
+    pub fn from_key_path(key_path: &Path) -> Result<Self> {
         let public_key = fs::read(key_path)?;
         Ok(Self::new(DecodingKey::from_ed_pem(&public_key)?))
     }

libs/utils/src/crashsafe.rs

@@ -1,14 +1,14 @@
 use std::{
     borrow::Cow,
+    ffi::OsStr,
     fs::{self, File},
     io,
+    path::{Path, PathBuf},
 };
 
-use camino::{Utf8Path, Utf8PathBuf};
-
 /// Similar to [`std::fs::create_dir`], except we fsync the
 /// created directory and its parent.
-pub fn create_dir(path: impl AsRef<Utf8Path>) -> io::Result<()> {
+pub fn create_dir(path: impl AsRef<Path>) -> io::Result<()> {
     let path = path.as_ref();
 
     fs::create_dir(path)?;
@@ -18,7 +18,7 @@ pub fn create_dir(path: impl AsRef<Utf8Path>) -> io::Result<()> {
 
 /// Similar to [`std::fs::create_dir_all`], except we fsync all
 /// newly created directories and the pre-existing parent.
-pub fn create_dir_all(path: impl AsRef<Utf8Path>) -> io::Result<()> {
+pub fn create_dir_all(path: impl AsRef<Path>) -> io::Result<()> {
     let mut path = path.as_ref();
 
     let mut dirs_to_create = Vec::new();
@@ -30,7 +30,7 @@ pub fn create_dir_all(path: impl AsRef<Utf8Path>) -> io::Result<()> {
             Ok(_) => {
                 return Err(io::Error::new(
                     io::ErrorKind::AlreadyExists,
-                    format!("non-directory found in path: {path}"),
+                    format!("non-directory found in path: {}", path.display()),
                 ));
             }
             Err(ref e) if e.kind() == io::ErrorKind::NotFound => {}
@@ -44,7 +44,7 @@ pub fn create_dir_all(path: impl AsRef<Utf8Path>) -> io::Result<()> {
             None => {
                 return Err(io::Error::new(
                     io::ErrorKind::InvalidInput,
-                    format!("can't find parent of path '{path}'"),
+                    format!("can't find parent of path '{}'", path.display()).as_str(),
                 ));
             }
         }
@@ -70,18 +70,21 @@ pub fn create_dir_all(path: impl AsRef<Utf8Path>) -> io::Result<()> {
 
 /// Adds a suffix to the file(directory) name, either appending the suffix to the end of its extension,
 /// or if there's no extension, creates one and puts a suffix there.
-pub fn path_with_suffix_extension(
-    original_path: impl AsRef<Utf8Path>,
-    suffix: &str,
-) -> Utf8PathBuf {
-    let new_extension = match original_path.as_ref().extension() {
+pub fn path_with_suffix_extension(original_path: impl AsRef<Path>, suffix: &str) -> PathBuf {
+    let new_extension = match original_path
+        .as_ref()
+        .extension()
+        .map(OsStr::to_string_lossy)
+    {
         Some(extension) => Cow::Owned(format!("{extension}.{suffix}")),
         None => Cow::Borrowed(suffix),
     };
-    original_path.as_ref().with_extension(new_extension)
+    original_path
+        .as_ref()
+        .with_extension(new_extension.as_ref())
 }
 
-pub fn fsync_file_and_parent(file_path: &Utf8Path) -> io::Result<()> {
+pub fn fsync_file_and_parent(file_path: &Path) -> io::Result<()> {
     let parent = file_path.parent().ok_or_else(|| {
         io::Error::new(
             io::ErrorKind::Other,
@@ -94,7 +97,7 @@ pub fn fsync_file_and_parent(file_path: &Utf8Path) -> io::Result<()> {
     Ok(())
 }
 
-pub fn fsync(path: &Utf8Path) -> io::Result<()> {
+pub fn fsync(path: &Path) -> io::Result<()> {
     File::open(path)
         .map_err(|e| io::Error::new(e.kind(), format!("Failed to open the file {path:?}: {e}")))
         .and_then(|file| {
@@ -108,18 +111,19 @@ pub fn fsync(path: &Utf8Path) -> io::Result<()> {
         .map_err(|e| io::Error::new(e.kind(), format!("Failed to fsync file {path:?}: {e}")))
 }
 
-pub async fn fsync_async(path: impl AsRef<Utf8Path>) -> Result<(), std::io::Error> {
-    tokio::fs::File::open(path.as_ref()).await?.sync_all().await
+pub async fn fsync_async(path: impl AsRef<std::path::Path>) -> Result<(), std::io::Error> {
+    tokio::fs::File::open(path).await?.sync_all().await
 }
 
 #[cfg(test)]
 mod tests {
+    use tempfile::tempdir;
 
     use super::*;
 
     #[test]
     fn test_create_dir_fsyncd() {
-        let dir = camino_tempfile::tempdir().unwrap();
+        let dir = tempdir().unwrap();
 
         let existing_dir_path = dir.path();
         let err = create_dir(existing_dir_path).unwrap_err();
@@ -135,7 +139,7 @@ mod tests {
 
     #[test]
     fn test_create_dir_all_fsyncd() {
-        let dir = camino_tempfile::tempdir().unwrap();
+        let dir = tempdir().unwrap();
 
         let existing_dir_path = dir.path();
         create_dir_all(existing_dir_path).unwrap();
@@ -162,29 +166,29 @@ mod tests {
 
     #[test]
     fn test_path_with_suffix_extension() {
-        let p = Utf8PathBuf::from("/foo/bar");
+        let p = PathBuf::from("/foo/bar");
         assert_eq!(
-            &path_with_suffix_extension(p, "temp").to_string(),
+            &path_with_suffix_extension(p, "temp").to_string_lossy(),
             "/foo/bar.temp"
         );
-        let p = Utf8PathBuf::from("/foo/bar");
+        let p = PathBuf::from("/foo/bar");
         assert_eq!(
-            &path_with_suffix_extension(p, "temp.temp").to_string(),
+            &path_with_suffix_extension(p, "temp.temp").to_string_lossy(),
             "/foo/bar.temp.temp"
         );
-        let p = Utf8PathBuf::from("/foo/bar.baz");
+        let p = PathBuf::from("/foo/bar.baz");
         assert_eq!(
-            &path_with_suffix_extension(p, "temp.temp").to_string(),
+            &path_with_suffix_extension(p, "temp.temp").to_string_lossy(),
             "/foo/bar.baz.temp.temp"
         );
-        let p = Utf8PathBuf::from("/foo/bar.baz");
+        let p = PathBuf::from("/foo/bar.baz");
         assert_eq!(
-            &path_with_suffix_extension(p, ".temp").to_string(),
+            &path_with_suffix_extension(p, ".temp").to_string_lossy(),
             "/foo/bar.baz..temp"
         );
-        let p = Utf8PathBuf::from("/foo/bar/dir/");
+        let p = PathBuf::from("/foo/bar/dir/");
         assert_eq!(
-            &path_with_suffix_extension(p, ".temp").to_string(),
+            &path_with_suffix_extension(p, ".temp").to_string_lossy(),
             "/foo/bar/dir..temp"
         );
     }

libs/utils/src/fs_ext.rs

@@ -55,6 +55,8 @@ where
 
 #[cfg(test)]
 mod test {
+    use std::path::PathBuf;
+
     use crate::fs_ext::{is_directory_empty, list_dir};
 
     use super::ignore_absent_files;
@@ -63,7 +65,7 @@ mod test {
     fn is_empty_dir() {
         use super::PathExt;
 
-        let dir = camino_tempfile::tempdir().unwrap();
+        let dir = tempfile::tempdir().unwrap();
         let dir_path = dir.path();
 
         // test positive case
@@ -73,7 +75,7 @@ mod test {
         );
 
         // invoke on a file to ensure it returns an error
-        let file_path = dir_path.join("testfile");
+        let file_path: PathBuf = dir_path.join("testfile");
         let f = std::fs::File::create(&file_path).unwrap();
         drop(f);
         assert!(file_path.is_empty_dir().is_err());
@@ -85,7 +87,7 @@ mod test {
 
     #[tokio::test]
     async fn is_empty_dir_async() {
-        let dir = camino_tempfile::tempdir().unwrap();
+        let dir = tempfile::tempdir().unwrap();
         let dir_path = dir.path();
 
         // test positive case
@@ -95,7 +97,7 @@ mod test {
         );
 
         // invoke on a file to ensure it returns an error
-        let file_path = dir_path.join("testfile");
+        let file_path: PathBuf = dir_path.join("testfile");
         let f = std::fs::File::create(&file_path).unwrap();
         drop(f);
         assert!(is_directory_empty(&file_path).await.is_err());
@@ -107,9 +109,10 @@ mod test {
 
     #[test]
     fn ignore_absent_files_works() {
-        let dir = camino_tempfile::tempdir().unwrap();
+        let dir = tempfile::tempdir().unwrap();
+        let dir_path = dir.path();
 
-        let file_path = dir.path().join("testfile");
+        let file_path: PathBuf = dir_path.join("testfile");
 
         ignore_absent_files(|| std::fs::remove_file(&file_path)).expect("should execute normally");
 
@@ -123,17 +126,17 @@ mod test {
 
     #[tokio::test]
     async fn list_dir_works() {
-        let dir = camino_tempfile::tempdir().unwrap();
+        let dir = tempfile::tempdir().unwrap();
         let dir_path = dir.path();
 
         assert!(list_dir(dir_path).await.unwrap().is_empty());
 
-        let file_path = dir_path.join("testfile");
+        let file_path: PathBuf = dir_path.join("testfile");
         let _ = std::fs::File::create(&file_path).unwrap();
 
         assert_eq!(&list_dir(dir_path).await.unwrap(), &["testfile"]);
 
-        let another_dir_path = dir_path.join("testdir");
+        let another_dir_path: PathBuf = dir_path.join("testdir");
         std::fs::create_dir(another_dir_path).unwrap();
 
         let expected = &["testdir", "testfile"];

libs/utils/src/http/error.rs

@@ -1,9 +1,8 @@
 use hyper::{header, Body, Response, StatusCode};
 use serde::{Deserialize, Serialize};
-use std::borrow::Cow;
 use std::error::Error as StdError;
 use thiserror::Error;
-use tracing::{error, info};
+use tracing::error;
 
 #[derive(Debug, Error)]
 pub enum ApiError {
@@ -25,9 +24,6 @@ pub enum ApiError {
     #[error("Precondition failed: {0}")]
     PreconditionFailed(Box<str>),
 
-    #[error("Resource temporarily unavailable: {0}")]
-    ResourceUnavailable(Cow<'static, str>),
-
     #[error("Shutting down")]
     ShuttingDown,
 
@@ -63,10 +59,6 @@ impl ApiError {
                 "Shutting down".to_string(),
                 StatusCode::SERVICE_UNAVAILABLE,
             ),
-            ApiError::ResourceUnavailable(err) => HttpErrorBody::response_from_msg_and_status(
-                err.to_string(),
-                StatusCode::SERVICE_UNAVAILABLE,
-            ),
             ApiError::InternalServerError(err) => HttpErrorBody::response_from_msg_and_status(
                 err.to_string(),
                 StatusCode::INTERNAL_SERVER_ERROR,
@@ -116,12 +108,10 @@ pub async fn route_error_handler(err: routerify::RouteError) -> Response<Body> {
 
 pub fn api_error_handler(api_error: ApiError) -> Response<Body> {
     // Print a stack trace for Internal Server errors
-
-    match api_error {
-        ApiError::ResourceUnavailable(_) => info!("Error processing HTTP request: {api_error:#}"),
-        ApiError::NotFound(_) => info!("Error processing HTTP request: {api_error:#}"),
-        ApiError::InternalServerError(_) => error!("Error processing HTTP request: {api_error:?}"),
-        _ => error!("Error processing HTTP request: {api_error:#}"),
+    if let ApiError::InternalServerError(_) = api_error {
+        error!("Error processing HTTP request: {api_error:?}");
+    } else {
+        error!("Error processing HTTP request: {api_error:#}");
     }
 
     api_error.into_response()

libs/utils/src/id.rs

@@ -1,3 +1,4 @@
+use std::ffi::OsStr;
 use std::{fmt, str::FromStr};
 
 use anyhow::Context;
@@ -214,11 +215,12 @@ pub struct TimelineId(Id);
 
 id_newtype!(TimelineId);
 
-impl TryFrom<Option<&str>> for TimelineId {
+impl TryFrom<Option<&OsStr>> for TimelineId {
     type Error = anyhow::Error;
 
-    fn try_from(value: Option<&str>) -> Result<Self, Self::Error> {
+    fn try_from(value: Option<&OsStr>) -> Result<Self, Self::Error> {
         value
+            .and_then(OsStr::to_str)
             .unwrap_or_default()
             .parse::<TimelineId>()
             .with_context(|| format!("Could not parse timeline id from {:?}", value))

libs/utils/src/lock_file.rs

@@ -11,10 +11,10 @@ use std::{
     io::{Read, Write},
     ops::Deref,
     os::unix::prelude::AsRawFd,
+    path::{Path, PathBuf},
 };
 
 use anyhow::Context;
-use camino::{Utf8Path, Utf8PathBuf};
 use nix::{errno::Errno::EAGAIN, fcntl};
 
 use crate::crashsafe;
@@ -23,7 +23,7 @@ use crate::crashsafe;
 /// Returned by [`create_exclusive`].
 #[must_use]
 pub struct UnwrittenLockFile {
-    path: Utf8PathBuf,
+    path: PathBuf,
     file: fs::File,
 }
 
@@ -60,7 +60,7 @@ impl UnwrittenLockFile {
 ///
 /// It is not an error if the file already exists.
 /// It is an error if the file is already locked.
-pub fn create_exclusive(lock_file_path: &Utf8Path) -> anyhow::Result<UnwrittenLockFile> {
+pub fn create_exclusive(lock_file_path: &Path) -> anyhow::Result<UnwrittenLockFile> {
     let lock_file = fs::OpenOptions::new()
         .create(true) // O_CREAT
         .write(true)
@@ -101,7 +101,7 @@ pub enum LockFileRead {
 /// Open & try to lock the lock file at the given `path`, returning a [handle][`LockFileRead`] to
 /// inspect its content. It is not an `Err(...)` if the file does not exist or is already locked.
 /// Check the [`LockFileRead`] variants for details.
-pub fn read_and_hold_lock_file(path: &Utf8Path) -> anyhow::Result<LockFileRead> {
+pub fn read_and_hold_lock_file(path: &Path) -> anyhow::Result<LockFileRead> {
     let res = fs::OpenOptions::new().read(true).open(path);
     let mut lock_file = match res {
         Ok(f) => f,

libs/utils/src/logging.rs

@@ -228,12 +228,6 @@ impl SecretString {
     }
 }
 
-impl From<String> for SecretString {
-    fn from(s: String) -> Self {
-        Self(s)
-    }
-}
-
 impl std::fmt::Debug for SecretString {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
         write!(f, "[SECRET]")

libs/utils/src/lsn.rs

@@ -1,9 +1,9 @@
 #![warn(missing_docs)]
 
-use camino::Utf8Path;
 use serde::{Deserialize, Serialize};
 use std::fmt;
 use std::ops::{Add, AddAssign};
+use std::path::Path;
 use std::str::FromStr;
 use std::sync::atomic::{AtomicU64, Ordering};
 
@@ -44,9 +44,11 @@ impl Lsn {
     /// Parse an LSN from a filename in the form `0000000000000000`
     pub fn from_filename<F>(filename: F) -> Result<Self, LsnParseError>
     where
-        F: AsRef<Utf8Path>,
+        F: AsRef<Path>,
     {
-        Lsn::from_hex(filename.as_ref().as_str())
+        let filename: &Path = filename.as_ref();
+        let filename = filename.to_str().ok_or(LsnParseError)?;
+        Lsn::from_hex(filename)
     }
 
     /// Parse an LSN from a string in the form `0000000000000000`

libs/utils/src/pid_file.rs

@@ -49,10 +49,9 @@
 //! At this point, `B` and `C` are running, which is hazardous.
 //! Morale of the story: don't unlink pidfiles, ever.
 
-use std::ops::Deref;
+use std::{ops::Deref, path::Path};
 
 use anyhow::Context;
-use camino::Utf8Path;
 use nix::unistd::Pid;
 
 use crate::lock_file::{self, LockFileRead};
@@ -85,7 +84,7 @@ impl Deref for PidFileGuard {
 /// The claim ends as soon as the returned guard object is dropped.
 /// To maintain the claim for the remaining lifetime of the current process,
 /// use [`std::mem::forget`] or similar.
-pub fn claim_for_current_process(path: &Utf8Path) -> anyhow::Result<PidFileGuard> {
+pub fn claim_for_current_process(path: &Path) -> anyhow::Result<PidFileGuard> {
     let unwritten_lock_file = lock_file::create_exclusive(path).context("lock file")?;
     // if any of the next steps fail, we drop the file descriptor and thereby release the lock
     let guard = unwritten_lock_file
@@ -133,7 +132,7 @@ pub enum PidFileRead {
 ///
 /// On success, this function returns a [`PidFileRead`].
 /// Check its docs for a description of the meaning of its different variants.
-pub fn read(pidfile: &Utf8Path) -> anyhow::Result<PidFileRead> {
+pub fn read(pidfile: &Path) -> anyhow::Result<PidFileRead> {
     let res = lock_file::read_and_hold_lock_file(pidfile).context("read and hold pid file")?;
     let ret = match res {
         LockFileRead::NotExist => PidFileRead::NotExist,

libs/utils/src/seqwait.rs

@@ -58,7 +58,7 @@ where
 // to get that.
 impl<T: Ord> PartialOrd for Waiter<T> {
     fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
-        Some(self.cmp(other))
+        other.wake_num.partial_cmp(&self.wake_num)
     }
 }
 

libs/vm_monitor/src/runner.rs

@@ -4,9 +4,9 @@
 //! This is the "Monitor" part of the monitor binary and is the main entrypoint for
 //! all functionality.
 
-use std::fmt::Debug;
 use std::sync::Arc;
 use std::time::{Duration, Instant};
+use std::{fmt::Debug, mem};
 
 use anyhow::{bail, Context};
 use axum::extract::ws::{Message, WebSocket};
@@ -141,6 +141,14 @@ impl Runner {
             );
 
             state.cgroup = Some(cgroup);
+        } else {
+            // *NOTE*: We need to forget the sender so that its drop impl does not get ran.
+            // This allows us to poll it in `Monitor::run` regardless of whether we
+            // are managing a cgroup or not. If we don't forget it, all receives will
+            // immediately return an error because the sender is droped and it will
+            // claim all select! statements, effectively turning `Monitor::run` into
+            // `loop { fail to receive }`.
+            mem::forget(requesting_send);
         }
 
         let mut file_cache_reserved_bytes = 0;
@@ -409,7 +417,7 @@ impl Runner {
                     }
                 }
                 // we need to propagate an upscale request
-                request = self.dispatcher.request_upscale_events.recv(), if self.cgroup.is_some() => {
+                request = self.dispatcher.request_upscale_events.recv() => {
                     if request.is_none() {
                         bail!("failed to listen for upscale event from cgroup")
                     }

pageserver/Cargo.toml

@@ -17,8 +17,6 @@ async-stream.workspace = true
 async-trait.workspace = true
 byteorder.workspace = true
 bytes.workspace = true
-camino.workspace = true
-camino-tempfile.workspace = true
 chrono = { workspace = true, features = ["serde"] }
 clap = { workspace = true, features = ["string"] }
 close_fds.workspace = true
@@ -82,6 +80,8 @@ enum-map.workspace = true
 enumset.workspace = true
 strum.workspace = true
 strum_macros.workspace = true
+tempfile.workspace = true
+async-channel = "1.9.0"
 
 [dev-dependencies]
 criterion.workspace = true

pageserver/benches/bench_walredo.rs

@@ -11,7 +11,10 @@ use std::sync::{Arc, Barrier};
 
 use bytes::{Buf, Bytes};
 use pageserver::{
-    config::PageServerConf, repository::Key, walrecord::NeonWalRecord, walredo::PostgresRedoManager,
+    config::PageServerConf,
+    repository::Key,
+    walrecord::NeonWalRecord,
+    walredo::{PostgresRedoManager, WalRedoError},
 };
 use utils::{id::TenantId, lsn::Lsn};
 
@@ -22,7 +25,7 @@ fn redo_scenarios(c: &mut Criterion) {
     // input to the stderr.
     // utils::logging::init(utils::logging::LogFormat::Plain).unwrap();
 
-    let repo_dir = camino_tempfile::tempdir_in(env!("CARGO_TARGET_TMPDIR")).unwrap();
+    let repo_dir = tempfile::tempdir_in(env!("CARGO_TARGET_TMPDIR")).unwrap();
 
     let conf = PageServerConf::dummy_conf(repo_dir.path().to_path_buf());
     let conf = Box::leak(Box::new(conf));
@@ -149,7 +152,7 @@ impl Drop for JoinOnDrop {
     }
 }
 
-fn execute_all<I>(input: I, manager: &PostgresRedoManager) -> anyhow::Result<()>
+fn execute_all<I>(input: I, manager: &PostgresRedoManager) -> Result<(), WalRedoError>
 where
     I: IntoIterator<Item = Request>,
 {
@@ -157,7 +160,7 @@ where
     input.into_iter().try_for_each(|req| {
         let page = req.execute(manager)?;
         assert_eq!(page.remaining(), 8192);
-        anyhow::Ok(())
+        Ok::<_, WalRedoError>(())
     })
 }
 
@@ -470,7 +473,7 @@ struct Request {
 }
 
 impl Request {
-    fn execute(self, manager: &PostgresRedoManager) -> anyhow::Result<Bytes> {
+    fn execute(self, manager: &PostgresRedoManager) -> Result<Bytes, WalRedoError> {
         use pageserver::walredo::WalRedoManager;
 
         let Request {

pageserver/ctl/Cargo.toml

@@ -9,7 +9,6 @@ license.workspace = true
 [dependencies]
 anyhow.workspace = true
 bytes.workspace = true
-camino.workspace = true
 clap = { workspace = true, features = ["string"] }
 git-version.workspace = true
 pageserver = { path = ".." }

pageserver/ctl/src/layer_map_analyzer.rs

@@ -3,14 +3,13 @@
 //! Currently it only analyzes holes, which are regions within the layer range that the layer contains no updates for. In the future it might do more analysis (maybe key quantiles?) but it should never return sensitive data.
 
 use anyhow::Result;
-use camino::{Utf8Path, Utf8PathBuf};
 use pageserver::context::{DownloadBehavior, RequestContext};
 use pageserver::task_mgr::TaskKind;
 use pageserver::tenant::{TENANTS_SEGMENT_NAME, TIMELINES_SEGMENT_NAME};
 use std::cmp::Ordering;
 use std::collections::BinaryHeap;
 use std::ops::Range;
-use std::{fs, str};
+use std::{fs, path::Path, str};
 
 use pageserver::page_cache::PAGE_SZ;
 use pageserver::repository::{Key, KEY_SIZE};
@@ -99,7 +98,7 @@ pub(crate) fn parse_filename(name: &str) -> Option<LayerFile> {
 }
 
 // Finds the max_holes largest holes, ignoring any that are smaller than MIN_HOLE_LENGTH"
-async fn get_holes(path: &Utf8Path, max_holes: usize, ctx: &RequestContext) -> Result<Vec<Hole>> {
+async fn get_holes(path: &Path, max_holes: usize, ctx: &RequestContext) -> Result<Vec<Hole>> {
     let file = FileBlockReader::new(VirtualFile::open(path).await?);
     let summary_blk = file.read_blk(0, ctx).await?;
     let actual_summary = Summary::des_prefix(summary_blk.as_ref())?;
@@ -168,9 +167,7 @@ pub(crate) async fn main(cmd: &AnalyzeLayerMapCmd) -> Result<()> {
                     parse_filename(&layer.file_name().into_string().unwrap())
                 {
                     if layer_file.is_delta {
-                        let layer_path =
-                            Utf8PathBuf::from_path_buf(layer.path()).expect("non-Unicode path");
-                        layer_file.holes = get_holes(&layer_path, max_holes, &ctx).await?;
+                        layer_file.holes = get_holes(&layer.path(), max_holes, &ctx).await?;
                         n_deltas += 1;
                     }
                     layers.push(layer_file);

pageserver/ctl/src/layers.rs

@@ -1,7 +1,6 @@
 use std::path::{Path, PathBuf};
 
 use anyhow::Result;
-use camino::Utf8Path;
 use clap::Subcommand;
 use pageserver::context::{DownloadBehavior, RequestContext};
 use pageserver::task_mgr::TaskKind;
@@ -48,7 +47,7 @@ pub(crate) enum LayerCmd {
 }
 
 async fn read_delta_file(path: impl AsRef<Path>, ctx: &RequestContext) -> Result<()> {
-    let path = Utf8Path::from_path(path.as_ref()).expect("non-Unicode path");
+    let path = path.as_ref();
     virtual_file::init(10);
     page_cache::init(100);
     let file = FileBlockReader::new(VirtualFile::open(path).await?);

pageserver/ctl/src/main.rs

@@ -8,7 +8,6 @@ mod draw_timeline_dir;
 mod layer_map_analyzer;
 mod layers;
 
-use camino::{Utf8Path, Utf8PathBuf};
 use clap::{Parser, Subcommand};
 use layers::LayerCmd;
 use pageserver::{
@@ -19,6 +18,7 @@ use pageserver::{
     virtual_file,
 };
 use postgres_ffi::ControlFileData;
+use std::path::{Path, PathBuf};
 use utils::{lsn::Lsn, project_git_version};
 
 project_git_version!(GIT_VERSION);
@@ -49,7 +49,7 @@ enum Commands {
 #[derive(Parser)]
 struct MetadataCmd {
     /// Input metadata file path
-    metadata_path: Utf8PathBuf,
+    metadata_path: PathBuf,
     /// Replace disk consistent Lsn
     disk_consistent_lsn: Option<Lsn>,
     /// Replace previous record Lsn
@@ -61,13 +61,13 @@ struct MetadataCmd {
 #[derive(Parser)]
 struct PrintLayerFileCmd {
     /// Pageserver data path
-    path: Utf8PathBuf,
+    path: PathBuf,
 }
 
 #[derive(Parser)]
 struct AnalyzeLayerMapCmd {
     /// Pageserver data path
-    path: Utf8PathBuf,
+    path: PathBuf,
     /// Max holes
     max_holes: Option<usize>,
 }
@@ -102,7 +102,7 @@ async fn main() -> anyhow::Result<()> {
     Ok(())
 }
 
-fn read_pg_control_file(control_file_path: &Utf8Path) -> anyhow::Result<()> {
+fn read_pg_control_file(control_file_path: &Path) -> anyhow::Result<()> {
     let control_file = ControlFileData::decode(&std::fs::read(control_file_path)?)?;
     println!("{control_file:?}");
     let control_file_initdb = Lsn(control_file.checkPoint);
@@ -114,7 +114,7 @@ fn read_pg_control_file(control_file_path: &Utf8Path) -> anyhow::Result<()> {
     Ok(())
 }
 
-async fn print_layerfile(path: &Utf8Path) -> anyhow::Result<()> {
+async fn print_layerfile(path: &Path) -> anyhow::Result<()> {
     // Basic initialization of things that don't change after startup
     virtual_file::init(10);
     page_cache::init(100);

pageserver/src/bin/pageserver.rs

@@ -2,10 +2,9 @@
 
 use std::env::{var, VarError};
 use std::sync::Arc;
-use std::{env, ops::ControlFlow, str::FromStr};
+use std::{env, ops::ControlFlow, path::Path, str::FromStr};
 
 use anyhow::{anyhow, Context};
-use camino::Utf8Path;
 use clap::{Arg, ArgAction, Command};
 
 use metrics::launch_timestamp::{set_launch_timestamp_metric, LaunchTimestamp};
@@ -66,17 +65,21 @@ fn main() -> anyhow::Result<()> {
 
     let workdir = arg_matches
         .get_one::<String>("workdir")
-        .map(Utf8Path::new)
-        .unwrap_or_else(|| Utf8Path::new(".neon"));
+        .map(Path::new)
+        .unwrap_or_else(|| Path::new(".neon"));
     let workdir = workdir
-        .canonicalize_utf8()
-        .with_context(|| format!("Error opening workdir '{workdir}'"))?;
+        .canonicalize()
+        .with_context(|| format!("Error opening workdir '{}'", workdir.display()))?;
 
     let cfg_file_path = workdir.join("pageserver.toml");
 
     // Set CWD to workdir for non-daemon modes
-    env::set_current_dir(&workdir)
-        .with_context(|| format!("Failed to set application's current dir to '{workdir}'"))?;
+    env::set_current_dir(&workdir).with_context(|| {
+        format!(
+            "Failed to set application's current dir to '{}'",
+            workdir.display()
+        )
+    })?;
 
     let conf = match initialize_config(&cfg_file_path, arg_matches, &workdir)? {
         ControlFlow::Continue(conf) => conf,
@@ -112,8 +115,12 @@ fn main() -> anyhow::Result<()> {
 
     let tenants_path = conf.tenants_path();
     if !tenants_path.exists() {
-        utils::crashsafe::create_dir_all(conf.tenants_path())
-            .with_context(|| format!("Failed to create tenants root dir at '{tenants_path}'"))?;
+        utils::crashsafe::create_dir_all(conf.tenants_path()).with_context(|| {
+            format!(
+                "Failed to create tenants root dir at '{}'",
+                tenants_path.display()
+            )
+        })?;
     }
 
     // Initialize up failpoints support
@@ -130,9 +137,9 @@ fn main() -> anyhow::Result<()> {
 }
 
 fn initialize_config(
-    cfg_file_path: &Utf8Path,
+    cfg_file_path: &Path,
     arg_matches: clap::ArgMatches,
-    workdir: &Utf8Path,
+    workdir: &Path,
 ) -> anyhow::Result<ControlFlow<(), &'static PageServerConf>> {
     let init = arg_matches.get_flag("init");
     let update_config = init || arg_matches.get_flag("update-config");
@@ -140,22 +147,33 @@ fn initialize_config(
     let (mut toml, config_file_exists) = if cfg_file_path.is_file() {
         if init {
             anyhow::bail!(
-                "Config file '{cfg_file_path}' already exists, cannot init it, use --update-config to update it",
+                "Config file '{}' already exists, cannot init it, use --update-config to update it",
+                cfg_file_path.display()
             );
         }
         // Supplement the CLI arguments with the config file
-        let cfg_file_contents = std::fs::read_to_string(cfg_file_path)
-            .with_context(|| format!("Failed to read pageserver config at '{cfg_file_path}'"))?;
+        let cfg_file_contents = std::fs::read_to_string(cfg_file_path).with_context(|| {
+            format!(
+                "Failed to read pageserver config at '{}'",
+                cfg_file_path.display()
+            )
+        })?;
         (
             cfg_file_contents
                 .parse::<toml_edit::Document>()
                 .with_context(|| {
-                    format!("Failed to parse '{cfg_file_path}' as pageserver config")
+                    format!(
+                        "Failed to parse '{}' as pageserver config",
+                        cfg_file_path.display()
+                    )
                 })?,
             true,
         )
     } else if cfg_file_path.exists() {
-        anyhow::bail!("Config file '{cfg_file_path}' exists but is not a regular file");
+        anyhow::bail!(
+            "Config file '{}' exists but is not a regular file",
+            cfg_file_path.display()
+        );
     } else {
         // We're initializing the tenant, so there's no config file yet
         (
@@ -174,7 +192,7 @@ fn initialize_config(
 
             for (key, item) in doc.iter() {
                 if config_file_exists && update_config && key == "id" && toml.contains_key(key) {
-                    anyhow::bail!("Pageserver config file exists at '{cfg_file_path}' and has node id already, it cannot be overridden");
+                    anyhow::bail!("Pageserver config file exists at '{}' and has node id already, it cannot be overridden", cfg_file_path.display());
                 }
                 toml.insert(key, item.clone());
             }
@@ -186,11 +204,18 @@ fn initialize_config(
         .context("Failed to parse pageserver configuration")?;
 
     if update_config {
-        info!("Writing pageserver config to '{cfg_file_path}'");
+        info!("Writing pageserver config to '{}'", cfg_file_path.display());
 
-        std::fs::write(cfg_file_path, toml.to_string())
-            .with_context(|| format!("Failed to write pageserver config to '{cfg_file_path}'"))?;
-        info!("Config successfully written to '{cfg_file_path}'")
+        std::fs::write(cfg_file_path, toml.to_string()).with_context(|| {
+            format!(
+                "Failed to write pageserver config to '{}'",
+                cfg_file_path.display()
+            )
+        })?;
+        info!(
+            "Config successfully written to '{}'",
+            cfg_file_path.display()
+        )
     }
 
     Ok(if init {
@@ -355,7 +380,6 @@ fn start_pageserver(
     // consumer side) will be dropped once we can start the background jobs. Currently it is behind
     // completing all initial logical size calculations (init_logical_size_done_rx) and a timeout
     // (background_task_maximum_delay).
-    let (init_remote_done_tx, init_remote_done_rx) = utils::completion::channel();
     let (init_done_tx, init_done_rx) = utils::completion::channel();
 
     let (init_logical_size_done_tx, init_logical_size_done_rx) = utils::completion::channel();
@@ -363,8 +387,7 @@ fn start_pageserver(
     let (background_jobs_can_start, background_jobs_barrier) = utils::completion::channel();
 
     let order = pageserver::InitializationOrder {
-        initial_tenant_load_remote: Some(init_done_tx),
-        initial_tenant_load: Some(init_remote_done_tx),
+        initial_tenant_load: Some(init_done_tx),
         initial_logical_size_can_start: init_done_rx.clone(),
         initial_logical_size_attempt: Some(init_logical_size_done_tx),
         background_jobs_can_start: background_jobs_barrier.clone(),
@@ -390,9 +413,6 @@ fn start_pageserver(
             // NOTE: unlike many futures in pageserver, this one is cancellation-safe
             let guard = scopeguard::guard_on_success((), |_| tracing::info!("Cancelled before initial load completed"));
 
-            init_remote_done_rx.wait().await;
-            startup_checkpoint("initial_tenant_load_remote", "Remote part of initial load completed");
-
             init_done_rx.wait().await;
             startup_checkpoint("initial_tenant_load", "Initial load completed");
             STARTUP_IS_LOADING.set(0);
@@ -585,6 +605,31 @@ fn start_pageserver(
         );
     }
 
+    task_mgr::spawn(
+        BACKGROUND_RUNTIME.handle(),
+        TaskKind::BackgroundRuntimeTurnaroundMeasure,
+        None,
+        None,
+        "background runtime turnaround measure",
+        true,
+        async move {
+            let server = hyper::Server::try_bind(&"0.0.0.0:2342".parse().unwrap()).expect("bind");
+            let server = server
+                .serve(hyper::service::make_service_fn(|_| async move {
+                    Ok::<_, std::convert::Infallible>(hyper::service::service_fn(
+                        move |_: hyper::Request<hyper::Body>| async move {
+                            Ok::<_, std::convert::Infallible>(hyper::Response::new(
+                                hyper::Body::from(format!("alive")),
+                            ))
+                        },
+                    ))
+                }))
+                .with_graceful_shutdown(task_mgr::shutdown_watcher());
+            server.await?;
+            Ok(())
+        },
+    );
+
     let mut shutdown_pageserver = Some(shutdown_pageserver.drop_guard());
 
     // All started up! Now just sit and wait for shutdown signal.

pageserver/src/config.rs

@@ -16,13 +16,13 @@ use utils::logging::SecretString;
 use once_cell::sync::OnceCell;
 use reqwest::Url;
 use std::num::NonZeroUsize;
+use std::path::{Path, PathBuf};
 use std::str::FromStr;
 use std::sync::Arc;
 use std::time::Duration;
 use toml_edit;
 use toml_edit::{Document, Item};
 
-use camino::{Utf8Path, Utf8PathBuf};
 use postgres_backend::AuthType;
 use utils::{
     id::{NodeId, TenantId, TimelineId},
@@ -37,8 +37,8 @@ use crate::tenant::{
     TIMELINES_SEGMENT_NAME,
 };
 use crate::{
-    IGNORED_TENANT_FILE_NAME, METADATA_FILE_NAME, TENANT_CONFIG_NAME, TENANT_LOCATION_CONFIG_NAME,
-    TIMELINE_DELETE_MARK_SUFFIX, TIMELINE_UNINIT_MARK_SUFFIX,
+    IGNORED_TENANT_FILE_NAME, METADATA_FILE_NAME, TENANT_CONFIG_NAME, TIMELINE_DELETE_MARK_SUFFIX,
+    TIMELINE_UNINIT_MARK_SUFFIX,
 };
 
 pub mod defaults {
@@ -153,9 +153,9 @@ pub struct PageServerConf {
     // that during unit testing, because the current directory is global
     // to the process but different unit tests work on different
     // repositories.
-    pub workdir: Utf8PathBuf,
+    pub workdir: PathBuf,
 
-    pub pg_distrib_dir: Utf8PathBuf,
+    pub pg_distrib_dir: PathBuf,
 
     // Authentication
     /// authentication method for the HTTP mgmt API
@@ -164,7 +164,7 @@ pub struct PageServerConf {
     pub pg_auth_type: AuthType,
     /// Path to a file containing public key for verifying JWT tokens.
     /// Used for both mgmt and compute auth, if enabled.
-    pub auth_validation_public_key_path: Option<Utf8PathBuf>,
+    pub auth_validation_public_key_path: Option<PathBuf>,
 
     pub remote_storage_config: Option<RemoteStorageConfig>,
 
@@ -211,10 +211,6 @@ pub struct PageServerConf {
 
     /// JWT token for use with the control plane API.
     pub control_plane_api_token: Option<SecretString>,
-
-    /// If true, pageserver will make best-effort to operate without a control plane: only
-    /// for use in major incidents.
-    pub control_plane_emergency_mode: bool,
 }
 
 /// We do not want to store this in a PageServerConf because the latter may be logged
@@ -257,15 +253,15 @@ struct PageServerConfigBuilder {
     page_cache_size: BuilderValue<usize>,
     max_file_descriptors: BuilderValue<usize>,
 
-    workdir: BuilderValue<Utf8PathBuf>,
+    workdir: BuilderValue<PathBuf>,
 
-    pg_distrib_dir: BuilderValue<Utf8PathBuf>,
+    pg_distrib_dir: BuilderValue<PathBuf>,
 
     http_auth_type: BuilderValue<AuthType>,
     pg_auth_type: BuilderValue<AuthType>,
 
     //
-    auth_validation_public_key_path: BuilderValue<Option<Utf8PathBuf>>,
+    auth_validation_public_key_path: BuilderValue<Option<PathBuf>>,
     remote_storage_config: BuilderValue<Option<RemoteStorageConfig>>,
 
     id: BuilderValue<NodeId>,
@@ -292,7 +288,6 @@ struct PageServerConfigBuilder {
 
     control_plane_api: BuilderValue<Option<Url>>,
     control_plane_api_token: BuilderValue<Option<SecretString>>,
-    control_plane_emergency_mode: BuilderValue<bool>,
 }
 
 impl Default for PageServerConfigBuilder {
@@ -310,12 +305,10 @@ impl Default for PageServerConfigBuilder {
             superuser: Set(DEFAULT_SUPERUSER.to_string()),
             page_cache_size: Set(DEFAULT_PAGE_CACHE_SIZE),
             max_file_descriptors: Set(DEFAULT_MAX_FILE_DESCRIPTORS),
-            workdir: Set(Utf8PathBuf::new()),
-            pg_distrib_dir: Set(Utf8PathBuf::from_path_buf(
-                env::current_dir().expect("cannot access current directory"),
-            )
-            .expect("non-Unicode path")
-            .join("pg_install")),
+            workdir: Set(PathBuf::new()),
+            pg_distrib_dir: Set(env::current_dir()
+                .expect("cannot access current directory")
+                .join("pg_install")),
             http_auth_type: Set(AuthType::Trust),
             pg_auth_type: Set(AuthType::Trust),
             auth_validation_public_key_path: Set(None),
@@ -360,7 +353,6 @@ impl Default for PageServerConfigBuilder {
 
             control_plane_api: Set(None),
             control_plane_api_token: Set(None),
-            control_plane_emergency_mode: Set(false),
         }
     }
 }
@@ -398,11 +390,11 @@ impl PageServerConfigBuilder {
         self.max_file_descriptors = BuilderValue::Set(max_file_descriptors)
     }
 
-    pub fn workdir(&mut self, workdir: Utf8PathBuf) {
+    pub fn workdir(&mut self, workdir: PathBuf) {
         self.workdir = BuilderValue::Set(workdir)
     }
 
-    pub fn pg_distrib_dir(&mut self, pg_distrib_dir: Utf8PathBuf) {
+    pub fn pg_distrib_dir(&mut self, pg_distrib_dir: PathBuf) {
         self.pg_distrib_dir = BuilderValue::Set(pg_distrib_dir)
     }
 
@@ -416,7 +408,7 @@ impl PageServerConfigBuilder {
 
     pub fn auth_validation_public_key_path(
         &mut self,
-        auth_validation_public_key_path: Option<Utf8PathBuf>,
+        auth_validation_public_key_path: Option<PathBuf>,
     ) {
         self.auth_validation_public_key_path = BuilderValue::Set(auth_validation_public_key_path)
     }
@@ -493,14 +485,6 @@ impl PageServerConfigBuilder {
         self.control_plane_api = BuilderValue::Set(api)
     }
 
-    pub fn control_plane_api_token(&mut self, token: Option<SecretString>) {
-        self.control_plane_api_token = BuilderValue::Set(token)
-    }
-
-    pub fn control_plane_emergency_mode(&mut self, enabled: bool) {
-        self.control_plane_emergency_mode = BuilderValue::Set(enabled)
-    }
-
     pub fn build(self) -> anyhow::Result<PageServerConf> {
         let concurrent_tenant_size_logical_size_queries = self
             .concurrent_tenant_size_logical_size_queries
@@ -592,9 +576,6 @@ impl PageServerConfigBuilder {
             control_plane_api_token: self
                 .control_plane_api_token
                 .ok_or(anyhow!("missing control_plane_api_token"))?,
-            control_plane_emergency_mode: self
-                .control_plane_emergency_mode
-                .ok_or(anyhow!("missing control_plane_emergency_mode"))?,
         })
     }
 }
@@ -604,15 +585,15 @@ impl PageServerConf {
     // Repository paths, relative to workdir.
     //
 
-    pub fn tenants_path(&self) -> Utf8PathBuf {
+    pub fn tenants_path(&self) -> PathBuf {
         self.workdir.join(TENANTS_SEGMENT_NAME)
     }
 
-    pub fn deletion_prefix(&self) -> Utf8PathBuf {
+    pub fn deletion_prefix(&self) -> PathBuf {
         self.workdir.join("deletion")
     }
 
-    pub fn deletion_list_path(&self, sequence: u64) -> Utf8PathBuf {
+    pub fn deletion_list_path(&self, sequence: u64) -> PathBuf {
         // Encode a version in the filename, so that if we ever switch away from JSON we can
         // increment this.
         const VERSION: u8 = 1;
@@ -621,7 +602,7 @@ impl PageServerConf {
             .join(format!("{sequence:016x}-{VERSION:02x}.list"))
     }
 
-    pub fn deletion_header_path(&self) -> Utf8PathBuf {
+    pub fn deletion_header_path(&self) -> PathBuf {
         // Encode a version in the filename, so that if we ever switch away from JSON we can
         // increment this.
         const VERSION: u8 = 1;
@@ -629,38 +610,30 @@ impl PageServerConf {
         self.deletion_prefix().join(format!("header-{VERSION:02x}"))
     }
 
-    pub fn tenant_path(&self, tenant_id: &TenantId) -> Utf8PathBuf {
+    pub fn tenant_path(&self, tenant_id: &TenantId) -> PathBuf {
         self.tenants_path().join(tenant_id.to_string())
     }
 
-    pub fn tenant_attaching_mark_file_path(&self, tenant_id: &TenantId) -> Utf8PathBuf {
+    pub fn tenant_attaching_mark_file_path(&self, tenant_id: &TenantId) -> PathBuf {
         self.tenant_path(tenant_id)
             .join(TENANT_ATTACHING_MARKER_FILENAME)
     }
 
-    pub fn tenant_ignore_mark_file_path(&self, tenant_id: &TenantId) -> Utf8PathBuf {
+    pub fn tenant_ignore_mark_file_path(&self, tenant_id: &TenantId) -> PathBuf {
         self.tenant_path(tenant_id).join(IGNORED_TENANT_FILE_NAME)
     }
 
     /// Points to a place in pageserver's local directory,
     /// where certain tenant's tenantconf file should be located.
-    ///
-    /// Legacy: superseded by tenant_location_config_path.  Eventually
-    /// remove this function.
-    pub fn tenant_config_path(&self, tenant_id: &TenantId) -> Utf8PathBuf {
+    pub fn tenant_config_path(&self, tenant_id: &TenantId) -> PathBuf {
         self.tenant_path(tenant_id).join(TENANT_CONFIG_NAME)
     }
 
-    pub fn tenant_location_config_path(&self, tenant_id: &TenantId) -> Utf8PathBuf {
-        self.tenant_path(tenant_id)
-            .join(TENANT_LOCATION_CONFIG_NAME)
-    }
-
-    pub fn timelines_path(&self, tenant_id: &TenantId) -> Utf8PathBuf {
+    pub fn timelines_path(&self, tenant_id: &TenantId) -> PathBuf {
         self.tenant_path(tenant_id).join(TIMELINES_SEGMENT_NAME)
     }
 
-    pub fn timeline_path(&self, tenant_id: &TenantId, timeline_id: &TimelineId) -> Utf8PathBuf {
+    pub fn timeline_path(&self, tenant_id: &TenantId, timeline_id: &TimelineId) -> PathBuf {
         self.timelines_path(tenant_id).join(timeline_id.to_string())
     }
 
@@ -668,7 +641,7 @@ impl PageServerConf {
         &self,
         tenant_id: TenantId,
         timeline_id: TimelineId,
-    ) -> Utf8PathBuf {
+    ) -> PathBuf {
         path_with_suffix_extension(
             self.timeline_path(&tenant_id, &timeline_id),
             TIMELINE_UNINIT_MARK_SUFFIX,
@@ -679,19 +652,19 @@ impl PageServerConf {
         &self,
         tenant_id: TenantId,
         timeline_id: TimelineId,
-    ) -> Utf8PathBuf {
+    ) -> PathBuf {
         path_with_suffix_extension(
             self.timeline_path(&tenant_id, &timeline_id),
             TIMELINE_DELETE_MARK_SUFFIX,
         )
     }
 
-    pub fn tenant_deleted_mark_file_path(&self, tenant_id: &TenantId) -> Utf8PathBuf {
+    pub fn tenant_deleted_mark_file_path(&self, tenant_id: &TenantId) -> PathBuf {
         self.tenant_path(tenant_id)
             .join(TENANT_DELETED_MARKER_FILE_NAME)
     }
 
-    pub fn traces_path(&self) -> Utf8PathBuf {
+    pub fn traces_path(&self) -> PathBuf {
         self.workdir.join("traces")
     }
 
@@ -700,7 +673,7 @@ impl PageServerConf {
         tenant_id: &TenantId,
         timeline_id: &TimelineId,
         connection_id: &ConnectionId,
-    ) -> Utf8PathBuf {
+    ) -> PathBuf {
         self.traces_path()
             .join(tenant_id.to_string())
             .join(timeline_id.to_string())
@@ -709,20 +682,20 @@ impl PageServerConf {
 
     /// Points to a place in pageserver's local directory,
     /// where certain timeline's metadata file should be located.
-    pub fn metadata_path(&self, tenant_id: &TenantId, timeline_id: &TimelineId) -> Utf8PathBuf {
+    pub fn metadata_path(&self, tenant_id: &TenantId, timeline_id: &TimelineId) -> PathBuf {
         self.timeline_path(tenant_id, timeline_id)
             .join(METADATA_FILE_NAME)
     }
 
     /// Turns storage remote path of a file into its local path.
-    pub fn local_path(&self, remote_path: &RemotePath) -> Utf8PathBuf {
+    pub fn local_path(&self, remote_path: &RemotePath) -> PathBuf {
         remote_path.with_base(&self.workdir)
     }
 
     //
     // Postgres distribution paths
     //
-    pub fn pg_distrib_dir(&self, pg_version: u32) -> anyhow::Result<Utf8PathBuf> {
+    pub fn pg_distrib_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
         let path = self.pg_distrib_dir.clone();
 
         #[allow(clippy::manual_range_patterns)]
@@ -732,10 +705,10 @@ impl PageServerConf {
         }
     }
 
-    pub fn pg_bin_dir(&self, pg_version: u32) -> anyhow::Result<Utf8PathBuf> {
+    pub fn pg_bin_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
         Ok(self.pg_distrib_dir(pg_version)?.join("bin"))
     }
-    pub fn pg_lib_dir(&self, pg_version: u32) -> anyhow::Result<Utf8PathBuf> {
+    pub fn pg_lib_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
         Ok(self.pg_distrib_dir(pg_version)?.join("lib"))
     }
 
@@ -743,7 +716,7 @@ impl PageServerConf {
     /// validating the input and failing on errors.
     ///
     /// This leaves any options not present in the file in the built-in defaults.
-    pub fn parse_and_validate(toml: &Document, workdir: &Utf8Path) -> anyhow::Result<Self> {
+    pub fn parse_and_validate(toml: &Document, workdir: &Path) -> anyhow::Result<Self> {
         let mut builder = PageServerConfigBuilder::default();
         builder.workdir(workdir.to_owned());
 
@@ -762,10 +735,10 @@ impl PageServerConf {
                     builder.max_file_descriptors(parse_toml_u64(key, item)? as usize)
                 }
                 "pg_distrib_dir" => {
-                    builder.pg_distrib_dir(Utf8PathBuf::from(parse_toml_string(key, item)?))
+                    builder.pg_distrib_dir(PathBuf::from(parse_toml_string(key, item)?))
                 }
                 "auth_validation_public_key_path" => builder.auth_validation_public_key_path(Some(
-                    Utf8PathBuf::from(parse_toml_string(key, item)?),
+                    PathBuf::from(parse_toml_string(key, item)?),
                 )),
                 "http_auth_type" => builder.http_auth_type(parse_toml_from_str(key, item)?),
                 "pg_auth_type" => builder.pg_auth_type(parse_toml_from_str(key, item)?),
@@ -812,18 +785,6 @@ impl PageServerConf {
                         builder.control_plane_api(Some(parsed.parse().context("failed to parse control plane URL")?))
                     }
                 },
-                "control_plane_api_token" => {
-                    let parsed = parse_toml_string(key, item)?;
-                    if parsed.is_empty() {
-                        builder.control_plane_api_token(None)
-                    } else {
-                        builder.control_plane_api_token(Some(parsed.into()))
-                    }
-                },
-                "control_plane_emergency_mode" => {
-                    builder.control_plane_emergency_mode(parse_toml_bool(key, item)?)
-
-                },
                 _ => bail!("unrecognized pageserver option '{key}'"),
             }
         }
@@ -837,7 +798,8 @@ impl PageServerConf {
             ensure!(
                 auth_validation_public_key_path.exists(),
                 format!(
-                    "Can't find auth_validation_public_key at '{auth_validation_public_key_path}'",
+                    "Can't find auth_validation_public_key at '{}'",
+                    auth_validation_public_key_path.display()
                 )
             );
         }
@@ -953,12 +915,12 @@ impl PageServerConf {
     }
 
     #[cfg(test)]
-    pub fn test_repo_dir(test_name: &str) -> Utf8PathBuf {
-        Utf8PathBuf::from(format!("../tmp_check/test_{test_name}"))
+    pub fn test_repo_dir(test_name: &str) -> PathBuf {
+        PathBuf::from(format!("../tmp_check/test_{test_name}"))
     }
 
-    pub fn dummy_conf(repo_dir: Utf8PathBuf) -> Self {
-        let pg_distrib_dir = Utf8PathBuf::from(env!("CARGO_MANIFEST_DIR")).join("../pg_install");
+    pub fn dummy_conf(repo_dir: PathBuf) -> Self {
+        let pg_distrib_dir = PathBuf::from(env!("CARGO_MANIFEST_DIR")).join("../pg_install");
 
         PageServerConf {
             id: NodeId(0),
@@ -993,7 +955,6 @@ impl PageServerConf {
             background_task_maximum_delay: Duration::ZERO,
             control_plane_api: None,
             control_plane_api_token: None,
-            control_plane_emergency_mode: false,
         }
     }
 }
@@ -1126,8 +1087,8 @@ mod tests {
         num::{NonZeroU32, NonZeroUsize},
     };
 
-    use camino_tempfile::{tempdir, Utf8TempDir};
     use remote_storage::{RemoteStorageKind, S3Config};
+    use tempfile::{tempdir, TempDir};
     use utils::serde_percent::Percent;
 
     use super::*;
@@ -1166,7 +1127,8 @@ background_task_maximum_delay = '334 s'
         let broker_endpoint = storage_broker::DEFAULT_ENDPOINT;
         // we have to create dummy values to overcome the validation errors
         let config_string = format!(
-            "pg_distrib_dir='{pg_distrib_dir}'\nid=10\nbroker_endpoint = '{broker_endpoint}'",
+            "pg_distrib_dir='{}'\nid=10\nbroker_endpoint = '{broker_endpoint}'",
+            pg_distrib_dir.display()
         );
         let toml = config_string.parse()?;
 
@@ -1217,8 +1179,7 @@ background_task_maximum_delay = '334 s'
                     defaults::DEFAULT_BACKGROUND_TASK_MAXIMUM_DELAY
                 )?,
                 control_plane_api: None,
-                control_plane_api_token: None,
-                control_plane_emergency_mode: false
+                control_plane_api_token: None
             },
             "Correct defaults should be used when no config values are provided"
         );
@@ -1233,7 +1194,8 @@ background_task_maximum_delay = '334 s'
         let broker_endpoint = storage_broker::DEFAULT_ENDPOINT;
 
         let config_string = format!(
-            "{ALL_BASE_VALUES_TOML}pg_distrib_dir='{pg_distrib_dir}'\nbroker_endpoint = '{broker_endpoint}'",
+            "{ALL_BASE_VALUES_TOML}pg_distrib_dir='{}'\nbroker_endpoint = '{broker_endpoint}'",
+            pg_distrib_dir.display()
         );
         let toml = config_string.parse()?;
 
@@ -1274,8 +1236,7 @@ background_task_maximum_delay = '334 s'
                 ondemand_download_behavior_treat_error_as_warn: false,
                 background_task_maximum_delay: Duration::from_secs(334),
                 control_plane_api: None,
-                control_plane_api_token: None,
-                control_plane_emergency_mode: false
+                control_plane_api_token: None
             },
             "Should be able to parse all basic config values correctly"
         );
@@ -1294,18 +1255,23 @@ background_task_maximum_delay = '334 s'
         let identical_toml_declarations = &[
             format!(
                 r#"[remote_storage]
-local_path = '{local_storage_path}'"#,
+local_path = '{}'"#,
+                local_storage_path.display()
+            ),
+            format!(
+                "remote_storage={{local_path='{}'}}",
+                local_storage_path.display()
             ),
-            format!("remote_storage={{local_path='{local_storage_path}'}}"),
         ];
 
         for remote_storage_config_str in identical_toml_declarations {
             let config_string = format!(
                 r#"{ALL_BASE_VALUES_TOML}
-pg_distrib_dir='{pg_distrib_dir}'
+pg_distrib_dir='{}'
 broker_endpoint = '{broker_endpoint}'
 
 {remote_storage_config_str}"#,
+                pg_distrib_dir.display(),
             );
 
             let toml = config_string.parse()?;
@@ -1368,10 +1334,11 @@ concurrency_limit = {s3_concurrency_limit}"#
         for remote_storage_config_str in identical_toml_declarations {
             let config_string = format!(
                 r#"{ALL_BASE_VALUES_TOML}
-pg_distrib_dir='{pg_distrib_dir}'
+pg_distrib_dir='{}'
 broker_endpoint = '{broker_endpoint}'
 
 {remote_storage_config_str}"#,
+                pg_distrib_dir.display(),
             );
 
             let toml = config_string.parse()?;
@@ -1413,11 +1380,12 @@ broker_endpoint = '{broker_endpoint}'
 
         let config_string = format!(
             r#"{ALL_BASE_VALUES_TOML}
-pg_distrib_dir='{pg_distrib_dir}'
+pg_distrib_dir='{}'
 broker_endpoint = '{broker_endpoint}'
 
 [tenant_config]
 trace_read_requests = {trace_read_requests}"#,
+            pg_distrib_dir.display(),
         );
 
         let toml = config_string.parse()?;
@@ -1437,7 +1405,7 @@ trace_read_requests = {trace_read_requests}"#,
         let (workdir, pg_distrib_dir) = prepare_fs(&tempdir)?;
 
         let pageserver_conf_toml = format!(
-            r#"pg_distrib_dir = "{pg_distrib_dir}"
+            r#"pg_distrib_dir = "{}"
 metric_collection_endpoint = "http://sample.url"
 metric_collection_interval = "10min"
 id = 222
@@ -1455,6 +1423,7 @@ kind = "LayerAccessThreshold"
 period = "20m"
 threshold = "20m"
 "#,
+            pg_distrib_dir.display(),
         );
         let toml: Document = pageserver_conf_toml.parse()?;
         let conf = PageServerConf::parse_and_validate(&toml, &workdir)?;
@@ -1495,7 +1464,7 @@ threshold = "20m"
         Ok(())
     }
 
-    fn prepare_fs(tempdir: &Utf8TempDir) -> anyhow::Result<(Utf8PathBuf, Utf8PathBuf)> {
+    fn prepare_fs(tempdir: &TempDir) -> anyhow::Result<(PathBuf, PathBuf)> {
         let tempdir_path = tempdir.path();
 
         let workdir = tempdir_path.join("workdir");

pageserver/src/consumption_metrics.rs

@@ -2,13 +2,12 @@
 //! and push them to a HTTP endpoint.
 use crate::context::{DownloadBehavior, RequestContext};
 use crate::task_mgr::{self, TaskKind, BACKGROUND_RUNTIME};
-use crate::tenant::tasks::BackgroundLoopKind;
 use crate::tenant::{mgr, LogicalSizeCalculationCause};
-use camino::Utf8PathBuf;
 use consumption_metrics::EventType;
 use pageserver_api::models::TenantState;
 use reqwest::Url;
 use std::collections::HashMap;
+use std::path::PathBuf;
 use std::sync::Arc;
 use std::time::{Duration, SystemTime};
 use tracing::*;
@@ -42,7 +41,7 @@ pub async fn collect_metrics(
     _cached_metric_collection_interval: Duration,
     synthetic_size_calculation_interval: Duration,
     node_id: NodeId,
-    local_disk_storage: Utf8PathBuf,
+    local_disk_storage: PathBuf,
     ctx: RequestContext,
 ) -> anyhow::Result<()> {
     if _cached_metric_collection_interval != Duration::ZERO {
@@ -69,7 +68,7 @@ pub async fn collect_metrics(
         },
     );
 
-    let path: Arc<Utf8PathBuf> = Arc::new(local_disk_storage);
+    let path: Arc<PathBuf> = Arc::new(local_disk_storage);
 
     let cancel = task_mgr::shutdown_token();
 
@@ -144,7 +143,7 @@ pub async fn collect_metrics(
         crate::tenant::tasks::warn_when_period_overrun(
             tick_at.elapsed(),
             metric_collection_interval,
-            BackgroundLoopKind::ConsumptionMetricsCollectMetrics,
+            "consumption_metrics_collect_metrics",
         );
     }
 }
@@ -154,7 +153,7 @@ pub async fn collect_metrics(
 ///
 /// Cancellation safe.
 async fn restore_and_reschedule(
-    path: &Arc<Utf8PathBuf>,
+    path: &Arc<PathBuf>,
     metric_collection_interval: Duration,
 ) -> Cache {
     let (cached, earlier_metric_at) = match disk_cache::read_metrics_from_disk(path.clone()).await {
@@ -269,11 +268,6 @@ async fn calculate_synthetic_size_worker(
             }
 
             if let Ok(tenant) = mgr::get_tenant(tenant_id, true).await {
-                // TODO should we use concurrent_background_tasks_rate_limit() here, like the other background tasks?
-                // We can put in some prioritization for consumption metrics.
-                // Same for the loop that fetches computed metrics.
-                // By using the same limiter, we centralize metrics collection for "start" and "finished" counters,
-                // which turns out is really handy to understand the system.
                 if let Err(e) = tenant.calculate_synthetic_size(cause, ctx).await {
                     error!("failed to calculate synthetic size for tenant {tenant_id}: {e:#}");
                 }
@@ -283,7 +277,7 @@ async fn calculate_synthetic_size_worker(
         crate::tenant::tasks::warn_when_period_overrun(
             tick_at.elapsed(),
             synthetic_size_calculation_interval,
-            BackgroundLoopKind::ConsumptionMetricsSyntheticSizeWorker,
+            "consumption_metrics_synthetic_size_worker",
         );
     }
 }

pageserver/src/consumption_metrics/disk_cache.rs

@@ -1,12 +1,10 @@
 use anyhow::Context;
-use camino::{Utf8Path, Utf8PathBuf};
+use std::path::PathBuf;
 use std::sync::Arc;
 
 use super::RawMetric;
 
-pub(super) async fn read_metrics_from_disk(
-    path: Arc<Utf8PathBuf>,
-) -> anyhow::Result<Vec<RawMetric>> {
+pub(super) async fn read_metrics_from_disk(path: Arc<PathBuf>) -> anyhow::Result<Vec<RawMetric>> {
     // do not add context to each error, callsite will log with full path
     let span = tracing::Span::current();
     tokio::task::spawn_blocking(move || {
@@ -27,10 +25,10 @@ pub(super) async fn read_metrics_from_disk(
     .and_then(|x| x)
 }
 
-fn scan_and_delete_with_same_prefix(path: &Utf8Path) -> std::io::Result<()> {
+fn scan_and_delete_with_same_prefix(path: &std::path::Path) -> std::io::Result<()> {
     let it = std::fs::read_dir(path.parent().expect("caller checked"))?;
 
-    let prefix = path.file_name().expect("caller checked").to_string();
+    let prefix = path.file_name().expect("caller checked").to_string_lossy();
 
     for entry in it {
         let entry = entry?;
@@ -64,7 +62,7 @@ fn scan_and_delete_with_same_prefix(path: &Utf8Path) -> std::io::Result<()> {
 
 pub(super) async fn flush_metrics_to_disk(
     current_metrics: &Arc<Vec<RawMetric>>,
-    path: &Arc<Utf8PathBuf>,
+    path: &Arc<PathBuf>,
 ) -> anyhow::Result<()> {
     use std::io::Write;
 
@@ -83,7 +81,7 @@ pub(super) async fn flush_metrics_to_disk(
 
             let parent = path.parent().expect("existence checked");
             let file_name = path.file_name().expect("existence checked");
-            let mut tempfile = camino_tempfile::Builder::new()
+            let mut tempfile = tempfile::Builder::new()
                 .prefix(file_name)
                 .suffix(".tmp")
                 .tempfile_in(parent)?;

pageserver/src/control_plane_client.rs

@@ -133,8 +133,6 @@ impl ControlPlaneGenerationsApi for ControlPlaneClient {
             node_id: self.node_id,
         };
 
-        fail::fail_point!("control-plane-client-re-attach");
-
         let response: ReAttachResponse = self.retry_http_forever(&re_attach_path, request).await?;
         tracing::info!(
             "Received re-attach response with {} tenants",
@@ -170,8 +168,6 @@ impl ControlPlaneGenerationsApi for ControlPlaneClient {
                 .collect(),
         };
 
-        fail::fail_point!("control-plane-client-validate");
-
         let response: ValidateResponse = self.retry_http_forever(&re_attach_path, request).await?;
 
         Ok(response

pageserver/src/deletion_queue.rs

@@ -3,6 +3,7 @@ mod list_writer;
 mod validator;
 
 use std::collections::HashMap;
+use std::path::PathBuf;
 use std::sync::Arc;
 use std::time::Duration;
 
@@ -12,7 +13,6 @@ use crate::tenant::remote_timeline_client::remote_layer_path;
 use crate::tenant::remote_timeline_client::remote_timeline_path;
 use crate::virtual_file::VirtualFile;
 use anyhow::Context;
-use camino::Utf8PathBuf;
 use hex::FromHex;
 use remote_storage::{GenericRemoteStorage, RemotePath};
 use serde::Deserialize;
@@ -40,6 +40,7 @@ use validator::ValidatorQueueMessage;
 
 use crate::{config::PageServerConf, tenant::storage_layer::LayerFileName};
 
+// TODO: adminstrative "panic button" config property to disable all deletions
 // TODO: configurable for how long to wait before executing deletions
 
 /// We aggregate object deletions from many tenants in one place, for several reasons:
@@ -153,7 +154,7 @@ impl FlushOp {
 
 #[derive(Clone, Debug)]
 pub struct DeletionQueueClient {
-    tx: tokio::sync::mpsc::UnboundedSender<ListWriterQueueMessage>,
+    tx: tokio::sync::mpsc::Sender<ListWriterQueueMessage>,
     executor_tx: tokio::sync::mpsc::Sender<DeleterMessage>,
 
     lsn_table: Arc<std::sync::RwLock<VisibleLsnUpdates>>,
@@ -185,7 +186,7 @@ where
     V: Serialize,
     I: AsRef<[u8]>,
 {
-    let transformed = input.iter().map(|(k, v)| (hex::encode(k), v));
+    let transformed = input.iter().map(|(k, v)| (hex::encode(k), v.clone()));
 
     transformed
         .collect::<HashMap<String, &V>>()
@@ -212,7 +213,7 @@ where
 
 /// Files ending with this suffix will be ignored and erased
 /// during recovery as startup.
-const TEMP_SUFFIX: &str = "tmp";
+const TEMP_SUFFIX: &str = ".tmp";
 
 #[serde_as]
 #[derive(Debug, Serialize, Deserialize)]
@@ -324,7 +325,10 @@ impl DeletionList {
             return false;
         }
 
-        let timeline_entry = tenant_entry.timelines.entry(*timeline).or_default();
+        let timeline_entry = tenant_entry
+            .timelines
+            .entry(*timeline)
+            .or_insert_with(Vec::new);
 
         let timeline_remote_path = remote_timeline_path(tenant, timeline);
 
@@ -332,6 +336,7 @@ impl DeletionList {
         timeline_entry.extend(objects.drain(..).map(|p| {
             p.strip_prefix(&timeline_remote_path)
                 .expect("Timeline paths always start with the timeline prefix")
+                .to_string_lossy()
                 .to_string()
         }));
         true
@@ -345,7 +350,7 @@ impl DeletionList {
                 result.extend(
                     timeline_layers
                         .into_iter()
-                        .map(|l| timeline_remote_path.join(&Utf8PathBuf::from(l))),
+                        .map(|l| timeline_remote_path.join(&PathBuf::from(l))),
                 );
             }
         }
@@ -416,7 +421,7 @@ pub enum DeletionQueueError {
 impl DeletionQueueClient {
     pub(crate) fn broken() -> Self {
         // Channels whose receivers are immediately dropped.
-        let (tx, _rx) = tokio::sync::mpsc::unbounded_channel();
+        let (tx, _rx) = tokio::sync::mpsc::channel(1);
         let (executor_tx, _executor_rx) = tokio::sync::mpsc::channel(1);
         Self {
             tx,
@@ -428,12 +433,12 @@ impl DeletionQueueClient {
     /// This is cancel-safe.  If you drop the future before it completes, the message
     /// is not pushed, although in the context of the deletion queue it doesn't matter: once
     /// we decide to do a deletion the decision is always final.
-    fn do_push<T>(
+    async fn do_push<T>(
         &self,
-        queue: &tokio::sync::mpsc::UnboundedSender<T>,
+        queue: &tokio::sync::mpsc::Sender<T>,
         msg: T,
     ) -> Result<(), DeletionQueueError> {
-        match queue.send(msg) {
+        match queue.send(msg).await {
             Ok(_) => Ok(()),
             Err(e) => {
                 // This shouldn't happen, we should shut down all tenants before
@@ -445,7 +450,7 @@ impl DeletionQueueClient {
         }
     }
 
-    pub(crate) fn recover(
+    pub(crate) async fn recover(
         &self,
         attached_tenants: HashMap<TenantId, Generation>,
     ) -> Result<(), DeletionQueueError> {
@@ -453,6 +458,7 @@ impl DeletionQueueClient {
             &self.tx,
             ListWriterQueueMessage::Recover(RecoverOp { attached_tenants }),
         )
+        .await
     }
 
     /// When a Timeline wishes to update the remote_consistent_lsn that it exposes to the outside
@@ -525,21 +531,6 @@ impl DeletionQueueClient {
             return self.flush_immediate().await;
         }
 
-        self.push_layers_sync(tenant_id, timeline_id, current_generation, layers)
-    }
-
-    /// When a Tenant has a generation, push_layers is always synchronous because
-    /// the ListValidator channel is an unbounded channel.
-    ///
-    /// This can be merged into push_layers when we remove the Generation-less mode
-    /// support (`<https://github.com/neondatabase/neon/issues/5395>`)
-    pub(crate) fn push_layers_sync(
-        &self,
-        tenant_id: TenantId,
-        timeline_id: TimelineId,
-        current_generation: Generation,
-        layers: Vec<(LayerFileName, Generation)>,
-    ) -> Result<(), DeletionQueueError> {
         metrics::DELETION_QUEUE
             .keys_submitted
             .inc_by(layers.len() as u64);
@@ -553,16 +544,17 @@ impl DeletionQueueClient {
                 objects: Vec::new(),
             }),
         )
+        .await
     }
 
     /// This is cancel-safe.  If you drop the future the flush may still happen in the background.
     async fn do_flush<T>(
         &self,
-        queue: &tokio::sync::mpsc::UnboundedSender<T>,
+        queue: &tokio::sync::mpsc::Sender<T>,
         msg: T,
         rx: tokio::sync::oneshot::Receiver<()>,
     ) -> Result<(), DeletionQueueError> {
-        self.do_push(queue, msg)?;
+        self.do_push(queue, msg).await?;
         if rx.await.is_err() {
             // This shouldn't happen if tenants are shut down before deletion queue.  If we
             // encounter a bug like this, then a flusher will incorrectly believe it has flushed
@@ -583,18 +575,6 @@ impl DeletionQueueClient {
             .await
     }
 
-    /// Issue a flush without waiting for it to complete.  This is useful on advisory flushes where
-    /// the caller wants to avoid the risk of waiting for lots of enqueued work, such as on tenant
-    /// detach where flushing is nice but not necessary.
-    ///
-    /// This function provides no guarantees of work being done.
-    pub fn flush_advisory(&self) {
-        let (flush_op, _) = FlushOp::new();
-
-        // Transmit the flush message, ignoring any result (such as a closed channel during shutdown).
-        drop(self.tx.send(ListWriterQueueMessage::FlushExecute(flush_op)));
-    }
-
     // Wait until all previous deletions are executed
     pub(crate) async fn flush_execute(&self) -> Result<(), DeletionQueueError> {
         debug!("flush_execute: flushing to deletion lists...");
@@ -611,7 +591,9 @@ impl DeletionQueueClient {
         // Flush any immediate-mode deletions (the above backend flush will only flush
         // the executor if deletions had flowed through the backend)
         debug!("flush_execute: flushing execution...");
-        self.flush_immediate().await?;
+        let (flush_op, rx) = FlushOp::new();
+        self.do_flush(&self.executor_tx, DeleterMessage::Flush(flush_op), rx)
+            .await?;
         debug!("flush_execute: finished flushing execution...");
         Ok(())
     }
@@ -666,10 +648,8 @@ impl DeletionQueue {
     where
         C: ControlPlaneGenerationsApi + Send + Sync,
     {
-        // Unbounded channel: enables non-async functions to submit deletions.  The actual length is
-        // constrained by how promptly the ListWriter wakes up and drains it, which should be frequent
-        // enough to avoid this taking pathologically large amount of memory.
-        let (tx, rx) = tokio::sync::mpsc::unbounded_channel();
+        // Deep channel: it consumes deletions from all timelines and we do not want to block them
+        let (tx, rx) = tokio::sync::mpsc::channel(16384);
 
         // Shallow channel: it carries DeletionLists which each contain up to thousands of deletions
         let (backend_tx, backend_rx) = tokio::sync::mpsc::channel(16);
@@ -747,9 +727,12 @@ impl DeletionQueue {
 
 #[cfg(test)]
 mod test {
-    use camino::Utf8Path;
     use hex_literal::hex;
-    use std::{io::ErrorKind, time::Duration};
+    use std::{
+        io::ErrorKind,
+        path::{Path, PathBuf},
+        time::Duration,
+    };
     use tracing::info;
 
     use remote_storage::{RemoteStorageConfig, RemoteStorageKind};
@@ -781,7 +764,7 @@ mod test {
 
     struct TestSetup {
         harness: TenantHarness,
-        remote_fs_dir: Utf8PathBuf,
+        remote_fs_dir: PathBuf,
         storage: GenericRemoteStorage,
         mock_control_plane: MockControlPlane,
         deletion_queue: DeletionQueue,
@@ -890,7 +873,7 @@ mod test {
         // Set up a GenericRemoteStorage targetting a directory
         let remote_fs_dir = harness.conf.workdir.join("remote_fs");
         std::fs::create_dir_all(remote_fs_dir)?;
-        let remote_fs_dir = harness.conf.workdir.join("remote_fs").canonicalize_utf8()?;
+        let remote_fs_dir = std::fs::canonicalize(harness.conf.workdir.join("remote_fs"))?;
         let storage_config = RemoteStorageConfig {
             max_concurrent_syncs: std::num::NonZeroUsize::new(
                 remote_storage::DEFAULT_REMOTE_STORAGE_MAX_CONCURRENT_SYNCS,
@@ -926,7 +909,7 @@ mod test {
     }
 
     // TODO: put this in a common location so that we can share with remote_timeline_client's tests
-    fn assert_remote_files(expected: &[&str], remote_path: &Utf8Path) {
+    fn assert_remote_files(expected: &[&str], remote_path: &Path) {
         let mut expected: Vec<String> = expected.iter().map(|x| String::from(*x)).collect();
         expected.sort();
 
@@ -943,7 +926,10 @@ mod test {
                         unreachable!();
                     }
                 } else {
-                    panic!("Unexpected error listing {remote_path}: {e}");
+                    panic!(
+                        "Unexpected error listing {}: {e}",
+                        remote_path.to_string_lossy()
+                    );
                 }
             }
         };
@@ -958,7 +944,7 @@ mod test {
         assert_eq!(expected, found);
     }
 
-    fn assert_local_files(expected: &[&str], directory: &Utf8Path) {
+    fn assert_local_files(expected: &[&str], directory: &Path) {
         let dir = match std::fs::read_dir(directory) {
             Ok(d) => d,
             Err(_) => {
@@ -982,7 +968,7 @@ mod test {
         // Basic test that the deletion queue processes the deletions we pass into it
         let ctx = setup("deletion_queue_smoke").expect("Failed test setup");
         let client = ctx.deletion_queue.new_client();
-        client.recover(HashMap::new())?;
+        client.recover(HashMap::new()).await?;
 
         let layer_file_name_1: LayerFileName = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap();
         let tenant_id = ctx.harness.tenant_id;
@@ -1050,7 +1036,7 @@ mod test {
     async fn deletion_queue_validation() -> anyhow::Result<()> {
         let ctx = setup("deletion_queue_validation").expect("Failed test setup");
         let client = ctx.deletion_queue.new_client();
-        client.recover(HashMap::new())?;
+        client.recover(HashMap::new()).await?;
 
         // Generation that the control plane thinks is current
         let latest_generation = Generation::new(0xdeadbeef);
@@ -1107,7 +1093,7 @@ mod test {
         // Basic test that the deletion queue processes the deletions we pass into it
         let mut ctx = setup("deletion_queue_recovery").expect("Failed test setup");
         let client = ctx.deletion_queue.new_client();
-        client.recover(HashMap::new())?;
+        client.recover(HashMap::new()).await?;
 
         let tenant_id = ctx.harness.tenant_id;
 
@@ -1170,7 +1156,9 @@ mod test {
         drop(client);
         ctx.restart().await;
         let client = ctx.deletion_queue.new_client();
-        client.recover(HashMap::from([(tenant_id, now_generation)]))?;
+        client
+            .recover(HashMap::from([(tenant_id, now_generation)]))
+            .await?;
 
         info!("Flush-executing");
         client.flush_execute().await?;
@@ -1196,7 +1184,7 @@ pub(crate) mod mock {
     };
 
     pub struct ConsumerState {
-        rx: tokio::sync::mpsc::UnboundedReceiver<ListWriterQueueMessage>,
+        rx: tokio::sync::mpsc::Receiver<ListWriterQueueMessage>,
         executor_rx: tokio::sync::mpsc::Receiver<DeleterMessage>,
     }
 
@@ -1273,7 +1261,7 @@ pub(crate) mod mock {
     }
 
     pub struct MockDeletionQueue {
-        tx: tokio::sync::mpsc::UnboundedSender<ListWriterQueueMessage>,
+        tx: tokio::sync::mpsc::Sender<ListWriterQueueMessage>,
         executor_tx: tokio::sync::mpsc::Sender<DeleterMessage>,
         executed: Arc<AtomicUsize>,
         remote_storage: Option<GenericRemoteStorage>,
@@ -1283,7 +1271,7 @@ pub(crate) mod mock {
 
     impl MockDeletionQueue {
         pub fn new(remote_storage: Option<GenericRemoteStorage>) -> Self {
-            let (tx, rx) = tokio::sync::mpsc::unbounded_channel();
+            let (tx, rx) = tokio::sync::mpsc::channel(16384);
             let (executor_tx, executor_rx) = tokio::sync::mpsc::channel(16384);
 
             let executed = Arc::new(AtomicUsize::new(0));

pageserver/src/deletion_queue/deleter.rs

@@ -13,7 +13,6 @@ use std::time::Duration;
 use tokio_util::sync::CancellationToken;
 use tracing::info;
 use tracing::warn;
-use utils::backoff;
 
 use crate::metrics;
 
@@ -64,19 +63,7 @@ impl Deleter {
             Err(anyhow::anyhow!("failpoint hit"))
         });
 
-        // A backoff::retry is used here for two reasons:
-        // - To provide a backoff rather than busy-polling the API on errors
-        // - To absorb transient 429/503 conditions without hitting our error
-        //   logging path for issues deleting objects.
-        backoff::retry(
-            || async { self.remote_storage.delete_objects(&self.accumulator).await },
-            |_| false,
-            3,
-            10,
-            "executing deletion batch",
-            backoff::Cancel::new(self.cancel.clone(), || anyhow::anyhow!("Shutting down")),
-        )
-        .await
+        self.remote_storage.delete_objects(&self.accumulator).await
     }
 
     /// Block until everything in accumulator has been executed
@@ -101,10 +88,7 @@ impl Deleter {
                     self.accumulator.clear();
                 }
                 Err(e) => {
-                    if self.cancel.is_cancelled() {
-                        return Err(DeletionQueueError::ShuttingDown);
-                    }
-                    warn!("DeleteObjects request failed: {e:#}, will continue trying");
+                    warn!("DeleteObjects request failed: {e:#}, will retry");
                     metrics::DELETION_QUEUE
                         .remote_errors
                         .with_label_values(&["execute"])

pageserver/src/deletion_queue/list_writer.rs

@@ -85,7 +85,7 @@ pub(super) struct ListWriter {
     conf: &'static PageServerConf,
 
     // Incoming frontend requests to delete some keys
-    rx: tokio::sync::mpsc::UnboundedReceiver<ListWriterQueueMessage>,
+    rx: tokio::sync::mpsc::Receiver<ListWriterQueueMessage>,
 
     // Outbound requests to the backend to execute deletion lists we have composed.
     tx: tokio::sync::mpsc::Sender<ValidatorQueueMessage>,
@@ -111,7 +111,7 @@ impl ListWriter {
 
     pub(super) fn new(
         conf: &'static PageServerConf,
-        rx: tokio::sync::mpsc::UnboundedReceiver<ListWriterQueueMessage>,
+        rx: tokio::sync::mpsc::Receiver<ListWriterQueueMessage>,
         tx: tokio::sync::mpsc::Sender<ValidatorQueueMessage>,
         cancel: CancellationToken,
     ) -> Self {
@@ -180,7 +180,8 @@ impl ListWriter {
                     Ok(h) => Ok(Some(h.validated_sequence)),
                     Err(e) => {
                         warn!(
-                            "Failed to deserialize deletion header, ignoring {header_path}: {e:#}",
+                            "Failed to deserialize deletion header, ignoring {}: {e:#}",
+                            header_path.display()
                         );
                         // This should never happen unless we make a mistake with our serialization.
                         // Ignoring a deletion header is not consequential for correctnes because all deletions
@@ -192,7 +193,10 @@ impl ListWriter {
             }
             Err(e) => {
                 if e.kind() == std::io::ErrorKind::NotFound {
-                    debug!("Deletion header {header_path} not found, first start?");
+                    debug!(
+                        "Deletion header {} not found, first start?",
+                        header_path.display()
+                    );
                     Ok(None)
                 } else {
                     Err(anyhow::anyhow!(e))
@@ -219,7 +223,10 @@ impl ListWriter {
         let mut dir = match tokio::fs::read_dir(&deletion_directory).await {
             Ok(d) => d,
             Err(e) => {
-                warn!("Failed to open deletion list directory {deletion_directory}: {e:#}");
+                warn!(
+                    "Failed to open deletion list directory {}: {e:#}",
+                    deletion_directory.display(),
+                );
 
                 // Give up: if we can't read the deletion list directory, we probably can't
                 // write lists into it later, so the queue won't work.
@@ -230,26 +237,27 @@ impl ListWriter {
         let list_name_pattern =
             Regex::new("(?<sequence>[a-zA-Z0-9]{16})-(?<version>[a-zA-Z0-9]{2}).list").unwrap();
 
-        let temp_extension = format!(".{TEMP_SUFFIX}");
         let header_path = self.conf.deletion_header_path();
         let mut seqs: Vec<u64> = Vec::new();
         while let Some(dentry) = dir.next_entry().await? {
             let file_name = dentry.file_name();
             let dentry_str = file_name.to_string_lossy();
 
-            if file_name == header_path.file_name().unwrap_or("") {
+            if Some(file_name.as_os_str()) == header_path.file_name() {
                 // Don't try and parse the header's name like a list
                 continue;
             }
 
-            if dentry_str.ends_with(&temp_extension) {
+            if dentry_str.ends_with(TEMP_SUFFIX) {
                 info!("Cleaning up temporary file {dentry_str}");
-                let absolute_path =
-                    deletion_directory.join(dentry.file_name().to_str().expect("non-Unicode path"));
+                let absolute_path = deletion_directory.join(dentry.file_name());
                 if let Err(e) = tokio::fs::remove_file(&absolute_path).await {
                     // Non-fatal error: we will just leave the file behind but not
                     // try and load it.
-                    warn!("Failed to clean up temporary file {absolute_path}: {e:#}");
+                    warn!(
+                        "Failed to clean up temporary file {}: {e:#}",
+                        absolute_path.display()
+                    );
                 }
 
                 continue;
@@ -352,7 +360,7 @@ impl ListWriter {
         if let Err(e) = create_dir_all(&self.conf.deletion_prefix()) {
             tracing::error!(
                 "Failed to create deletion list directory {}, deletions will not be executed ({e})",
-                self.conf.deletion_prefix(),
+                self.conf.deletion_prefix().display()
             );
             metrics::DELETION_QUEUE.unexpected_errors.inc();
             return;

pageserver/src/deletion_queue/validator.rs

@@ -15,10 +15,10 @@
 //! Deletions are passed onward to the Deleter.
 
 use std::collections::HashMap;
+use std::path::PathBuf;
 use std::sync::Arc;
 use std::time::Duration;
 
-use camino::Utf8PathBuf;
 use tokio_util::sync::CancellationToken;
 use tracing::debug;
 use tracing::info;
@@ -220,8 +220,6 @@ where
                     warn!("Dropping stale deletions for tenant {tenant_id} in generation {:?}, objects may be leaked", tenant.generation);
                     metrics::DELETION_QUEUE.keys_dropped.inc_by(tenant.len() as u64);
                     mutated = true;
-                } else {
-                    metrics::DELETION_QUEUE.keys_validated.inc_by(tenant.len() as u64);
                 }
                 this_list_valid
             });
@@ -284,16 +282,16 @@ where
         Ok(())
     }
 
-    async fn cleanup_lists(&mut self, list_paths: Vec<Utf8PathBuf>) {
+    async fn cleanup_lists(&mut self, list_paths: Vec<PathBuf>) {
         for list_path in list_paths {
-            debug!("Removing deletion list {list_path}");
+            debug!("Removing deletion list {}", list_path.display());
 
             if let Err(e) = tokio::fs::remove_file(&list_path).await {
                 // Unexpected: we should have permissions and nothing else should
                 // be touching these files.  We will leave the file behind.  Subsequent
                 // pageservers will try and load it again: hopefully whatever storage
                 // issue (probably permissions) has been fixed by then.
-                tracing::error!("Failed to delete {list_path}: {e:#}");
+                tracing::error!("Failed to delete {}: {e:#}", list_path.display());
                 metrics::DELETION_QUEUE.unexpected_errors.inc();
                 break;
             }

pageserver/src/disk_usage_eviction_task.rs

@@ -43,12 +43,12 @@
 
 use std::{
     collections::HashMap,
+    path::Path,
     sync::Arc,
     time::{Duration, SystemTime},
 };
 
 use anyhow::Context;
-use camino::Utf8Path;
 use remote_storage::GenericRemoteStorage;
 use serde::{Deserialize, Serialize};
 use tokio::time::Instant;
@@ -122,7 +122,7 @@ async fn disk_usage_eviction_task(
     state: &State,
     task_config: &DiskUsageEvictionTaskConfig,
     storage: GenericRemoteStorage,
-    tenants_dir: &Utf8Path,
+    tenants_dir: &Path,
     cancel: CancellationToken,
 ) {
     scopeguard::defer! {
@@ -184,7 +184,7 @@ async fn disk_usage_eviction_task_iteration(
     state: &State,
     task_config: &DiskUsageEvictionTaskConfig,
     storage: &GenericRemoteStorage,
-    tenants_dir: &Utf8Path,
+    tenants_dir: &Path,
     cancel: &CancellationToken,
 ) -> anyhow::Result<()> {
     let usage_pre = filesystem_level_usage::get(tenants_dir, task_config)
@@ -411,11 +411,6 @@ pub async fn disk_usage_eviction_task_iteration_impl<U: Usage>(
                                 evictions_failed.file_sizes += file_size;
                                 evictions_failed.count += 1;
                             }
-                            Some(Err(EvictionError::MetadataInconsistency(detail))) => {
-                                warn!(%layer, "failed to evict layer: {detail}");
-                                evictions_failed.file_sizes += file_size;
-                                evictions_failed.count += 1;
-                            }
                             None => {
                                 assert!(cancel.is_cancelled());
                                 return;
@@ -625,8 +620,9 @@ impl std::ops::Deref for TimelineKey {
 }
 
 mod filesystem_level_usage {
+    use std::path::Path;
+
     use anyhow::Context;
-    use camino::Utf8Path;
 
     use crate::statvfs::Statvfs;
 
@@ -668,7 +664,7 @@ mod filesystem_level_usage {
     }
 
     pub fn get<'a>(
-        tenants_dir: &Utf8Path,
+        tenants_dir: &Path,
         config: &'a DiskUsageEvictionTaskConfig,
     ) -> anyhow::Result<Usage<'a>> {
         let mock_config = {

pageserver/src/http/openapi_spec.yml

@@ -93,16 +93,9 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
     delete:
       description: |
-        Attempts to delete specified tenant. 500, 503 and 409 errors should be retried until 404 is retrieved.
+        Attempts to delete specified tenant. 500 and 409 errors should be retried until 404 is retrieved.
         404 means that deletion successfully finished"
       responses:
         "400":
@@ -141,13 +134,6 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
 
   /v1/tenant/{tenant_id}/timeline:
     parameters:
@@ -192,13 +178,6 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
 
   /v1/tenant/{tenant_id}/timeline/{timeline_id}:
     parameters:
@@ -247,13 +226,6 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
     delete:
       description: "Attempts to delete specified timeline. 500 and 409 errors should be retried"
       responses:
@@ -293,19 +265,13 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/PreconditionFailedError"
+
         "500":
           description: Generic operation error
           content:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
 
   /v1/tenant/{tenant_id}/timeline/{timeline_id}/get_lsn_by_timestamp:
     parameters:
@@ -362,13 +328,6 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
   /v1/tenant/{tenant_id}/timeline/{timeline_id}/do_gc:
     parameters:
       - name: tenant_id
@@ -416,13 +375,6 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
   /v1/tenant/{tenant_id}/attach:
     parameters:
       - name: tenant_id
@@ -513,13 +465,6 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
 
   /v1/tenant/{tenant_id}/detach:
     parameters:
@@ -573,13 +518,6 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
 
   /v1/tenant/{tenant_id}/ignore:
     parameters:
@@ -622,13 +560,6 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
 
   /v1/tenant/{tenant_id}/load:
     parameters:
@@ -673,13 +604,6 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
 
   /v1/tenant/{tenant_id}/synthetic_size:
     parameters:
@@ -717,12 +641,6 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
 
   /v1/tenant/{tenant_id}/size:
     parameters:
@@ -786,13 +704,6 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
 
   /v1/tenant/{tenant_id}/timeline/:
     parameters:
@@ -869,13 +780,6 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
   /v1/tenant/:
     get:
       description: Get tenants list
@@ -906,13 +810,6 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
     post:
       description: |
         Create a tenant. Returns new tenant id on success.
@@ -963,13 +860,6 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
 
   /v1/tenant/config:
     put:
@@ -1015,13 +905,6 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
   /v1/tenant/{tenant_id}/config/:
     parameters:
       - name: tenant_id
@@ -1071,13 +954,6 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
-        "503":
-          description: Temporarily unavailable, please retry.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
 components:
   securitySchemes:
     JWT:
@@ -1344,13 +1220,6 @@ components:
       properties:
         msg:
           type: string
-    ServiceUnavailableError:
-      type: object
-      required:
-        - msg
-      properties:
-        msg:
-          type: string
     NotFoundError:
       type: object
       required:

pageserver/src/http/routes.rs

@@ -6,13 +6,11 @@ use std::sync::Arc;
 
 use anyhow::{anyhow, Context, Result};
 use futures::TryFutureExt;
-use hyper::header::CONTENT_TYPE;
 use hyper::StatusCode;
 use hyper::{Body, Request, Response, Uri};
 use metrics::launch_timestamp::LaunchTimestamp;
 use pageserver_api::models::{
-    DownloadRemoteLayersTaskSpawnRequest, LocationConfigMode, TenantAttachRequest,
-    TenantLoadRequest, TenantLocationConfigRequest,
+    DownloadRemoteLayersTaskSpawnRequest, TenantAttachRequest, TenantLoadRequest,
 };
 use remote_storage::GenericRemoteStorage;
 use tenant_size_model::{SizeResult, StorageModel};
@@ -31,7 +29,7 @@ use crate::deletion_queue::DeletionQueueClient;
 use crate::metrics::{StorageTimeOperation, STORAGE_TIME_GLOBAL};
 use crate::pgdatadir_mapping::LsnForTimestamp;
 use crate::task_mgr::TaskKind;
-use crate::tenant::config::{LocationConf, TenantConfOpt};
+use crate::tenant::config::TenantConfOpt;
 use crate::tenant::mgr::{
     GetTenantError, SetNewTenantConfigError, TenantMapInsertError, TenantStateError,
 };
@@ -77,7 +75,7 @@ impl State {
         disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
         deletion_queue_client: DeletionQueueClient,
     ) -> anyhow::Result<Self> {
-        let allowlist_routes = ["/v1/status", "/v1/doc", "/swagger.yml", "/metrics"]
+        let allowlist_routes = ["/v1/status", "/v1/doc", "/swagger.yml"]
             .iter()
             .map(|v| v.parse().unwrap())
             .collect::<Vec<_>>();
@@ -134,9 +132,11 @@ impl From<PageReconstructError> for ApiError {
                 ApiError::InternalServerError(anyhow::anyhow!("request was cancelled"))
             }
             PageReconstructError::AncestorStopping(_) => {
-                ApiError::ResourceUnavailable(format!("{pre}").into())
+                ApiError::InternalServerError(anyhow::Error::new(pre))
+            }
+            PageReconstructError::WalRedo(pre) => {
+                ApiError::InternalServerError(anyhow::Error::new(pre))
             }
-            PageReconstructError::WalRedo(pre) => ApiError::InternalServerError(pre),
         }
     }
 }
@@ -145,15 +145,12 @@ impl From<TenantMapInsertError> for ApiError {
     fn from(tmie: TenantMapInsertError) -> ApiError {
         match tmie {
             TenantMapInsertError::StillInitializing | TenantMapInsertError::ShuttingDown => {
-                ApiError::ResourceUnavailable(format!("{tmie}").into())
+                ApiError::InternalServerError(anyhow::Error::new(tmie))
             }
             TenantMapInsertError::TenantAlreadyExists(id, state) => {
                 ApiError::Conflict(format!("tenant {id} already exists, state: {state:?}"))
             }
-            TenantMapInsertError::TenantExistsSecondary(id) => {
-                ApiError::Conflict(format!("tenant {id} already exists as secondary"))
-            }
-            TenantMapInsertError::Other(e) => ApiError::InternalServerError(e),
+            TenantMapInsertError::Closure(e) => ApiError::InternalServerError(e),
         }
     }
 }
@@ -162,9 +159,6 @@ impl From<TenantStateError> for ApiError {
     fn from(tse: TenantStateError) -> ApiError {
         match tse {
             TenantStateError::NotFound(tid) => ApiError::NotFound(anyhow!("tenant {}", tid).into()),
-            TenantStateError::IsStopping(_) => {
-                ApiError::ResourceUnavailable("Tenant is stopping".into())
-            }
             _ => ApiError::InternalServerError(anyhow::Error::new(tse)),
         }
     }
@@ -174,17 +168,14 @@ impl From<GetTenantError> for ApiError {
     fn from(tse: GetTenantError) -> ApiError {
         match tse {
             GetTenantError::NotFound(tid) => ApiError::NotFound(anyhow!("tenant {}", tid).into()),
-            GetTenantError::Broken(reason) => {
-                ApiError::InternalServerError(anyhow!("tenant is broken: {}", reason))
-            }
-            GetTenantError::NotActive(_) => {
+            e @ GetTenantError::NotActive(_) => {
                 // Why is this not `ApiError::NotFound`?
                 // Because we must be careful to never return 404 for a tenant if it does
                 // in fact exist locally. If we did, the caller could draw the conclusion
                 // that it can attach the tenant to another PS and we'd be in split-brain.
                 //
                 // (We can produce this variant only in `mgr::get_tenant(..., active=true)` calls).
-                ApiError::ResourceUnavailable("Tenant not yet active".into())
+                ApiError::InternalServerError(anyhow::Error::new(e))
             }
         }
     }
@@ -391,9 +382,6 @@ async fn timeline_create_handler(
                     format!("{err:#}")
                 ))
             }
-            Err(e @ tenant::CreateTimelineError::AncestorNotActive) => {
-                json_response(StatusCode::SERVICE_UNAVAILABLE, HttpErrorBody::from_msg(e.to_string()))
-            }
             Err(tenant::CreateTimelineError::Other(err)) => Err(ApiError::InternalServerError(err)),
         }
     }
@@ -570,14 +558,9 @@ async fn tenant_detach_handler(
 
     let state = get_state(&request);
     let conf = state.conf;
-    mgr::detach_tenant(
-        conf,
-        tenant_id,
-        detach_ignored.unwrap_or(false),
-        &state.deletion_queue_client,
-    )
-    .instrument(info_span!("tenant_detach", %tenant_id))
-    .await?;
+    mgr::detach_tenant(conf, tenant_id, detach_ignored.unwrap_or(false))
+        .instrument(info_span!("tenant_detach", %tenant_id))
+        .await?;
 
     json_response(StatusCode::OK, ())
 }
@@ -639,9 +622,8 @@ async fn tenant_list_handler(
     let response_data = mgr::list_tenants()
         .instrument(info_span!("tenant_list"))
         .await
-        .map_err(|_| {
-            ApiError::ResourceUnavailable("Tenant map is initializing or shutting down".into())
-        })?
+        .map_err(anyhow::Error::new)
+        .map_err(ApiError::InternalServerError)?
         .iter()
         .map(|(id, state)| TenantInfo {
             id: *id,
@@ -1019,48 +1001,6 @@ async fn update_tenant_config_handler(
     json_response(StatusCode::OK, ())
 }
 
-async fn put_tenant_location_config_handler(
-    mut request: Request<Body>,
-    _cancel: CancellationToken,
-) -> Result<Response<Body>, ApiError> {
-    let request_data: TenantLocationConfigRequest = json_request(&mut request).await?;
-    let tenant_id = request_data.tenant_id;
-    check_permission(&request, Some(tenant_id))?;
-
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
-    let state = get_state(&request);
-    let conf = state.conf;
-
-    // The `Detached` state is special, it doesn't upsert a tenant, it removes
-    // its local disk content and drops it from memory.
-    if let LocationConfigMode::Detached = request_data.config.mode {
-        mgr::detach_tenant(conf, tenant_id, true, &state.deletion_queue_client)
-            .instrument(info_span!("tenant_detach", %tenant_id))
-            .await?;
-        return json_response(StatusCode::OK, ());
-    }
-
-    let location_conf =
-        LocationConf::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
-
-    mgr::upsert_location(
-        state.conf,
-        tenant_id,
-        location_conf,
-        state.broker_client.clone(),
-        state.remote_storage.clone(),
-        state.deletion_queue_client.clone(),
-        &ctx,
-    )
-    .await
-    // TODO: badrequest assumes the caller was asking for something unreasonable, but in
-    // principle we might have hit something like concurrent API calls to the same tenant,
-    // which is not a 400 but a 409.
-    .map_err(ApiError::BadRequest)?;
-
-    json_response(StatusCode::OK, ())
-}
-
 /// Testing helper to transition a tenant to [`crate::tenant::TenantState::Broken`].
 async fn handle_tenant_break(
     r: Request<Body>,
@@ -1240,136 +1180,6 @@ async fn deletion_queue_flush(
     }
 }
 
-/// Try if `GetPage@Lsn` is successful, useful for manual debugging.
-async fn getpage_at_lsn_handler(
-    request: Request<Body>,
-    _cancel: CancellationToken,
-) -> Result<Response<Body>, ApiError> {
-    let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
-    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
-    check_permission(&request, Some(tenant_id))?;
-
-    struct Key(crate::repository::Key);
-
-    impl std::str::FromStr for Key {
-        type Err = anyhow::Error;
-
-        fn from_str(s: &str) -> std::result::Result<Self, Self::Err> {
-            crate::repository::Key::from_hex(s).map(Key)
-        }
-    }
-
-    let key: Key = parse_query_param(&request, "key")?
-        .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'key' query parameter")))?;
-    let lsn: Lsn = parse_query_param(&request, "lsn")?
-        .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'lsn' query parameter")))?;
-
-    async {
-        let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
-        let timeline = active_timeline_of_active_tenant(tenant_id, timeline_id).await?;
-
-        let page = timeline.get(key.0, lsn, &ctx).await?;
-
-        Result::<_, ApiError>::Ok(
-            Response::builder()
-                .status(StatusCode::OK)
-                .header(CONTENT_TYPE, "application/octet-stream")
-                .body(hyper::Body::from(page))
-                .unwrap(),
-        )
-    }
-    .instrument(info_span!("timeline_get", %tenant_id, %timeline_id))
-    .await
-}
-
-async fn timeline_collect_keyspace(
-    request: Request<Body>,
-    _cancel: CancellationToken,
-) -> Result<Response<Body>, ApiError> {
-    let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
-    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
-    check_permission(&request, Some(tenant_id))?;
-
-    struct Partitioning {
-        keys: crate::keyspace::KeySpace,
-
-        at_lsn: Lsn,
-    }
-
-    impl serde::Serialize for Partitioning {
-        fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
-        where
-            S: serde::Serializer,
-        {
-            use serde::ser::SerializeMap;
-            let mut map = serializer.serialize_map(Some(2))?;
-            map.serialize_key("keys")?;
-            map.serialize_value(&KeySpace(&self.keys))?;
-            map.serialize_key("at_lsn")?;
-            map.serialize_value(&WithDisplay(&self.at_lsn))?;
-            map.end()
-        }
-    }
-
-    struct WithDisplay<'a, T>(&'a T);
-
-    impl<'a, T: std::fmt::Display> serde::Serialize for WithDisplay<'a, T> {
-        fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
-        where
-            S: serde::Serializer,
-        {
-            serializer.collect_str(&self.0)
-        }
-    }
-
-    struct KeySpace<'a>(&'a crate::keyspace::KeySpace);
-
-    impl<'a> serde::Serialize for KeySpace<'a> {
-        fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
-        where
-            S: serde::Serializer,
-        {
-            use serde::ser::SerializeSeq;
-            let mut seq = serializer.serialize_seq(Some(self.0.ranges.len()))?;
-            for kr in &self.0.ranges {
-                seq.serialize_element(&KeyRange(kr))?;
-            }
-            seq.end()
-        }
-    }
-
-    struct KeyRange<'a>(&'a std::ops::Range<crate::repository::Key>);
-
-    impl<'a> serde::Serialize for KeyRange<'a> {
-        fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
-        where
-            S: serde::Serializer,
-        {
-            use serde::ser::SerializeTuple;
-            let mut t = serializer.serialize_tuple(2)?;
-            t.serialize_element(&WithDisplay(&self.0.start))?;
-            t.serialize_element(&WithDisplay(&self.0.end))?;
-            t.end()
-        }
-    }
-
-    let at_lsn: Option<Lsn> = parse_query_param(&request, "at_lsn")?;
-
-    async {
-        let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
-        let timeline = active_timeline_of_active_tenant(tenant_id, timeline_id).await?;
-        let at_lsn = at_lsn.unwrap_or_else(|| timeline.get_last_record_lsn());
-        let keys = timeline
-            .collect_keyspace(at_lsn, &ctx)
-            .await
-            .map_err(ApiError::InternalServerError)?;
-
-        json_response(StatusCode::OK, Partitioning { keys, at_lsn })
-    }
-    .instrument(info_span!("timeline_collect_keyspace", %tenant_id, %timeline_id))
-    .await
-}
-
 async fn active_timeline_of_active_tenant(
     tenant_id: TenantId,
     timeline_id: TimelineId,
@@ -1644,9 +1454,6 @@ pub fn make_router(
         .get("/v1/tenant/:tenant_id/config", |r| {
             api_handler(r, get_tenant_config_handler)
         })
-        .put("/v1/tenant/:tenant_id/location_config", |r| {
-            api_handler(r, put_tenant_location_config_handler)
-        })
         .get("/v1/tenant/:tenant_id/timeline", |r| {
             api_handler(r, timeline_list_handler)
         })
@@ -1717,12 +1524,5 @@ pub fn make_router(
         .post("/v1/tracing/event", |r| {
             testing_api_handler("emit a tracing event", r, post_tracing_event_handler)
         })
-        .get("/v1/tenant/:tenant_id/timeline/:timeline_id/getpage", |r| {
-            testing_api_handler("getpage@lsn", r, getpage_at_lsn_handler)
-        })
-        .get(
-            "/v1/tenant/:tenant_id/timeline/:timeline_id/keyspace",
-            |r| testing_api_handler("read out the keyspace", r, timeline_collect_keyspace),
-        )
         .any(handler_404))
 }

pageserver/src/import_datadir.rs

@@ -6,7 +6,6 @@ use std::path::{Path, PathBuf};
 
 use anyhow::{bail, ensure, Context, Result};
 use bytes::Bytes;
-use camino::Utf8Path;
 use futures::StreamExt;
 use tokio::io::{AsyncRead, AsyncReadExt};
 use tokio_tar::Archive;
@@ -30,7 +29,7 @@ use postgres_ffi::{BLCKSZ, WAL_SEGMENT_SIZE};
 use utils::lsn::Lsn;
 
 // Returns checkpoint LSN from controlfile
-pub fn get_lsn_from_controlfile(path: &Utf8Path) -> Result<Lsn> {
+pub fn get_lsn_from_controlfile(path: &Path) -> Result<Lsn> {
     // Read control file to extract the LSN
     let controlfile_path = path.join("global").join("pg_control");
     let controlfile = ControlFileData::decode(&std::fs::read(controlfile_path)?)?;
@@ -47,7 +46,7 @@ pub fn get_lsn_from_controlfile(path: &Utf8Path) -> Result<Lsn> {
 /// cluster was not shut down cleanly.
 pub async fn import_timeline_from_postgres_datadir(
     tline: &Timeline,
-    pgdata_path: &Utf8Path,
+    pgdata_path: &Path,
     pgdata_lsn: Lsn,
     ctx: &RequestContext,
 ) -> Result<()> {
@@ -257,7 +256,7 @@ async fn import_slru(
 /// Scan PostgreSQL WAL files in given directory and load all records between
 /// 'startpoint' and 'endpoint' into the repository.
 async fn import_wal(
-    walpath: &Utf8Path,
+    walpath: &Path,
     tline: &Timeline,
     startpoint: Lsn,
     endpoint: Lsn,

pageserver/src/lib.rs

@@ -25,8 +25,9 @@ pub mod walredo;
 
 pub mod failpoint_support;
 
+use std::path::Path;
+
 use crate::task_mgr::TaskKind;
-use camino::Utf8Path;
 use deletion_queue::DeletionQueue;
 use tracing::info;
 
@@ -112,10 +113,6 @@ pub const METADATA_FILE_NAME: &str = "metadata";
 /// Full path: `tenants/<tenant_id>/config`.
 pub const TENANT_CONFIG_NAME: &str = "config";
 
-/// Per-tenant configuration file.
-/// Full path: `tenants/<tenant_id>/config`.
-pub const TENANT_LOCATION_CONFIG_NAME: &str = "config-v1";
-
 /// A suffix used for various temporary files. Any temporary files found in the
 /// data directory at pageserver startup can be automatically removed.
 pub const TEMP_FILE_SUFFIX: &str = "___temp";
@@ -135,25 +132,25 @@ pub const TIMELINE_DELETE_MARK_SUFFIX: &str = "___delete";
 /// Full path: `tenants/<tenant_id>/___ignored_tenant`.
 pub const IGNORED_TENANT_FILE_NAME: &str = "___ignored_tenant";
 
-pub fn is_temporary(path: &Utf8Path) -> bool {
+pub fn is_temporary(path: &Path) -> bool {
     match path.file_name() {
-        Some(name) => name.ends_with(TEMP_FILE_SUFFIX),
+        Some(name) => name.to_string_lossy().ends_with(TEMP_FILE_SUFFIX),
         None => false,
     }
 }
 
-fn ends_with_suffix(path: &Utf8Path, suffix: &str) -> bool {
+fn ends_with_suffix(path: &Path, suffix: &str) -> bool {
     match path.file_name() {
-        Some(name) => name.ends_with(suffix),
+        Some(name) => name.to_string_lossy().ends_with(suffix),
         None => false,
     }
 }
 
-pub fn is_uninit_mark(path: &Utf8Path) -> bool {
+pub fn is_uninit_mark(path: &Path) -> bool {
     ends_with_suffix(path, TIMELINE_UNINIT_MARK_SUFFIX)
 }
 
-pub fn is_delete_mark(path: &Utf8Path) -> bool {
+pub fn is_delete_mark(path: &Path) -> bool {
     ends_with_suffix(path, TIMELINE_DELETE_MARK_SUFFIX)
 }
 
@@ -173,9 +170,6 @@ fn is_walkdir_io_not_found(e: &walkdir::Error) -> bool {
 /// delaying is needed.
 #[derive(Clone)]
 pub struct InitializationOrder {
-    /// Each initial tenant load task carries this until it is done loading timelines from remote storage
-    pub initial_tenant_load_remote: Option<utils::completion::Completion>,
-
     /// Each initial tenant load task carries this until completion.
     pub initial_tenant_load: Option<utils::completion::Completion>,
 

pageserver/src/metrics.rs

@@ -94,35 +94,15 @@ pub(crate) static READ_NUM_FS_LAYERS: Lazy<Histogram> = Lazy::new(|| {
 });
 
 // Metrics collected on operations on the storage repository.
-
-pub(crate) struct ReconstructTimeMetrics {
-    ok: Histogram,
-    err: Histogram,
-}
-
-pub(crate) static RECONSTRUCT_TIME: Lazy<ReconstructTimeMetrics> = Lazy::new(|| {
-    let inner = register_histogram_vec!(
+pub(crate) static RECONSTRUCT_TIME: Lazy<Histogram> = Lazy::new(|| {
+    register_histogram!(
         "pageserver_getpage_reconstruct_seconds",
         "Time spent in reconstruct_value (reconstruct a page from deltas)",
-        &["result"],
         CRITICAL_OP_BUCKETS.into(),
     )
-    .expect("failed to define a metric");
-    ReconstructTimeMetrics {
-        ok: inner.get_metric_with_label_values(&["ok"]).unwrap(),
-        err: inner.get_metric_with_label_values(&["err"]).unwrap(),
-    }
+    .expect("failed to define a metric")
 });
 
-impl ReconstructTimeMetrics {
-    pub(crate) fn for_result<T, E>(&self, result: &Result<T, E>) -> &Histogram {
-        match result {
-            Ok(_) => &self.ok,
-            Err(_) => &self.err,
-        }
-    }
-}
-
 pub(crate) static MATERIALIZED_PAGE_CACHE_HIT_DIRECT: Lazy<IntCounter> = Lazy::new(|| {
     register_int_counter!(
         "pageserver_materialized_cache_hits_direct_total",
@@ -691,9 +671,10 @@ impl StorageIoTime {
         .expect("failed to define a metric");
         let metrics = std::array::from_fn(|i| {
             let op = StorageIoOperation::from_repr(i).unwrap();
-            storage_io_histogram_vec
+            let metric = storage_io_histogram_vec
                 .get_metric_with_label_values(&[op.as_str()])
-                .unwrap()
+                .unwrap();
+            metric
         });
         Self { metrics }
     }
@@ -966,7 +947,6 @@ pub(crate) struct DeletionQueueMetrics {
     pub(crate) keys_submitted: IntCounter,
     pub(crate) keys_dropped: IntCounter,
     pub(crate) keys_executed: IntCounter,
-    pub(crate) keys_validated: IntCounter,
     pub(crate) dropped_lsn_updates: IntCounter,
     pub(crate) unexpected_errors: IntCounter,
     pub(crate) remote_errors: IntCounterVec,
@@ -988,13 +968,7 @@ pub(crate) static DELETION_QUEUE: Lazy<DeletionQueueMetrics> = Lazy::new(|| {
 
     keys_executed: register_int_counter!(
         "pageserver_deletion_queue_executed_total",
-        "Number of objects deleted. Only includes objects that we actually deleted, sum with pageserver_deletion_queue_dropped_total for the total number of keys processed to completion"
-    )
-    .expect("failed to define a metric"),
-
-    keys_validated: register_int_counter!(
-        "pageserver_deletion_queue_validated_total",
-        "Number of keys validated for deletion.  Sum with pageserver_deletion_queue_dropped_total for the total number of keys that have passed through the validation stage."
+        "Number of objects deleted. Only includes objects that we actually deleted, sum with pageserver_deletion_queue_dropped_total for the total number of keys processed."
     )
     .expect("failed to define a metric"),
 
@@ -1067,26 +1041,6 @@ pub(crate) static TENANT_TASK_EVENTS: Lazy<IntCounterVec> = Lazy::new(|| {
     .expect("Failed to register tenant_task_events metric")
 });
 
-pub(crate) static BACKGROUND_LOOP_SEMAPHORE_WAIT_START_COUNT: Lazy<IntCounterVec> =
-    Lazy::new(|| {
-        register_int_counter_vec!(
-            "pageserver_background_loop_semaphore_wait_start_count",
-            "Counter for background loop concurrency-limiting semaphore acquire calls started",
-            &["task"],
-        )
-        .unwrap()
-    });
-
-pub(crate) static BACKGROUND_LOOP_SEMAPHORE_WAIT_FINISH_COUNT: Lazy<IntCounterVec> =
-    Lazy::new(|| {
-        register_int_counter_vec!(
-            "pageserver_background_loop_semaphore_wait_finish_count",
-            "Counter for background loop concurrency-limiting semaphore acquire calls finished",
-            &["task"],
-        )
-        .unwrap()
-    });
-
 pub(crate) static BACKGROUND_LOOP_PERIOD_OVERRUN_COUNT: Lazy<IntCounterVec> = Lazy::new(|| {
     register_int_counter_vec!(
         "pageserver_background_loop_period_overrun_count",
@@ -1902,6 +1856,7 @@ pub fn preinitialize_metrics() {
     // histograms
     [
         &READ_NUM_FS_LAYERS,
+        &RECONSTRUCT_TIME,
         &WAIT_LSN_TIME,
         &WAL_REDO_TIME,
         &WAL_REDO_WAIT_TIME,
@@ -1912,7 +1867,4 @@ pub fn preinitialize_metrics() {
     .for_each(|h| {
         Lazy::force(h);
     });
-
-    // Custom
-    Lazy::force(&RECONSTRUCT_TIME);
 }

pageserver/src/page_cache.rs

@@ -66,7 +66,8 @@
 //! inserted to the mapping, but you must hold the write-lock on the slot until
 //! the contents are valid. If you need to release the lock without initializing
 //! the contents, you must remove the mapping first. We make that easy for the
-//! callers with PageWriteGuard: the caller must explicitly call guard.mark_valid() after it has
+//! callers with PageWriteGuard: when lock_for_write() returns an uninitialized
+//! page, the caller must explicitly call guard.mark_valid() after it has
 //! initialized it. If the guard is dropped without calling mark_valid(), the
 //! mapping is automatically removed and the slot is marked free.
 //!
@@ -78,6 +79,7 @@ use std::{
         atomic::{AtomicU64, AtomicU8, AtomicUsize, Ordering},
         Arc, Weak,
     },
+    task::Poll,
     time::Duration,
 };
 
@@ -251,6 +253,11 @@ pub struct PageCache {
     /// This is interpreted modulo the page cache size.
     next_evict_slot: AtomicUsize,
 
+    find_victim_sender:
+        async_channel::Sender<(usize, tokio::sync::RwLockWriteGuard<'static, SlotInner>)>,
+    find_victim_waiters:
+        async_channel::Receiver<(usize, tokio::sync::RwLockWriteGuard<'static, SlotInner>)>,
+
     size_metrics: &'static PageCacheSizeMetrics,
 }
 
@@ -285,7 +292,10 @@ impl AsRef<[u8; PAGE_SZ]> for PageReadGuard<'_> {
 ///
 /// Counterintuitively, this is used even for a read, if the requested page is not
 /// currently found in the page cache. In that case, the caller of lock_for_read()
-/// is expected to fill in the page contents and call mark_valid().
+/// is expected to fill in the page contents and call mark_valid(). Similarly
+/// lock_for_write() can return an invalid buffer that the caller is expected to
+/// to initialize.
+///
 pub struct PageWriteGuard<'i> {
     state: PageWriteGuardState<'i>,
 }
@@ -301,7 +311,7 @@ enum PageWriteGuardState<'i> {
 impl std::ops::DerefMut for PageWriteGuard<'_> {
     fn deref_mut(&mut self) -> &mut Self::Target {
         match &mut self.state {
-            PageWriteGuardState::Invalid { inner, _permit } => inner.buf,
+            PageWriteGuardState::Invalid { inner, _permit } => &mut inner.buf,
             PageWriteGuardState::Downgraded => unreachable!(),
         }
     }
@@ -312,7 +322,7 @@ impl std::ops::Deref for PageWriteGuard<'_> {
 
     fn deref(&self) -> &Self::Target {
         match &self.state {
-            PageWriteGuardState::Invalid { inner, _permit } => inner.buf,
+            PageWriteGuardState::Invalid { inner, _permit } => &inner.buf,
             PageWriteGuardState::Downgraded => unreachable!(),
         }
     }
@@ -321,8 +331,8 @@ impl std::ops::Deref for PageWriteGuard<'_> {
 impl AsMut<[u8; PAGE_SZ]> for PageWriteGuard<'_> {
     fn as_mut(&mut self) -> &mut [u8; PAGE_SZ] {
         match &mut self.state {
-            PageWriteGuardState::Invalid { inner, _permit } => inner.buf,
-            PageWriteGuardState::Downgraded => unreachable!(),
+            PageWriteGuardState::Invalid { inner, _permit } => &mut inner.buf,
+            PageWriteGuardState::Downgraded => todo!(),
         }
     }
 }
@@ -354,7 +364,6 @@ impl Drop for PageWriteGuard<'_> {
     fn drop(&mut self) {
         match &mut self.state {
             PageWriteGuardState::Invalid { inner, _permit } => {
-                assert!(inner.key.is_some());
                 let self_key = inner.key.as_ref().unwrap();
                 PAGE_CACHE.get().unwrap().remove_mapping(self_key);
                 inner.key = None;
@@ -370,6 +379,12 @@ pub enum ReadBufResult<'a> {
     NotFound(PageWriteGuard<'a>),
 }
 
+/// lock_for_write() return value
+pub enum WriteBufResult<'a> {
+    Found(PageReadGuard<'a>),
+    NotFound(PageWriteGuard<'a>),
+}
+
 impl PageCache {
     //
     // Section 1.1: Public interface functions for looking up and memorizing materialized page
@@ -440,7 +455,7 @@ impl PageCache {
     /// Store an image of the given page in the cache.
     ///
     pub async fn memorize_materialized_page(
-        &self,
+        &'static self,
         tenant_id: TenantId,
         timeline_id: TimelineId,
         key: Key,
@@ -456,83 +471,26 @@ impl PageCache {
             lsn,
         };
 
-        let mut permit = Some(self.try_get_pinned_slot_permit().await?);
-        loop {
-            // First check if the key already exists in the cache.
-            if let Some(slot_idx) = self.search_mapping_exact(&cache_key) {
-                // The page was found in the mapping. Lock the slot, and re-check
-                // that it's still what we expected (because we don't released the mapping
-                // lock already, another thread could have evicted the page)
-                let slot = &self.slots[slot_idx];
-                let inner = slot.inner.write().await;
-                if inner.key.as_ref() == Some(&cache_key) {
-                    slot.inc_usage_count();
-                    debug_assert!(
-                        {
-                            let guard = inner.permit.lock().unwrap();
-                            guard.upgrade().is_none()
-                        },
-                        "we hold a write lock, so, no one else should have a permit"
-                    );
-                    debug_assert_eq!(inner.buf.len(), img.len());
-                    // We already had it in cache. Another thread must've put it there
-                    // concurrently. Check that it had the same contents that we
-                    // replayed.
-                    assert!(inner.buf == img);
-                    return Ok(());
-                }
+        match self.lock_for_write(&cache_key).await? {
+            WriteBufResult::Found(read_guard) => {
+                // We already had it in cache. Another thread must've put it there
+                // concurrently. Check that it had the same contents that we
+                // replayed.
+                assert!(*read_guard == img);
             }
-            debug_assert!(permit.is_some());
-
-            // Not found. Find a victim buffer
-            let (slot_idx, mut inner) = self
-                .find_victim(permit.as_ref().unwrap())
-                .await
-                .context("Failed to find evict victim")?;
-
-            // Insert mapping for this. At this point, we may find that another
-            // thread did the same thing concurrently. In that case, we evicted
-            // our victim buffer unnecessarily. Put it into the free list and
-            // continue with the slot that the other thread chose.
-            if let Some(_existing_slot_idx) = self.try_insert_mapping(&cache_key, slot_idx) {
-                // TODO: put to free list
-
-                // We now just loop back to start from beginning. This is not
-                // optimal, we'll perform the lookup in the mapping again, which
-                // is not really necessary because we already got
-                // 'existing_slot_idx'.  But this shouldn't happen often enough
-                // to matter much.
-                continue;
+            WriteBufResult::NotFound(mut write_guard) => {
+                write_guard.copy_from_slice(img);
+                let _ = write_guard.mark_valid();
             }
-
-            // Make the slot ready
-            let slot = &self.slots[slot_idx];
-            inner.key = Some(cache_key.clone());
-            slot.set_usage_count(1);
-            // Create a write guard for the slot so we go through the expected motions.
-            debug_assert!(
-                {
-                    let guard = inner.permit.lock().unwrap();
-                    guard.upgrade().is_none()
-                },
-                "we hold a write lock, so, no one else should have a permit"
-            );
-            let mut write_guard = PageWriteGuard {
-                state: PageWriteGuardState::Invalid {
-                    _permit: permit.take().unwrap(),
-                    inner,
-                },
-            };
-            write_guard.copy_from_slice(img);
-            let _ = write_guard.mark_valid();
-            return Ok(());
         }
+
+        Ok(())
     }
 
     // Section 1.2: Public interface functions for working with immutable file pages.
 
     pub async fn read_immutable_buf(
-        &self,
+        &'static self,
         file_id: FileId,
         blkno: u32,
         ctx: &RequestContext,
@@ -551,26 +509,13 @@ impl PageCache {
     // not require changes.
 
     async fn try_get_pinned_slot_permit(&self) -> anyhow::Result<PinnedSlotsPermit> {
-        let timer = crate::metrics::PAGE_CACHE_ACQUIRE_PINNED_SLOT_TIME.start_timer();
-        match tokio::time::timeout(
-            // Choose small timeout, neon_smgr does its own retries.
-            // https://neondb.slack.com/archives/C04DGM6SMTM/p1694786876476869
-            Duration::from_secs(10),
-            Arc::clone(&self.pinned_slots).acquire_owned(),
-        )
-        .await
-        {
-            Ok(res) => Ok(PinnedSlotsPermit(
-                res.expect("this semaphore is never closed"),
-            )),
-            Err(_timeout) => {
-                timer.stop_and_discard();
-                crate::metrics::page_cache_errors_inc(
-                    crate::metrics::PageCacheErrorKind::AcquirePinnedSlotTimeout,
-                );
-                anyhow::bail!("timeout: there were page guards alive for all page cache slots")
-            }
-        }
+        let _timer = crate::metrics::PAGE_CACHE_ACQUIRE_PINNED_SLOT_TIME.start_timer();
+        Ok(PinnedSlotsPermit(
+            Arc::clone(&self.pinned_slots)
+                .acquire_owned()
+                .await
+                .unwrap(),
+        ))
     }
 
     /// Look up a page in the cache.
@@ -638,7 +583,7 @@ impl PageCache {
     /// ```
     ///
     async fn lock_for_read(
-        &self,
+        &'static self,
         cache_key: &mut CacheKey,
         ctx: &RequestContext,
     ) -> anyhow::Result<ReadBufResult> {
@@ -713,6 +658,86 @@ impl PageCache {
         }
     }
 
+    // FIXME: the name is wrong.
+    async fn try_lock_for_write(
+        &self,
+        cache_key: &CacheKey,
+        permit: &mut Option<PinnedSlotsPermit>,
+    ) -> Option<PageReadGuard> {
+        if let Some(slot_idx) = self.search_mapping_for_write(cache_key) {
+            // The page was found in the mapping. Lock the slot, and re-check
+            // that it's still what we expected (because we don't released the mapping
+            // lock already, another thread could have evicted the page)
+            let slot = &self.slots[slot_idx];
+            let inner = slot.inner.read().await;
+            if inner.key.as_ref() == Some(cache_key) {
+                slot.inc_usage_count();
+                return Some(PageReadGuard {
+                    _permit: inner.coalesce_readers_permit(permit.take().unwrap()),
+                    slot_guard: inner,
+                });
+            }
+        }
+        None
+    }
+
+    /// Return a write-locked buffer for given block.
+    ///
+    /// Similar to lock_for_read(), but the returned buffer is write-locked and
+    /// may be modified by the caller even if it's already found in the cache.
+    async fn lock_for_write(&'static self, cache_key: &CacheKey) -> anyhow::Result<WriteBufResult> {
+        let mut permit = Some(self.try_get_pinned_slot_permit().await?);
+        loop {
+            // First check if the key already exists in the cache.
+            if let Some(write_guard) = self.try_lock_for_write(cache_key, &mut permit).await {
+                debug_assert!(permit.is_none());
+                return Ok(WriteBufResult::Found(write_guard));
+            }
+            debug_assert!(permit.is_some());
+
+            // Not found. Find a victim buffer
+            let (slot_idx, mut inner) = self
+                .find_victim(permit.as_ref().unwrap())
+                .await
+                .context("Failed to find evict victim")?;
+
+            // Insert mapping for this. At this point, we may find that another
+            // thread did the same thing concurrently. In that case, we evicted
+            // our victim buffer unnecessarily. Put it into the free list and
+            // continue with the slot that the other thread chose.
+            if let Some(_existing_slot_idx) = self.try_insert_mapping(cache_key, slot_idx) {
+                // TODO: put to free list
+
+                // We now just loop back to start from beginning. This is not
+                // optimal, we'll perform the lookup in the mapping again, which
+                // is not really necessary because we already got
+                // 'existing_slot_idx'.  But this shouldn't happen often enough
+                // to matter much.
+                continue;
+            }
+
+            // Make the slot ready
+            let slot = &self.slots[slot_idx];
+            inner.key = Some(cache_key.clone());
+            slot.set_usage_count(1);
+
+            debug_assert!(
+                {
+                    let guard = inner.permit.lock().unwrap();
+                    guard.upgrade().is_none()
+                },
+                "we hold a write lock, so, no one else should have a permit"
+            );
+
+            return Ok(WriteBufResult::NotFound(PageWriteGuard {
+                state: PageWriteGuardState::Invalid {
+                    _permit: permit.take().unwrap(),
+                    inner,
+                },
+            }));
+        }
+    }
+
     //
     // Section 3: Mapping functions
     //
@@ -753,7 +778,7 @@ impl PageCache {
     ///
     /// Like 'search_mapping, but performs an "exact" search. Used for
     /// allocating a new buffer.
-    fn search_mapping_exact(&self, key: &CacheKey) -> Option<usize> {
+    fn search_mapping_for_write(&self, key: &CacheKey) -> Option<usize> {
         match key {
             CacheKey::MaterializedPage { hash_key, lsn } => {
                 let map = self.materialized_page_map.read().unwrap();
@@ -860,10 +885,20 @@ impl PageCache {
     ///
     /// On return, the slot is empty and write-locked.
     async fn find_victim(
-        &self,
+        &'static self,
         _permit_witness: &PinnedSlotsPermit,
     ) -> anyhow::Result<(usize, tokio::sync::RwLockWriteGuard<SlotInner>)> {
-        let iter_limit = self.slots.len() * 10;
+        // Get in line.
+        let mut receiver = self.find_victim_waiters.recv();
+        // If we get cancelled at the receiver.await below, the victim slot
+        // remains in the channel. Consume these first before going into
+        // the loop below.
+        match futures::poll!(&mut receiver) {
+            Poll::Ready(Ok(res)) => return Ok(res),
+            Poll::Ready(Err(_closed)) => unreachable!("we never close the channel"),
+            Poll::Pending => {} // the regular case where we aren't cancelled below
+        };
+
         let mut iters = 0;
         loop {
             iters += 1;
@@ -875,41 +910,8 @@ impl PageCache {
                 let mut inner = match slot.inner.try_write() {
                     Ok(inner) => inner,
                     Err(_err) => {
-                        if iters > iter_limit {
-                            // NB: Even with the permits, there's no hard guarantee that we will find a slot with
-                            // any particular number of iterations: other threads might race ahead and acquire and
-                            // release pins just as we're scanning the array.
-                            //
-                            // Imagine that nslots is 2, and as starting point, usage_count==1 on all
-                            // slots. There are two threads running concurrently, A and B. A has just
-                            // acquired the permit from the semaphore.
-                            //
-                            //   A: Look at slot 1. Its usage_count == 1, so decrement it to zero, and continue the search
-                            //   B: Acquire permit.
-                            //   B: Look at slot 2, decrement its usage_count to zero and continue the search
-                            //   B: Look at slot 1. Its usage_count is zero, so pin it and bump up its usage_count to 1.
-                            //   B: Release pin and permit again
-                            //   B: Acquire permit.
-                            //   B: Look at slot 2. Its usage_count is zero, so pin it and bump up its usage_count to 1.
-                            //   B: Release pin and permit again
-                            //
-                            // Now we're back in the starting situation that both slots have
-                            // usage_count 1, but A has now been through one iteration of the
-                            // find_victim() loop. This can repeat indefinitely and on each
-                            // iteration, A's iteration count increases by one.
-                            //
-                            // So, even though the semaphore for the permits is fair, the victim search
-                            // itself happens in parallel and is not fair.
-                            // Hence even with a permit, a task can theoretically be starved.
-                            // To avoid this, we'd need tokio to give priority to tasks that are holding
-                            // permits for longer.
-                            // Note that just yielding to tokio during iteration without such
-                            // priority boosting is likely counter-productive. We'd just give more opportunities
-                            // for B to bump usage count, further starving A.
-                            crate::metrics::page_cache_errors_inc(
-                                crate::metrics::PageCacheErrorKind::EvictIterLimit,
-                            );
-                            anyhow::bail!("exceeded evict iter limit");
+                        if iters > self.slots.len() * (MAX_USAGE_COUNT as usize) {
+                            unreachable!("find_victim_waiters prevents starvation");
                         }
                         continue;
                     }
@@ -920,7 +922,10 @@ impl PageCache {
                     inner.key = None;
                 }
                 crate::metrics::PAGE_CACHE_FIND_VICTIMS_ITERS_TOTAL.inc_by(iters as u64);
-                return Ok((slot_idx, inner));
+                self.find_victim_sender
+                    .try_send((slot_idx, inner))
+                    .expect("we always get in line first");
+                return Ok(receiver.await.unwrap());
             }
         }
     }
@@ -957,6 +962,7 @@ impl PageCache {
             })
             .collect();
 
+        let (find_victim_sender, find_victim_waiters) = async_channel::bounded(num_pages);
         Self {
             materialized_page_map: Default::default(),
             immutable_page_map: Default::default(),
@@ -964,6 +970,8 @@ impl PageCache {
             next_evict_slot: AtomicUsize::new(0),
             size_metrics,
             pinned_slots: Arc::new(tokio::sync::Semaphore::new(num_pages)),
+            find_victim_sender,
+            find_victim_waiters,
         }
     }
 }

pageserver/src/page_service.rs

@@ -35,7 +35,6 @@ use std::time::Duration;
 use tokio::io::AsyncWriteExt;
 use tokio::io::{AsyncRead, AsyncWrite};
 use tokio_util::io::StreamReader;
-use tokio_util::sync::CancellationToken;
 use tracing::field;
 use tracing::*;
 use utils::id::ConnectionId;
@@ -65,6 +64,69 @@ use crate::trace::Tracer;
 use postgres_ffi::pg_constants::DEFAULTTABLESPACE_OID;
 use postgres_ffi::BLCKSZ;
 
+fn copyin_stream<IO>(pgb: &mut PostgresBackend<IO>) -> impl Stream<Item = io::Result<Bytes>> + '_
+where
+    IO: AsyncRead + AsyncWrite + Unpin,
+{
+    async_stream::try_stream! {
+        loop {
+            let msg = tokio::select! {
+                biased;
+
+                _ = task_mgr::shutdown_watcher() => {
+                    // We were requested to shut down.
+                    let msg = "pageserver is shutting down";
+                    let _ = pgb.write_message_noflush(&BeMessage::ErrorResponse(msg, None));
+                    Err(QueryError::Other(anyhow::anyhow!(msg)))
+                }
+
+                msg = pgb.read_message() => { msg.map_err(QueryError::from)}
+            };
+
+            match msg {
+                Ok(Some(message)) => {
+                    let copy_data_bytes = match message {
+                        FeMessage::CopyData(bytes) => bytes,
+                        FeMessage::CopyDone => { break },
+                        FeMessage::Sync => continue,
+                        FeMessage::Terminate => {
+                            let msg = "client terminated connection with Terminate message during COPY";
+                            let query_error = QueryError::Disconnected(ConnectionError::Io(io::Error::new(io::ErrorKind::ConnectionReset, msg)));
+                            // error can't happen here, ErrorResponse serialization should be always ok
+                            pgb.write_message_noflush(&BeMessage::ErrorResponse(msg, Some(query_error.pg_error_code()))).map_err(|e| e.into_io_error())?;
+                            Err(io::Error::new(io::ErrorKind::ConnectionReset, msg))?;
+                            break;
+                        }
+                        m => {
+                            let msg = format!("unexpected message {m:?}");
+                            // error can't happen here, ErrorResponse serialization should be always ok
+                            pgb.write_message_noflush(&BeMessage::ErrorResponse(&msg, None)).map_err(|e| e.into_io_error())?;
+                            Err(io::Error::new(io::ErrorKind::Other, msg))?;
+                            break;
+                        }
+                    };
+
+                    yield copy_data_bytes;
+                }
+                Ok(None) => {
+                    let msg = "client closed connection during COPY";
+                    let query_error = QueryError::Disconnected(ConnectionError::Io(io::Error::new(io::ErrorKind::ConnectionReset, msg)));
+                    // error can't happen here, ErrorResponse serialization should be always ok
+                    pgb.write_message_noflush(&BeMessage::ErrorResponse(msg, Some(query_error.pg_error_code()))).map_err(|e| e.into_io_error())?;
+                    pgb.flush().await?;
+                    Err(io::Error::new(io::ErrorKind::ConnectionReset, msg))?;
+                }
+                Err(QueryError::Disconnected(ConnectionError::Io(io_error))) => {
+                    Err(io_error)?;
+                }
+                Err(other) => {
+                    Err(io::Error::new(io::ErrorKind::Other, other.to_string()))?;
+                }
+            };
+        }
+    }
+}
+
 /// Read the end of a tar archive.
 ///
 /// A tar archive normally ends with two consecutive blocks of zeros, 512 bytes each.
@@ -222,13 +284,7 @@ async fn page_service_conn_main(
     // and create a child per-query context when it invokes process_query.
     // But it's in a shared crate, so, we store connection_ctx inside PageServerHandler
     // and create the per-query context in process_query ourselves.
-    let mut conn_handler = PageServerHandler::new(
-        conf,
-        broker_client,
-        auth,
-        connection_ctx,
-        task_mgr::shutdown_token(),
-    );
+    let mut conn_handler = PageServerHandler::new(conf, broker_client, auth, connection_ctx);
     let pgbackend = PostgresBackend::new_from_io(socket, peer_addr, auth_type, None)?;
 
     match pgbackend
@@ -262,10 +318,6 @@ struct PageServerHandler {
     /// For each query received over the connection,
     /// `process_query` creates a child context from this one.
     connection_ctx: RequestContext,
-
-    /// A token that should fire when the tenant transitions from
-    /// attached state, or when the pageserver is shutting down.
-    cancel: CancellationToken,
 }
 
 impl PageServerHandler {
@@ -274,7 +326,6 @@ impl PageServerHandler {
         broker_client: storage_broker::BrokerClientChannel,
         auth: Option<Arc<JwtAuth>>,
         connection_ctx: RequestContext,
-        cancel: CancellationToken,
     ) -> Self {
         PageServerHandler {
             _conf: conf,
@@ -282,91 +333,6 @@ impl PageServerHandler {
             auth,
             claims: None,
             connection_ctx,
-            cancel,
-        }
-    }
-
-    /// Wrap PostgresBackend::flush to respect our CancellationToken: it is important to use
-    /// this rather than naked flush() in order to shut down promptly.  Without this, we would
-    /// block shutdown of a tenant if a postgres client was failing to consume bytes we send
-    /// in the flush.
-    async fn flush_cancellable<IO>(&self, pgb: &mut PostgresBackend<IO>) -> Result<(), QueryError>
-    where
-        IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
-    {
-        tokio::select!(
-            flush_r = pgb.flush() => {
-                Ok(flush_r?)
-            },
-            _ = self.cancel.cancelled() => {
-                Err(QueryError::Other(anyhow::anyhow!("Shutting down")))
-            }
-        )
-    }
-
-    fn copyin_stream<'a, IO>(
-        &'a self,
-        pgb: &'a mut PostgresBackend<IO>,
-    ) -> impl Stream<Item = io::Result<Bytes>> + 'a
-    where
-        IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
-    {
-        async_stream::try_stream! {
-            loop {
-                let msg = tokio::select! {
-                    biased;
-
-                    _ = task_mgr::shutdown_watcher() => {
-                        // We were requested to shut down.
-                        let msg = "pageserver is shutting down";
-                        let _ = pgb.write_message_noflush(&BeMessage::ErrorResponse(msg, None));
-                        Err(QueryError::Other(anyhow::anyhow!(msg)))
-                    }
-
-                    msg = pgb.read_message() => { msg.map_err(QueryError::from)}
-                };
-
-                match msg {
-                    Ok(Some(message)) => {
-                        let copy_data_bytes = match message {
-                            FeMessage::CopyData(bytes) => bytes,
-                            FeMessage::CopyDone => { break },
-                            FeMessage::Sync => continue,
-                            FeMessage::Terminate => {
-                                let msg = "client terminated connection with Terminate message during COPY";
-                                let query_error = QueryError::Disconnected(ConnectionError::Io(io::Error::new(io::ErrorKind::ConnectionReset, msg)));
-                                // error can't happen here, ErrorResponse serialization should be always ok
-                                pgb.write_message_noflush(&BeMessage::ErrorResponse(msg, Some(query_error.pg_error_code()))).map_err(|e| e.into_io_error())?;
-                                Err(io::Error::new(io::ErrorKind::ConnectionReset, msg))?;
-                                break;
-                            }
-                            m => {
-                                let msg = format!("unexpected message {m:?}");
-                                // error can't happen here, ErrorResponse serialization should be always ok
-                                pgb.write_message_noflush(&BeMessage::ErrorResponse(&msg, None)).map_err(|e| e.into_io_error())?;
-                                Err(io::Error::new(io::ErrorKind::Other, msg))?;
-                                break;
-                            }
-                        };
-
-                        yield copy_data_bytes;
-                    }
-                    Ok(None) => {
-                        let msg = "client closed connection during COPY";
-                        let query_error = QueryError::Disconnected(ConnectionError::Io(io::Error::new(io::ErrorKind::ConnectionReset, msg)));
-                        // error can't happen here, ErrorResponse serialization should be always ok
-                        pgb.write_message_noflush(&BeMessage::ErrorResponse(msg, Some(query_error.pg_error_code()))).map_err(|e| e.into_io_error())?;
-                        self.flush_cancellable(pgb).await.map_err(|e| io::Error::new(io::ErrorKind::Other, e.to_string()))?;
-                        Err(io::Error::new(io::ErrorKind::ConnectionReset, msg))?;
-                    }
-                    Err(QueryError::Disconnected(ConnectionError::Io(io_error))) => {
-                        Err(io_error)?;
-                    }
-                    Err(other) => {
-                        Err(io::Error::new(io::ErrorKind::Other, other.to_string()))?;
-                    }
-                };
-            }
         }
     }
 
@@ -406,7 +372,7 @@ impl PageServerHandler {
 
         // switch client to COPYBOTH
         pgb.write_message_noflush(&BeMessage::CopyBothResponse)?;
-        self.flush_cancellable(pgb).await?;
+        pgb.flush().await?;
 
         let metrics = metrics::SmgrQueryTimePerTimeline::new(&tenant_id, &timeline_id);
 
@@ -446,60 +412,38 @@ impl PageServerHandler {
             // TODO: We could create a new per-request context here, with unique ID.
             // Currently we use the same per-timeline context for all requests
 
-            let (response, span) = match neon_fe_msg {
+            let response = match neon_fe_msg {
                 PagestreamFeMessage::Exists(req) => {
                     let _timer = metrics.start_timer(metrics::SmgrQueryType::GetRelExists);
-                    let span = tracing::info_span!("handle_get_rel_exists_request", rel = %req.rel, req_lsn = %req.lsn);
-                    (
-                        self.handle_get_rel_exists_request(&timeline, &req, &ctx)
-                            .instrument(span.clone())
-                            .await,
-                        span,
-                    )
+                    self.handle_get_rel_exists_request(&timeline, &req, &ctx)
+                        .await
                 }
                 PagestreamFeMessage::Nblocks(req) => {
                     let _timer = metrics.start_timer(metrics::SmgrQueryType::GetRelSize);
-                    let span = tracing::info_span!("handle_get_nblocks_request", rel = %req.rel, req_lsn = %req.lsn);
-                    (
-                        self.handle_get_nblocks_request(&timeline, &req, &ctx)
-                            .instrument(span.clone())
-                            .await,
-                        span,
-                    )
+                    self.handle_get_nblocks_request(&timeline, &req, &ctx).await
                 }
                 PagestreamFeMessage::GetPage(req) => {
                     let _timer = metrics.start_timer(metrics::SmgrQueryType::GetPageAtLsn);
-                    let span = tracing::info_span!("handle_get_page_at_lsn_request", rel = %req.rel, blkno = %req.blkno, req_lsn = %req.lsn);
-                    (
-                        self.handle_get_page_at_lsn_request(&timeline, &req, &ctx)
-                            .instrument(span.clone())
-                            .await,
-                        span,
-                    )
+                    self.handle_get_page_at_lsn_request(&timeline, &req, &ctx)
+                        .await
                 }
                 PagestreamFeMessage::DbSize(req) => {
                     let _timer = metrics.start_timer(metrics::SmgrQueryType::GetDbSize);
-                    let span = tracing::info_span!("handle_db_size_request", dbnode = %req.dbnode, req_lsn = %req.lsn);
-                    (
-                        self.handle_db_size_request(&timeline, &req, &ctx)
-                            .instrument(span.clone())
-                            .await,
-                        span,
-                    )
+                    self.handle_db_size_request(&timeline, &req, &ctx).await
                 }
             };
 
             let response = response.unwrap_or_else(|e| {
                 // print the all details to the log with {:#}, but for the client the
                 // error message is enough
-                span.in_scope(|| error!("error reading relation or page version: {:#}", e));
+                error!("error reading relation or page version: {:?}", e);
                 PagestreamBeMessage::Error(PagestreamErrorResponse {
                     message: e.to_string(),
                 })
             });
 
             pgb.write_message_noflush(&BeMessage::CopyData(&response.serialize()))?;
-            self.flush_cancellable(pgb).await?;
+            pgb.flush().await?;
         }
         Ok(())
     }
@@ -542,9 +486,9 @@ impl PageServerHandler {
         // Import basebackup provided via CopyData
         info!("importing basebackup");
         pgb.write_message_noflush(&BeMessage::CopyInResponse)?;
-        self.flush_cancellable(pgb).await?;
+        pgb.flush().await?;
 
-        let mut copyin_reader = pin!(StreamReader::new(self.copyin_stream(pgb)));
+        let mut copyin_reader = pin!(StreamReader::new(copyin_stream(pgb)));
         timeline
             .import_basebackup_from_tar(
                 &mut copyin_reader,
@@ -597,8 +541,8 @@ impl PageServerHandler {
         // Import wal provided via CopyData
         info!("importing wal");
         pgb.write_message_noflush(&BeMessage::CopyInResponse)?;
-        self.flush_cancellable(pgb).await?;
-        let mut copyin_reader = pin!(StreamReader::new(self.copyin_stream(pgb)));
+        pgb.flush().await?;
+        let mut copyin_reader = pin!(StreamReader::new(copyin_stream(pgb)));
         import_wal_from_tar(&timeline, &mut copyin_reader, start_lsn, end_lsn, &ctx).await?;
         info!("wal import complete");
 
@@ -683,6 +627,7 @@ impl PageServerHandler {
         Ok(lsn)
     }
 
+    #[instrument(skip(self, timeline, req, ctx), fields(rel = %req.rel, req_lsn = %req.lsn))]
     async fn handle_get_rel_exists_request(
         &self,
         timeline: &Timeline,
@@ -703,6 +648,7 @@ impl PageServerHandler {
         }))
     }
 
+    #[instrument(skip(self, timeline, req, ctx), fields(rel = %req.rel, req_lsn = %req.lsn))]
     async fn handle_get_nblocks_request(
         &self,
         timeline: &Timeline,
@@ -721,6 +667,7 @@ impl PageServerHandler {
         }))
     }
 
+    #[instrument(skip(self, timeline, req, ctx), fields(dbnode = %req.dbnode, req_lsn = %req.lsn))]
     async fn handle_db_size_request(
         &self,
         timeline: &Timeline,
@@ -742,6 +689,7 @@ impl PageServerHandler {
         }))
     }
 
+    #[instrument(skip(self, timeline, req, ctx), fields(rel = %req.rel, blkno = %req.blkno, req_lsn = %req.lsn))]
     async fn handle_get_page_at_lsn_request(
         &self,
         timeline: &Timeline,
@@ -806,7 +754,7 @@ impl PageServerHandler {
 
         // switch client to COPYOUT
         pgb.write_message_noflush(&BeMessage::CopyOutResponse)?;
-        self.flush_cancellable(pgb).await?;
+        pgb.flush().await?;
 
         // Send a tarball of the latest layer on the timeline. Compress if not
         // fullbackup. TODO Compress in that case too (tests need to be updated)
@@ -858,7 +806,7 @@ impl PageServerHandler {
         }
 
         pgb.write_message_noflush(&BeMessage::CopyDone)?;
-        self.flush_cancellable(pgb).await?;
+        pgb.flush().await?;
 
         let basebackup_after = started
             .elapsed()
@@ -1317,10 +1265,7 @@ async fn get_active_tenant_with_timeout(
         Ok(tenant) => tenant,
         Err(e @ GetTenantError::NotFound(_)) => return Err(GetActiveTenantError::NotFound(e)),
         Err(GetTenantError::NotActive(_)) => {
-            unreachable!("we're calling get_tenant with active_only=false")
-        }
-        Err(GetTenantError::Broken(_)) => {
-            unreachable!("we're calling get_tenant with active_only=false")
+            unreachable!("we're calling get_tenant with active=false")
         }
     };
     let wait_time = Duration::from_secs(30);

pageserver/src/statvfs.rs

@@ -1,6 +1,6 @@
 //! Wrapper around nix::sys::statvfs::Statvfs that allows for mocking.
 
-use camino::Utf8Path;
+use std::path::Path;
 
 pub enum Statvfs {
     Real(nix::sys::statvfs::Statvfs),
@@ -12,13 +12,11 @@ pub enum Statvfs {
 // Sincce it should only be a problem on > 2TiB disks, let's ignore
 // the problem for now and upcast to u64.
 impl Statvfs {
-    pub fn get(tenants_dir: &Utf8Path, mocked: Option<&mock::Behavior>) -> nix::Result<Self> {
+    pub fn get(tenants_dir: &Path, mocked: Option<&mock::Behavior>) -> nix::Result<Self> {
         if let Some(mocked) = mocked {
             Ok(Statvfs::Mock(mock::get(tenants_dir, mocked)?))
         } else {
-            Ok(Statvfs::Real(nix::sys::statvfs::statvfs(
-                tenants_dir.as_std_path(),
-            )?))
+            Ok(Statvfs::Real(nix::sys::statvfs::statvfs(tenants_dir)?))
         }
     }
 
@@ -57,8 +55,8 @@ impl Statvfs {
 
 pub mod mock {
     use anyhow::Context;
-    use camino::Utf8Path;
     use regex::Regex;
+    use std::path::Path;
     use tracing::log::info;
 
     #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
@@ -88,7 +86,7 @@ pub mod mock {
         }
     }
 
-    pub fn get(tenants_dir: &Utf8Path, behavior: &Behavior) -> nix::Result<Statvfs> {
+    pub fn get(tenants_dir: &Path, behavior: &Behavior) -> nix::Result<Statvfs> {
         info!("running mocked statvfs");
 
         match behavior {
@@ -121,7 +119,7 @@ pub mod mock {
         }
     }
 
-    fn walk_dir_disk_usage(path: &Utf8Path, name_filter: Option<&Regex>) -> anyhow::Result<u64> {
+    fn walk_dir_disk_usage(path: &Path, name_filter: Option<&Regex>) -> anyhow::Result<u64> {
         let mut total = 0;
         for entry in walkdir::WalkDir::new(path) {
             let entry = entry?;

pageserver/src/task_mgr.rs

@@ -293,6 +293,8 @@ pub enum TaskKind {
 
     DebugTool,
 
+    BackgroundRuntimeTurnaroundMeasure,
+
     #[cfg(test)]
     UnitTest,
 }

pageserver/src/tenant/blob_io.rs

@@ -238,14 +238,14 @@ mod tests {
     use rand::{Rng, SeedableRng};
 
     async fn round_trip_test<const BUFFERED: bool>(blobs: &[Vec<u8>]) -> Result<(), Error> {
-        let temp_dir = camino_tempfile::tempdir()?;
-        let pathbuf = temp_dir.path().join("file");
+        let temp_dir = tempfile::tempdir()?;
+        let path = temp_dir.path().join("file");
         let ctx = RequestContext::new(TaskKind::UnitTest, DownloadBehavior::Error);
 
         // Write part (in block to drop the file)
         let mut offsets = Vec::new();
         {
-            let file = VirtualFile::create(pathbuf.as_path()).await?;
+            let file = VirtualFile::create(&path).await?;
             let mut wtr = BlobWriter::<BUFFERED>::new(file, 0);
             for blob in blobs.iter() {
                 let offs = wtr.write_blob(blob).await?;
@@ -258,7 +258,7 @@ mod tests {
             wtr.flush_buffer().await?;
         }
 
-        let file = VirtualFile::open(pathbuf.as_path()).await?;
+        let file = VirtualFile::open(&path).await?;
         let rdr = BlockReaderRef::VirtualFile(&file);
         let rdr = BlockCursor::new(rdr);
         for (idx, (blob, offset)) in blobs.iter().zip(offsets.iter()).enumerate() {

pageserver/src/tenant/block_io.rs

@@ -195,13 +195,13 @@ impl FileBlockReader {
                     format!("Failed to read immutable buf: {e:#}"),
                 )
             })? {
-            ReadBufResult::Found(guard) => Ok(guard.into()),
+            ReadBufResult::Found(guard) => return Ok(guard.into()),
             ReadBufResult::NotFound(mut write_guard) => {
                 // Read the page from disk into the buffer
                 self.fill_buffer(write_guard.deref_mut(), blknum).await?;
-                Ok(write_guard.mark_valid().into())
+                return Ok(write_guard.mark_valid().into());
             }
-        }
+        };
     }
 }
 

pageserver/src/tenant/config.rs

@@ -13,7 +13,6 @@ use pageserver_api::models;
 use serde::{Deserialize, Serialize};
 use std::num::NonZeroU64;
 use std::time::Duration;
-use utils::generation::Generation;
 
 pub mod defaults {
     // FIXME: This current value is very low. I would imagine something like 1 GB or 10 GB
@@ -45,211 +44,7 @@ pub mod defaults {
     pub const DEFAULT_EVICTIONS_LOW_RESIDENCE_DURATION_METRIC_THRESHOLD: &str = "24 hour";
 }
 
-#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
-pub(crate) enum AttachmentMode {
-    /// Our generation is current as far as we know, and as far as we know we are the only attached
-    /// pageserver.  This is the "normal" attachment mode.
-    Single,
-    /// Our generation number is current as far as we know, but we are advised that another
-    /// pageserver is still attached, and therefore to avoid executing deletions.   This is
-    /// the attachment mode of a pagesever that is the destination of a migration.
-    Multi,
-    /// Our generation number is superseded, or about to be superseded.  We are advised
-    /// to avoid remote storage writes if possible, and to avoid sending billing data.  This
-    /// is the attachment mode of a pageserver that is the origin of a migration.
-    Stale,
-}
-
-#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
-pub(crate) struct AttachedLocationConfig {
-    pub(crate) generation: Generation,
-    pub(crate) attach_mode: AttachmentMode,
-    // TODO: add a flag to override AttachmentMode's policies under
-    // disk pressure (i.e. unblock uploads under disk pressure in Stale
-    // state, unblock deletions after timeout in Multi state)
-}
-
-#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
-pub(crate) struct SecondaryLocationConfig {
-    /// If true, keep the local cache warm by polling remote storage
-    pub(crate) warm: bool,
-}
-
-#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
-pub(crate) enum LocationMode {
-    Attached(AttachedLocationConfig),
-    Secondary(SecondaryLocationConfig),
-}
-
-/// Per-tenant, per-pageserver configuration.  All pageservers use the same TenantConf,
-/// but have distinct LocationConf.
-#[derive(Clone, PartialEq, Eq, Serialize, Deserialize)]
-pub(crate) struct LocationConf {
-    /// The location-specific part of the configuration, describes the operating
-    /// mode of this pageserver for this tenant.
-    pub(crate) mode: LocationMode,
-    /// The pan-cluster tenant configuration, the same on all locations
-    pub(crate) tenant_conf: TenantConfOpt,
-}
-
-impl std::fmt::Debug for LocationConf {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        match &self.mode {
-            LocationMode::Attached(conf) => {
-                write!(
-                    f,
-                    "Attached {:?}, gen={:?}",
-                    conf.attach_mode, conf.generation
-                )
-            }
-            LocationMode::Secondary(conf) => {
-                write!(f, "Secondary, warm={}", conf.warm)
-            }
-        }
-    }
-}
-
-impl AttachedLocationConfig {
-    /// Consult attachment mode to determine whether we are currently permitted
-    /// to delete layers.  This is only advisory, not required for data safety.
-    /// See [`AttachmentMode`] for more context.
-    pub(crate) fn may_delete_layers_hint(&self) -> bool {
-        // TODO: add an override for disk pressure in AttachedLocationConfig,
-        // and respect it here.
-        match &self.attach_mode {
-            AttachmentMode::Single => true,
-            AttachmentMode::Multi | AttachmentMode::Stale => {
-                // In Multi mode we avoid doing deletions because some other
-                // attached pageserver might get 404 while trying to read
-                // a layer we delete which is still referenced in their metadata.
-                //
-                // In Stale mode, we avoid doing deletions because we expect
-                // that they would ultimately fail validation in the deletion
-                // queue due to our stale generation.
-                false
-            }
-        }
-    }
-
-    /// Whether we are currently hinted that it is worthwhile to upload layers.
-    /// This is only advisory, not required for data safety.
-    /// See [`AttachmentMode`] for more context.
-    pub(crate) fn may_upload_layers_hint(&self) -> bool {
-        // TODO: add an override for disk pressure in AttachedLocationConfig,
-        // and respect it here.
-        match &self.attach_mode {
-            AttachmentMode::Single | AttachmentMode::Multi => true,
-            AttachmentMode::Stale => {
-                // In Stale mode, we avoid doing uploads because we expect that
-                // our replacement pageserver will already have started its own
-                // IndexPart that will never reference layers we upload: it is
-                // wasteful.
-                false
-            }
-        }
-    }
-}
-
-impl LocationConf {
-    /// For use when loading from a legacy configuration: presence of a tenant
-    /// implies it is in AttachmentMode::Single, which used to be the only
-    /// possible state.  This function should eventually be removed.
-    pub(crate) fn attached_single(tenant_conf: TenantConfOpt, generation: Generation) -> Self {
-        Self {
-            mode: LocationMode::Attached(AttachedLocationConfig {
-                generation,
-                attach_mode: AttachmentMode::Single,
-            }),
-            tenant_conf,
-        }
-    }
-
-    /// For use when attaching/re-attaching: update the generation stored in this
-    /// structure.  If we were in a secondary state, promote to attached (posession
-    /// of a fresh generation implies this).
-    pub(crate) fn attach_in_generation(&mut self, generation: Generation) {
-        match &mut self.mode {
-            LocationMode::Attached(attach_conf) => {
-                attach_conf.generation = generation;
-            }
-            LocationMode::Secondary(_) => {
-                // We are promoted to attached by the control plane's re-attach response
-                self.mode = LocationMode::Attached(AttachedLocationConfig {
-                    generation,
-                    attach_mode: AttachmentMode::Single,
-                })
-            }
-        }
-    }
-
-    pub(crate) fn try_from(conf: &'_ models::LocationConfig) -> anyhow::Result<Self> {
-        let tenant_conf = TenantConfOpt::try_from(&conf.tenant_conf)?;
-
-        fn get_generation(conf: &'_ models::LocationConfig) -> Result<Generation, anyhow::Error> {
-            conf.generation
-                .ok_or_else(|| anyhow::anyhow!("Generation must be set when attaching"))
-        }
-
-        let mode = match &conf.mode {
-            models::LocationConfigMode::AttachedMulti => {
-                LocationMode::Attached(AttachedLocationConfig {
-                    generation: get_generation(conf)?,
-                    attach_mode: AttachmentMode::Multi,
-                })
-            }
-            models::LocationConfigMode::AttachedSingle => {
-                LocationMode::Attached(AttachedLocationConfig {
-                    generation: get_generation(conf)?,
-                    attach_mode: AttachmentMode::Single,
-                })
-            }
-            models::LocationConfigMode::AttachedStale => {
-                LocationMode::Attached(AttachedLocationConfig {
-                    generation: get_generation(conf)?,
-                    attach_mode: AttachmentMode::Stale,
-                })
-            }
-            models::LocationConfigMode::Secondary => {
-                anyhow::ensure!(conf.generation.is_none());
-
-                let warm = conf
-                    .secondary_conf
-                    .as_ref()
-                    .map(|c| c.warm)
-                    .unwrap_or(false);
-                LocationMode::Secondary(SecondaryLocationConfig { warm })
-            }
-            models::LocationConfigMode::Detached => {
-                // Should not have been called: API code should translate this mode
-                // into a detach rather than trying to decode it as a LocationConf
-                return Err(anyhow::anyhow!("Cannot decode a Detached configuration"));
-            }
-        };
-
-        Ok(Self { mode, tenant_conf })
-    }
-}
-
-impl Default for LocationConf {
-    // TODO: this should be removed once tenant loading can guarantee that we are never
-    // loading from a directory without a configuration.
-    // => tech debt since https://github.com/neondatabase/neon/issues/1555
-    fn default() -> Self {
-        Self {
-            mode: LocationMode::Attached(AttachedLocationConfig {
-                generation: Generation::none(),
-                attach_mode: AttachmentMode::Single,
-            }),
-            tenant_conf: TenantConfOpt::default(),
-        }
-    }
-}
-
-/// A tenant's calcuated configuration, which is the result of merging a
-/// tenant's TenantConfOpt with the global TenantConf from PageServerConf.
-///
-/// For storing and transmitting individual tenant's configuration, see
-/// TenantConfOpt.
+/// Per-tenant configuration options
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
 pub struct TenantConf {
     // Flush out an inmemory layer, if it's holding WAL older than this

pageserver/src/tenant/delete.rs

@@ -1,7 +1,9 @@
-use std::sync::Arc;
+use std::{
+    path::{Path, PathBuf},
+    sync::Arc,
+};
 
 use anyhow::Context;
-use camino::{Utf8Path, Utf8PathBuf};
 use pageserver_api::models::TenantState;
 use remote_storage::{DownloadError, GenericRemoteStorage, RemotePath};
 use tokio::sync::OwnedMutexGuard;
@@ -31,7 +33,7 @@ use super::{
 const SHOULD_RESUME_DELETION_FETCH_MARK_ATTEMPTS: u32 = 3;
 
 #[derive(Debug, thiserror::Error)]
-pub(crate) enum DeleteTenantError {
+pub enum DeleteTenantError {
     #[error("GetTenant {0}")]
     Get(#[from] GetTenantError),
 
@@ -60,7 +62,7 @@ fn remote_tenant_delete_mark_path(
         .context("Failed to strip workdir prefix")
         .and_then(RemotePath::new)
         .context("tenant path")?;
-    Ok(tenant_remote_path.join(Utf8Path::new("deleted")))
+    Ok(tenant_remote_path.join(Path::new("deleted")))
 }
 
 async fn create_remote_delete_mark(
@@ -146,7 +148,7 @@ async fn schedule_ordered_timeline_deletions(
     Ok(already_running_deletions)
 }
 
-async fn ensure_timelines_dir_empty(timelines_path: &Utf8Path) -> Result<(), DeleteTenantError> {
+async fn ensure_timelines_dir_empty(timelines_path: &Path) -> Result<(), DeleteTenantError> {
     // Assert timelines dir is empty.
     if !fs_ext::is_directory_empty(timelines_path).await? {
         // Display first 10 items in directory
@@ -186,18 +188,20 @@ async fn cleanup_remaining_fs_traces(
     conf: &PageServerConf,
     tenant_id: &TenantId,
 ) -> Result<(), DeleteTenantError> {
-    let rm = |p: Utf8PathBuf, is_dir: bool| async move {
+    let rm = |p: PathBuf, is_dir: bool| async move {
         if is_dir {
             tokio::fs::remove_dir(&p).await
         } else {
             tokio::fs::remove_file(&p).await
         }
         .or_else(fs_ext::ignore_not_found)
-        .with_context(|| format!("failed to delete {p}"))
+        .with_context(|| {
+            let to_display = p.display();
+            format!("failed to delete {to_display}")
+        })
     };
 
     rm(conf.tenant_config_path(tenant_id), false).await?;
-    rm(conf.tenant_location_config_path(tenant_id), false).await?;
 
     fail::fail_point!("tenant-delete-before-remove-timelines-dir", |_| {
         Err(anyhow::anyhow!(
@@ -376,7 +380,7 @@ impl DeleteTenantFlow {
         Ok(())
     }
 
-    pub(crate) async fn should_resume_deletion(
+    pub async fn should_resume_deletion(
         conf: &'static PageServerConf,
         remote_storage: Option<&GenericRemoteStorage>,
         tenant: &Tenant,
@@ -432,7 +436,7 @@ impl DeleteTenantFlow {
         // Tenant may not be loadable if we fail late in cleanup_remaining_fs_traces (e g remove timelines dir)
         let timelines_path = tenant.conf.timelines_path(&tenant.tenant_id);
         if timelines_path.exists() {
-            tenant.load(init_order, None, ctx).await.context("load")?;
+            tenant.load(init_order, ctx).await.context("load")?;
         }
 
         Self::background(

pageserver/src/tenant/ephemeral_file.rs

@@ -6,11 +6,11 @@ use crate::context::RequestContext;
 use crate::page_cache::{self, PAGE_SZ};
 use crate::tenant::block_io::{BlockCursor, BlockLease, BlockReader};
 use crate::virtual_file::VirtualFile;
-use camino::Utf8PathBuf;
 use std::cmp::min;
 use std::fs::OpenOptions;
 use std::io::{self, ErrorKind};
 use std::ops::DerefMut;
+use std::path::PathBuf;
 use std::sync::atomic::AtomicU64;
 use tracing::*;
 use utils::id::{TenantId, TimelineId};
@@ -40,9 +40,7 @@ impl EphemeralFile {
 
         let filename = conf
             .timeline_path(&tenant_id, &timeline_id)
-            .join(Utf8PathBuf::from(format!(
-                "ephemeral-{filename_disambiguator}"
-            )));
+            .join(PathBuf::from(format!("ephemeral-{filename_disambiguator}")));
 
         let file = VirtualFile::open_with_options(
             &filename,
@@ -81,7 +79,9 @@ impl EphemeralFile {
                         // order path before error because error is anyhow::Error => might have many contexts
                         format!(
                             "ephemeral file: read immutable page #{}: {}: {:#}",
-                            blknum, self.file.path, e,
+                            blknum,
+                            self.file.path.display(),
+                            e,
                         ),
                     )
                 })? {
@@ -191,7 +191,7 @@ impl EphemeralFile {
                                         "ephemeral_file: write_blob: write-back full tail blk #{}: {:#}: {}",
                                         self.blknum,
                                         e,
-                                        self.ephemeral_file.file.path,
+                                        self.ephemeral_file.file.path.display(),
                                     ),
                                 ));
                             }
@@ -254,7 +254,8 @@ impl Drop for EphemeralFile {
                 // not found files might also be related to https://github.com/neondatabase/neon/issues/2442
                 error!(
                     "could not remove ephemeral file '{}': {}",
-                    self.file.path, e
+                    self.file.path.display(),
+                    e
                 );
             }
         }

pageserver/src/tenant/par_fsync.rs

@@ -1,17 +1,16 @@
 use std::{
     io,
+    path::{Path, PathBuf},
     sync::atomic::{AtomicUsize, Ordering},
 };
 
-use camino::{Utf8Path, Utf8PathBuf};
-
-fn fsync_path(path: &Utf8Path) -> io::Result<()> {
+fn fsync_path(path: &Path) -> io::Result<()> {
     // TODO use VirtualFile::fsync_all once we fully go async.
     let file = std::fs::File::open(path)?;
     file.sync_all()
 }
 
-fn parallel_worker(paths: &[Utf8PathBuf], next_path_idx: &AtomicUsize) -> io::Result<()> {
+fn parallel_worker(paths: &[PathBuf], next_path_idx: &AtomicUsize) -> io::Result<()> {
     while let Some(path) = paths.get(next_path_idx.fetch_add(1, Ordering::Relaxed)) {
         fsync_path(path)?;
     }
@@ -19,7 +18,7 @@ fn parallel_worker(paths: &[Utf8PathBuf], next_path_idx: &AtomicUsize) -> io::Re
     Ok(())
 }
 
-fn fsync_in_thread_pool(paths: &[Utf8PathBuf]) -> io::Result<()> {
+fn fsync_in_thread_pool(paths: &[PathBuf]) -> io::Result<()> {
     // TODO: remove this function in favor of `par_fsync_async` once we asyncify everything.
 
     /// Use at most this number of threads.
@@ -48,7 +47,7 @@ fn fsync_in_thread_pool(paths: &[Utf8PathBuf]) -> io::Result<()> {
 }
 
 /// Parallel fsync all files. Can be used in non-async context as it is using rayon thread pool.
-pub fn par_fsync(paths: &[Utf8PathBuf]) -> io::Result<()> {
+pub fn par_fsync(paths: &[PathBuf]) -> io::Result<()> {
     if paths.len() == 1 {
         fsync_path(&paths[0])?;
         return Ok(());
@@ -59,7 +58,7 @@ pub fn par_fsync(paths: &[Utf8PathBuf]) -> io::Result<()> {
 
 /// Parallel fsync asynchronously. If number of files are less than PARALLEL_PATH_THRESHOLD, fsync is done in the current
 /// execution thread. Otherwise, we will spawn_blocking and run it in tokio.
-pub async fn par_fsync_async(paths: &[Utf8PathBuf]) -> io::Result<()> {
+pub async fn par_fsync_async(paths: &[PathBuf]) -> io::Result<()> {
     const MAX_CONCURRENT_FSYNC: usize = 64;
     let mut next = paths.iter().peekable();
     let mut js = tokio::task::JoinSet::new();

pageserver/src/tenant/remote_timeline_client.rs

@@ -209,7 +209,6 @@ pub mod index;
 mod upload;
 
 use anyhow::Context;
-use camino::Utf8Path;
 use chrono::{NaiveDateTime, Utc};
 // re-export these
 pub use download::{is_temp_download_file, list_remote_timelines};
@@ -220,6 +219,7 @@ use utils::backoff::{
 };
 
 use std::collections::{HashMap, VecDeque};
+use std::path::{Path, PathBuf};
 use std::sync::atomic::{AtomicU32, Ordering};
 use std::sync::{Arc, Mutex};
 
@@ -901,27 +901,9 @@ impl RemoteTimelineClient {
         .await
         .context("list prefixes")?;
 
-        // We will delete the current index_part object last, since it acts as a deletion
-        // marker via its deleted_at attribute
-        let latest_index = remaining
-            .iter()
-            .filter(|p| {
-                p.object_name()
-                    .map(|n| n.starts_with(IndexPart::FILE_NAME))
-                    .unwrap_or(false)
-            })
-            .filter_map(|path| parse_remote_index_path(path.clone()).map(|gen| (path, gen)))
-            .max_by_key(|i| i.1)
-            .map(|i| i.0.clone())
-            .unwrap_or(
-                // No generation-suffixed indices, assume we are dealing with
-                // a legacy index.
-                remote_index_path(&self.tenant_id, &self.timeline_id, Generation::none()),
-            );
-
-        let remaining_layers: Vec<RemotePath> = remaining
+        let remaining: Vec<RemotePath> = remaining
             .into_iter()
-            .filter(|p| p!= &latest_index)
+            .filter(|p| p.object_name() != Some(IndexPart::FILE_NAME))
             .inspect(|path| {
                 if let Some(name) = path.object_name() {
                     info!(%name, "deleting a file not referenced from index_part.json");
@@ -931,11 +913,9 @@ impl RemoteTimelineClient {
             })
             .collect();
 
-        let not_referenced_count = remaining_layers.len();
-        if !remaining_layers.is_empty() {
-            self.deletion_queue_client
-                .push_immediate(remaining_layers)
-                .await?;
+        let not_referenced_count = remaining.len();
+        if !remaining.is_empty() {
+            self.deletion_queue_client.push_immediate(remaining).await?;
         }
 
         fail::fail_point!("timeline-delete-before-index-delete", |_| {
@@ -944,9 +924,11 @@ impl RemoteTimelineClient {
             ))?
         });
 
+        let index_file_path = timeline_storage_path.join(Path::new(IndexPart::FILE_NAME));
+
         debug!("enqueuing index part deletion");
         self.deletion_queue_client
-            .push_immediate([latest_index].to_vec())
+            .push_immediate([index_file_path].to_vec())
             .await?;
 
         // Timeline deletion is rare and we have probably emitted a reasonably number of objects: wait
@@ -1419,13 +1401,6 @@ impl RemoteTimelineClient {
             }
         }
     }
-
-    pub(crate) fn get_layer_metadata(
-        &self,
-        name: &LayerFileName,
-    ) -> anyhow::Result<Option<LayerFileMetadata>> {
-        self.upload_queue.lock().unwrap().get_layer_metadata(name)
-    }
 }
 
 pub fn remote_timelines_path(tenant_id: &TenantId) -> RemotePath {
@@ -1434,7 +1409,7 @@ pub fn remote_timelines_path(tenant_id: &TenantId) -> RemotePath {
 }
 
 pub fn remote_timeline_path(tenant_id: &TenantId, timeline_id: &TimelineId) -> RemotePath {
-    remote_timelines_path(tenant_id).join(Utf8Path::new(&timeline_id.to_string()))
+    remote_timelines_path(tenant_id).join(&PathBuf::from(timeline_id.to_string()))
 }
 
 pub fn remote_layer_path(
@@ -1477,7 +1452,14 @@ pub(crate) fn parse_remote_index_path(path: RemotePath) -> Option<Generation> {
         }
     };
 
-    match file_name.split_once('-') {
+    let file_name_str = match file_name.to_str() {
+        Some(s) => s,
+        None => {
+            tracing::warn!("Malformed index key {:?}", path);
+            return None;
+        }
+    };
+    match file_name_str.split_once('-') {
         Some((_, gen_suffix)) => Generation::parse_suffix(gen_suffix),
         None => None,
     }
@@ -1489,16 +1471,20 @@ pub(crate) fn parse_remote_index_path(path: RemotePath) -> Option<Generation> {
 /// Errors if the path provided does not start from pageserver's workdir.
 pub fn remote_path(
     conf: &PageServerConf,
-    local_path: &Utf8Path,
+    local_path: &Path,
     generation: Generation,
 ) -> anyhow::Result<RemotePath> {
     let stripped = local_path
         .strip_prefix(&conf.workdir)
         .context("Failed to strip workdir prefix")?;
 
-    let suffixed = format!("{0}{1}", stripped, generation.get_suffix());
+    let suffixed = format!(
+        "{0}{1}",
+        stripped.to_string_lossy(),
+        generation.get_suffix()
+    );
 
-    RemotePath::new(Utf8Path::new(&suffixed)).with_context(|| {
+    RemotePath::new(&PathBuf::from(suffixed)).with_context(|| {
         format!(
             "to resolve remote part of path {:?} for base {:?}",
             local_path, conf.workdir
@@ -1518,7 +1504,7 @@ mod tests {
         DEFAULT_PG_VERSION,
     };
 
-    use std::collections::HashSet;
+    use std::{collections::HashSet, path::Path};
     use utils::lsn::Lsn;
 
     pub(super) fn dummy_contents(name: &str) -> Vec<u8> {
@@ -1552,7 +1538,7 @@ mod tests {
         assert_eq!(avec, bvec);
     }
 
-    fn assert_remote_files(expected: &[&str], remote_path: &Utf8Path, generation: Generation) {
+    fn assert_remote_files(expected: &[&str], remote_path: &Path, generation: Generation) {
         let mut expected: Vec<String> = expected
             .iter()
             .map(|x| format!("{}{}", x, generation.get_suffix()))
@@ -1671,12 +1657,12 @@ mod tests {
 
         let timeline_path = harness.timeline_path(&TIMELINE_ID);
 
-        println!("workdir: {}", harness.conf.workdir);
+        println!("workdir: {}", harness.conf.workdir.display());
 
         let remote_timeline_dir = harness
             .remote_fs_dir
             .join(timeline_path.strip_prefix(&harness.conf.workdir).unwrap());
-        println!("remote_timeline_dir: {remote_timeline_dir}");
+        println!("remote_timeline_dir: {}", remote_timeline_dir.display());
 
         let generation = harness.generation;
 
@@ -1923,7 +1909,7 @@ mod tests {
         let index_path = test_state.harness.remote_fs_dir.join(
             remote_index_path(&test_state.harness.tenant_id, &TIMELINE_ID, generation).get_path(),
         );
-        eprintln!("Writing {index_path}");
+        eprintln!("Writing {}", index_path.display());
         std::fs::write(&index_path, index_part_bytes).unwrap();
         example_index_part
     }

pageserver/src/tenant/remote_timeline_client/download.rs

@@ -5,10 +5,10 @@
 
 use std::collections::HashSet;
 use std::future::Future;
+use std::path::Path;
 use std::time::Duration;
 
 use anyhow::{anyhow, Context};
-use camino::Utf8Path;
 use tokio::fs;
 use tokio::io::AsyncWriteExt;
 use tokio_util::sync::CancellationToken;
@@ -74,7 +74,12 @@ pub async fn download_layer_file<'a>(
             // TODO: this doesn't use the cached fd for some reason?
             let mut destination_file = fs::File::create(&temp_file_path)
                 .await
-                .with_context(|| format!("create a destination file for layer '{temp_file_path}'"))
+                .with_context(|| {
+                    format!(
+                        "create a destination file for layer '{}'",
+                        temp_file_path.display()
+                    )
+                })
                 .map_err(DownloadError::Other)?;
             let mut download = storage
                 .download(&remote_path)
@@ -116,7 +121,7 @@ pub async fn download_layer_file<'a>(
     destination_file
         .flush()
         .await
-        .with_context(|| format!("flush source file at {temp_file_path}"))
+        .with_context(|| format!("flush source file at {}", temp_file_path.display()))
         .map_err(DownloadError::Other)?;
 
     let expected = layer_metadata.file_size();
@@ -130,7 +135,12 @@ pub async fn download_layer_file<'a>(
     destination_file
         .sync_all()
         .await
-        .with_context(|| format!("failed to fsync source file at {temp_file_path}"))
+        .with_context(|| {
+            format!(
+                "failed to fsync source file at {}",
+                temp_file_path.display()
+            )
+        })
         .map_err(DownloadError::Other)?;
     drop(destination_file);
 
@@ -142,23 +152,27 @@ pub async fn download_layer_file<'a>(
 
     fs::rename(&temp_file_path, &local_path)
         .await
-        .with_context(|| format!("rename download layer file to {local_path}"))
+        .with_context(|| format!("rename download layer file to {}", local_path.display(),))
         .map_err(DownloadError::Other)?;
 
     crashsafe::fsync_async(&local_path)
         .await
-        .with_context(|| format!("fsync layer file {local_path}"))
+        .with_context(|| format!("fsync layer file {}", local_path.display(),))
         .map_err(DownloadError::Other)?;
 
-    tracing::debug!("download complete: {local_path}");
+    tracing::debug!("download complete: {}", local_path.display());
 
     Ok(bytes_amount)
 }
 
 const TEMP_DOWNLOAD_EXTENSION: &str = "temp_download";
 
-pub fn is_temp_download_file(path: &Utf8Path) -> bool {
-    let extension = path.extension();
+pub fn is_temp_download_file(path: &Path) -> bool {
+    let extension = path.extension().map(|pname| {
+        pname
+            .to_str()
+            .expect("paths passed to this function must be valid Rust strings")
+    });
     match extension {
         Some(TEMP_DOWNLOAD_EXTENSION) => true,
         Some(_) => false,

pageserver/src/tenant/remote_timeline_client/upload.rs

@@ -1,9 +1,8 @@
 //! Helper functions to upload files to remote storage with a RemoteStorage
 
 use anyhow::{bail, Context};
-use camino::Utf8Path;
 use fail::fail_point;
-use std::io::ErrorKind;
+use std::{io::ErrorKind, path::Path};
 use tokio::fs;
 
 use super::Generation;
@@ -31,7 +30,6 @@ pub(super) async fn upload_index_part<'a>(
     fail_point!("before-upload-index", |_| {
         bail!("failpoint before-upload-index")
     });
-    pausable_failpoint!("before-upload-index-pausable");
 
     let index_part_bytes =
         serde_json::to_vec(&index_part).context("serialize index part file into bytes")?;
@@ -52,7 +50,7 @@ pub(super) async fn upload_index_part<'a>(
 pub(super) async fn upload_timeline_layer<'a>(
     conf: &'static PageServerConf,
     storage: &'a GenericRemoteStorage,
-    source_path: &'a Utf8Path,
+    source_path: &'a Path,
     known_metadata: &'a LayerFileMetadata,
     generation: Generation,
 ) -> anyhow::Result<()> {
@@ -70,7 +68,7 @@ pub(super) async fn upload_timeline_layer<'a>(
             // upload. However, a nonexistent file can also be indicative of
             // something worse, like when a file is scheduled for upload before
             // it has been written to disk yet.
-            info!(path = %source_path, "File to upload doesn't exist. Likely the file has been deleted and an upload is not required any more.");
+            info!(path = %source_path.display(), "File to upload doesn't exist. Likely the file has been deleted and an upload is not required any more.");
             return Ok(());
         }
         Err(e) => {
@@ -95,7 +93,7 @@ pub(super) async fn upload_timeline_layer<'a>(
     storage
         .upload(source_file, fs_size, &storage_path, None)
         .await
-        .with_context(|| format!("upload layer from local path '{source_path}'"))?;
+        .with_context(|| format!("upload layer from local path '{}'", source_path.display()))?;
 
     Ok(())
 }

pageserver/src/tenant/storage_layer.rs

@@ -14,7 +14,6 @@ use crate::task_mgr::TaskKind;
 use crate::walrecord::NeonWalRecord;
 use anyhow::Result;
 use bytes::Bytes;
-use camino::Utf8PathBuf;
 use enum_map::EnumMap;
 use enumset::EnumSet;
 use once_cell::sync::Lazy;
@@ -23,6 +22,7 @@ use pageserver_api::models::{
     HistoricLayerInfo, LayerResidenceEvent, LayerResidenceEventReason, LayerResidenceStatus,
 };
 use std::ops::Range;
+use std::path::PathBuf;
 use std::sync::{Arc, Mutex};
 use std::time::{Duration, SystemTime, UNIX_EPOCH};
 use tracing::warn;
@@ -378,7 +378,7 @@ pub trait PersistentLayer: Layer + AsLayerDesc {
 
     // Path to the layer file in the local filesystem.
     // `None` for `RemoteLayer`.
-    fn local_path(&self) -> Option<Utf8PathBuf>;
+    fn local_path(&self) -> Option<PathBuf>;
 
     /// Permanently remove this layer from disk.
     fn delete_resident_layer_file(&self) -> Result<()>;
@@ -456,7 +456,7 @@ pub mod tests {
 /// config. In that case, we use the Path variant to hold the full path to the file on
 /// disk.
 enum PathOrConf {
-    Path(Utf8PathBuf),
+    Path(PathBuf),
     Conf(&'static PageServerConf),
 }
 

pageserver/src/tenant/storage_layer/delta_layer.rs

@@ -41,7 +41,6 @@ use crate::virtual_file::VirtualFile;
 use crate::{walrecord, TEMP_FILE_SUFFIX};
 use crate::{DELTA_FILE_MAGIC, STORAGE_FORMAT_VERSION};
 use anyhow::{bail, ensure, Context, Result};
-use camino::{Utf8Path, Utf8PathBuf};
 use pageserver_api::models::{HistoricLayerInfo, LayerAccessKind};
 use rand::{distributions::Alphanumeric, Rng};
 use serde::{Deserialize, Serialize};
@@ -49,6 +48,7 @@ use std::fs::{self, File};
 use std::io::SeekFrom;
 use std::ops::Range;
 use std::os::unix::fs::FileExt;
+use std::path::{Path, PathBuf};
 use std::sync::Arc;
 use tokio::sync::OnceCell;
 use tracing::*;
@@ -267,7 +267,7 @@ impl PersistentLayer for DeltaLayer {
         Some(self)
     }
 
-    fn local_path(&self) -> Option<Utf8PathBuf> {
+    fn local_path(&self) -> Option<PathBuf> {
         self.local_path()
     }
 
@@ -374,7 +374,7 @@ impl DeltaLayer {
             .await
     }
 
-    pub(crate) fn local_path(&self) -> Option<Utf8PathBuf> {
+    pub(crate) fn local_path(&self) -> Option<PathBuf> {
         Some(self.path())
     }
 
@@ -409,7 +409,7 @@ impl DeltaLayer {
         tenant_id: &TenantId,
         timeline_id: &TimelineId,
         fname: &DeltaFileName,
-    ) -> Utf8PathBuf {
+    ) -> PathBuf {
         match path_or_conf {
             PathOrConf::Path(path) => path.clone(),
             PathOrConf::Conf(conf) => conf
@@ -424,7 +424,7 @@ impl DeltaLayer {
         timeline_id: &TimelineId,
         key_start: Key,
         lsn_range: &Range<Lsn>,
-    ) -> Utf8PathBuf {
+    ) -> PathBuf {
         let rand_string: String = rand::thread_rng()
             .sample_iter(&Alphanumeric)
             .take(8)
@@ -455,7 +455,7 @@ impl DeltaLayer {
         self.inner
             .get_or_try_init(|| self.load_inner(ctx))
             .await
-            .with_context(|| format!("Failed to load delta layer {}", self.path()))
+            .with_context(|| format!("Failed to load delta layer {}", self.path().display()))
     }
 
     async fn load_inner(&self, ctx: &RequestContext) -> Result<Arc<DeltaLayerInner>> {
@@ -471,7 +471,7 @@ impl DeltaLayer {
         if let PathOrConf::Path(ref path) = self.path_or_conf {
             // not production code
 
-            let actual_filename = path.file_name().unwrap().to_owned();
+            let actual_filename = path.file_name().unwrap().to_str().unwrap().to_owned();
             let expected_filename = self.filename().file_name();
 
             if actual_filename != expected_filename {
@@ -510,8 +510,9 @@ impl DeltaLayer {
     /// Create a DeltaLayer struct representing an existing file on disk.
     ///
     /// This variant is only used for debugging purposes, by the 'pagectl' binary.
-    pub fn new_for_path(path: &Utf8Path, file: File) -> Result<Self> {
-        let mut summary_buf = vec![0; PAGE_SZ];
+    pub fn new_for_path(path: &Path, file: File) -> Result<Self> {
+        let mut summary_buf = Vec::new();
+        summary_buf.resize(PAGE_SZ, 0);
         file.read_exact_at(&mut summary_buf, 0)?;
         let summary = Summary::des_prefix(&summary_buf)?;
 
@@ -537,7 +538,7 @@ impl DeltaLayer {
         self.desc.delta_file_name()
     }
     /// Path to the layer file in pageserver workdir.
-    pub fn path(&self) -> Utf8PathBuf {
+    pub fn path(&self) -> PathBuf {
         Self::path_for(
             &self.path_or_conf,
             &self.desc.tenant_id,
@@ -572,7 +573,7 @@ impl DeltaLayer {
 ///
 struct DeltaLayerWriterInner {
     conf: &'static PageServerConf,
-    pub path: Utf8PathBuf,
+    pub path: PathBuf,
     timeline_id: TimelineId,
     tenant_id: TenantId,
 
@@ -710,7 +711,7 @@ impl DeltaLayerWriterInner {
         ensure!(
             metadata.len() <= S3_UPLOAD_LIMIT,
             "Created delta layer file at {} of size {} above limit {S3_UPLOAD_LIMIT}!",
-            file.path,
+            file.path.display(),
             metadata.len()
         );
 
@@ -747,7 +748,7 @@ impl DeltaLayerWriterInner {
         );
         std::fs::rename(self.path, &final_path)?;
 
-        trace!("created delta layer {final_path}");
+        trace!("created delta layer {}", final_path.display());
 
         Ok(layer)
     }
@@ -846,13 +847,13 @@ impl Drop for DeltaLayerWriter {
 
 impl DeltaLayerInner {
     pub(super) async fn load(
-        path: &Utf8Path,
+        path: &std::path::Path,
         summary: Option<Summary>,
         ctx: &RequestContext,
     ) -> anyhow::Result<Self> {
         let file = VirtualFile::open(path)
             .await
-            .with_context(|| format!("Failed to open file '{path}'"))?;
+            .with_context(|| format!("Failed to open file '{}'", path.display()))?;
         let file = FileBlockReader::new(file);
 
         let summary_blk = file.read_blk(0, ctx).await?;
@@ -863,11 +864,11 @@ impl DeltaLayerInner {
             expected_summary.index_start_blk = actual_summary.index_start_blk;
             expected_summary.index_root_blk = actual_summary.index_root_blk;
             if actual_summary != expected_summary {
-                bail!(
-                    "in-file summary does not match expected summary. actual = {:?} expected = {:?}",
-                    actual_summary,
-                    expected_summary
-                );
+                // bail!(
+                //     "in-file summary does not match expected summary. actual = {:?} expected = {:?}",
+                //     actual_summary,
+                //     expected_summary
+                // );
             }
         }
 
@@ -932,12 +933,15 @@ impl DeltaLayerInner {
                 .read_blob_into_buf(pos, &mut buf, ctx)
                 .await
                 .with_context(|| {
-                    format!("Failed to read blob from virtual file {}", file.file.path)
+                    format!(
+                        "Failed to read blob from virtual file {}",
+                        file.file.path.display()
+                    )
                 })?;
             let val = Value::des(&buf).with_context(|| {
                 format!(
                     "Failed to deserialize file blob from virtual file {}",
-                    file.file.path
+                    file.file.path.display()
                 )
             })?;
             match val {

pageserver/src/tenant/storage_layer/image_layer.rs

@@ -37,7 +37,6 @@ use crate::virtual_file::VirtualFile;
 use crate::{IMAGE_FILE_MAGIC, STORAGE_FORMAT_VERSION, TEMP_FILE_SUFFIX};
 use anyhow::{bail, ensure, Context, Result};
 use bytes::Bytes;
-use camino::{Utf8Path, Utf8PathBuf};
 use hex;
 use pageserver_api::models::{HistoricLayerInfo, LayerAccessKind};
 use rand::{distributions::Alphanumeric, Rng};
@@ -46,6 +45,7 @@ use std::fs::{self, File};
 use std::io::SeekFrom;
 use std::ops::Range;
 use std::os::unix::prelude::FileExt;
+use std::path::{Path, PathBuf};
 use tokio::sync::OnceCell;
 use tracing::*;
 
@@ -195,7 +195,7 @@ impl AsLayerDesc for ImageLayer {
 }
 
 impl PersistentLayer for ImageLayer {
-    fn local_path(&self) -> Option<Utf8PathBuf> {
+    fn local_path(&self) -> Option<PathBuf> {
         self.local_path()
     }
 
@@ -269,10 +269,10 @@ impl ImageLayer {
             .get_value_reconstruct_data(key, reconstruct_state, ctx)
             .await
             // FIXME: makes no sense to dump paths
-            .with_context(|| format!("read {}", self.path()))
+            .with_context(|| format!("read {}", self.path().display()))
     }
 
-    pub(crate) fn local_path(&self) -> Option<Utf8PathBuf> {
+    pub(crate) fn local_path(&self) -> Option<PathBuf> {
         Some(self.path())
     }
 
@@ -304,7 +304,7 @@ impl ImageLayer {
         timeline_id: TimelineId,
         tenant_id: TenantId,
         fname: &ImageFileName,
-    ) -> Utf8PathBuf {
+    ) -> PathBuf {
         match path_or_conf {
             PathOrConf::Path(path) => path.to_path_buf(),
             PathOrConf::Conf(conf) => conf
@@ -318,7 +318,7 @@ impl ImageLayer {
         timeline_id: TimelineId,
         tenant_id: TenantId,
         fname: &ImageFileName,
-    ) -> Utf8PathBuf {
+    ) -> PathBuf {
         let rand_string: String = rand::thread_rng()
             .sample_iter(&Alphanumeric)
             .take(8)
@@ -342,7 +342,7 @@ impl ImageLayer {
         self.inner
             .get_or_try_init(|| self.load_inner(ctx))
             .await
-            .with_context(|| format!("Failed to load image layer {}", self.path()))
+            .with_context(|| format!("Failed to load image layer {}", self.path().display()))
     }
 
     async fn load_inner(&self, ctx: &RequestContext) -> Result<ImageLayerInner> {
@@ -359,7 +359,7 @@ impl ImageLayer {
 
         if let PathOrConf::Path(ref path) = self.path_or_conf {
             // not production code
-            let actual_filename = path.file_name().unwrap().to_owned();
+            let actual_filename = path.file_name().unwrap().to_str().unwrap().to_owned();
             let expected_filename = self.filename().file_name();
 
             if actual_filename != expected_filename {
@@ -399,8 +399,9 @@ impl ImageLayer {
     /// Create an ImageLayer struct representing an existing file on disk.
     ///
     /// This variant is only used for debugging purposes, by the 'pagectl' binary.
-    pub fn new_for_path(path: &Utf8Path, file: File) -> Result<ImageLayer> {
-        let mut summary_buf = vec![0; PAGE_SZ];
+    pub fn new_for_path(path: &Path, file: File) -> Result<ImageLayer> {
+        let mut summary_buf = Vec::new();
+        summary_buf.resize(PAGE_SZ, 0);
         file.read_exact_at(&mut summary_buf, 0)?;
         let summary = Summary::des_prefix(&summary_buf)?;
         let metadata = file
@@ -426,7 +427,7 @@ impl ImageLayer {
     }
 
     /// Path to the layer file in pageserver workdir.
-    pub fn path(&self) -> Utf8PathBuf {
+    pub fn path(&self) -> PathBuf {
         Self::path_for(
             &self.path_or_conf,
             self.desc.timeline_id,
@@ -438,14 +439,14 @@ impl ImageLayer {
 
 impl ImageLayerInner {
     pub(super) async fn load(
-        path: &Utf8Path,
+        path: &std::path::Path,
         lsn: Lsn,
         summary: Option<Summary>,
         ctx: &RequestContext,
     ) -> anyhow::Result<Self> {
         let file = VirtualFile::open(path)
             .await
-            .with_context(|| format!("Failed to open file '{}'", path))?;
+            .with_context(|| format!("Failed to open file '{}'", path.display()))?;
         let file = FileBlockReader::new(file);
         let summary_blk = file.read_blk(0, ctx).await?;
         let actual_summary = Summary::des_prefix(summary_blk.as_ref())?;
@@ -456,11 +457,11 @@ impl ImageLayerInner {
             expected_summary.index_root_blk = actual_summary.index_root_blk;
 
             if actual_summary != expected_summary {
-                bail!(
-                    "in-file summary does not match expected summary. actual = {:?} expected = {:?}",
-                    actual_summary,
-                    expected_summary
-                );
+                // bail!(
+                //     "in-file summary does not match expected summary. actual = {:?} expected = {:?}",
+                //     actual_summary,
+                //     expected_summary
+                // );
             }
         }
 
@@ -525,7 +526,7 @@ impl ImageLayerInner {
 ///
 struct ImageLayerWriterInner {
     conf: &'static PageServerConf,
-    path: Utf8PathBuf,
+    path: PathBuf,
     timeline_id: TimelineId,
     tenant_id: TenantId,
     key_range: Range<Key>,
@@ -557,7 +558,7 @@ impl ImageLayerWriterInner {
                 lsn,
             },
         );
-        info!("new image layer {path}");
+        info!("new image layer {}", path.display());
         let mut file = VirtualFile::open_with_options(
             &path,
             std::fs::OpenOptions::new().write(true).create_new(true),
@@ -684,7 +685,7 @@ impl ImageLayerWriterInner {
         );
         std::fs::rename(self.path, final_path)?;
 
-        trace!("created image layer {}", layer.path());
+        trace!("created image layer {}", layer.path().display());
 
         Ok(layer)
     }

pageserver/src/tenant/storage_layer/remote_layer.rs

@@ -8,9 +8,9 @@ use crate::tenant::remote_timeline_client::index::LayerFileMetadata;
 use crate::tenant::storage_layer::{Layer, ValueReconstructResult, ValueReconstructState};
 use crate::tenant::timeline::layer_manager::LayerManager;
 use anyhow::{bail, Result};
-use camino::Utf8PathBuf;
 use pageserver_api::models::HistoricLayerInfo;
 use std::ops::Range;
+use std::path::PathBuf;
 use std::sync::Arc;
 
 use utils::{
@@ -92,7 +92,7 @@ impl AsLayerDesc for RemoteLayer {
 }
 
 impl PersistentLayer for RemoteLayer {
-    fn local_path(&self) -> Option<Utf8PathBuf> {
+    fn local_path(&self) -> Option<PathBuf> {
         None
     }
 

pageserver/src/tenant/tasks.rs

@@ -14,73 +14,6 @@ use tokio_util::sync::CancellationToken;
 use tracing::*;
 use utils::completion;
 
-static CONCURRENT_BACKGROUND_TASKS: once_cell::sync::Lazy<tokio::sync::Semaphore> =
-    once_cell::sync::Lazy::new(|| {
-        let total_threads = *task_mgr::BACKGROUND_RUNTIME_WORKER_THREADS;
-        let permits = usize::max(
-            1,
-            // while a lot of the work is done on spawn_blocking, we still do
-            // repartitioning in the async context. this should give leave us some workers
-            // unblocked to be blocked on other work, hopefully easing any outside visible
-            // effects of restarts.
-            //
-            // 6/8 is a guess; previously we ran with unlimited 8 and more from
-            // spawn_blocking.
-            (total_threads * 3).checked_div(4).unwrap_or(0),
-        );
-        assert_ne!(permits, 0, "we will not be adding in permits later");
-        assert!(
-            permits < total_threads,
-            "need threads avail for shorter work"
-        );
-        tokio::sync::Semaphore::new(permits)
-    });
-
-#[derive(Debug, PartialEq, Eq, Clone, Copy, strum_macros::IntoStaticStr)]
-#[strum(serialize_all = "snake_case")]
-pub(crate) enum BackgroundLoopKind {
-    Compaction,
-    Gc,
-    Eviction,
-    ConsumptionMetricsCollectMetrics,
-    ConsumptionMetricsSyntheticSizeWorker,
-}
-
-impl BackgroundLoopKind {
-    fn as_static_str(&self) -> &'static str {
-        let s: &'static str = self.into();
-        s
-    }
-}
-
-pub(crate) enum RateLimitError {
-    Cancelled,
-}
-
-pub(crate) async fn concurrent_background_tasks_rate_limit(
-    loop_kind: BackgroundLoopKind,
-    _ctx: &RequestContext,
-    cancel: &CancellationToken,
-) -> Result<impl Drop, RateLimitError> {
-    crate::metrics::BACKGROUND_LOOP_SEMAPHORE_WAIT_START_COUNT
-        .with_label_values(&[loop_kind.as_static_str()])
-        .inc();
-    scopeguard::defer!(
-        crate::metrics::BACKGROUND_LOOP_SEMAPHORE_WAIT_FINISH_COUNT.with_label_values(&[loop_kind.as_static_str()]).inc();
-    );
-    tokio::select! {
-        permit = CONCURRENT_BACKGROUND_TASKS.acquire() => {
-            match permit {
-                Ok(permit) => Ok(permit),
-                Err(_closed) => unreachable!("we never close the semaphore"),
-            }
-        },
-        _ = cancel.cancelled() => {
-            Err(RateLimitError::Cancelled)
-        }
-    }
-}
-
 /// Start per tenant background loops: compaction and gc.
 pub fn start_background_loops(
     tenant: &Arc<Tenant>,
@@ -183,7 +116,7 @@ async fn compaction_loop(tenant: Arc<Tenant>, cancel: CancellationToken) {
                 }
             };
 
-            warn_when_period_overrun(started_at.elapsed(), period, BackgroundLoopKind::Compaction);
+            warn_when_period_overrun(started_at.elapsed(), period, "compaction");
 
             // Sleep
             if tokio::time::timeout(sleep_duration, cancel.cancelled())
@@ -251,7 +184,7 @@ async fn gc_loop(tenant: Arc<Tenant>, cancel: CancellationToken) {
                 }
             };
 
-            warn_when_period_overrun(started_at.elapsed(), period, BackgroundLoopKind::Gc);
+            warn_when_period_overrun(started_at.elapsed(), period, "gc");
 
             // Sleep
             if tokio::time::timeout(sleep_duration, cancel.cancelled())
@@ -325,11 +258,7 @@ pub(crate) async fn random_init_delay(
 }
 
 /// Attention: the `task` and `period` beocme labels of a pageserver-wide prometheus metric.
-pub(crate) fn warn_when_period_overrun(
-    elapsed: Duration,
-    period: Duration,
-    task: BackgroundLoopKind,
-) {
+pub(crate) fn warn_when_period_overrun(elapsed: Duration, period: Duration, task: &str) {
     // Duration::ZERO will happen because it's the "disable [bgtask]" value.
     if elapsed >= period && period != Duration::ZERO {
         // humantime does no significant digits clamping whereas Duration's debug is a bit more
@@ -338,11 +267,11 @@ pub(crate) fn warn_when_period_overrun(
         warn!(
             ?elapsed,
             period = %humantime::format_duration(period),
-            ?task,
+            task,
             "task iteration took longer than the configured period"
         );
         crate::metrics::BACKGROUND_LOOP_PERIOD_OVERRUN_COUNT
-            .with_label_values(&[task.as_static_str(), &format!("{}", period.as_secs())])
+            .with_label_values(&[task, &format!("{}", period.as_secs())])
             .inc();
     }
 }

pageserver/src/tenant/timeline.rs

@@ -9,7 +9,6 @@ mod walreceiver;
 
 use anyhow::{anyhow, bail, ensure, Context, Result};
 use bytes::Bytes;
-use camino::{Utf8Path, Utf8PathBuf};
 use fail::fail_point;
 use futures::StreamExt;
 use itertools::Itertools;
@@ -30,6 +29,7 @@ use utils::id::TenantTimelineId;
 use std::cmp::{max, min, Ordering};
 use std::collections::{BinaryHeap, HashMap, HashSet};
 use std::ops::{Deref, Range};
+use std::path::{Path, PathBuf};
 use std::pin::pin;
 use std::sync::atomic::Ordering as AtomicOrdering;
 use std::sync::{Arc, Mutex, RwLock, Weak};
@@ -44,7 +44,6 @@ use crate::tenant::storage_layer::delta_layer::DeltaEntry;
 use crate::tenant::storage_layer::{
     DeltaLayerWriter, ImageLayerWriter, InMemoryLayer, LayerAccessStats, LayerFileName, RemoteLayer,
 };
-use crate::tenant::tasks::{BackgroundLoopKind, RateLimitError};
 use crate::tenant::timeline::logical_size::CurrentLogicalSize;
 use crate::tenant::{
     layer_map::{LayerMap, SearchResult},
@@ -57,7 +56,7 @@ use crate::config::PageServerConf;
 use crate::keyspace::{KeyPartitioning, KeySpace, KeySpaceRandomAccum};
 use crate::metrics::{
     TimelineMetrics, MATERIALIZED_PAGE_CACHE_HIT, MATERIALIZED_PAGE_CACHE_HIT_DIRECT,
-    UNEXPECTED_ONDEMAND_DOWNLOADS,
+    RECONSTRUCT_TIME, UNEXPECTED_ONDEMAND_DOWNLOADS,
 };
 use crate::pgdatadir_mapping::LsnForTimestamp;
 use crate::pgdatadir_mapping::{is_rel_fsm_block_key, is_rel_vm_block_key};
@@ -92,12 +91,12 @@ use self::logical_size::LogicalSize;
 use self::walreceiver::{WalReceiver, WalReceiverConf};
 
 use super::config::TenantConf;
+use super::debug_assert_current_span_has_tenant_and_timeline_id;
 use super::remote_timeline_client::index::IndexPart;
 use super::remote_timeline_client::RemoteTimelineClient;
 use super::storage_layer::{
     AsLayerDesc, DeltaLayer, ImageLayer, LayerAccessStatsReset, PersistentLayerDesc,
 };
-use super::{debug_assert_current_span_has_tenant_and_timeline_id, AttachedTenantConf};
 
 #[derive(Debug, PartialEq, Eq, Clone, Copy)]
 pub(super) enum FlushLoopState {
@@ -150,7 +149,7 @@ pub struct TimelineResources {
 
 pub struct Timeline {
     conf: &'static PageServerConf,
-    tenant_conf: Arc<RwLock<AttachedTenantConf>>,
+    tenant_conf: Arc<RwLock<TenantConfOpt>>,
 
     myself: Weak<Self>,
 
@@ -159,9 +158,6 @@ pub struct Timeline {
 
     /// The generation of the tenant that instantiated us: this is used for safety when writing remote objects.
     /// Never changes for the lifetime of this [`Timeline`] object.
-    ///
-    /// This duplicates the generation stored in LocationConf, but that structure is mutable:
-    /// this copy enforces the invariant that generatio doesn't change during a Tenant's lifetime.
     generation: Generation,
 
     pub pg_version: u32,
@@ -371,7 +367,7 @@ pub enum PageReconstructError {
 
     /// An error happened replaying WAL records
     #[error(transparent)]
-    WalRedo(anyhow::Error),
+    WalRedo(#[from] crate::walredo::WalRedoError),
 }
 
 impl std::fmt::Debug for PageReconstructError {
@@ -500,39 +496,13 @@ impl Timeline {
         };
 
         let timer = crate::metrics::GET_RECONSTRUCT_DATA_TIME.start_timer();
-        let path = self
-            .get_reconstruct_data(key, lsn, &mut reconstruct_state, ctx)
+        self.get_reconstruct_data(key, lsn, &mut reconstruct_state, ctx)
             .await?;
         timer.stop_and_record();
 
-        let start = Instant::now();
-        let res = self.reconstruct_value(key, lsn, reconstruct_state).await;
-        let elapsed = start.elapsed();
-        crate::metrics::RECONSTRUCT_TIME
-            .for_result(&res)
-            .observe(elapsed.as_secs_f64());
-
-        if cfg!(feature = "testing") && res.is_err() {
-            // it can only be walredo issue
-            use std::fmt::Write;
-
-            let mut msg = String::new();
-
-            path.into_iter().for_each(|(res, cont_lsn, layer)| {
-                writeln!(
-                    msg,
-                    "- layer traversal: result {res:?}, cont_lsn {cont_lsn}, layer: {}",
-                    layer(),
-                )
-                .expect("string grows")
-            });
-
-            // this is to rule out or provide evidence that we could in some cases read a duplicate
-            // walrecord
-            tracing::info!("walredo failed, path:\n{msg}");
-        }
-
-        res
+        RECONSTRUCT_TIME
+            .observe_closure_duration(|| self.reconstruct_value(key, lsn, reconstruct_state))
+            .await
     }
 
     /// Get last or prev record separately. Same as get_last_record_rlsn().last/prev.
@@ -685,18 +655,38 @@ impl Timeline {
     ) -> anyhow::Result<()> {
         const ROUNDS: usize = 2;
 
-        // this wait probably never needs any "long time spent" logging, because we already nag if
-        // compaction task goes over it's period (20s) which is quite often in production.
-        let _permit = match super::tasks::concurrent_background_tasks_rate_limit(
-            BackgroundLoopKind::Compaction,
-            ctx,
-            cancel,
-        )
-        .await
-        {
-            Ok(permit) => permit,
-            Err(RateLimitError::Cancelled) => return Ok(()),
-        };
+        // static CONCURRENT_COMPACTIONS: once_cell::sync::Lazy<tokio::sync::Semaphore> =
+        //     once_cell::sync::Lazy::new(|| {
+        //         let total_threads = *task_mgr::BACKGROUND_RUNTIME_WORKER_THREADS;
+        //         let permits = usize::max(
+        //             1,
+        //             // while a lot of the work is done on spawn_blocking, we still do
+        //             // repartitioning in the async context. this should give leave us some workers
+        //             // unblocked to be blocked on other work, hopefully easing any outside visible
+        //             // effects of restarts.
+        //             //
+        //             // 6/8 is a guess; previously we ran with unlimited 8 and more from
+        //             // spawn_blocking.
+        //             (total_threads * 3).checked_div(4).unwrap_or(0),
+        //         );
+        //         assert_ne!(permits, 0, "we will not be adding in permits later");
+        //         assert!(
+        //             permits < total_threads,
+        //             "need threads avail for shorter work"
+        //         );
+        //         tokio::sync::Semaphore::new(permits)
+        //     });
+
+        // // this wait probably never needs any "long time spent" logging, because we already nag if
+        // // compaction task goes over it's period (20s) which is quite often in production.
+        // let _permit = tokio::select! {
+        //     permit = CONCURRENT_COMPACTIONS.acquire() => {
+        //         permit
+        //     },
+        //     _ = cancel.cancelled() => {
+        //         return Ok(());
+        //     }
+        // };
 
         let last_record_lsn = self.get_last_record_lsn();
 
@@ -1275,23 +1265,7 @@ impl Timeline {
                 Ok(delta) => Some(delta),
             };
 
-        // RemoteTimelineClient holds the metadata on layers' remote generations, so
-        // query it to construct a RemoteLayer.
-        let layer_metadata = self
-            .remote_client
-            .as_ref()
-            .expect("Eviction is not called without remote storage")
-            .get_layer_metadata(&local_layer.filename())
-            .map_err(EvictionError::LayerNotFound)?
-            .ok_or_else(|| {
-                EvictionError::LayerNotFound(anyhow::anyhow!("Layer not in remote metadata"))
-            })?;
-        if layer_metadata.file_size() != layer_file_size {
-            return Err(EvictionError::MetadataInconsistency(format!(
-                "Layer size {layer_file_size} doesn't match remote metadata file size {}",
-                layer_metadata.file_size()
-            )));
-        }
+        let layer_metadata = LayerFileMetadata::new(layer_file_size, self.generation);
 
         let new_remote_layer = Arc::new(match local_layer.filename() {
             LayerFileName::Image(image_name) => RemoteLayer::new_img(
@@ -1370,10 +1344,6 @@ pub(crate) enum EvictionError {
     /// different objects in memory.
     #[error("layer was no longer part of LayerMap")]
     LayerNotFound(#[source] anyhow::Error),
-
-    /// This should never happen
-    #[error("Metadata inconsistency")]
-    MetadataInconsistency(String),
 }
 
 /// Number of times we will compute partition within a checkpoint distance.
@@ -1382,42 +1352,42 @@ const REPARTITION_FREQ_IN_CHECKPOINT_DISTANCE: u64 = 10;
 // Private functions
 impl Timeline {
     fn get_checkpoint_distance(&self) -> u64 {
-        let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
+        let tenant_conf = self.tenant_conf.read().unwrap();
         tenant_conf
             .checkpoint_distance
             .unwrap_or(self.conf.default_tenant_conf.checkpoint_distance)
     }
 
     fn get_checkpoint_timeout(&self) -> Duration {
-        let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
+        let tenant_conf = self.tenant_conf.read().unwrap();
         tenant_conf
             .checkpoint_timeout
             .unwrap_or(self.conf.default_tenant_conf.checkpoint_timeout)
     }
 
     fn get_compaction_target_size(&self) -> u64 {
-        let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
+        let tenant_conf = self.tenant_conf.read().unwrap();
         tenant_conf
             .compaction_target_size
             .unwrap_or(self.conf.default_tenant_conf.compaction_target_size)
     }
 
     fn get_compaction_threshold(&self) -> usize {
-        let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
+        let tenant_conf = self.tenant_conf.read().unwrap();
         tenant_conf
             .compaction_threshold
             .unwrap_or(self.conf.default_tenant_conf.compaction_threshold)
     }
 
     fn get_image_creation_threshold(&self) -> usize {
-        let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
+        let tenant_conf = self.tenant_conf.read().unwrap();
         tenant_conf
             .image_creation_threshold
             .unwrap_or(self.conf.default_tenant_conf.image_creation_threshold)
     }
 
     fn get_eviction_policy(&self) -> EvictionPolicy {
-        let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
+        let tenant_conf = self.tenant_conf.read().unwrap();
         tenant_conf
             .eviction_policy
             .unwrap_or(self.conf.default_tenant_conf.eviction_policy)
@@ -1433,7 +1403,7 @@ impl Timeline {
     }
 
     fn get_gc_feedback(&self) -> bool {
-        let tenant_conf = &self.tenant_conf.read().unwrap().tenant_conf;
+        let tenant_conf = self.tenant_conf.read().unwrap();
         tenant_conf
             .gc_feedback
             .unwrap_or(self.conf.default_tenant_conf.gc_feedback)
@@ -1446,7 +1416,7 @@ impl Timeline {
         // The threshold is embedded in the metric. So, we need to update it.
         {
             let new_threshold = Self::get_evictions_low_residence_duration_metric_threshold(
-                &self.tenant_conf.read().unwrap().tenant_conf,
+                &self.tenant_conf.read().unwrap(),
                 &self.conf.default_tenant_conf,
             );
             let tenant_id_str = self.tenant_id.to_string();
@@ -1465,7 +1435,7 @@ impl Timeline {
     #[allow(clippy::too_many_arguments)]
     pub(super) fn new(
         conf: &'static PageServerConf,
-        tenant_conf: Arc<RwLock<AttachedTenantConf>>,
+        tenant_conf: Arc<RwLock<TenantConfOpt>>,
         metadata: &TimelineMetadata,
         ancestor: Option<Arc<Timeline>>,
         timeline_id: TimelineId,
@@ -1488,7 +1458,7 @@ impl Timeline {
 
         let evictions_low_residence_duration_metric_threshold =
             Self::get_evictions_low_residence_duration_metric_threshold(
-                &tenant_conf_guard.tenant_conf,
+                &tenant_conf_guard,
                 &conf.default_tenant_conf,
             );
         drop(tenant_conf_guard);
@@ -1653,15 +1623,12 @@ impl Timeline {
 
         let tenant_conf_guard = self.tenant_conf.read().unwrap();
         let wal_connect_timeout = tenant_conf_guard
-            .tenant_conf
             .walreceiver_connect_timeout
             .unwrap_or(self.conf.default_tenant_conf.walreceiver_connect_timeout);
         let lagging_wal_timeout = tenant_conf_guard
-            .tenant_conf
             .lagging_wal_timeout
             .unwrap_or(self.conf.default_tenant_conf.lagging_wal_timeout);
         let max_lsn_wal_lag = tenant_conf_guard
-            .tenant_conf
             .max_lsn_wal_lag
             .unwrap_or(self.conf.default_tenant_conf.max_lsn_wal_lag);
         drop(tenant_conf_guard);
@@ -1743,7 +1710,7 @@ impl Timeline {
                         Discovered::Temporary(name) => (name, "temporary timeline file"),
                         Discovered::TemporaryDownload(name) => (name, "temporary download"),
                     };
-                    path.push(Utf8Path::new(&name));
+                    path.push(name);
                     init::cleanup(&path, kind)?;
                     path.pop();
                 }
@@ -2224,10 +2191,10 @@ impl TraversalLayerExt for Arc<dyn PersistentLayer> {
         let timeline_id = self.layer_desc().timeline_id;
         match self.local_path() {
             Some(local_path) => {
-                debug_assert!(local_path.to_string().contains(&format!("{}", timeline_id)),
+                debug_assert!(local_path.to_str().unwrap().contains(&format!("{}", timeline_id)),
                     "need timeline ID to uniquely identify the layer when traversal crosses ancestor boundary",
                 );
-                format!("{local_path}")
+                format!("{}", local_path.display())
             }
             None => {
                 format!("remote {}/{self}", timeline_id)
@@ -2257,7 +2224,7 @@ impl Timeline {
         request_lsn: Lsn,
         reconstruct_state: &mut ValueReconstructState,
         ctx: &RequestContext,
-    ) -> Result<Vec<TraversalPathItem>, PageReconstructError> {
+    ) -> Result<(), PageReconstructError> {
         // Start from the current timeline.
         let mut timeline_owned;
         let mut timeline = self;
@@ -2288,12 +2255,12 @@ impl Timeline {
             // The function should have updated 'state'
             //info!("CALLED for {} at {}: {:?} with {} records, cached {}", key, cont_lsn, result, reconstruct_state.records.len(), cached_lsn);
             match result {
-                ValueReconstructResult::Complete => return Ok(traversal_path),
+                ValueReconstructResult::Complete => return Ok(()),
                 ValueReconstructResult::Continue => {
                     // If we reached an earlier cached page image, we're done.
                     if cont_lsn == cached_lsn + 1 {
                         MATERIALIZED_PAGE_CACHE_HIT.inc_by(1);
-                        return Ok(traversal_path);
+                        return Ok(());
                     }
                     if prev_lsn <= cont_lsn {
                         // Didn't make any progress in last iteration. Error out to avoid
@@ -2364,7 +2331,7 @@ impl Timeline {
                 // during branch creation.
                 match ancestor.wait_to_become_active(ctx).await {
                     Ok(()) => {}
-                    Err(TimelineState::Stopping) => {
+                    Err(state) if state == TimelineState::Stopping => {
                         return Err(PageReconstructError::AncestorStopping(ancestor.timeline_id));
                     }
                     Err(state) => {
@@ -3729,11 +3696,6 @@ impl Timeline {
             });
 
             writer.as_mut().unwrap().put_value(key, lsn, value).await?;
-
-            if !new_layers.is_empty() {
-                fail_point!("after-timeline-compacted-first-L1");
-            }
-
             prev_key = Some(key);
         }
         if let Some(writer) = writer {
@@ -3755,7 +3717,7 @@ impl Timeline {
                     );
                 }
             }
-            let mut layer_paths: Vec<Utf8PathBuf> = new_layers.iter().map(|l| l.path()).collect();
+            let mut layer_paths: Vec<PathBuf> = new_layers.iter().map(|l| l.path()).collect();
 
             // Fsync all the layer files and directory using multiple threads to
             // minimize latency.
@@ -3865,7 +3827,10 @@ impl Timeline {
             let new_delta_path = l.path();
 
             let metadata = new_delta_path.metadata().with_context(|| {
-                format!("read file metadata for new created layer {new_delta_path}")
+                format!(
+                    "read file metadata for new created layer {}",
+                    new_delta_path.display()
+                )
             })?;
 
             if let Some(remote_client) = &self.remote_client {
@@ -3888,7 +3853,6 @@ impl Timeline {
             );
             let l = l as Arc<dyn PersistentLayer>;
             if guard.contains(&l) {
-                tracing::error!(layer=%l, "duplicated L1 layer");
                 duplicated_layers.insert(l.layer_desc().key());
             } else {
                 if LayerMap::is_l0(l.layer_desc()) {
@@ -4800,10 +4764,11 @@ fn is_send() {
 
 /// Add a suffix to a layer file's name: .{num}.old
 /// Uses the first available num (starts at 0)
-fn rename_to_backup(path: &Utf8Path) -> anyhow::Result<()> {
+fn rename_to_backup(path: &Path) -> anyhow::Result<()> {
     let filename = path
         .file_name()
-        .ok_or_else(|| anyhow!("Path {path} don't have a file name"))?;
+        .ok_or_else(|| anyhow!("Path {} don't have a file name", path.display()))?
+        .to_string_lossy();
     let mut new_path = path.to_owned();
 
     for i in 0u32.. {

pageserver/src/tenant/timeline/eviction_task.rs

@@ -30,7 +30,6 @@ use crate::{
     tenant::{
         config::{EvictionPolicy, EvictionPolicyLayerAccessThreshold},
         storage_layer::PersistentLayer,
-        tasks::{BackgroundLoopKind, RateLimitError},
         timeline::EvictionError,
         LogicalSizeCalculationCause, Tenant,
     },
@@ -130,11 +129,7 @@ impl Timeline {
                     ControlFlow::Continue(()) => (),
                 }
                 let elapsed = start.elapsed();
-                crate::tenant::tasks::warn_when_period_overrun(
-                    elapsed,
-                    p.period,
-                    BackgroundLoopKind::Eviction,
-                );
+                crate::tenant::tasks::warn_when_period_overrun(elapsed, p.period, "eviction");
                 crate::metrics::EVICTION_ITERATION_DURATION
                     .get_metric_with_label_values(&[
                         &format!("{}", p.period.as_secs()),
@@ -155,17 +150,6 @@ impl Timeline {
     ) -> ControlFlow<()> {
         let now = SystemTime::now();
 
-        let _permit = match crate::tenant::tasks::concurrent_background_tasks_rate_limit(
-            BackgroundLoopKind::Eviction,
-            ctx,
-            cancel,
-        )
-        .await
-        {
-            Ok(permit) => permit,
-            Err(RateLimitError::Cancelled) => return ControlFlow::Break(()),
-        };
-
         // If we evict layers but keep cached values derived from those layers, then
         // we face a storm of on-demand downloads after pageserver restart.
         // The reason is that the restart empties the caches, and so, the values
@@ -301,10 +285,6 @@ impl Timeline {
                     warn!(layer = %l, "failed to evict layer: {e}");
                     stats.not_evictable += 1;
                 }
-                Some(Err(EvictionError::MetadataInconsistency(detail))) => {
-                    warn!(layer = %l, "failed to evict layer: {detail}");
-                    stats.not_evictable += 1;
-                }
             }
         }
         if stats.candidates == stats.not_evictable {

pageserver/src/tenant/timeline/init.rs

@@ -12,8 +12,7 @@ use crate::{
     METADATA_FILE_NAME,
 };
 use anyhow::Context;
-use camino::Utf8Path;
-use std::{collections::HashMap, str::FromStr};
+use std::{collections::HashMap, ffi::OsString, path::Path, str::FromStr};
 use utils::lsn::Lsn;
 
 /// Identified files in the timeline directory.
@@ -21,43 +20,46 @@ pub(super) enum Discovered {
     /// The only one we care about
     Layer(LayerFileName, u64),
     /// Old ephmeral files from previous launches, should be removed
-    Ephemeral(String),
+    Ephemeral(OsString),
     /// Old temporary timeline files, unsure what these really are, should be removed
-    Temporary(String),
+    Temporary(OsString),
     /// Temporary on-demand download files, should be removed
-    TemporaryDownload(String),
+    TemporaryDownload(OsString),
     /// "metadata" file we persist locally and include in `index_part.json`
     Metadata,
     /// Backup file from previously future layers
     IgnoredBackup,
     /// Unrecognized, warn about these
-    Unknown(String),
+    Unknown(OsString),
 }
 
 /// Scans the timeline directory for interesting files.
-pub(super) fn scan_timeline_dir(path: &Utf8Path) -> anyhow::Result<Vec<Discovered>> {
+pub(super) fn scan_timeline_dir(path: &Path) -> anyhow::Result<Vec<Discovered>> {
     let mut ret = Vec::new();
 
-    for direntry in path.read_dir_utf8()? {
+    for direntry in std::fs::read_dir(path)? {
         let direntry = direntry?;
-        let file_name = direntry.file_name().to_string();
+        let direntry_path = direntry.path();
+        let file_name = direntry.file_name();
 
-        let discovered = match LayerFileName::from_str(&file_name) {
+        let fname = file_name.to_string_lossy();
+
+        let discovered = match LayerFileName::from_str(&fname) {
             Ok(file_name) => {
                 let file_size = direntry.metadata()?.len();
                 Discovered::Layer(file_name, file_size)
             }
             Err(_) => {
-                if file_name == METADATA_FILE_NAME {
+                if fname == METADATA_FILE_NAME {
                     Discovered::Metadata
-                } else if file_name.ends_with(".old") {
+                } else if fname.ends_with(".old") {
                     // ignore these
                     Discovered::IgnoredBackup
-                } else if remote_timeline_client::is_temp_download_file(direntry.path()) {
+                } else if remote_timeline_client::is_temp_download_file(&direntry_path) {
                     Discovered::TemporaryDownload(file_name)
-                } else if is_ephemeral_file(&file_name) {
+                } else if is_ephemeral_file(&fname) {
                     Discovered::Ephemeral(file_name)
-                } else if is_temporary(direntry.path()) {
+                } else if is_temporary(&direntry_path) {
                     Discovered::Temporary(file_name)
                 } else {
                     Discovered::Unknown(file_name)
@@ -160,14 +162,15 @@ pub(super) fn reconcile(
         .collect::<Vec<_>>()
 }
 
-pub(super) fn cleanup(path: &Utf8Path, kind: &str) -> anyhow::Result<()> {
+pub(super) fn cleanup(path: &Path, kind: &str) -> anyhow::Result<()> {
     let file_name = path.file_name().expect("must be file path");
     tracing::debug!(kind, ?file_name, "cleaning up");
-    std::fs::remove_file(path).with_context(|| format!("failed to remove {kind} at {path}"))
+    std::fs::remove_file(path)
+        .with_context(|| format!("failed to remove {kind} at {}", path.display()))
 }
 
 pub(super) fn cleanup_local_file_for_remote(
-    path: &Utf8Path,
+    path: &Path,
     local: &LayerFileMetadata,
     remote: &LayerFileMetadata,
 ) -> anyhow::Result<()> {
@@ -179,7 +182,8 @@ pub(super) fn cleanup_local_file_for_remote(
     if let Err(err) = crate::tenant::timeline::rename_to_backup(path) {
         assert!(
             path.exists(),
-            "we would leave the local_layer without a file if this does not hold: {path}",
+            "we would leave the local_layer without a file if this does not hold: {}",
+            path.display()
         );
         Err(err)
     } else {
@@ -188,7 +192,7 @@ pub(super) fn cleanup_local_file_for_remote(
 }
 
 pub(super) fn cleanup_future_layer(
-    path: &Utf8Path,
+    path: &Path,
     name: &LayerFileName,
     disk_consistent_lsn: Lsn,
 ) -> anyhow::Result<()> {

pageserver/src/tenant/timeline/uninit.rs

@@ -1,7 +1,6 @@
-use std::{collections::hash_map::Entry, fs, sync::Arc};
+use std::{collections::hash_map::Entry, fs, path::PathBuf, sync::Arc};
 
 use anyhow::Context;
-use camino::Utf8PathBuf;
 use tracing::{error, info, info_span, warn};
 use utils::{crashsafe, fs_ext, id::TimelineId, lsn::Lsn};
 
@@ -156,12 +155,12 @@ pub(crate) fn cleanup_timeline_directory(uninit_mark: TimelineUninitMark) {
 #[must_use]
 pub(crate) struct TimelineUninitMark {
     uninit_mark_deleted: bool,
-    uninit_mark_path: Utf8PathBuf,
-    pub(crate) timeline_path: Utf8PathBuf,
+    uninit_mark_path: PathBuf,
+    pub(crate) timeline_path: PathBuf,
 }
 
 impl TimelineUninitMark {
-    pub(crate) fn new(uninit_mark_path: Utf8PathBuf, timeline_path: Utf8PathBuf) -> Self {
+    pub(crate) fn new(uninit_mark_path: PathBuf, timeline_path: PathBuf) -> Self {
         Self {
             uninit_mark_deleted: false,
             uninit_mark_path,
@@ -198,13 +197,14 @@ impl Drop for TimelineUninitMark {
             if self.timeline_path.exists() {
                 error!(
                     "Uninit mark {} is not removed, timeline {} stays uninitialized",
-                    self.uninit_mark_path, self.timeline_path
+                    self.uninit_mark_path.display(),
+                    self.timeline_path.display()
                 )
             } else {
                 // unblock later timeline creation attempts
                 warn!(
                     "Removing intermediate uninit mark file {}",
-                    self.uninit_mark_path
+                    self.uninit_mark_path.display()
                 );
                 if let Err(e) = self.delete_mark_file_if_present() {
                     error!("Failed to remove the uninit mark file: {e}")

pageserver/src/tenant/upload_queue.rs

@@ -203,18 +203,6 @@ impl UploadQueue {
             UploadQueue::Stopped(stopped) => Ok(stopped),
         }
     }
-
-    pub(crate) fn get_layer_metadata(
-        &self,
-        name: &LayerFileName,
-    ) -> anyhow::Result<Option<LayerFileMetadata>> {
-        match self {
-            UploadQueue::Stopped(_) | UploadQueue::Uninitialized => {
-                anyhow::bail!("queue is in state {}", self.as_str())
-            }
-            UploadQueue::Initialized(inner) => Ok(inner.latest_files.get(name).cloned()),
-        }
-    }
 }
 
 /// An in-progress upload or delete task.
@@ -265,7 +253,7 @@ impl std::fmt::Display for UploadOp {
                 write!(f, "UploadMetadata(lsn: {})", lsn)
             }
             UploadOp::Delete(delete) => {
-                write!(f, "Delete({} layers)", delete.layers.len())
+                write!(f, "Delete({} layers)", delete.layers.len(),)
             }
             UploadOp::Barrier(_) => write!(f, "Barrier"),
         }

pageserver/src/trace.rs

@@ -1,8 +1,8 @@
 use bytes::Bytes;
-use camino::Utf8PathBuf;
 use std::{
     fs::{create_dir_all, File},
     io::{BufWriter, Write},
+    path::PathBuf,
 };
 
 pub struct Tracer {
@@ -16,7 +16,7 @@ impl Drop for Tracer {
 }
 
 impl Tracer {
-    pub fn new(path: Utf8PathBuf) -> Self {
+    pub fn new(path: PathBuf) -> Self {
         let parent = path.parent().expect("failed to parse parent path");
         create_dir_all(parent).expect("failed to create trace dir");
 

pageserver/src/virtual_file.rs

@@ -12,13 +12,14 @@
 //!
 use crate::metrics::{StorageIoOperation, STORAGE_IO_SIZE, STORAGE_IO_TIME_METRIC};
 use crate::tenant::TENANTS_SEGMENT_NAME;
-use camino::{Utf8Path, Utf8PathBuf};
 use once_cell::sync::OnceCell;
 use std::fs::{self, File, OpenOptions};
 use std::io::{Error, ErrorKind, Seek, SeekFrom};
 use std::os::unix::fs::FileExt;
+use std::path::{Path, PathBuf};
 use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
-use std::sync::{RwLock, RwLockWriteGuard};
+use tokio::sync::{RwLock, RwLockReadGuard, RwLockWriteGuard};
+use tokio::time::Instant;
 
 ///
 /// A virtual file descriptor. You can use this just like std::fs::File, but internally
@@ -51,7 +52,7 @@ pub struct VirtualFile {
     /// if a new file is created, we only pass the create flag when it's initially
     /// opened, in the VirtualFile::create() function, and strip the flag before
     /// storing it here.
-    pub path: Utf8PathBuf,
+    pub path: PathBuf,
     open_options: OpenOptions,
 
     // These are strings becase we only use them for metrics, and those expect strings.
@@ -110,7 +111,7 @@ impl OpenFiles {
     ///
     /// On return, we hold a lock on the slot, and its 'tag' has been updated
     /// recently_used has been set. It's all ready for reuse.
-    fn find_victim_slot(&self) -> (SlotHandle, RwLockWriteGuard<SlotInner>) {
+    async fn find_victim_slot(&self) -> (SlotHandle, RwLockWriteGuard<SlotInner>) {
         //
         // Run the clock algorithm to find a slot to replace.
         //
@@ -142,7 +143,7 @@ impl OpenFiles {
                 }
                 retries += 1;
             } else {
-                slot_guard = slot.inner.write().unwrap();
+                slot_guard = slot.inner.write().await;
                 index = next;
                 break;
             }
@@ -153,7 +154,7 @@ impl OpenFiles {
         // old file.
         //
         if let Some(old_file) = slot_guard.file.take() {
-            // the normal path of dropping VirtualFile uses "close", use "close-by-replace" here to
+            // the normal path of dropping VirtualFile uses `Close`, use `CloseByReplace` here to
             // distinguish the two.
             STORAGE_IO_TIME_METRIC
                 .get(StorageIoOperation::CloseByReplace)
@@ -177,19 +178,19 @@ impl OpenFiles {
 pub enum CrashsafeOverwriteError {
     #[error("final path has no parent dir")]
     FinalPathHasNoParentDir,
-    #[error("remove tempfile")]
+    #[error("remove tempfile: {0}")]
     RemovePreviousTempfile(#[source] std::io::Error),
-    #[error("create tempfile")]
+    #[error("create tempfile: {0}")]
     CreateTempfile(#[source] std::io::Error),
-    #[error("write tempfile")]
+    #[error("write tempfile: {0}")]
     WriteContents(#[source] std::io::Error),
-    #[error("sync tempfile")]
+    #[error("sync tempfile: {0}")]
     SyncTempfile(#[source] std::io::Error),
-    #[error("rename tempfile to final path")]
+    #[error("rename tempfile to final path: {0}")]
     RenameTempfileToFinalPath(#[source] std::io::Error),
-    #[error("open final path parent dir")]
+    #[error("open final path parent dir: {0}")]
     OpenFinalPathParentDir(#[source] std::io::Error),
-    #[error("sync final path parent dir")]
+    #[error("sync final path parent dir: {0}")]
     SyncFinalPathParentDir(#[source] std::io::Error),
 }
 impl CrashsafeOverwriteError {
@@ -208,15 +209,38 @@ impl CrashsafeOverwriteError {
     }
 }
 
+/// Observe duration for the given storage I/O operation
+///
+/// Unlike `observe_closure_duration`, this supports async,
+/// where "support" means that we measure wall clock time.
+macro_rules! observe_duration {
+    ($op:expr, $($body:tt)*) => {{
+        let instant = Instant::now();
+        let result = $($body)*;
+        let elapsed = instant.elapsed().as_secs_f64();
+        STORAGE_IO_TIME_METRIC
+            .get($op)
+            .observe(elapsed);
+        result
+    }}
+}
+
+macro_rules! with_file {
+    ($this:expr, $op:expr, | $ident:ident | $($body:tt)*) => {{
+        let $ident = $this.lock_file().await?;
+        observe_duration!($op, $($body)*)
+    }};
+}
+
 impl VirtualFile {
     /// Open a file in read-only mode. Like File::open.
-    pub async fn open(path: &Utf8Path) -> Result<VirtualFile, std::io::Error> {
+    pub async fn open(path: &Path) -> Result<VirtualFile, std::io::Error> {
         Self::open_with_options(path, OpenOptions::new().read(true)).await
     }
 
     /// Create a new file for writing. If the file exists, it will be truncated.
     /// Like File::create.
-    pub async fn create(path: &Utf8Path) -> Result<VirtualFile, std::io::Error> {
+    pub async fn create(path: &Path) -> Result<VirtualFile, std::io::Error> {
         Self::open_with_options(
             path,
             OpenOptions::new().write(true).create(true).truncate(true),
@@ -230,10 +254,10 @@ impl VirtualFile {
     /// they will be applied also when the file is subsequently re-opened, not only
     /// on the first time. Make sure that's sane!
     pub async fn open_with_options(
-        path: &Utf8Path,
+        path: &Path,
         open_options: &OpenOptions,
     ) -> Result<VirtualFile, std::io::Error> {
-        let path_str = path.to_string();
+        let path_str = path.to_string_lossy();
         let parts = path_str.split('/').collect::<Vec<&str>>();
         let tenant_id;
         let timeline_id;
@@ -244,11 +268,9 @@ impl VirtualFile {
             tenant_id = "*".to_string();
             timeline_id = "*".to_string();
         }
-        let (handle, mut slot_guard) = get_open_files().find_victim_slot();
+        let (handle, mut slot_guard) = get_open_files().find_victim_slot().await;
 
-        let file = STORAGE_IO_TIME_METRIC
-            .get(StorageIoOperation::Open)
-            .observe_closure_duration(|| open_options.open(path))?;
+        let file = observe_duration!(StorageIoOperation::Open, open_options.open(path))?;
 
         // Strip all options other than read and write.
         //
@@ -281,8 +303,8 @@ impl VirtualFile {
     /// atomic, a crash during the write operation will never leave behind a
     /// partially written file.
     pub async fn crashsafe_overwrite(
-        final_path: &Utf8Path,
-        tmp_path: &Utf8Path,
+        final_path: &Path,
+        tmp_path: &Path,
         content: &[u8],
     ) -> Result<(), CrashsafeOverwriteError> {
         let Some(final_path_parent) = final_path.parent() else {
@@ -331,22 +353,24 @@ impl VirtualFile {
 
     /// Call File::sync_all() on the underlying File.
     pub async fn sync_all(&self) -> Result<(), Error> {
-        self.with_file(StorageIoOperation::Fsync, |file| file.sync_all())
-            .await?
+        with_file!(self, StorageIoOperation::Fsync, |file| file
+            .as_ref()
+            .sync_all())
     }
 
     pub async fn metadata(&self) -> Result<fs::Metadata, Error> {
-        self.with_file(StorageIoOperation::Metadata, |file| file.metadata())
-            .await?
+        with_file!(self, StorageIoOperation::Metadata, |file| file
+            .as_ref()
+            .metadata())
     }
 
-    /// Helper function that looks up the underlying File for this VirtualFile,
-    /// opening it and evicting some other File if necessary. It calls 'func'
-    /// with the physical File.
-    async fn with_file<F, R>(&self, op: StorageIoOperation, mut func: F) -> Result<R, Error>
-    where
-        F: FnMut(&File) -> R,
-    {
+    /// Helper function internal to `VirtualFile` that looks up the underlying File,
+    /// opens it and evicts some other File if necessary. The passed parameter is
+    /// assumed to be a function available for the physical `File`.
+    ///
+    /// We are doing it via a macro as Rust doesn't support async closures that
+    /// take on parameters with lifetimes.
+    async fn lock_file(&self) -> Result<FileGuard<'_>, Error> {
         let open_files = get_open_files();
 
         let mut handle_guard = {
@@ -356,27 +380,23 @@ impl VirtualFile {
             // We only need to hold the handle lock while we read the current handle. If
             // another thread closes the file and recycles the slot for a different file,
             // we will notice that the handle we read is no longer valid and retry.
-            let mut handle = *self.handle.read().unwrap();
+            let mut handle = *self.handle.read().await;
             loop {
                 // Check if the slot contains our File
                 {
                     let slot = &open_files.slots[handle.index];
-                    let slot_guard = slot.inner.read().unwrap();
-                    if slot_guard.tag == handle.tag {
-                        if let Some(file) = &slot_guard.file {
-                            // Found a cached file descriptor.
-                            slot.recently_used.store(true, Ordering::Relaxed);
-                            return Ok(STORAGE_IO_TIME_METRIC
-                                .get(op)
-                                .observe_closure_duration(|| func(file)));
-                        }
+                    let slot_guard = slot.inner.read().await;
+                    if slot_guard.tag == handle.tag && slot_guard.file.is_some() {
+                        // Found a cached file descriptor.
+                        slot.recently_used.store(true, Ordering::Relaxed);
+                        return Ok(FileGuard { slot_guard });
                     }
                 }
 
                 // The slot didn't contain our File. We will have to open it ourselves,
                 // but before that, grab a write lock on handle in the VirtualFile, so
                 // that no other thread will try to concurrently open the same file.
-                let handle_guard = self.handle.write().unwrap();
+                let handle_guard = self.handle.write().await;
 
                 // If another thread changed the handle while we were not holding the lock,
                 // then the handle might now be valid again. Loop back to retry.
@@ -390,17 +410,10 @@ impl VirtualFile {
 
         // We need to open the file ourselves. The handle in the VirtualFile is
         // now locked in write-mode. Find a free slot to put it in.
-        let (handle, mut slot_guard) = open_files.find_victim_slot();
+        let (handle, mut slot_guard) = open_files.find_victim_slot().await;
 
         // Open the physical file
-        let file = STORAGE_IO_TIME_METRIC
-            .get(StorageIoOperation::Open)
-            .observe_closure_duration(|| self.open_options.open(&self.path))?;
-
-        // Perform the requested operation on it
-        let result = STORAGE_IO_TIME_METRIC
-            .get(op)
-            .observe_closure_duration(|| func(&file));
+        let file = observe_duration!(StorageIoOperation::Open, self.open_options.open(&self.path))?;
 
         // Store the File in the slot and update the handle in the VirtualFile
         // to point to it.
@@ -408,7 +421,9 @@ impl VirtualFile {
 
         *handle_guard = handle;
 
-        Ok(result)
+        return Ok(FileGuard {
+            slot_guard: slot_guard.downgrade(),
+        });
     }
 
     pub fn remove(self) {
@@ -423,11 +438,9 @@ impl VirtualFile {
                 self.pos = offset;
             }
             SeekFrom::End(offset) => {
-                self.pos = self
-                    .with_file(StorageIoOperation::Seek, |mut file| {
-                        file.seek(SeekFrom::End(offset))
-                    })
-                    .await??
+                self.pos = with_file!(self, StorageIoOperation::Seek, |file| file
+                    .as_ref()
+                    .seek(SeekFrom::End(offset)))?
             }
             SeekFrom::Current(offset) => {
                 let pos = self.pos as i128 + offset as i128;
@@ -515,9 +528,9 @@ impl VirtualFile {
     }
 
     pub async fn read_at(&self, buf: &mut [u8], offset: u64) -> Result<usize, Error> {
-        let result = self
-            .with_file(StorageIoOperation::Read, |file| file.read_at(buf, offset))
-            .await?;
+        let result = with_file!(self, StorageIoOperation::Read, |file| file
+            .as_ref()
+            .read_at(buf, offset));
         if let Ok(size) = result {
             STORAGE_IO_SIZE
                 .with_label_values(&["read", &self.tenant_id, &self.timeline_id])
@@ -527,9 +540,9 @@ impl VirtualFile {
     }
 
     async fn write_at(&self, buf: &[u8], offset: u64) -> Result<usize, Error> {
-        let result = self
-            .with_file(StorageIoOperation::Write, |file| file.write_at(buf, offset))
-            .await?;
+        let result = with_file!(self, StorageIoOperation::Write, |file| file
+            .as_ref()
+            .write_at(buf, offset));
         if let Ok(size) = result {
             STORAGE_IO_SIZE
                 .with_label_values(&["write", &self.tenant_id, &self.timeline_id])
@@ -539,6 +552,18 @@ impl VirtualFile {
     }
 }
 
+struct FileGuard<'a> {
+    slot_guard: RwLockReadGuard<'a, SlotInner>,
+}
+
+impl<'a> AsRef<File> for FileGuard<'a> {
+    fn as_ref(&self) -> &File {
+        // This unwrap is safe because we only create `FileGuard`s
+        // if we know that the file is Some.
+        self.slot_guard.file.as_ref().unwrap()
+    }
+}
+
 #[cfg(test)]
 impl VirtualFile {
     pub(crate) async fn read_blk(
@@ -571,20 +596,39 @@ impl VirtualFile {
 impl Drop for VirtualFile {
     /// If a VirtualFile is dropped, close the underlying file if it was open.
     fn drop(&mut self) {
-        let handle = self.handle.get_mut().unwrap();
+        let handle = self.handle.get_mut();
 
-        // We could check with a read-lock first, to avoid waiting on an
-        // unrelated I/O.
-        let slot = &get_open_files().slots[handle.index];
-        let mut slot_guard = slot.inner.write().unwrap();
-        if slot_guard.tag == handle.tag {
-            slot.recently_used.store(false, Ordering::Relaxed);
-            // there is also operation "close-by-replace" for closes done on eviction for
-            // comparison.
-            STORAGE_IO_TIME_METRIC
-                .get(StorageIoOperation::Close)
-                .observe_closure_duration(|| drop(slot_guard.file.take()));
+        fn clean_slot(slot: &Slot, mut slot_guard: RwLockWriteGuard<'_, SlotInner>, tag: u64) {
+            if slot_guard.tag == tag {
+                slot.recently_used.store(false, Ordering::Relaxed);
+                // there is also the `CloseByReplace` operation for closes done on eviction for
+                // comparison.
+                STORAGE_IO_TIME_METRIC
+                    .get(StorageIoOperation::Close)
+                    .observe_closure_duration(|| drop(slot_guard.file.take()));
+            }
         }
+
+        // We don't have async drop so we cannot directly await the lock here.
+        // Instead, first do a best-effort attempt at closing the underlying
+        // file descriptor by using `try_write`, and if that fails, spawn
+        // a tokio task to do it asynchronously: we just want it to be
+        // cleaned up eventually.
+        // Most of the time, the `try_lock` should succeed though,
+        // as we have `&mut self` access. In other words, if the slot
+        // is still occupied by our file, there should be no access from
+        // other I/O operations; the only other possible place to lock
+        // the slot is the lock algorithm looking for free slots.
+        let slot = &get_open_files().slots[handle.index];
+        if let Ok(slot_guard) = slot.inner.try_write() {
+            clean_slot(slot, slot_guard, handle.tag);
+        } else {
+            let tag = handle.tag;
+            tokio::spawn(async move {
+                let slot_guard = slot.inner.write().await;
+                clean_slot(slot, slot_guard, tag);
+            });
+        };
     }
 }
 
@@ -734,7 +778,7 @@ mod tests {
 
     async fn test_files<OF, FT>(testname: &str, openfunc: OF) -> Result<(), Error>
     where
-        OF: Fn(Utf8PathBuf, OpenOptions) -> FT,
+        OF: Fn(PathBuf, OpenOptions) -> FT,
         FT: Future<Output = Result<MaybeVirtualFile, std::io::Error>>,
     {
         let testdir = crate::config::PageServerConf::test_repo_dir(testname);

pageserver/src/walredo.rs

@@ -18,13 +18,11 @@
 //! any WAL records, so that even if an attacker hijacks the Postgres
 //! process, he cannot escape out of it.
 //!
-use anyhow::Context;
 use byteorder::{ByteOrder, LittleEndian};
 use bytes::{BufMut, Bytes, BytesMut};
 use nix::poll::*;
 use serde::Serialize;
 use std::collections::VecDeque;
-use std::io;
 use std::io::prelude::*;
 use std::io::{Error, ErrorKind};
 use std::ops::{Deref, DerefMut};
@@ -35,13 +33,11 @@ use std::process::{Child, ChildStderr, ChildStdin, ChildStdout, Command};
 use std::sync::{Mutex, MutexGuard};
 use std::time::Duration;
 use std::time::Instant;
+use std::{fs, io};
 use tracing::*;
+use utils::crashsafe::path_with_suffix_extension;
 use utils::{bin_ser::BeSer, id::TenantId, lsn::Lsn, nonblock::set_nonblock};
 
-#[cfg(feature = "testing")]
-use std::sync::atomic::{AtomicUsize, Ordering};
-
-use crate::config::PageServerConf;
 use crate::metrics::{
     WAL_REDO_BYTES_HISTOGRAM, WAL_REDO_RECORDS_HISTOGRAM, WAL_REDO_RECORD_COUNTER, WAL_REDO_TIME,
     WAL_REDO_WAIT_TIME,
@@ -50,6 +46,7 @@ use crate::pgdatadir_mapping::{key_to_rel_block, key_to_slru_block};
 use crate::repository::Key;
 use crate::task_mgr::BACKGROUND_RUNTIME;
 use crate::walrecord::NeonWalRecord;
+use crate::{config::PageServerConf, TEMP_FILE_SUFFIX};
 use pageserver_api::reltag::{RelTag, SlruKind};
 use postgres_ffi::pg_constants;
 use postgres_ffi::relfile_utils::VISIBILITYMAP_FORKNUM;
@@ -66,7 +63,7 @@ use postgres_ffi::BLCKSZ;
 /// [See more related comments here](https://github.com/postgres/postgres/blob/99c5852e20a0987eca1c38ba0c09329d4076b6a0/src/include/storage/buf_internals.h#L91).
 ///
 #[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Clone, Copy, Serialize)]
-pub(crate) struct BufferTag {
+pub struct BufferTag {
     pub rel: RelTag,
     pub blknum: u32,
 }
@@ -89,7 +86,7 @@ pub trait WalRedoManager: Send + Sync {
         base_img: Option<(Lsn, Bytes)>,
         records: Vec<(Lsn, NeonWalRecord)>,
         pg_version: u32,
-    ) -> anyhow::Result<Bytes>;
+    ) -> Result<Bytes, WalRedoError>;
 }
 
 struct ProcessInput {
@@ -116,9 +113,6 @@ struct ProcessOutput {
 pub struct PostgresRedoManager {
     tenant_id: TenantId,
     conf: &'static PageServerConf,
-    /// Counter to separate same sized walredo inputs failing at the same millisecond.
-    #[cfg(feature = "testing")]
-    dump_sequence: AtomicUsize,
 
     stdout: Mutex<Option<ProcessOutput>>,
     stdin: Mutex<Option<ProcessInput>>,
@@ -140,6 +134,20 @@ fn can_apply_in_neon(rec: &NeonWalRecord) -> bool {
     }
 }
 
+/// An error happened in WAL redo
+#[derive(Debug, thiserror::Error)]
+pub enum WalRedoError {
+    #[error(transparent)]
+    IoError(#[from] std::io::Error),
+
+    #[error("cannot perform WAL redo now")]
+    InvalidState,
+    #[error("cannot perform WAL redo for this request")]
+    InvalidRequest,
+    #[error("cannot perform WAL redo for this record")]
+    InvalidRecord,
+}
+
 ///
 /// Public interface of WAL redo manager
 ///
@@ -157,9 +165,10 @@ impl WalRedoManager for PostgresRedoManager {
         base_img: Option<(Lsn, Bytes)>,
         records: Vec<(Lsn, NeonWalRecord)>,
         pg_version: u32,
-    ) -> anyhow::Result<Bytes> {
+    ) -> Result<Bytes, WalRedoError> {
         if records.is_empty() {
-            anyhow::bail!("invalid WAL redo request with no records");
+            error!("invalid WAL redo request with no records");
+            return Err(WalRedoError::InvalidRequest);
         }
 
         let base_img_lsn = base_img.as_ref().map(|p| p.0).unwrap_or(Lsn::INVALID);
@@ -215,14 +224,21 @@ impl PostgresRedoManager {
         PostgresRedoManager {
             tenant_id,
             conf,
-            #[cfg(feature = "testing")]
-            dump_sequence: AtomicUsize::default(),
             stdin: Mutex::new(None),
             stdout: Mutex::new(None),
             stderr: Mutex::new(None),
         }
     }
 
+    /// Launch process pre-emptively. Should not be needed except for benchmarking.
+    pub fn launch_process(&self, pg_version: u32) -> anyhow::Result<()> {
+        let mut proc = self.stdin.lock().unwrap();
+        if proc.is_none() {
+            self.launch(&mut proc, pg_version)?;
+        }
+        Ok(())
+    }
+
     ///
     /// Process one request for WAL redo using wal-redo postgres
     ///
@@ -236,8 +252,8 @@ impl PostgresRedoManager {
         records: &[(Lsn, NeonWalRecord)],
         wal_redo_timeout: Duration,
         pg_version: u32,
-    ) -> anyhow::Result<Bytes> {
-        let (rel, blknum) = key_to_rel_block(key).context("invalid record")?;
+    ) -> Result<Bytes, WalRedoError> {
+        let (rel, blknum) = key_to_rel_block(key).or(Err(WalRedoError::InvalidRecord))?;
         const MAX_RETRY_ATTEMPTS: u32 = 1;
         let start_time = Instant::now();
         let mut n_attempts = 0u32;
@@ -247,8 +263,7 @@ impl PostgresRedoManager {
 
             // launch the WAL redo process on first use
             if proc.is_none() {
-                self.launch(&mut proc, pg_version)
-                    .context("launch process")?;
+                self.launch(&mut proc, pg_version)?;
             }
             WAL_REDO_WAIT_TIME.observe(lock_time.duration_since(start_time).as_secs_f64());
 
@@ -256,7 +271,7 @@ impl PostgresRedoManager {
             let buf_tag = BufferTag { rel, blknum };
             let result = self
                 .apply_wal_records(proc, buf_tag, &base_img, records, wal_redo_timeout)
-                .context("apply_wal_records");
+                .map_err(WalRedoError::IoError);
 
             let end_time = Instant::now();
             let duration = end_time.duration_since(lock_time);
@@ -275,27 +290,25 @@ impl PostgresRedoManager {
             WAL_REDO_BYTES_HISTOGRAM.observe(nbytes as f64);
 
             debug!(
-                "postgres applied {} WAL records ({} bytes) in {} us to reconstruct page image at LSN {}",
-                len,
-                nbytes,
-                duration.as_micros(),
-                lsn
-            );
+				"postgres applied {} WAL records ({} bytes) in {} us to reconstruct page image at LSN {}",
+				len,
+				nbytes,
+				duration.as_micros(),
+				lsn
+			);
 
             // If something went wrong, don't try to reuse the process. Kill it, and
             // next request will launch a new one.
-            if let Err(e) = result.as_ref() {
+            if result.is_err() {
                 error!(
-                    "error applying {} WAL records {}..{} ({} bytes) to base image with LSN {} to reconstruct page image at LSN {} n_attempts={}: {:?}",
-                    records.len(),
-                    records.first().map(|p| p.0).unwrap_or(Lsn(0)),
-                    records.last().map(|p| p.0).unwrap_or(Lsn(0)),
-                    nbytes,
-                    base_img_lsn,
-                    lsn,
-                    n_attempts,
-                    e,
-                );
+                "error applying {} WAL records {}..{} ({} bytes) to base image with LSN {} to reconstruct page image at LSN {}",
+                records.len(),
+				records.first().map(|p| p.0).unwrap_or(Lsn(0)),
+				records.last().map(|p| p.0).unwrap_or(Lsn(0)),
+                nbytes,
+				base_img_lsn,
+                lsn
+            );
                 // self.stdin only holds stdin & stderr as_raw_fd().
                 // Dropping it as part of take() doesn't close them.
                 // The owning objects (ChildStdout and ChildStderr) are stored in
@@ -312,8 +325,6 @@ impl PostgresRedoManager {
                 if let Some(proc) = self.stdin.lock().unwrap().take() {
                     proc.child.kill_and_wait();
                 }
-            } else if n_attempts != 0 {
-                info!(n_attempts, "retried walredo succeeded");
             }
             n_attempts += 1;
             if n_attempts > MAX_RETRY_ATTEMPTS || result.is_ok() {
@@ -331,7 +342,7 @@ impl PostgresRedoManager {
         lsn: Lsn,
         base_img: Option<Bytes>,
         records: &[(Lsn, NeonWalRecord)],
-    ) -> anyhow::Result<Bytes> {
+    ) -> Result<Bytes, WalRedoError> {
         let start_time = Instant::now();
 
         let mut page = BytesMut::new();
@@ -340,7 +351,8 @@ impl PostgresRedoManager {
             page.extend_from_slice(&fpi[..]);
         } else {
             // All the current WAL record types that we can handle require a base image.
-            anyhow::bail!("invalid neon WAL redo request with no base image");
+            error!("invalid neon WAL redo request with no base image");
+            return Err(WalRedoError::InvalidRequest);
         }
 
         // Apply all the WAL records in the batch
@@ -368,13 +380,14 @@ impl PostgresRedoManager {
         page: &mut BytesMut,
         _record_lsn: Lsn,
         record: &NeonWalRecord,
-    ) -> anyhow::Result<()> {
+    ) -> Result<(), WalRedoError> {
         match record {
             NeonWalRecord::Postgres {
                 will_init: _,
                 rec: _,
             } => {
-                anyhow::bail!("tried to pass postgres wal record to neon WAL redo");
+                error!("tried to pass postgres wal record to neon WAL redo");
+                return Err(WalRedoError::InvalidRequest);
             }
             NeonWalRecord::ClearVisibilityMapFlags {
                 new_heap_blkno,
@@ -382,7 +395,7 @@ impl PostgresRedoManager {
                 flags,
             } => {
                 // sanity check that this is modifying the correct relation
-                let (rel, blknum) = key_to_rel_block(key).context("invalid record")?;
+                let (rel, blknum) = key_to_rel_block(key).or(Err(WalRedoError::InvalidRecord))?;
                 assert!(
                     rel.forknum == VISIBILITYMAP_FORKNUM,
                     "ClearVisibilityMapFlags record on unexpected rel {}",
@@ -420,7 +433,7 @@ impl PostgresRedoManager {
             // same effects as the corresponding Postgres WAL redo function.
             NeonWalRecord::ClogSetCommitted { xids, timestamp } => {
                 let (slru_kind, segno, blknum) =
-                    key_to_slru_block(key).context("invalid record")?;
+                    key_to_slru_block(key).or(Err(WalRedoError::InvalidRecord))?;
                 assert_eq!(
                     slru_kind,
                     SlruKind::Clog,
@@ -470,7 +483,7 @@ impl PostgresRedoManager {
             }
             NeonWalRecord::ClogSetAborted { xids } => {
                 let (slru_kind, segno, blknum) =
-                    key_to_slru_block(key).context("invalid record")?;
+                    key_to_slru_block(key).or(Err(WalRedoError::InvalidRecord))?;
                 assert_eq!(
                     slru_kind,
                     SlruKind::Clog,
@@ -501,7 +514,7 @@ impl PostgresRedoManager {
             }
             NeonWalRecord::MultixactOffsetCreate { mid, moff } => {
                 let (slru_kind, segno, blknum) =
-                    key_to_slru_block(key).context("invalid record")?;
+                    key_to_slru_block(key).or(Err(WalRedoError::InvalidRecord))?;
                 assert_eq!(
                     slru_kind,
                     SlruKind::MultiXactOffsets,
@@ -534,7 +547,7 @@ impl PostgresRedoManager {
             }
             NeonWalRecord::MultixactMembersCreate { moff, members } => {
                 let (slru_kind, segno, blknum) =
-                    key_to_slru_block(key).context("invalid record")?;
+                    key_to_slru_block(key).or(Err(WalRedoError::InvalidRecord))?;
                 assert_eq!(
                     slru_kind,
                     SlruKind::MultiXactMembers,
@@ -624,6 +637,26 @@ impl PostgresRedoManager {
         input: &mut MutexGuard<Option<ProcessInput>>,
         pg_version: u32,
     ) -> Result<(), Error> {
+        // Previous versions of wal-redo required data directory and that directories
+        // occupied some space on disk. Remove it if we face it.
+        //
+        // This code could be dropped after one release cycle.
+        let legacy_datadir = path_with_suffix_extension(
+            self.conf
+                .tenant_path(&self.tenant_id)
+                .join("wal-redo-datadir"),
+            TEMP_FILE_SUFFIX,
+        );
+        if legacy_datadir.exists() {
+            info!("legacy wal-redo datadir {legacy_datadir:?} exists, removing");
+            fs::remove_dir_all(&legacy_datadir).map_err(|e| {
+                Error::new(
+                    e.kind(),
+                    format!("legacy wal-redo datadir {legacy_datadir:?} removal failure: {e}"),
+                )
+            })?;
+        }
+
         let pg_bin_dir_path = self
             .conf
             .pg_bin_dir(pg_version)
@@ -709,12 +742,12 @@ impl PostgresRedoManager {
     #[instrument(skip_all, fields(tenant_id=%self.tenant_id, pid=%input.as_ref().unwrap().child.id()))]
     fn apply_wal_records(
         &self,
-        input: MutexGuard<Option<ProcessInput>>,
+        mut input: MutexGuard<Option<ProcessInput>>,
         tag: BufferTag,
         base_img: &Option<Bytes>,
         records: &[(Lsn, NeonWalRecord)],
         wal_redo_timeout: Duration,
-    ) -> anyhow::Result<Bytes> {
+    ) -> Result<Bytes, std::io::Error> {
         // Serialize all the messages to send the WAL redo process first.
         //
         // This could be problematic if there are millions of records to replay,
@@ -737,29 +770,15 @@ impl PostgresRedoManager {
             {
                 build_apply_record_msg(*lsn, postgres_rec, &mut writebuf);
             } else {
-                anyhow::bail!("tried to pass neon wal record to postgres WAL redo");
+                return Err(Error::new(
+                    ErrorKind::Other,
+                    "tried to pass neon wal record to postgres WAL redo",
+                ));
             }
         }
         build_get_page_msg(tag, &mut writebuf);
         WAL_REDO_RECORD_COUNTER.inc_by(records.len() as u64);
 
-        let res = self.apply_wal_records0(&writebuf, input, wal_redo_timeout);
-
-        if res.is_err() {
-            // not all of these can be caused by this particular input, however these are so rare
-            // in tests so capture all.
-            self.record_and_log(&writebuf);
-        }
-
-        res
-    }
-
-    fn apply_wal_records0(
-        &self,
-        writebuf: &[u8],
-        mut input: MutexGuard<Option<ProcessInput>>,
-        wal_redo_timeout: Duration,
-    ) -> anyhow::Result<Bytes> {
         let proc = input.as_mut().unwrap();
         let mut nwrite = 0usize;
         let stdout_fd = proc.stdout_fd;
@@ -777,13 +796,13 @@ impl PostgresRedoManager {
         while nwrite < writebuf.len() {
             let n = loop {
                 match nix::poll::poll(&mut pollfds[0..2], wal_redo_timeout.as_millis() as i32) {
-                    Err(nix::errno::Errno::EINTR) => continue,
+                    Err(e) if e == nix::errno::Errno::EINTR => continue,
                     res => break res,
                 }
             }?;
 
             if n == 0 {
-                anyhow::bail!("WAL redo timed out");
+                return Err(Error::new(ErrorKind::Other, "WAL redo timed out"));
             }
 
             // If we have some messages in stderr, forward them to the log.
@@ -807,7 +826,10 @@ impl PostgresRedoManager {
                     continue;
                 }
             } else if err_revents.contains(PollFlags::POLLHUP) {
-                anyhow::bail!("WAL redo process closed its stderr unexpectedly");
+                return Err(Error::new(
+                    ErrorKind::BrokenPipe,
+                    "WAL redo process closed its stderr unexpectedly",
+                ));
             }
 
             // If 'stdin' is writeable, do write.
@@ -816,7 +838,10 @@ impl PostgresRedoManager {
                 nwrite += proc.stdin.write(&writebuf[nwrite..])?;
             } else if in_revents.contains(PollFlags::POLLHUP) {
                 // We still have more data to write, but the process closed the pipe.
-                anyhow::bail!("WAL redo process closed its stdin unexpectedly");
+                return Err(Error::new(
+                    ErrorKind::BrokenPipe,
+                    "WAL redo process closed its stdin unexpectedly",
+                ));
             }
         }
         let request_no = proc.n_requests;
@@ -847,7 +872,10 @@ impl PostgresRedoManager {
             //
             // Cross-read this with the comment in apply_batch_postgres if result.is_err().
             // That's where we kill the child process.
-            anyhow::bail!("WAL redo process closed its stdout unexpectedly");
+            return Err(Error::new(
+                ErrorKind::BrokenPipe,
+                "WAL redo process closed its stdout unexpectedly",
+            ));
         }
         let n_processed_responses = output.n_processed_responses;
         while n_processed_responses + output.pending_responses.len() <= request_no {
@@ -860,13 +888,13 @@ impl PostgresRedoManager {
                 // and forward any logging information that the child writes to its stderr to the page server's log.
                 let n = loop {
                     match nix::poll::poll(&mut pollfds[1..3], wal_redo_timeout.as_millis() as i32) {
-                        Err(nix::errno::Errno::EINTR) => continue,
+                        Err(e) if e == nix::errno::Errno::EINTR => continue,
                         res => break res,
                     }
                 }?;
 
                 if n == 0 {
-                    anyhow::bail!("WAL redo timed out");
+                    return Err(Error::new(ErrorKind::Other, "WAL redo timed out"));
                 }
 
                 // If we have some messages in stderr, forward them to the log.
@@ -890,7 +918,10 @@ impl PostgresRedoManager {
                         continue;
                     }
                 } else if err_revents.contains(PollFlags::POLLHUP) {
-                    anyhow::bail!("WAL redo process closed its stderr unexpectedly");
+                    return Err(Error::new(
+                        ErrorKind::BrokenPipe,
+                        "WAL redo process closed its stderr unexpectedly",
+                    ));
                 }
 
                 // If we have some data in stdout, read it to the result buffer.
@@ -898,7 +929,10 @@ impl PostgresRedoManager {
                 if out_revents & (PollFlags::POLLERR | PollFlags::POLLIN) != PollFlags::empty() {
                     nresult += output.stdout.read(&mut resultbuf[nresult..])?;
                 } else if out_revents.contains(PollFlags::POLLHUP) {
-                    anyhow::bail!("WAL redo process closed its stdout unexpectedly");
+                    return Err(Error::new(
+                        ErrorKind::BrokenPipe,
+                        "WAL redo process closed its stdout unexpectedly",
+                    ));
                 }
             }
             output
@@ -950,38 +984,6 @@ impl PostgresRedoManager {
         }
         Ok(res)
     }
-
-    #[cfg(feature = "testing")]
-    fn record_and_log(&self, writebuf: &[u8]) {
-        let millis = std::time::SystemTime::now()
-            .duration_since(std::time::SystemTime::UNIX_EPOCH)
-            .unwrap()
-            .as_millis();
-
-        let seq = self.dump_sequence.fetch_add(1, Ordering::Relaxed);
-
-        // these files will be collected to an allure report
-        let filename = format!("walredo-{millis}-{}-{seq}.walredo", writebuf.len());
-
-        let path = self.conf.tenant_path(&self.tenant_id).join(&filename);
-
-        let res = std::fs::OpenOptions::new()
-            .write(true)
-            .create_new(true)
-            .read(true)
-            .open(path)
-            .and_then(|mut f| f.write_all(writebuf));
-
-        // trip up allowed_errors
-        if let Err(e) = res {
-            tracing::error!(target=%filename, length=writebuf.len(), "failed to write out the walredo errored input: {e}");
-        } else {
-            tracing::error!(filename, "erroring walredo input saved");
-        }
-    }
-
-    #[cfg(not(feature = "testing"))]
-    fn record_and_log(&self, _: &[u8]) {}
 }
 
 /// Wrapper type around `std::process::Child` which guarantees that the child
@@ -1215,13 +1217,13 @@ mod tests {
 
     struct RedoHarness {
         // underscored because unused, except for removal at drop
-        _repo_dir: camino_tempfile::Utf8TempDir,
+        _repo_dir: tempfile::TempDir,
         manager: PostgresRedoManager,
     }
 
     impl RedoHarness {
         fn new() -> anyhow::Result<Self> {
-            let repo_dir = camino_tempfile::tempdir()?;
+            let repo_dir = tempfile::tempdir()?;
             let conf = PageServerConf::dummy_conf(repo_dir.path().to_path_buf());
             let conf = Box::leak(Box::new(conf));
             let tenant_id = TenantId::generate();

pgxn/neon/Makefile

@@ -7,12 +7,12 @@ OBJS = \
 	extension_server.o \
 	file_cache.o \
 	libpagestore.o \
+	libpqwalproposer.o \
 	neon.o \
-	neon_utils.o \
 	pagestore_smgr.o \
 	relsize_cache.o \
 	walproposer.o \
-	walproposer_pg.o \
+	walproposer_utils.o \
 	control_plane_connector.o
 
 PG_CPPFLAGS = -I$(libpq_srcdir)

pgxn/neon/control_plane_connector.c

@@ -741,6 +741,13 @@ NeonProcessUtility(
 			break;
 		case T_DropdbStmt:
 			HandleDropDb(castNode(DropdbStmt, parseTree));
+			/*
+			 * We do this here to hack around the fact that Postgres performs the drop
+			 * INSIDE of standard_ProcessUtility, which means that if we try to
+			 * abort the drop normally it'll be too late. DROP DATABASE can't be inside
+			 * of a transaction block anyway, so this should be fine to do.
+			 */
+			NeonXactCallback(XACT_EVENT_PRE_COMMIT, NULL);
 			break;
 		case T_CreateRoleStmt:
 			HandleCreateRole(castNode(CreateRoleStmt, parseTree));

pgxn/neon/libpagestore.c

@@ -30,7 +30,7 @@
 
 #include "neon.h"
 #include "walproposer.h"
-#include "neon_utils.h"
+#include "walproposer_utils.h"
 
 #define PageStoreTrace DEBUG5
 

pgxn/neon/libpqwalproposer.c

@@ -0,0 +1,424 @@
+#include "postgres.h"
+
+#include "libpq-fe.h"
+#include "neon.h"
+#include "walproposer.h"
+
+/* Header in walproposer.h -- Wrapper struct to abstract away the libpq connection */
+struct WalProposerConn
+{
+	PGconn	   *pg_conn;
+	bool		is_nonblocking; /* whether the connection is non-blocking */
+	char	   *recvbuf;		/* last received data from
+								 * walprop_async_read */
+};
+
+/* Helper function */
+static bool
+ensure_nonblocking_status(WalProposerConn *conn, bool is_nonblocking)
+{
+	/* If we're already correctly blocking or nonblocking, all good */
+	if (is_nonblocking == conn->is_nonblocking)
+		return true;
+
+	/* Otherwise, set it appropriately */
+	if (PQsetnonblocking(conn->pg_conn, is_nonblocking) == -1)
+		return false;
+
+	conn->is_nonblocking = is_nonblocking;
+	return true;
+}
+
+/* Exported function definitions */
+char *
+walprop_error_message(WalProposerConn *conn)
+{
+	return PQerrorMessage(conn->pg_conn);
+}
+
+WalProposerConnStatusType
+walprop_status(WalProposerConn *conn)
+{
+	switch (PQstatus(conn->pg_conn))
+	{
+		case CONNECTION_OK:
+			return WP_CONNECTION_OK;
+		case CONNECTION_BAD:
+			return WP_CONNECTION_BAD;
+		default:
+			return WP_CONNECTION_IN_PROGRESS;
+	}
+}
+
+WalProposerConn *
+walprop_connect_start(char *conninfo, char *password)
+{
+	WalProposerConn *conn;
+	PGconn	   *pg_conn;
+	const char *keywords[3];
+	const char *values[3];
+	int			n;
+
+	/*
+	 * Connect using the given connection string. If the
+	 * NEON_AUTH_TOKEN environment variable was set, use that as
+	 * the password.
+	 *
+	 * The connection options are parsed in the order they're given, so
+	 * when we set the password before the connection string, the
+	 * connection string can override the password from the env variable.
+	 * Seems useful, although we don't currently use that capability
+	 * anywhere.
+	 */
+	n = 0;
+	if (password)
+	{
+		keywords[n] = "password";
+		values[n] = password;
+		n++;
+	}
+	keywords[n] = "dbname";
+	values[n] = conninfo;
+	n++;
+	keywords[n] = NULL;
+	values[n] = NULL;
+	n++;
+	pg_conn = PQconnectStartParams(keywords, values, 1);
+
+	/*
+	 * Allocation of a PQconn can fail, and will return NULL. We want to fully
+	 * replicate the behavior of PQconnectStart here.
+	 */
+	if (!pg_conn)
+		return NULL;
+
+	/*
+	 * And in theory this allocation can fail as well, but it's incredibly
+	 * unlikely if we just successfully allocated a PGconn.
+	 *
+	 * palloc will exit on failure though, so there's not much we could do if
+	 * it *did* fail.
+	 */
+	conn = palloc(sizeof(WalProposerConn));
+	conn->pg_conn = pg_conn;
+	conn->is_nonblocking = false;	/* connections always start in blocking
+									 * mode */
+	conn->recvbuf = NULL;
+	return conn;
+}
+
+WalProposerConnectPollStatusType
+walprop_connect_poll(WalProposerConn *conn)
+{
+	WalProposerConnectPollStatusType return_val;
+
+	switch (PQconnectPoll(conn->pg_conn))
+	{
+		case PGRES_POLLING_FAILED:
+			return_val = WP_CONN_POLLING_FAILED;
+			break;
+		case PGRES_POLLING_READING:
+			return_val = WP_CONN_POLLING_READING;
+			break;
+		case PGRES_POLLING_WRITING:
+			return_val = WP_CONN_POLLING_WRITING;
+			break;
+		case PGRES_POLLING_OK:
+			return_val = WP_CONN_POLLING_OK;
+			break;
+
+			/*
+			 * There's a comment at its source about this constant being
+			 * unused. We'll expect it's never returned.
+			 */
+		case PGRES_POLLING_ACTIVE:
+			elog(FATAL, "Unexpected PGRES_POLLING_ACTIVE returned from PQconnectPoll");
+
+			/*
+			 * This return is never actually reached, but it's here to make
+			 * the compiler happy
+			 */
+			return WP_CONN_POLLING_FAILED;
+
+		default:
+			Assert(false);
+			return_val = WP_CONN_POLLING_FAILED;	/* keep the compiler quiet */
+	}
+
+	return return_val;
+}
+
+bool
+walprop_send_query(WalProposerConn *conn, char *query)
+{
+	/*
+	 * We need to be in blocking mode for sending the query to run without
+	 * requiring a call to PQflush
+	 */
+	if (!ensure_nonblocking_status(conn, false))
+		return false;
+
+	/* PQsendQuery returns 1 on success, 0 on failure */
+	if (!PQsendQuery(conn->pg_conn, query))
+		return false;
+
+	return true;
+}
+
+WalProposerExecStatusType
+walprop_get_query_result(WalProposerConn *conn)
+{
+	PGresult   *result;
+	WalProposerExecStatusType return_val;
+
+	/* Marker variable if we need to log an unexpected success result */
+	char	   *unexpected_success = NULL;
+
+	/* Consume any input that we might be missing */
+	if (!PQconsumeInput(conn->pg_conn))
+		return WP_EXEC_FAILED;
+
+	if (PQisBusy(conn->pg_conn))
+		return WP_EXEC_NEEDS_INPUT;
+
+
+	result = PQgetResult(conn->pg_conn);
+
+	/*
+	 * PQgetResult returns NULL only if getting the result was successful &
+	 * there's no more of the result to get.
+	 */
+	if (!result)
+	{
+		elog(WARNING, "[libpqwalproposer] Unexpected successful end of command results");
+		return WP_EXEC_UNEXPECTED_SUCCESS;
+	}
+
+	/* Helper macro to reduce boilerplate */
+#define UNEXPECTED_SUCCESS(msg) \
+		return_val = WP_EXEC_UNEXPECTED_SUCCESS; \
+		unexpected_success = msg; \
+		break;
+
+
+	switch (PQresultStatus(result))
+	{
+			/* "true" success case */
+		case PGRES_COPY_BOTH:
+			return_val = WP_EXEC_SUCCESS_COPYBOTH;
+			break;
+
+			/* Unexpected success case */
+		case PGRES_EMPTY_QUERY:
+			UNEXPECTED_SUCCESS("empty query return");
+		case PGRES_COMMAND_OK:
+			UNEXPECTED_SUCCESS("data-less command end");
+		case PGRES_TUPLES_OK:
+			UNEXPECTED_SUCCESS("tuples return");
+		case PGRES_COPY_OUT:
+			UNEXPECTED_SUCCESS("'Copy Out' response");
+		case PGRES_COPY_IN:
+			UNEXPECTED_SUCCESS("'Copy In' response");
+		case PGRES_SINGLE_TUPLE:
+			UNEXPECTED_SUCCESS("single tuple return");
+		case PGRES_PIPELINE_SYNC:
+			UNEXPECTED_SUCCESS("pipeline sync point");
+
+			/* Failure cases */
+		case PGRES_BAD_RESPONSE:
+		case PGRES_NONFATAL_ERROR:
+		case PGRES_FATAL_ERROR:
+		case PGRES_PIPELINE_ABORTED:
+			return_val = WP_EXEC_FAILED;
+			break;
+
+		default:
+			Assert(false);
+			return_val = WP_EXEC_FAILED;	/* keep the compiler quiet */
+	}
+
+	if (unexpected_success)
+		elog(WARNING, "[libpqwalproposer] Unexpected successful %s", unexpected_success);
+
+	return return_val;
+}
+
+pgsocket
+walprop_socket(WalProposerConn *conn)
+{
+	return PQsocket(conn->pg_conn);
+}
+
+int
+walprop_flush(WalProposerConn *conn)
+{
+	return (PQflush(conn->pg_conn));
+}
+
+void
+walprop_finish(WalProposerConn *conn)
+{
+	if (conn->recvbuf != NULL)
+		PQfreemem(conn->recvbuf);
+	PQfinish(conn->pg_conn);
+	pfree(conn);
+}
+
+/*
+ * Receive a message from the safekeeper.
+ *
+ * On success, the data is placed in *buf. It is valid until the next call
+ * to this function.
+ */
+PGAsyncReadResult
+walprop_async_read(WalProposerConn *conn, char **buf, int *amount)
+{
+	int			result;
+
+	if (conn->recvbuf != NULL)
+	{
+		PQfreemem(conn->recvbuf);
+		conn->recvbuf = NULL;
+	}
+
+	/* Call PQconsumeInput so that we have the data we need */
+	if (!PQconsumeInput(conn->pg_conn))
+	{
+		*amount = 0;
+		*buf = NULL;
+		return PG_ASYNC_READ_FAIL;
+	}
+
+	/*
+	 * The docs for PQgetCopyData list the return values as: 0 if the copy is
+	 * still in progress, but no "complete row" is available -1 if the copy is
+	 * done -2 if an error occurred (> 0) if it was successful; that value is
+	 * the amount transferred.
+	 *
+	 * The protocol we use between walproposer and safekeeper means that we
+	 * *usually* wouldn't expect to see that the copy is done, but this can
+	 * sometimes be triggered by the server returning an ErrorResponse (which
+	 * also happens to have the effect that the copy is done).
+	 */
+	switch (result = PQgetCopyData(conn->pg_conn, &conn->recvbuf, true))
+	{
+		case 0:
+			*amount = 0;
+			*buf = NULL;
+			return PG_ASYNC_READ_TRY_AGAIN;
+		case -1:
+			{
+				/*
+				 * If we get -1, it's probably because of a server error; the
+				 * safekeeper won't normally send a CopyDone message.
+				 *
+				 * We can check PQgetResult to make sure that the server
+				 * failed; it'll always result in PGRES_FATAL_ERROR
+				 */
+				ExecStatusType status = PQresultStatus(PQgetResult(conn->pg_conn));
+
+				if (status != PGRES_FATAL_ERROR)
+					elog(FATAL, "unexpected result status %d after failed PQgetCopyData", status);
+
+				/*
+				 * If there was actually an error, it'll be properly reported
+				 * by calls to PQerrorMessage -- we don't have to do anything
+				 * else
+				 */
+				*amount = 0;
+				*buf = NULL;
+				return PG_ASYNC_READ_FAIL;
+			}
+		case -2:
+			*amount = 0;
+			*buf = NULL;
+			return PG_ASYNC_READ_FAIL;
+		default:
+			/* Positive values indicate the size of the returned result */
+			*amount = result;
+			*buf = conn->recvbuf;
+			return PG_ASYNC_READ_SUCCESS;
+	}
+}
+
+PGAsyncWriteResult
+walprop_async_write(WalProposerConn *conn, void const *buf, size_t size)
+{
+	int			result;
+
+	/* If we aren't in non-blocking mode, switch to it. */
+	if (!ensure_nonblocking_status(conn, true))
+		return PG_ASYNC_WRITE_FAIL;
+
+	/*
+	 * The docs for PQputcopyData list the return values as: 1 if the data was
+	 * queued, 0 if it was not queued because of full buffers, or -1 if an
+	 * error occurred
+	 */
+	result = PQputCopyData(conn->pg_conn, buf, size);
+
+	/*
+	 * We won't get a result of zero because walproposer always empties the
+	 * connection's buffers before sending more
+	 */
+	Assert(result != 0);
+
+	switch (result)
+	{
+		case 1:
+			/* good -- continue */
+			break;
+		case -1:
+			return PG_ASYNC_WRITE_FAIL;
+		default:
+			elog(FATAL, "invalid return %d from PQputCopyData", result);
+	}
+
+	/*
+	 * After queueing the data, we still need to flush to get it to send. This
+	 * might take multiple tries, but we don't want to wait around until it's
+	 * done.
+	 *
+	 * PQflush has the following returns (directly quoting the docs): 0 if
+	 * sucessful, 1 if it was unable to send all the data in the send queue
+	 * yet -1 if it failed for some reason
+	 */
+	switch (result = PQflush(conn->pg_conn))
+	{
+		case 0:
+			return PG_ASYNC_WRITE_SUCCESS;
+		case 1:
+			return PG_ASYNC_WRITE_TRY_FLUSH;
+		case -1:
+			return PG_ASYNC_WRITE_FAIL;
+		default:
+			elog(FATAL, "invalid return %d from PQflush", result);
+	}
+}
+
+/*
+ * This function is very similar to walprop_async_write. For more
+ * information, refer to the comments there.
+ */
+bool
+walprop_blocking_write(WalProposerConn *conn, void const *buf, size_t size)
+{
+	int			result;
+
+	/* If we are in non-blocking mode, switch out of it. */
+	if (!ensure_nonblocking_status(conn, false))
+		return false;
+
+	if ((result = PQputCopyData(conn->pg_conn, buf, size)) == -1)
+		return false;
+
+	Assert(result == 1);
+
+	/* Because the connection is non-blocking, flushing returns 0 or -1 */
+
+	if ((result = PQflush(conn->pg_conn)) == -1)
+		return false;
+
+	Assert(result == 0);
+	return true;
+}

pgxn/neon/neon.h

@@ -18,10 +18,6 @@ extern char *neon_auth_token;
 extern char *neon_timeline;
 extern char *neon_tenant;
 
-extern char *wal_acceptors_list;
-extern int	wal_acceptor_reconnect_timeout;
-extern int	wal_acceptor_connection_timeout;
-
 extern void pg_init_libpagestore(void);
 extern void pg_init_walproposer(void);
 
@@ -34,10 +30,4 @@ extern void pg_init_extension_server(void);
 extern bool neon_redo_read_buffer_filter(XLogReaderState *record, uint8 block_id);
 extern bool	(*old_redo_read_buffer_filter) (XLogReaderState *record, uint8 block_id);
 
-extern uint64 BackpressureThrottlingTime(void);
-extern void replication_feedback_get_lsns(XLogRecPtr *writeLsn, XLogRecPtr *flushLsn, XLogRecPtr *applyLsn);
-
-extern void PGDLLEXPORT WalProposerSync(int argc, char *argv[]);
-extern void PGDLLEXPORT WalProposerMain(Datum main_arg);
-
 #endif							/* NEON_H */

pgxn/neon/neon_utils.c

@@ -1,116 +0,0 @@
-#include "postgres.h"
-
-#include "access/timeline.h"
-#include "access/xlogutils.h"
-#include "common/logging.h"
-#include "common/ip.h"
-#include "funcapi.h"
-#include "libpq/libpq.h"
-#include "libpq/pqformat.h"
-#include "miscadmin.h"
-#include "postmaster/interrupt.h"
-#include "replication/slot.h"
-#include "replication/walsender_private.h"
-
-#include "storage/ipc.h"
-#include "utils/builtins.h"
-#include "utils/ps_status.h"
-
-#include "libpq-fe.h"
-#include <netinet/tcp.h>
-#include <unistd.h>
-
-#if PG_VERSION_NUM >= 150000
-#include "access/xlogutils.h"
-#include "access/xlogrecovery.h"
-#endif
-#if PG_MAJORVERSION_NUM >= 16
-#include "utils/guc.h"
-#endif
-
-/*
- * Convert a character which represents a hexadecimal digit to an integer.
- *
- * Returns -1 if the character is not a hexadecimal digit.
- */
-int
-HexDecodeChar(char c)
-{
-	if (c >= '0' && c <= '9')
-		return c - '0';
-	if (c >= 'a' && c <= 'f')
-		return c - 'a' + 10;
-	if (c >= 'A' && c <= 'F')
-		return c - 'A' + 10;
-
-	return -1;
-}
-
-/*
- * Decode a hex string into a byte string, 2 hex chars per byte.
- *
- * Returns false if invalid characters are encountered; otherwise true.
- */
-bool
-HexDecodeString(uint8 *result, char *input, int nbytes)
-{
-	int			i;
-
-	for (i = 0; i < nbytes; ++i)
-	{
-		int			n1 = HexDecodeChar(input[i * 2]);
-		int			n2 = HexDecodeChar(input[i * 2 + 1]);
-
-		if (n1 < 0 || n2 < 0)
-			return false;
-		result[i] = n1 * 16 + n2;
-	}
-
-	return true;
-}
-
-/* --------------------------------
- *		pq_getmsgint32_le	- get a binary 4-byte int from a message buffer in native (LE) order
- * --------------------------------
- */
-uint32
-pq_getmsgint32_le(StringInfo msg)
-{
-	uint32		n32;
-
-	pq_copymsgbytes(msg, (char *) &n32, sizeof(n32));
-
-	return n32;
-}
-
-/* --------------------------------
- *		pq_getmsgint64	- get a binary 8-byte int from a message buffer in native (LE) order
- * --------------------------------
- */
-uint64
-pq_getmsgint64_le(StringInfo msg)
-{
-	uint64		n64;
-
-	pq_copymsgbytes(msg, (char *) &n64, sizeof(n64));
-
-	return n64;
-}
-
-/* append a binary [u]int32 to a StringInfo buffer in native (LE) order */
-void
-pq_sendint32_le(StringInfo buf, uint32 i)
-{
-	enlargeStringInfo(buf, sizeof(uint32));
-	memcpy(buf->data + buf->len, &i, sizeof(uint32));
-	buf->len += sizeof(uint32);
-}
-
-/* append a binary [u]int64 to a StringInfo buffer in native (LE) order */
-void
-pq_sendint64_le(StringInfo buf, uint64 i)
-{
-	enlargeStringInfo(buf, sizeof(uint64));
-	memcpy(buf->data + buf->len, &i, sizeof(uint64));
-	buf->len += sizeof(uint64);
-}

pgxn/neon/neon_utils.h

@@ -1,12 +0,0 @@
-#ifndef __NEON_UTILS_H__
-#define __NEON_UTILS_H__
-
-#include "postgres.h"
-
-bool		HexDecodeString(uint8 *result, char *input, int nbytes);
-uint32		pq_getmsgint32_le(StringInfo msg);
-uint64		pq_getmsgint64_le(StringInfo msg);
-void		pq_sendint32_le(StringInfo buf, uint32 i);
-void		pq_sendint64_le(StringInfo buf, uint64 i);
-
-#endif							/* __NEON_UTILS_H__ */

pgxn/neon/pagestore_smgr.c

@@ -721,7 +721,7 @@ prefetch_register_buffer(BufferTag tag, bool *force_latest, XLogRecPtr *force_ls
 
 	/* use an intermediate PrefetchRequest struct to ensure correct alignment */
 	req.buftag = tag;
-  Retry:
+	
 	entry = prfh_lookup(MyPState->prf_hash, (PrefetchRequest *) &req);
 
 	if (entry != NULL)
@@ -858,11 +858,7 @@ prefetch_register_buffer(BufferTag tag, bool *force_latest, XLogRecPtr *force_ls
 	if (flush_every_n_requests > 0 &&
 		MyPState->ring_unused - MyPState->ring_flush >= flush_every_n_requests)
 	{
-		if (!page_server->flush())
-		{
-			/* Prefetch set is reset in case of error, so we should try to register our request once again */
-			goto Retry;
-		}
+		page_server->flush();
 		MyPState->ring_flush = MyPState->ring_unused;
 	}