Fix unrelated test

.github/workflows/build_and_test.yml

@@ -623,6 +623,51 @@ jobs:
       - name: Cleanup ECR folder
         run: rm -rf ~/.ecr
 
+
+  neon-image-depot:
+    # For testing this will run side-by-side for a few merges.
+    # This action is not really optimized yet, but gets the job done
+    runs-on: [ self-hosted, gen3, large ]
+    needs: [ tag ]
+    container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/base:pinned
+    permissions:
+      contents: read
+      id-token: write
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+        with:
+          submodules: true
+          fetch-depth: 0
+
+      - name: Setup go
+        uses: actions/setup-go@v3
+        with:
+          go-version: '1.19'
+
+      - name: Set up Depot CLI
+        uses: depot/setup-action@v1
+
+      - name: Install Crane & ECR helper
+        run: go install github.com/awslabs/amazon-ecr-credential-helper/ecr-login/cli/docker-credential-ecr-login@69c85dc22db6511932bbf119e1a0cc5c90c69a7f # v0.6.0
+
+      - name: Configure ECR login
+        run: |
+          mkdir /github/home/.docker/
+          echo "{\"credsStore\":\"ecr-login\"}" > /github/home/.docker/config.json
+
+      - name: Build and push
+        uses: depot/build-push-action@v1
+        with:
+          # if no depot.json file is at the root of your repo, you must specify the project id
+          project: nrdv0s4kcs
+          push: true
+          tags: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/neon:depot-${{needs.tag.outputs.build-tag}}
+          build-args: |
+            GIT_VERSION=${{ github.sha }}
+            REPOSITORY=369495373322.dkr.ecr.eu-central-1.amazonaws.com
+
   compute-tools-image:
     runs-on: [ self-hosted, gen3, large ]
     needs: [ tag ]
@@ -914,20 +959,6 @@ jobs:
             exit 1
           fi
 
-      - name: Create tag "release-${{ needs.tag.outputs.build-tag }}"
-        if: github.ref_name == 'release'
-        uses: actions/github-script@v6
-        with:
-          # Retry script for 5XX server errors: https://github.com/actions/github-script#retries
-          retries: 5
-          script: |
-            github.rest.git.createRef({
-              owner: context.repo.owner,
-              repo: context.repo.repo,
-              ref: "refs/tags/release-${{ needs.tag.outputs.build-tag }}",
-              sha: context.sha,
-            })
-
   promote-compatibility-data:
     runs-on: [ self-hosted, gen3, small ]
     container:

.github/workflows/release.yml

@@ -3,7 +3,6 @@ name: Create Release Branch
 on:
   schedule:
     - cron: '0 10 * * 2'
-  workflow_dispatch:
 
 jobs:
   create_release_branch:

Cargo.lock

@@ -2770,7 +2770,7 @@ dependencies = [
 [[package]]
 name = "postgres"
 version = "0.19.4"
-source = "git+https://github.com/neondatabase/rust-postgres.git?rev=1aaedab101b23f7612042850d8f2036810fa7c7f#1aaedab101b23f7612042850d8f2036810fa7c7f"
+source = "git+https://github.com/neondatabase/rust-postgres.git?rev=f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c#f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c"
 dependencies = [
  "bytes",
  "fallible-iterator",
@@ -2783,7 +2783,7 @@ dependencies = [
 [[package]]
 name = "postgres-native-tls"
 version = "0.5.0"
-source = "git+https://github.com/neondatabase/rust-postgres.git?rev=1aaedab101b23f7612042850d8f2036810fa7c7f#1aaedab101b23f7612042850d8f2036810fa7c7f"
+source = "git+https://github.com/neondatabase/rust-postgres.git?rev=f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c#f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c"
 dependencies = [
  "native-tls",
  "tokio",
@@ -2794,7 +2794,7 @@ dependencies = [
 [[package]]
 name = "postgres-protocol"
 version = "0.6.4"
-source = "git+https://github.com/neondatabase/rust-postgres.git?rev=1aaedab101b23f7612042850d8f2036810fa7c7f#1aaedab101b23f7612042850d8f2036810fa7c7f"
+source = "git+https://github.com/neondatabase/rust-postgres.git?rev=f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c#f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c"
 dependencies = [
  "base64 0.20.0",
  "byteorder",
@@ -2812,7 +2812,7 @@ dependencies = [
 [[package]]
 name = "postgres-types"
 version = "0.2.4"
-source = "git+https://github.com/neondatabase/rust-postgres.git?rev=1aaedab101b23f7612042850d8f2036810fa7c7f#1aaedab101b23f7612042850d8f2036810fa7c7f"
+source = "git+https://github.com/neondatabase/rust-postgres.git?rev=f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c#f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c"
 dependencies = [
  "bytes",
  "fallible-iterator",
@@ -4272,7 +4272,7 @@ dependencies = [
 [[package]]
 name = "tokio-postgres"
 version = "0.7.7"
-source = "git+https://github.com/neondatabase/rust-postgres.git?rev=1aaedab101b23f7612042850d8f2036810fa7c7f#1aaedab101b23f7612042850d8f2036810fa7c7f"
+source = "git+https://github.com/neondatabase/rust-postgres.git?rev=f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c#f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c"
 dependencies = [
  "async-trait",
  "byteorder",

Cargo.toml

@@ -140,11 +140,11 @@ env_logger = "0.10"
 log = "0.4"
 
 ## Libraries from neondatabase/ git forks, ideally with changes to be upstreamed
-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="1aaedab101b23f7612042850d8f2036810fa7c7f" }
-postgres-native-tls = { git = "https://github.com/neondatabase/rust-postgres.git", rev="1aaedab101b23f7612042850d8f2036810fa7c7f" }
-postgres-protocol = { git = "https://github.com/neondatabase/rust-postgres.git", rev="1aaedab101b23f7612042850d8f2036810fa7c7f" }
-postgres-types = { git = "https://github.com/neondatabase/rust-postgres.git", rev="1aaedab101b23f7612042850d8f2036810fa7c7f" }
-tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="1aaedab101b23f7612042850d8f2036810fa7c7f" }
+postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c" }
+postgres-native-tls = { git = "https://github.com/neondatabase/rust-postgres.git", rev="f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c" }
+postgres-protocol = { git = "https://github.com/neondatabase/rust-postgres.git", rev="f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c" }
+postgres-types = { git = "https://github.com/neondatabase/rust-postgres.git", rev="f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c" }
+tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c" }
 tokio-tar = { git = "https://github.com/neondatabase/tokio-tar.git", rev="404df61437de0feef49ba2ccdbdd94eb8ad6e142" }
 
 ## Other git libraries
@@ -180,7 +180,7 @@ tonic-build = "0.9"
 
 # This is only needed for proxy's tests.
 # TODO: we should probably fork `tokio-postgres-rustls` instead.
-tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="1aaedab101b23f7612042850d8f2036810fa7c7f" }
+tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="f6ec31df3bcce89cb34f300f17c8a8c031c5ee8c" }
 
 # Changes the MAX_THREADS limit from 4096 to 32768.
 # This is a temporary workaround for using tracing from many threads in safekeepers code,

Dockerfile.compute-node

@@ -67,7 +67,7 @@ RUN apt update && \
 RUN wget https://gitlab.com/Oslandia/SFCGAL/-/archive/v1.3.10/SFCGAL-v1.3.10.tar.gz -O SFCGAL.tar.gz && \
     echo "4e39b3b2adada6254a7bdba6d297bb28e1a9835a9f879b74f37e2dab70203232 SFCGAL.tar.gz" | sha256sum --check && \
     mkdir sfcgal-src && cd sfcgal-src && tar xvzf ../SFCGAL.tar.gz --strip-components=1 -C . && \
-    cmake -DCMAKE_BUILD_TYPE=Release . && make -j $(getconf _NPROCESSORS_ONLN) && \
+    cmake . && make -j $(getconf _NPROCESSORS_ONLN) && \
     DESTDIR=/sfcgal make install -j $(getconf _NPROCESSORS_ONLN) && \
     make clean && cp -R /sfcgal/* /
 
@@ -95,7 +95,7 @@ RUN wget https://github.com/pgRouting/pgrouting/archive/v3.4.2.tar.gz -O pgrouti
     mkdir pgrouting-src && cd pgrouting-src && tar xvzf ../pgrouting.tar.gz --strip-components=1 -C . && \
     mkdir build && \
     cd build && \
-    cmake -DCMAKE_BUILD_TYPE=Release .. && \
+    cmake .. && \
     make -j $(getconf _NPROCESSORS_ONLN) && \
     make -j $(getconf _NPROCESSORS_ONLN) install && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgrouting.control
@@ -355,7 +355,7 @@ RUN apt-get update && \
     wget https://github.com/timescale/timescaledb/archive/refs/tags/2.10.1.tar.gz -O timescaledb.tar.gz && \
     echo "6fca72a6ed0f6d32d2b3523951ede73dc5f9b0077b38450a029a5f411fdb8c73 timescaledb.tar.gz" | sha256sum --check && \
     mkdir timescaledb-src && cd timescaledb-src && tar xvzf ../timescaledb.tar.gz --strip-components=1 -C . && \
-    ./bootstrap -DSEND_TELEMETRY_DEFAULT:BOOL=OFF -DUSE_TELEMETRY:BOOL=OFF -DAPACHE_ONLY:BOOL=ON -DCMAKE_BUILD_TYPE=Release && \
+    ./bootstrap -DSEND_TELEMETRY_DEFAULT:BOOL=OFF -DUSE_TELEMETRY:BOOL=OFF -DAPACHE_ONLY:BOOL=ON && \
     cd build && \
     make -j $(getconf _NPROCESSORS_ONLN) && \
     make install -j $(getconf _NPROCESSORS_ONLN) && \
@@ -410,7 +410,7 @@ RUN apt-get update && \
     mkdir kq_imcx-src && cd kq_imcx-src && tar xvzf ../kq_imcx.tar.gz --strip-components=1 -C . && \
     mkdir build && \
     cd build && \
-    cmake -DCMAKE_BUILD_TYPE=Release .. && \
+    cmake .. && \
     make -j $(getconf _NPROCESSORS_ONLN) && \
     make -j $(getconf _NPROCESSORS_ONLN) install && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/kq_imcx.control
@@ -432,54 +432,6 @@ RUN wget https://github.com/citusdata/pg_cron/archive/refs/tags/v1.5.2.tar.gz -O
     make -j $(getconf _NPROCESSORS_ONLN) install && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/pg_cron.control
 
-#########################################################################################
-#
-# Layer "rdkit-pg-build"
-# compile rdkit extension
-#
-#########################################################################################
-FROM build-deps AS rdkit-pg-build
-COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
-
-RUN apt-get update && \
-    apt-get install -y \
-        cmake \
-        libboost-iostreams1.74-dev \
-        libboost-regex1.74-dev \
-        libboost-serialization1.74-dev \
-        libboost-system1.74-dev \
-        libeigen3-dev \
-        libfreetype6-dev
-
-ENV PATH "/usr/local/pgsql/bin/:/usr/local/pgsql/:$PATH"
-RUN wget https://github.com/rdkit/rdkit/archive/refs/tags/Release_2023_03_1.tar.gz -O rdkit.tar.gz && \
-    echo "db346afbd0ba52c843926a2a62f8a38c7b774ffab37eaf382d789a824f21996c rdkit.tar.gz" | sha256sum --check && \
-    mkdir rdkit-src && cd rdkit-src && tar xvzf ../rdkit.tar.gz --strip-components=1 -C . && \
-    cmake \
-        -D RDK_BUILD_CAIRO_SUPPORT=OFF \
-        -D RDK_BUILD_INCHI_SUPPORT=ON \
-        -D RDK_BUILD_AVALON_SUPPORT=ON \
-        -D RDK_BUILD_PYTHON_WRAPPERS=OFF \
-        -D RDK_BUILD_DESCRIPTORS3D=OFF \
-        -D RDK_BUILD_FREESASA_SUPPORT=OFF \
-        -D RDK_BUILD_COORDGEN_SUPPORT=ON \
-        -D RDK_BUILD_MOLINTERCHANGE_SUPPORT=OFF \
-        -D RDK_BUILD_YAEHMOP_SUPPORT=OFF \
-        -D RDK_BUILD_STRUCTCHECKER_SUPPORT=OFF \
-        -D RDK_USE_URF=OFF \
-        -D RDK_BUILD_PGSQL=ON \
-        -D RDK_PGSQL_STATIC=ON \
-        -D PostgreSQL_CONFIG=pg_config \
-        -D PostgreSQL_INCLUDE_DIR=`pg_config --includedir` \
-        -D PostgreSQL_TYPE_INCLUDE_DIR=`pg_config --includedir-server` \
-        -D PostgreSQL_LIBRARY_DIR=`pg_config --libdir` \
-        -D RDK_INSTALL_INTREE=OFF \
-        -D CMAKE_BUILD_TYPE=Release \
-        . && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/rdkit.control
-
 #########################################################################################
 #
 # Layer "rust extensions"
@@ -612,7 +564,6 @@ COPY --from=pg-hint-plan-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=kq-imcx-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-cron-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-pgx-ulid-build /usr/local/pgsql/ /usr/local/pgsql/
-COPY --from=rdkit-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY pgxn/ pgxn/
 
 RUN make -j $(getconf _NPROCESSORS_ONLN) \
@@ -686,19 +637,14 @@ COPY --from=compute-tools --chown=postgres /home/nonroot/target/release-line-deb
 # libgeos, libgdal, libsfcgal1, libproj and libprotobuf-c1 for PostGIS
 # libxml2, libxslt1.1 for xml2
 # libzstd1 for zstd
-# libboost*, libfreetype6, and zlib1g for rdkit
 RUN apt update &&  \
     apt install --no-install-recommends -y \
         gdb \
+        locales \
         libicu67 \
         liblz4-1 \
         libreadline8 \
-        libboost-iostreams1.74.0 \
-        libboost-regex1.74.0 \
-        libboost-serialization1.74.0 \
-        libboost-system1.74.0 \
         libossp-uuid16 \
-        libfreetype6 \
         libgeos-c1v5 \
         libgdal28 \
         libproj19 \
@@ -708,9 +654,7 @@ RUN apt update &&  \
         libxslt1.1 \
         libzstd1 \
         libcurl4-openssl-dev \
-        locales \
-        procps \
-        zlib1g && \
+        procps && \
     rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/* && \
     localedef -i en_US -c -f UTF-8 -A /usr/share/locale/locale.alias en_US.UTF-8
 

compute_tools/src/compute.rs

@@ -370,6 +370,11 @@ impl ComputeNode {
         // 'Close' connection
         drop(client);
 
+        info!(
+            "finished configuration of compute for project {}",
+            spec.cluster.cluster_id.as_deref().unwrap_or("None")
+        );
+
         Ok(())
     }
 
@@ -422,23 +427,57 @@ impl ComputeNode {
     #[instrument(skip(self))]
     pub fn start_compute(&self) -> Result<std::process::Child> {
         let compute_state = self.state.lock().unwrap().clone();
-        let pspec = compute_state.pspec.as_ref().expect("spec must be set");
+        let spec = compute_state.pspec.as_ref().expect("spec must be set");
         info!(
             "starting compute for project {}, operation {}, tenant {}, timeline {}",
-            pspec.spec.cluster.cluster_id.as_deref().unwrap_or("None"),
-            pspec.spec.operation_uuid.as_deref().unwrap_or("None"),
-            pspec.tenant_id,
-            pspec.timeline_id,
+            spec.spec.cluster.cluster_id.as_deref().unwrap_or("None"),
+            spec.spec.operation_uuid.as_deref().unwrap_or("None"),
+            spec.tenant_id,
+            spec.timeline_id,
         );
 
         self.prepare_pgdata(&compute_state)?;
 
         let start_time = Utc::now();
 
-        let pg = self.start_postgres(pspec.storage_auth_token.clone())?;
+        let pg = self.start_postgres(spec.storage_auth_token.clone())?;
 
-        if pspec.spec.mode == ComputeMode::Primary && !pspec.spec.skip_pg_catalog_updates {
-            self.apply_config(&compute_state)?;
+        // Maybe apply the spec
+        if spec.spec.mode == ComputeMode::Primary {
+            let spec = &compute_state.pspec.as_ref().expect("spec must be set").spec;
+
+            // Get spec_id or make it up by hashing
+            //
+            // TODO Make spec_id required so there would be no need to hash.
+            let spec_id = spec.operation_uuid.clone().unwrap_or_else(|| {
+                use std::collections::hash_map::DefaultHasher;
+                use std::hash::{Hash, Hasher};
+
+                // HACK Exclude postgresql.conf because it doesn't need
+                //      to be applied like the other fields in the spec
+                let mut spec_no_conf = spec.clone();
+                spec_no_conf.cluster.postgresql_conf = None;
+
+                let json = serde_json::to_vec(&spec_no_conf).unwrap();
+                let mut hasher = DefaultHasher::new();
+                json.hash(&mut hasher);
+                let hash = hasher.finish();
+                format!("{:x}", hash)
+            });
+
+            // Get current spec_id
+            let path = Path::new(&self.pgdata).join("neon_compute_spec_id.txt");
+            let current_spec_id = std::fs::read_to_string(path).ok();
+
+            // Respec if needed
+            if current_spec_id == Some(spec_id.clone()) {
+                info!("no need to respec");
+            } else {
+                info!("respeccing {:?} {:?}", current_spec_id, &spec_id);
+
+                self.apply_config(&compute_state)?;
+                self.cache_spec_id(&compute_state, spec_id)?;
+            }
         }
 
         let startup_end_time = Utc::now();
@@ -457,14 +496,31 @@ impl ComputeNode {
         }
         self.set_status(ComputeStatus::Running);
 
-        info!(
-            "finished configuration of compute for project {}",
-            pspec.spec.cluster.cluster_id.as_deref().unwrap_or("None")
-        );
-
         Ok(pg)
     }
 
+    fn cache_spec_id(&self, compute_state: &ComputeState, spec_id: String) -> anyhow::Result<()> {
+        let spec = &compute_state.pspec.as_ref().expect("spec must be set");
+        let cmd = format!(
+            "set_compute_spec_id {} {} {}",
+            spec.tenant_id, spec.timeline_id, spec_id,
+        );
+        let mut config = postgres::Config::from_str(&spec.pageserver_connstr)?;
+
+        // Use the storage auth token from the config file, if given.
+        // Note: this overrides any password set in the connection string.
+        if let Some(storage_auth_token) = &spec.storage_auth_token {
+            info!("Got storage auth token from spec file");
+            config.password(storage_auth_token);
+        } else {
+            info!("Storage auth token not set");
+        }
+        let mut client = config.connect(NoTls)?;
+        client.simple_query(&cmd)?;
+
+        Ok(())
+    }
+
     // Look for core dumps and collect backtraces.
     //
     // EKS worker nodes have following core dump settings:

control_plane/src/endpoint.rs

@@ -450,7 +450,6 @@ impl Endpoint {
 
         // Create spec file
         let spec = ComputeSpec {
-            skip_pg_catalog_updates: false,
             format_version: 1.0,
             operation_uuid: None,
             cluster: Cluster {

libs/compute_api/src/spec.rs

@@ -27,12 +27,6 @@ pub struct ComputeSpec {
     pub cluster: Cluster,
     pub delta_operations: Option<Vec<DeltaOp>>,
 
-    /// An optinal hint that can be passed to speed up startup time if we know
-    /// that no pg catalog mutations (like role creation, database creation,
-    /// extension creation) need to be done on the actual database to start.
-    #[serde(default)] // Default false
-    pub skip_pg_catalog_updates: bool,
-
     // Information needed to connect to the storage layer.
     //
     // `tenant_id`, `timeline_id` and `pageserver_connstring` are always needed.
@@ -148,14 +142,4 @@ mod tests {
         let file = File::open("tests/cluster_spec.json").unwrap();
         let _spec: ComputeSpec = serde_json::from_reader(file).unwrap();
     }
-
-    #[test]
-    fn parse_unknown_fields() {
-        // Forward compatibility test
-        let file = File::open("tests/cluster_spec.json").unwrap();
-        let mut json: serde_json::Value = serde_json::from_reader(file).unwrap();
-        let ob = json.as_object_mut().unwrap();
-        ob.insert("unknown_field_123123123".into(), "hello".into());
-        let _spec: ComputeSpec = serde_json::from_value(json).unwrap();
-    }
 }

libs/remote_storage/src/lib.rs

@@ -111,8 +111,6 @@ pub trait RemoteStorage: Send + Sync + 'static {
     ) -> Result<Download, DownloadError>;
 
     async fn delete(&self, path: &RemotePath) -> anyhow::Result<()>;
-
-    async fn delete_objects<'a>(&self, paths: &'a [RemotePath]) -> anyhow::Result<()>;
 }
 
 pub struct Download {
@@ -225,14 +223,6 @@ impl GenericRemoteStorage {
             Self::Unreliable(s) => s.delete(path).await,
         }
     }
-
-    pub async fn delete_objects<'a>(&self, paths: &'a [RemotePath]) -> anyhow::Result<()> {
-        match self {
-            Self::LocalFs(s) => s.delete_objects(paths).await,
-            Self::AwsS3(s) => s.delete_objects(paths).await,
-            Self::Unreliable(s) => s.delete_objects(paths).await,
-        }
-    }
 }
 
 impl GenericRemoteStorage {

libs/remote_storage/src/local_fs.rs

@@ -320,13 +320,6 @@ impl RemoteStorage for LocalFs {
             .await
             .map_err(|e| anyhow::anyhow!(e))?)
     }
-
-    async fn delete_objects<'a>(&self, paths: &'a [RemotePath]) -> anyhow::Result<()> {
-        for path in paths {
-            self.delete(path).await?
-        }
-        Ok(())
-    }
 }
 
 fn storage_metadata_path(original_path: &Path) -> PathBuf {

libs/remote_storage/src/s3_bucket.rs

@@ -17,7 +17,6 @@ use aws_sdk_s3::{
     error::SdkError,
     operation::get_object::GetObjectError,
     primitives::ByteStream,
-    types::{Delete, ObjectIdentifier},
     Client,
 };
 use aws_smithy_http::body::SdkBody;
@@ -34,8 +33,6 @@ use crate::{
     Download, DownloadError, RemotePath, RemoteStorage, S3Config, REMOTE_STORAGE_PREFIX_SEPARATOR,
 };
 
-const MAX_DELETE_OBJECTS_REQUEST_SIZE: usize = 1000;
-
 pub(super) mod metrics {
     use metrics::{register_int_counter_vec, IntCounterVec};
     use once_cell::sync::Lazy;
@@ -84,24 +81,12 @@ pub(super) mod metrics {
             .inc();
     }
 
-    pub fn inc_delete_objects(count: u64) {
-        S3_REQUESTS_COUNT
-            .with_label_values(&["delete_object"])
-            .inc_by(count);
-    }
-
     pub fn inc_delete_object_fail() {
         S3_REQUESTS_FAIL_COUNT
             .with_label_values(&["delete_object"])
             .inc();
     }
 
-    pub fn inc_delete_objects_fail(count: u64) {
-        S3_REQUESTS_FAIL_COUNT
-            .with_label_values(&["delete_object"])
-            .inc_by(count);
-    }
-
     pub fn inc_list_objects() {
         S3_REQUESTS_COUNT.with_label_values(&["list_objects"]).inc();
     }
@@ -411,50 +396,6 @@ impl RemoteStorage for S3Bucket {
         })
         .await
     }
-    async fn delete_objects<'a>(&self, paths: &'a [RemotePath]) -> anyhow::Result<()> {
-        let _guard = self
-            .concurrency_limiter
-            .acquire()
-            .await
-            .context("Concurrency limiter semaphore got closed during S3 delete")?;
-
-        let mut delete_objects = Vec::with_capacity(paths.len());
-        for path in paths {
-            let obj_id = ObjectIdentifier::builder()
-                .set_key(Some(self.relative_path_to_s3_object(path)))
-                .build();
-            delete_objects.push(obj_id);
-        }
-
-        for chunk in delete_objects.chunks(MAX_DELETE_OBJECTS_REQUEST_SIZE) {
-            metrics::inc_delete_objects(chunk.len() as u64);
-
-            let resp = self
-                .client
-                .delete_objects()
-                .bucket(self.bucket_name.clone())
-                .delete(Delete::builder().set_objects(Some(chunk.to_vec())).build())
-                .send()
-                .await;
-
-            match resp {
-                Ok(resp) => {
-                    if let Some(errors) = resp.errors {
-                        metrics::inc_delete_objects_fail(errors.len() as u64);
-                        return Err(anyhow::format_err!(
-                            "Failed to delete {} objects",
-                            errors.len()
-                        ));
-                    }
-                }
-                Err(e) => {
-                    metrics::inc_delete_objects_fail(chunk.len() as u64);
-                    return Err(e.into());
-                }
-            }
-        }
-        Ok(())
-    }
 
     async fn delete(&self, path: &RemotePath) -> anyhow::Result<()> {
         let _guard = self

libs/remote_storage/src/simulate_failures.rs

@@ -24,7 +24,6 @@ enum RemoteOp {
     Upload(RemotePath),
     Download(RemotePath),
     Delete(RemotePath),
-    DeleteObjects(Vec<RemotePath>),
 }
 
 impl UnreliableWrapper {
@@ -120,21 +119,4 @@ impl RemoteStorage for UnreliableWrapper {
         self.attempt(RemoteOp::Delete(path.clone()))?;
         self.inner.delete(path).await
     }
-
-    async fn delete_objects<'a>(&self, paths: &'a [RemotePath]) -> anyhow::Result<()> {
-        self.attempt(RemoteOp::DeleteObjects(paths.to_vec()))?;
-        let mut error_counter = 0;
-        for path in paths {
-            if (self.delete(path).await).is_err() {
-                error_counter += 1;
-            }
-        }
-        if error_counter > 0 {
-            return Err(anyhow::anyhow!(
-                "failed to delete {} objects",
-                error_counter
-            ));
-        }
-        Ok(())
-    }
 }

libs/remote_storage/tests/test_real_s3.rs

@@ -107,37 +107,6 @@ async fn s3_delete_non_exising_works(ctx: &mut MaybeEnabledS3) -> anyhow::Result
     Ok(())
 }
 
-#[test_context(MaybeEnabledS3)]
-#[tokio::test]
-async fn s3_delete_objects_works(ctx: &mut MaybeEnabledS3) -> anyhow::Result<()> {
-    let ctx = match ctx {
-        MaybeEnabledS3::Enabled(ctx) => ctx,
-        MaybeEnabledS3::Disabled => return Ok(()),
-    };
-
-    let path1 = RemotePath::new(&PathBuf::from(format!("{}/path1", ctx.base_prefix,)))
-        .with_context(|| "RemotePath conversion")?;
-
-    let path2 = RemotePath::new(&PathBuf::from(format!("{}/path2", ctx.base_prefix,)))
-        .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();
-    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.delete_objects(&[path1, path2]).await?;
-
-    Ok(())
-}
-
 fn ensure_logging_ready() {
     LOGGING_DONE.get_or_init(|| {
         utils::logging::init(

libs/utils/src/http/endpoint.rs

@@ -1,18 +1,19 @@
 use crate::auth::{Claims, JwtAuth};
 use crate::http::error::{api_error_handler, route_error_handler, ApiError};
-use anyhow::Context;
+use anyhow::{anyhow, Context};
 use hyper::header::{HeaderName, AUTHORIZATION};
 use hyper::http::HeaderValue;
 use hyper::Method;
-use hyper::{header::CONTENT_TYPE, Body, Request, Response};
+use hyper::{header::CONTENT_TYPE, Body, Request, Response, Server};
 use metrics::{register_int_counter, Encoder, IntCounter, TextEncoder};
 use once_cell::sync::Lazy;
 use routerify::ext::RequestExt;
-use routerify::{Middleware, RequestInfo, Router, RouterBuilder};
+use routerify::{Middleware, RequestInfo, Router, RouterBuilder, RouterService};
 use tokio::task::JoinError;
 use tracing::{self, debug, info, info_span, warn, Instrument};
 
 use std::future::Future;
+use std::net::TcpListener;
 use std::str::FromStr;
 
 static SERVE_METRICS_COUNT: Lazy<IntCounter> = Lazy::new(|| {
@@ -347,6 +348,40 @@ pub fn check_permission_with(
     }
 }
 
+///
+/// Start listening for HTTP requests on given socket.
+///
+/// 'shutdown_future' can be used to stop. If the Future becomes
+/// ready, we stop listening for new requests, and the function returns.
+///
+pub fn serve_thread_main<S>(
+    router_builder: RouterBuilder<hyper::Body, ApiError>,
+    listener: TcpListener,
+    shutdown_future: S,
+) -> anyhow::Result<()>
+where
+    S: Future<Output = ()> + Send + Sync,
+{
+    info!("Starting an HTTP endpoint at {}", listener.local_addr()?);
+
+    // Create a Service from the router above to handle incoming requests.
+    let service = RouterService::new(router_builder.build().map_err(|err| anyhow!(err))?).unwrap();
+
+    // Enter a single-threaded tokio runtime bound to the current thread
+    let runtime = tokio::runtime::Builder::new_current_thread()
+        .enable_all()
+        .build()?;
+
+    let _guard = runtime.enter();
+
+    let server = Server::from_tcp(listener)?
+        .serve(service)
+        .with_graceful_shutdown(shutdown_future);
+
+    runtime.block_on(server)?;
+
+    Ok(())
+}
 #[cfg(test)]
 mod tests {
     use super::*;

pageserver/src/basebackup.rs

@@ -417,6 +417,16 @@ where
     // Also send zenith.signal file with extra bootstrap data.
     //
     async fn add_pgcontrol_file(&mut self) -> anyhow::Result<()> {
+        // Add neon_compute_spec_id.txt
+        if let Some(spec_id) = &self.timeline.compute_spec_id.lock().await.clone() {
+            self.ar
+                .append(
+                    &new_tar_header("neon_compute_spec_id.txt", spec_id.len() as u64)?,
+                    spec_id.as_bytes(),
+                )
+                .await?;
+        }
+
         // add zenith.signal file
         let mut zenith_signal = String::new();
         if self.prev_record_lsn == Lsn(0) {

pageserver/src/disk_usage_eviction_task.rs

@@ -516,7 +516,7 @@ async fn collect_eviction_candidates(
             if !tl.is_active() {
                 continue;
             }
-            let info = tl.get_local_layers_for_disk_usage_eviction().await;
+            let info = tl.get_local_layers_for_disk_usage_eviction();
             debug!(tenant_id=%tl.tenant_id, timeline_id=%tl.timeline_id, "timeline resident layers count: {}", info.resident_layers.len());
             tenant_candidates.extend(
                 info.resident_layers

pageserver/src/http/routes.rs

@@ -215,7 +215,7 @@ async fn build_timeline_info(
 ) -> anyhow::Result<TimelineInfo> {
     crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
 
-    let mut info = build_timeline_info_common(timeline, ctx).await?;
+    let mut info = build_timeline_info_common(timeline, ctx)?;
     if include_non_incremental_logical_size {
         // XXX we should be using spawn_ondemand_logical_size_calculation here.
         // Otherwise, if someone deletes the timeline / detaches the tenant while
@@ -233,7 +233,7 @@ async fn build_timeline_info(
     Ok(info)
 }
 
-async fn build_timeline_info_common(
+fn build_timeline_info_common(
     timeline: &Arc<Timeline>,
     ctx: &RequestContext,
 ) -> anyhow::Result<TimelineInfo> {
@@ -264,7 +264,7 @@ async fn build_timeline_info_common(
             None
         }
     };
-    let current_physical_size = Some(timeline.layer_size_sum().await);
+    let current_physical_size = Some(timeline.layer_size_sum());
     let state = timeline.current_state();
     let remote_consistent_lsn = timeline.get_remote_consistent_lsn().unwrap_or(Lsn(0));
 
@@ -330,7 +330,6 @@ async fn timeline_create_handler(
             Ok(Some(new_timeline)) => {
                 // Created. Construct a TimelineInfo for it.
                 let timeline_info = build_timeline_info_common(&new_timeline, &ctx)
-                    .await
                     .map_err(ApiError::InternalServerError)?;
                 json_response(StatusCode::CREATED, timeline_info)
             }
@@ -592,7 +591,7 @@ async fn tenant_status(
         // Calculate total physical size of all timelines
         let mut current_physical_size = 0;
         for timeline in tenant.list_timelines().iter() {
-            current_physical_size += timeline.layer_size_sum().await;
+            current_physical_size += timeline.layer_size_sum();
         }
 
         let state = tenant.current_state();
@@ -702,7 +701,7 @@ async fn layer_map_info_handler(
     check_permission(&request, Some(tenant_id))?;
 
     let timeline = active_timeline_of_active_tenant(tenant_id, timeline_id).await?;
-    let layer_map_info = timeline.layer_map_info(reset).await;
+    let layer_map_info = timeline.layer_map_info(reset);
 
     json_response(StatusCode::OK, layer_map_info)
 }

pageserver/src/import_datadir.rs

@@ -75,12 +75,12 @@ pub async fn import_timeline_from_postgres_datadir(
             {
                 pg_control = Some(control_file);
             }
-            modification.flush().await?;
+            modification.flush()?;
         }
     }
 
     // We're done importing all the data files.
-    modification.commit().await?;
+    modification.commit()?;
 
     // We expect the Postgres server to be shut down cleanly.
     let pg_control = pg_control.context("pg_control file not found")?;
@@ -148,17 +148,17 @@ async fn import_rel(
     // because there is no guarantee about the order in which we are processing segments.
     // ignore "relation already exists" error
     //
-    // FIXME: Keep track of which relations we've already created?
+    // FIXME: use proper error type for this, instead of parsing the error message.
+    // Or better yet, keep track of which relations we've already created
     // https://github.com/neondatabase/neon/issues/3309
     if let Err(e) = modification
         .put_rel_creation(rel, nblocks as u32, ctx)
         .await
     {
-        match e {
-            RelationError::AlreadyExists => {
-                debug!("Relation {} already exist. We must be extending it.", rel)
-            }
-            _ => return Err(e.into()),
+        if e.to_string().contains("already exists") {
+            debug!("relation {} already exists. we must be extending it", rel);
+        } else {
+            return Err(e);
         }
     }
 
@@ -359,7 +359,7 @@ pub async fn import_basebackup_from_tar(
                     // We found the pg_control file.
                     pg_control = Some(res);
                 }
-                modification.flush().await?;
+                modification.flush()?;
             }
             tokio_tar::EntryType::Directory => {
                 debug!("directory {:?}", file_path);
@@ -377,7 +377,7 @@ pub async fn import_basebackup_from_tar(
     // sanity check: ensure that pg_control is loaded
     let _pg_control = pg_control.context("pg_control file not found")?;
 
-    modification.commit().await?;
+    modification.commit()?;
     Ok(())
 }
 
@@ -594,7 +594,7 @@ async fn import_file(
         // zenith.signal is not necessarily the last file, that we handle
         // but it is ok to call `finish_write()`, because final `modification.commit()`
         // will update lsn once more to the final one.
-        let writer = modification.tline.writer().await;
+        let writer = modification.tline.writer();
         writer.finish_write(prev_lsn);
 
         debug!("imported zenith signal {}", prev_lsn);

pageserver/src/metrics.rs

@@ -1,3 +1,4 @@
+use metrics::core::{AtomicU64, GenericCounter};
 use metrics::{
     register_counter_vec, register_histogram, register_histogram_vec, register_int_counter,
     register_int_counter_vec, register_int_gauge, register_int_gauge_vec, register_uint_gauge_vec,
@@ -94,19 +95,21 @@ static READ_NUM_FS_LAYERS: Lazy<HistogramVec> = Lazy::new(|| {
 });
 
 // Metrics collected on operations on the storage repository.
-pub static RECONSTRUCT_TIME: Lazy<Histogram> = Lazy::new(|| {
-    register_histogram!(
+static RECONSTRUCT_TIME: Lazy<HistogramVec> = Lazy::new(|| {
+    register_histogram_vec!(
         "pageserver_getpage_reconstruct_seconds",
-        "Time spent in reconstruct_value (reconstruct a page from deltas)",
+        "Time spent in reconstruct_value",
+        &["tenant_id", "timeline_id"],
         CRITICAL_OP_BUCKETS.into(),
     )
     .expect("failed to define a metric")
 });
 
-pub static MATERIALIZED_PAGE_CACHE_HIT_DIRECT: Lazy<IntCounter> = Lazy::new(|| {
-    register_int_counter!(
+static MATERIALIZED_PAGE_CACHE_HIT_DIRECT: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
         "pageserver_materialized_cache_hits_direct_total",
         "Number of cache hits from materialized page cache without redo",
+        &["tenant_id", "timeline_id"]
     )
     .expect("failed to define a metric")
 });
@@ -121,10 +124,11 @@ static GET_RECONSTRUCT_DATA_TIME: Lazy<HistogramVec> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
-pub static MATERIALIZED_PAGE_CACHE_HIT: Lazy<IntCounter> = Lazy::new(|| {
-    register_int_counter!(
+static MATERIALIZED_PAGE_CACHE_HIT: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
         "pageserver_materialized_cache_hits_total",
         "Number of cache hits from materialized page cache",
+        &["tenant_id", "timeline_id"]
     )
     .expect("failed to define a metric")
 });
@@ -748,7 +752,10 @@ impl StorageTimeMetrics {
 pub struct TimelineMetrics {
     tenant_id: String,
     timeline_id: String,
+    pub reconstruct_time_histo: Histogram,
     pub get_reconstruct_data_time_histo: Histogram,
+    pub materialized_page_cache_hit_counter: GenericCounter<AtomicU64>,
+    pub materialized_page_cache_hit_upon_request_counter: GenericCounter<AtomicU64>,
     pub flush_time_histo: StorageTimeMetrics,
     pub compact_time_histo: StorageTimeMetrics,
     pub create_images_time_histo: StorageTimeMetrics,
@@ -776,9 +783,15 @@ impl TimelineMetrics {
     ) -> Self {
         let tenant_id = tenant_id.to_string();
         let timeline_id = timeline_id.to_string();
+        let reconstruct_time_histo = RECONSTRUCT_TIME
+            .get_metric_with_label_values(&[&tenant_id, &timeline_id])
+            .unwrap();
         let get_reconstruct_data_time_histo = GET_RECONSTRUCT_DATA_TIME
             .get_metric_with_label_values(&[&tenant_id, &timeline_id])
             .unwrap();
+        let materialized_page_cache_hit_counter = MATERIALIZED_PAGE_CACHE_HIT
+            .get_metric_with_label_values(&[&tenant_id, &timeline_id])
+            .unwrap();
         let flush_time_histo =
             StorageTimeMetrics::new(StorageTimeOperation::LayerFlush, &tenant_id, &timeline_id);
         let compact_time_histo =
@@ -820,18 +833,19 @@ impl TimelineMetrics {
         let read_num_fs_layers = READ_NUM_FS_LAYERS
             .get_metric_with_label_values(&[&tenant_id, &timeline_id])
             .unwrap();
+        let materialized_page_cache_hit_upon_request_counter = MATERIALIZED_PAGE_CACHE_HIT_DIRECT
+            .get_metric_with_label_values(&[&tenant_id, &timeline_id])
+            .unwrap();
         let evictions_with_low_residence_duration =
             evictions_with_low_residence_duration_builder.build(&tenant_id, &timeline_id);
 
-        // TODO(chi): remove this once we remove Lazy for all metrics. Otherwise this will not appear in the exporter
-        // and integration test will error.
-        MATERIALIZED_PAGE_CACHE_HIT_DIRECT.get();
-        MATERIALIZED_PAGE_CACHE_HIT.get();
-
         TimelineMetrics {
             tenant_id,
             timeline_id,
+            reconstruct_time_histo,
             get_reconstruct_data_time_histo,
+            materialized_page_cache_hit_counter,
+            materialized_page_cache_hit_upon_request_counter,
             flush_time_histo,
             compact_time_histo,
             create_images_time_histo,
@@ -858,7 +872,10 @@ impl Drop for TimelineMetrics {
     fn drop(&mut self) {
         let tenant_id = &self.tenant_id;
         let timeline_id = &self.timeline_id;
+        let _ = RECONSTRUCT_TIME.remove_label_values(&[tenant_id, timeline_id]);
         let _ = GET_RECONSTRUCT_DATA_TIME.remove_label_values(&[tenant_id, timeline_id]);
+        let _ = MATERIALIZED_PAGE_CACHE_HIT.remove_label_values(&[tenant_id, timeline_id]);
+        let _ = MATERIALIZED_PAGE_CACHE_HIT_DIRECT.remove_label_values(&[tenant_id, timeline_id]);
         let _ = LAST_RECORD_LSN.remove_label_values(&[tenant_id, timeline_id]);
         let _ = WAIT_LSN_TIME.remove_label_values(&[tenant_id, timeline_id]);
         let _ = RESIDENT_PHYSICAL_SIZE.remove_label_values(&[tenant_id, timeline_id]);

pageserver/src/page_service.rs

@@ -915,6 +915,27 @@ where
             self.handle_basebackup_request(pgb, tenant_id, timeline_id, lsn, None, false, ctx)
                 .await?;
             pgb.write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
+        } else if query_string.starts_with("set_compute_spec_id ") {
+            let (_, params_raw) = query_string.split_at("set_compute_spec_id ".len());
+            let params = params_raw.split_whitespace().collect::<Vec<_>>();
+
+            if params.len() != 3 {
+                return Err(QueryError::Other(anyhow::anyhow!(
+                    "invalid param number for set_compute_spec_id command"
+                )));
+            }
+
+            let tenant_id = TenantId::from_str(params[0])
+                .with_context(|| format!("Failed to parse tenant id from {}", params[0]))?;
+            let timeline_id = TimelineId::from_str(params[1])
+                .with_context(|| format!("Failed to parse timeline id from {}", params[1]))?;
+            let spec_id = params[2].to_string();
+
+            self.check_permission(Some(tenant_id))?;
+            let timeline = get_active_tenant_timeline(tenant_id, timeline_id, &ctx).await?;
+            *timeline.compute_spec_id.lock().await = Some(spec_id);
+
+            pgb.write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
         }
         // return pair of prev_lsn and last_lsn
         else if query_string.starts_with("get_last_record_rlsn ") {

pageserver/src/pgdatadir_mapping.rs

@@ -43,16 +43,6 @@ pub enum CalculateLogicalSizeError {
     Other(#[from] anyhow::Error),
 }
 
-#[derive(Debug, thiserror::Error)]
-pub enum RelationError {
-    #[error("Relation Already Exists")]
-    AlreadyExists,
-    #[error("invalid relnode")]
-    InvalidRelnode,
-    #[error(transparent)]
-    Other(#[from] anyhow::Error),
-}
-
 ///
 /// This impl provides all the functionality to store PostgreSQL relations, SLRUs,
 /// and other special kinds of files, in a versioned key-value store. The
@@ -111,9 +101,9 @@ impl Timeline {
         ctx: &RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         if tag.relnode == 0 {
-            return Err(PageReconstructError::Other(
-                RelationError::InvalidRelnode.into(),
-            ));
+            return Err(PageReconstructError::Other(anyhow::anyhow!(
+                "invalid relnode"
+            )));
         }
 
         let nblocks = self.get_rel_size(tag, lsn, latest, ctx).await?;
@@ -158,9 +148,9 @@ impl Timeline {
         ctx: &RequestContext,
     ) -> Result<BlockNumber, PageReconstructError> {
         if tag.relnode == 0 {
-            return Err(PageReconstructError::Other(
-                RelationError::InvalidRelnode.into(),
-            ));
+            return Err(PageReconstructError::Other(anyhow::anyhow!(
+                "invalid relnode"
+            )));
         }
 
         if let Some(nblocks) = self.get_cached_rel_size(&tag, lsn) {
@@ -203,9 +193,9 @@ impl Timeline {
         ctx: &RequestContext,
     ) -> Result<bool, PageReconstructError> {
         if tag.relnode == 0 {
-            return Err(PageReconstructError::Other(
-                RelationError::InvalidRelnode.into(),
-            ));
+            return Err(PageReconstructError::Other(anyhow::anyhow!(
+                "invalid relnode"
+            )));
         }
 
         // first try to lookup relation in cache
@@ -709,20 +699,6 @@ impl<'a> DatadirModification<'a> {
         Ok(())
     }
 
-    #[cfg(test)]
-    pub fn init_empty_test_timeline(&mut self) -> anyhow::Result<()> {
-        self.init_empty()?;
-        self.put_control_file(bytes::Bytes::from_static(
-            b"control_file contents do not matter",
-        ))
-        .context("put_control_file")?;
-        self.put_checkpoint(bytes::Bytes::from_static(
-            b"checkpoint_file contents do not matter",
-        ))
-        .context("put_checkpoint_file")?;
-        Ok(())
-    }
-
     /// Put a new page version that can be constructed from a WAL record
     ///
     /// NOTE: this will *not* implicitly extend the relation, if the page is beyond the
@@ -734,7 +710,7 @@ impl<'a> DatadirModification<'a> {
         blknum: BlockNumber,
         rec: NeonWalRecord,
     ) -> anyhow::Result<()> {
-        anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
+        anyhow::ensure!(rel.relnode != 0, "invalid relnode");
         self.put(rel_block_to_key(rel, blknum), Value::WalRecord(rec));
         Ok(())
     }
@@ -761,7 +737,7 @@ impl<'a> DatadirModification<'a> {
         blknum: BlockNumber,
         img: Bytes,
     ) -> anyhow::Result<()> {
-        anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
+        anyhow::ensure!(rel.relnode != 0, "invalid relnode");
         self.put(rel_block_to_key(rel, blknum), Value::Image(img));
         Ok(())
     }
@@ -885,38 +861,32 @@ impl<'a> DatadirModification<'a> {
         rel: RelTag,
         nblocks: BlockNumber,
         ctx: &RequestContext,
-    ) -> Result<(), RelationError> {
-        if rel.relnode == 0 {
-            return Err(RelationError::AlreadyExists);
-        }
+    ) -> anyhow::Result<()> {
+        anyhow::ensure!(rel.relnode != 0, "invalid relnode");
         // It's possible that this is the first rel for this db in this
         // tablespace.  Create the reldir entry for it if so.
-        let mut dbdir = DbDirectory::des(&self.get(DBDIR_KEY, ctx).await.context("read db")?)
-            .context("deserialize db")?;
+        let mut dbdir = DbDirectory::des(&self.get(DBDIR_KEY, ctx).await?)?;
         let rel_dir_key = rel_dir_to_key(rel.spcnode, rel.dbnode);
         let mut rel_dir = if dbdir.dbdirs.get(&(rel.spcnode, rel.dbnode)).is_none() {
             // Didn't exist. Update dbdir
             dbdir.dbdirs.insert((rel.spcnode, rel.dbnode), false);
-            let buf = DbDirectory::ser(&dbdir).context("serialize db")?;
+            let buf = DbDirectory::ser(&dbdir)?;
             self.put(DBDIR_KEY, Value::Image(buf.into()));
 
             // and create the RelDirectory
             RelDirectory::default()
         } else {
             // reldir already exists, fetch it
-            RelDirectory::des(&self.get(rel_dir_key, ctx).await.context("read db")?)
-                .context("deserialize db")?
+            RelDirectory::des(&self.get(rel_dir_key, ctx).await?)?
         };
 
         // Add the new relation to the rel directory entry, and write it back
         if !rel_dir.rels.insert((rel.relnode, rel.forknum)) {
-            return Err(RelationError::AlreadyExists);
+            anyhow::bail!("rel {rel} already exists");
         }
         self.put(
             rel_dir_key,
-            Value::Image(Bytes::from(
-                RelDirectory::ser(&rel_dir).context("serialize")?,
-            )),
+            Value::Image(Bytes::from(RelDirectory::ser(&rel_dir)?)),
         );
 
         // Put size
@@ -941,7 +911,7 @@ impl<'a> DatadirModification<'a> {
         nblocks: BlockNumber,
         ctx: &RequestContext,
     ) -> anyhow::Result<()> {
-        anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
+        anyhow::ensure!(rel.relnode != 0, "invalid relnode");
         let last_lsn = self.tline.get_last_record_lsn();
         if self.tline.get_rel_exists(rel, last_lsn, true, ctx).await? {
             let size_key = rel_size_to_key(rel);
@@ -972,7 +942,7 @@ impl<'a> DatadirModification<'a> {
         nblocks: BlockNumber,
         ctx: &RequestContext,
     ) -> anyhow::Result<()> {
-        anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
+        anyhow::ensure!(rel.relnode != 0, "invalid relnode");
 
         // Put size
         let size_key = rel_size_to_key(rel);
@@ -993,7 +963,7 @@ impl<'a> DatadirModification<'a> {
 
     /// Drop a relation.
     pub async fn put_rel_drop(&mut self, rel: RelTag, ctx: &RequestContext) -> anyhow::Result<()> {
-        anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
+        anyhow::ensure!(rel.relnode != 0, "invalid relnode");
 
         // Remove it from the directory entry
         let dir_key = rel_dir_to_key(rel.spcnode, rel.dbnode);
@@ -1138,7 +1108,7 @@ impl<'a> DatadirModification<'a> {
     /// retains all the metadata, but data pages are flushed. That's again OK
     /// for bulk import, where you are just loading data pages and won't try to
     /// modify the same pages twice.
-    pub async fn flush(&mut self) -> anyhow::Result<()> {
+    pub fn flush(&mut self) -> anyhow::Result<()> {
         // Unless we have accumulated a decent amount of changes, it's not worth it
         // to scan through the pending_updates list.
         let pending_nblocks = self.pending_nblocks;
@@ -1146,20 +1116,19 @@ impl<'a> DatadirModification<'a> {
             return Ok(());
         }
 
-        let writer = self.tline.writer().await;
+        let writer = self.tline.writer();
 
         // Flush relation and  SLRU data blocks, keep metadata.
-        let mut retained_pending_updates = HashMap::new();
-        for (key, value) in self.pending_updates.drain() {
-            if is_rel_block_key(key) || is_slru_block_key(key) {
-                // This bails out on first error without modifying pending_updates.
-                // That's Ok, cf this function's doc comment.
-                writer.put(key, self.lsn, &value).await?;
+        let mut result: anyhow::Result<()> = Ok(());
+        self.pending_updates.retain(|&key, value| {
+            if result.is_ok() && (is_rel_block_key(key) || is_slru_block_key(key)) {
+                result = writer.put(key, self.lsn, value);
+                false
             } else {
-                retained_pending_updates.insert(key, value);
+                true
             }
-        }
-        self.pending_updates.extend(retained_pending_updates);
+        });
+        result?;
 
         if pending_nblocks != 0 {
             writer.update_current_logical_size(pending_nblocks * i64::from(BLCKSZ));
@@ -1174,17 +1143,17 @@ impl<'a> DatadirModification<'a> {
     /// underlying timeline.
     /// All the modifications in this atomic update are stamped by the specified LSN.
     ///
-    pub async fn commit(&mut self) -> anyhow::Result<()> {
-        let writer = self.tline.writer().await;
+    pub fn commit(&mut self) -> anyhow::Result<()> {
+        let writer = self.tline.writer();
         let lsn = self.lsn;
         let pending_nblocks = self.pending_nblocks;
         self.pending_nblocks = 0;
 
         for (key, value) in self.pending_updates.drain() {
-            writer.put(key, lsn, &value).await?;
+            writer.put(key, lsn, &value)?;
         }
         for key_range in self.pending_deletions.drain(..) {
-            writer.delete(key_range, lsn).await?;
+            writer.delete(key_range, lsn)?;
         }
 
         writer.finish_write(lsn);
@@ -1624,6 +1593,20 @@ fn is_slru_block_key(key: Key) -> bool {
         && key.field6 != 0xffffffff // and not SlruSegSize
 }
 
+#[cfg(test)]
+pub fn create_test_timeline(
+    tenant: &crate::tenant::Tenant,
+    timeline_id: utils::id::TimelineId,
+    pg_version: u32,
+    ctx: &RequestContext,
+) -> anyhow::Result<std::sync::Arc<Timeline>> {
+    let tline = tenant.create_test_timeline(timeline_id, Lsn(8), pg_version, ctx)?;
+    let mut m = tline.begin_modification(Lsn(8));
+    m.init_empty()?;
+    m.commit()?;
+    Ok(tline)
+}
+
 #[allow(clippy::bool_assert_comparison)]
 #[cfg(test)]
 mod tests {

pageserver/src/tenant/config.rs

@@ -38,8 +38,8 @@ pub mod defaults {
     pub const DEFAULT_GC_PERIOD: &str = "1 hr";
     pub const DEFAULT_IMAGE_CREATION_THRESHOLD: usize = 3;
     pub const DEFAULT_PITR_INTERVAL: &str = "7 days";
-    pub const DEFAULT_WALRECEIVER_CONNECT_TIMEOUT: &str = "10 seconds";
-    pub const DEFAULT_WALRECEIVER_LAGGING_WAL_TIMEOUT: &str = "10 seconds";
+    pub const DEFAULT_WALRECEIVER_CONNECT_TIMEOUT: &str = "2 seconds";
+    pub const DEFAULT_WALRECEIVER_LAGGING_WAL_TIMEOUT: &str = "3 seconds";
     pub const DEFAULT_MAX_WALRECEIVER_LSN_WAL_LAG: u64 = 10 * 1024 * 1024;
     pub const DEFAULT_EVICTIONS_LOW_RESIDENCE_DURATION_METRIC_THRESHOLD: &str = "24 hour";
 }

pageserver/src/tenant/manifest.rs

@@ -1,325 +0,0 @@
-//! This module contains the encoding and decoding of the local manifest file.
-//!
-//! MANIFEST is a write-ahead log which is stored locally to each timeline. It
-//! records the state of the storage engine. It contains a snapshot of the
-//! state and all operations proceeding that snapshot. The file begins with a
-//! header recording MANIFEST version number. After that, it contains a snapshot.
-//! The snapshot is followed by a list of operations. Each operation is a list
-//! of records. Each record is either an addition or a removal of a layer.
-//!
-//! With MANIFEST, we can:
-//!
-//! 1. recover state quickly by reading the file, potentially boosting the
-//!    startup speed.
-//! 2. ensure all operations are atomic and avoid corruption, solving issues
-//!    like redundant image layer and preparing us for future compaction
-//!    strategies.
-//!
-//! There is also a format for storing all layer files on S3, called
-//! `index_part.json`. Compared with index_part, MANIFEST is an WAL which
-//! records all operations as logs, and therefore we can easily replay the
-//! operations when recovering from crash, while ensuring those operations
-//! are atomic upon restart.
-//!
-//! Currently, this is not used in the system. Future refactors will ensure
-//! the storage state will be recorded in this file, and the system can be
-//! recovered from this file. This is tracked in
-//! https://github.com/neondatabase/neon/issues/4418
-
-use std::io::{self, Read, Write};
-
-use crate::virtual_file::VirtualFile;
-use anyhow::Result;
-use bytes::{Buf, BufMut, Bytes, BytesMut};
-use crc32c::crc32c;
-use serde::{Deserialize, Serialize};
-use tracing::log::warn;
-use utils::lsn::Lsn;
-
-use super::storage_layer::PersistentLayerDesc;
-
-pub struct Manifest {
-    file: VirtualFile,
-}
-
-#[derive(Clone, Serialize, Deserialize, PartialEq, Eq, Debug)]
-pub struct Snapshot {
-    pub layers: Vec<PersistentLayerDesc>,
-}
-
-/// serde by default encode this in tagged enum, and therefore it will be something
-/// like `{ "AddLayer": { ... } }`.
-#[derive(Clone, Serialize, Deserialize, PartialEq, Eq, Debug)]
-pub enum Record {
-    AddLayer(PersistentLayerDesc),
-    RemoveLayer(PersistentLayerDesc),
-}
-
-/// `echo neon.manifest | sha1sum` and take the leading 8 bytes.
-const MANIFEST_MAGIC_NUMBER: u64 = 0xf5c44592b806109c;
-const MANIFEST_VERSION: u64 = 1;
-
-#[derive(Clone, Serialize, Deserialize, PartialEq, Eq, Debug)]
-pub struct ManifestHeader {
-    magic_number: u64,
-    version: u64,
-}
-
-const MANIFEST_HEADER_LEN: usize = 16;
-
-impl ManifestHeader {
-    fn encode(&self) -> BytesMut {
-        let mut buf = BytesMut::with_capacity(MANIFEST_HEADER_LEN);
-        buf.put_u64(self.magic_number);
-        buf.put_u64(self.version);
-        buf
-    }
-
-    fn decode(mut buf: &[u8]) -> Self {
-        assert!(buf.len() == MANIFEST_HEADER_LEN, "invalid header");
-        Self {
-            magic_number: buf.get_u64(),
-            version: buf.get_u64(),
-        }
-    }
-}
-
-#[derive(Clone, Serialize, Deserialize, PartialEq, Eq, Debug)]
-pub enum Operation {
-    /// A snapshot of the current state.
-    ///
-    /// Lsn field represents the LSN that is persisted to disk for this snapshot.
-    Snapshot(Snapshot, Lsn),
-    /// An atomic operation that changes the state.
-    ///
-    /// Lsn field represents the LSN that is persisted to disk after the operation is done.
-    /// This will only change when new L0 is flushed to the disk.
-    Operation(Vec<Record>, Lsn),
-}
-
-struct RecordHeader {
-    size: u32,
-    checksum: u32,
-}
-
-const RECORD_HEADER_LEN: usize = 8;
-
-impl RecordHeader {
-    fn encode(&self) -> BytesMut {
-        let mut buf = BytesMut::with_capacity(RECORD_HEADER_LEN);
-        buf.put_u32(self.size);
-        buf.put_u32(self.checksum);
-        buf
-    }
-
-    fn decode(mut buf: &[u8]) -> Self {
-        assert!(buf.len() == RECORD_HEADER_LEN, "invalid header");
-        Self {
-            size: buf.get_u32(),
-            checksum: buf.get_u32(),
-        }
-    }
-}
-
-#[derive(Debug, thiserror::Error)]
-pub enum ManifestLoadError {
-    #[error("manifest header is corrupted")]
-    CorruptedManifestHeader,
-    #[error("unsupported manifest version: got {0}, expected {1}")]
-    UnsupportedVersion(u64, u64),
-    #[error("error when decoding record: {0}")]
-    DecodeRecord(serde_json::Error),
-    #[error("I/O error: {0}")]
-    Io(io::Error),
-}
-
-#[must_use = "Should check if the manifest is partially corrupted"]
-pub struct ManifestPartiallyCorrupted(bool);
-
-impl Manifest {
-    /// Create a new manifest by writing the manifest header and a snapshot record to the given file.
-    pub fn init(file: VirtualFile, snapshot: Snapshot, lsn: Lsn) -> Result<Self> {
-        let mut manifest = Self { file };
-        manifest.append_manifest_header(ManifestHeader {
-            magic_number: MANIFEST_MAGIC_NUMBER,
-            version: MANIFEST_VERSION,
-        })?;
-        manifest.append_operation(Operation::Snapshot(snapshot, lsn))?;
-        Ok(manifest)
-    }
-
-    /// Load a manifest. Returns the manifest and a list of operations. If the manifest is corrupted,
-    /// the bool flag will be set to true and the user is responsible to reconstruct a new manifest and
-    /// backup the current one.
-    pub fn load(
-        mut file: VirtualFile,
-    ) -> Result<(Self, Vec<Operation>, ManifestPartiallyCorrupted), ManifestLoadError> {
-        let mut buf = vec![];
-        file.read_to_end(&mut buf).map_err(ManifestLoadError::Io)?;
-
-        // Read manifest header
-        let mut buf = Bytes::from(buf);
-        if buf.remaining() < MANIFEST_HEADER_LEN {
-            return Err(ManifestLoadError::CorruptedManifestHeader);
-        }
-        let header = ManifestHeader::decode(&buf[..MANIFEST_HEADER_LEN]);
-        buf.advance(MANIFEST_HEADER_LEN);
-        if header.version != MANIFEST_VERSION {
-            return Err(ManifestLoadError::UnsupportedVersion(
-                header.version,
-                MANIFEST_VERSION,
-            ));
-        }
-
-        // Read operations
-        let mut operations = Vec::new();
-        let corrupted = loop {
-            if buf.remaining() == 0 {
-                break false;
-            }
-            if buf.remaining() < RECORD_HEADER_LEN {
-                warn!("incomplete header when decoding manifest, could be corrupted");
-                break true;
-            }
-            let RecordHeader { size, checksum } = RecordHeader::decode(&buf[..RECORD_HEADER_LEN]);
-            let size = size as usize;
-            buf.advance(RECORD_HEADER_LEN);
-            if buf.remaining() < size {
-                warn!("incomplete data when decoding manifest, could be corrupted");
-                break true;
-            }
-            let data = &buf[..size];
-            if crc32c(data) != checksum {
-                warn!("checksum mismatch when decoding manifest, could be corrupted");
-                break true;
-            }
-            // if the following decode fails, we cannot use the manifest or safely ignore any record.
-            operations.push(serde_json::from_slice(data).map_err(ManifestLoadError::DecodeRecord)?);
-            buf.advance(size);
-        };
-        Ok((
-            Self { file },
-            operations,
-            ManifestPartiallyCorrupted(corrupted),
-        ))
-    }
-
-    fn append_data(&mut self, data: &[u8]) -> Result<()> {
-        if data.len() >= u32::MAX as usize {
-            panic!("data too large");
-        }
-        let header = RecordHeader {
-            size: data.len() as u32,
-            checksum: crc32c(data),
-        };
-        let header = header.encode();
-        self.file.write_all(&header)?;
-        self.file.write_all(data)?;
-        self.file.sync_all()?;
-        Ok(())
-    }
-
-    fn append_manifest_header(&mut self, header: ManifestHeader) -> Result<()> {
-        let encoded = header.encode();
-        self.file.write_all(&encoded)?;
-        Ok(())
-    }
-
-    /// Add an operation to the manifest. The operation will be appended to the end of the file,
-    /// and the file will fsync.
-    pub fn append_operation(&mut self, operation: Operation) -> Result<()> {
-        let encoded = Vec::from(serde_json::to_string(&operation)?);
-        self.append_data(&encoded)
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use std::fs::OpenOptions;
-
-    use crate::repository::Key;
-
-    use super::*;
-
-    #[test]
-    fn test_read_manifest() {
-        let testdir = crate::config::PageServerConf::test_repo_dir("test_read_manifest");
-        std::fs::create_dir_all(&testdir).unwrap();
-        let file = VirtualFile::create(&testdir.join("MANIFEST")).unwrap();
-        let layer1 = PersistentLayerDesc::new_test(Key::from_i128(0)..Key::from_i128(233));
-        let layer2 = PersistentLayerDesc::new_test(Key::from_i128(233)..Key::from_i128(2333));
-        let layer3 = PersistentLayerDesc::new_test(Key::from_i128(2333)..Key::from_i128(23333));
-        let layer4 = PersistentLayerDesc::new_test(Key::from_i128(23333)..Key::from_i128(233333));
-
-        // Write a manifest with a snapshot and some operations
-        let snapshot = Snapshot {
-            layers: vec![layer1, layer2],
-        };
-        let mut manifest = Manifest::init(file, snapshot.clone(), Lsn::from(0)).unwrap();
-        manifest
-            .append_operation(Operation::Operation(
-                vec![Record::AddLayer(layer3.clone())],
-                Lsn::from(1),
-            ))
-            .unwrap();
-        drop(manifest);
-
-        // Open the second time and write
-        let file = VirtualFile::open_with_options(
-            &testdir.join("MANIFEST"),
-            OpenOptions::new()
-                .read(true)
-                .write(true)
-                .create_new(false)
-                .truncate(false),
-        )
-        .unwrap();
-        let (mut manifest, operations, corrupted) = Manifest::load(file).unwrap();
-        assert!(!corrupted.0);
-        assert_eq!(operations.len(), 2);
-        assert_eq!(
-            &operations[0],
-            &Operation::Snapshot(snapshot.clone(), Lsn::from(0))
-        );
-        assert_eq!(
-            &operations[1],
-            &Operation::Operation(vec![Record::AddLayer(layer3.clone())], Lsn::from(1))
-        );
-        manifest
-            .append_operation(Operation::Operation(
-                vec![
-                    Record::RemoveLayer(layer3.clone()),
-                    Record::AddLayer(layer4.clone()),
-                ],
-                Lsn::from(2),
-            ))
-            .unwrap();
-        drop(manifest);
-
-        // Open the third time and verify
-        let file = VirtualFile::open_with_options(
-            &testdir.join("MANIFEST"),
-            OpenOptions::new()
-                .read(true)
-                .write(true)
-                .create_new(false)
-                .truncate(false),
-        )
-        .unwrap();
-        let (_manifest, operations, corrupted) = Manifest::load(file).unwrap();
-        assert!(!corrupted.0);
-        assert_eq!(operations.len(), 3);
-        assert_eq!(&operations[0], &Operation::Snapshot(snapshot, Lsn::from(0)));
-        assert_eq!(
-            &operations[1],
-            &Operation::Operation(vec![Record::AddLayer(layer3.clone())], Lsn::from(1))
-        );
-        assert_eq!(
-            &operations[2],
-            &Operation::Operation(
-                vec![Record::RemoveLayer(layer3), Record::AddLayer(layer4)],
-                Lsn::from(2)
-            )
-        );
-    }
-}

pageserver/src/tenant/remote_timeline_client.rs

@@ -753,18 +753,22 @@ impl RemoteTimelineClient {
 
         // Have a failpoint that can use the `pause` failpoint action.
         // We don't want to block the executor thread, hence, spawn_blocking + await.
-        if cfg!(feature = "testing") {
-            tokio::task::spawn_blocking({
-                let current = tracing::Span::current();
-                move || {
-                    let _entered = current.entered();
-                    tracing::info!("at failpoint persist_deleted_index_part");
-                    fail::fail_point!("persist_deleted_index_part");
-                }
-            })
-            .await
-            .expect("spawn_blocking");
-        }
+        #[cfg(feature = "testing")]
+        tokio::task::spawn_blocking({
+            let current = tracing::Span::current();
+            move || {
+                let _entered = current.entered();
+                tracing::info!(
+                    "at failpoint persist_index_part_with_deleted_flag_after_set_before_upload_pause"
+                );
+                fail::fail_point!(
+                    "persist_index_part_with_deleted_flag_after_set_before_upload_pause"
+                );
+            }
+        })
+        .await
+        .expect("spawn_blocking");
+
         upload::upload_index_part(
             self.conf,
             &self.storage_impl,
@@ -1388,12 +1392,7 @@ mod tests {
             let harness = TenantHarness::create(test_name)?;
             let (tenant, ctx) = runtime.block_on(harness.load());
             // create an empty timeline directory
-            let _ = runtime.block_on(tenant.create_test_timeline(
-                TIMELINE_ID,
-                Lsn(8),
-                DEFAULT_PG_VERSION,
-                &ctx,
-            ))?;
+            let _ = tenant.create_test_timeline(TIMELINE_ID, Lsn(0), DEFAULT_PG_VERSION, &ctx)?;
 
             let remote_fs_dir = harness.conf.workdir.join("remote_fs");
             std::fs::create_dir_all(remote_fs_dir)?;

pageserver/src/tenant/storage_layer.rs

@@ -389,10 +389,10 @@ pub trait Layer: std::fmt::Debug + Send + Sync {
 }
 
 /// Returned by [`Layer::iter`]
-pub type LayerIter<'i> = Box<dyn Iterator<Item = Result<(Key, Lsn, Value)>> + 'i + Send>;
+pub type LayerIter<'i> = Box<dyn Iterator<Item = Result<(Key, Lsn, Value)>> + 'i>;
 
 /// Returned by [`Layer::key_iter`]
-pub type LayerKeyIter<'i> = Box<dyn Iterator<Item = (Key, Lsn, u64)> + 'i + Send>;
+pub type LayerKeyIter<'i> = Box<dyn Iterator<Item = (Key, Lsn, u64)> + 'i>;
 
 /// A Layer contains all data in a "rectangle" consisting of a range of keys and
 /// range of LSNs.

pageserver/src/tenant/storage_layer/delta_layer.rs

@@ -37,7 +37,6 @@ 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 once_cell::sync::OnceCell;
 use pageserver_api::models::{HistoricLayerInfo, LayerAccessKind};
 use rand::{distributions::Alphanumeric, Rng};
 use serde::{Deserialize, Serialize};
@@ -47,6 +46,7 @@ use std::io::{Seek, SeekFrom};
 use std::ops::Range;
 use std::os::unix::fs::FileExt;
 use std::path::{Path, PathBuf};
+use std::sync::{RwLock, RwLockReadGuard, RwLockWriteGuard};
 use tracing::*;
 
 use utils::{
@@ -184,7 +184,7 @@ pub struct DeltaLayer {
 
     access_stats: LayerAccessStats,
 
-    inner: OnceCell<DeltaLayerInner>,
+    inner: RwLock<DeltaLayerInner>,
 }
 
 impl std::fmt::Debug for DeltaLayer {
@@ -201,17 +201,21 @@ impl std::fmt::Debug for DeltaLayer {
 }
 
 pub struct DeltaLayerInner {
+    /// If false, the fields below have not been loaded into memory yet.
+    loaded: bool,
+
     // values copied from summary
     index_start_blk: u32,
     index_root_blk: u32,
 
-    /// Reader object for reading blocks from the file.
-    file: FileBlockReader<VirtualFile>,
+    /// Reader object for reading blocks from the file. (None if not loaded yet)
+    file: Option<FileBlockReader<VirtualFile>>,
 }
 
 impl std::fmt::Debug for DeltaLayerInner {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
         f.debug_struct("DeltaLayerInner")
+            .field("loaded", &self.loaded)
             .field("index_start_blk", &self.index_start_blk)
             .field("index_root_blk", &self.index_root_blk)
             .finish()
@@ -242,7 +246,7 @@ impl Layer for DeltaLayer {
             inner.index_start_blk, inner.index_root_blk
         );
 
-        let file = &inner.file;
+        let file = inner.file.as_ref().unwrap();
         let tree_reader = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
             inner.index_start_blk,
             inner.index_root_blk,
@@ -311,7 +315,7 @@ impl Layer for DeltaLayer {
             let inner = self.load(LayerAccessKind::GetValueReconstructData, ctx)?;
 
             // Scan the page versions backwards, starting from `lsn`.
-            let file = &inner.file;
+            let file = inner.file.as_ref().unwrap();
             let tree_reader = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
                 inner.index_start_blk,
                 inner.index_root_blk,
@@ -496,22 +500,51 @@ impl DeltaLayer {
     /// Open the underlying file and read the metadata into memory, if it's
     /// not loaded already.
     ///
-    fn load(&self, access_kind: LayerAccessKind, ctx: &RequestContext) -> Result<&DeltaLayerInner> {
+    fn load(
+        &self,
+        access_kind: LayerAccessKind,
+        ctx: &RequestContext,
+    ) -> Result<RwLockReadGuard<DeltaLayerInner>> {
         self.access_stats
             .record_access(access_kind, ctx.task_kind());
-        // Quick exit if already loaded
-        self.inner
-            .get_or_try_init(|| self.load_inner())
-            .with_context(|| format!("Failed to load delta layer {}", self.path().display()))
+        loop {
+            // Quick exit if already loaded
+            let inner = self.inner.read().unwrap();
+            if inner.loaded {
+                return Ok(inner);
+            }
+
+            // Need to open the file and load the metadata. Upgrade our lock to
+            // a write lock. (Or rather, release and re-lock in write mode.)
+            drop(inner);
+            let inner = self.inner.write().unwrap();
+            if !inner.loaded {
+                self.load_inner(inner).with_context(|| {
+                    format!("Failed to load delta layer {}", self.path().display())
+                })?;
+            } else {
+                // Another thread loaded it while we were not holding the lock.
+            }
+
+            // We now have the file open and loaded. There's no function to do
+            // that in the std library RwLock, so we have to release and re-lock
+            // in read mode. (To be precise, the lock guard was moved in the
+            // above call to `load_inner`, so it's already been released). And
+            // while we do that, another thread could unload again, so we have
+            // to re-check and retry if that happens.
+        }
     }
 
-    fn load_inner(&self) -> Result<DeltaLayerInner> {
+    fn load_inner(&self, mut inner: RwLockWriteGuard<DeltaLayerInner>) -> Result<()> {
         let path = self.path();
 
-        let file = VirtualFile::open(&path)
-            .with_context(|| format!("Failed to open file '{}'", path.display()))?;
-        let file = FileBlockReader::new(file);
-
+        // Open the file if it's not open already.
+        if inner.file.is_none() {
+            let file = VirtualFile::open(&path)
+                .with_context(|| format!("Failed to open file '{}'", path.display()))?;
+            inner.file = Some(FileBlockReader::new(file));
+        }
+        let file = inner.file.as_mut().unwrap();
         let summary_blk = file.read_blk(0)?;
         let actual_summary = Summary::des_prefix(summary_blk.as_ref())?;
 
@@ -538,13 +571,13 @@ impl DeltaLayer {
             }
         }
 
+        inner.index_start_blk = actual_summary.index_start_blk;
+        inner.index_root_blk = actual_summary.index_root_blk;
+
         debug!("loaded from {}", &path.display());
 
-        Ok(DeltaLayerInner {
-            file,
-            index_start_blk: actual_summary.index_start_blk,
-            index_root_blk: actual_summary.index_root_blk,
-        })
+        inner.loaded = true;
+        Ok(())
     }
 
     /// Create a DeltaLayer struct representing an existing file on disk.
@@ -566,7 +599,12 @@ impl DeltaLayer {
                 file_size,
             ),
             access_stats,
-            inner: once_cell::sync::OnceCell::new(),
+            inner: RwLock::new(DeltaLayerInner {
+                loaded: false,
+                file: None,
+                index_start_blk: 0,
+                index_root_blk: 0,
+            }),
         }
     }
 
@@ -593,7 +631,12 @@ impl DeltaLayer {
                 metadata.len(),
             ),
             access_stats: LayerAccessStats::empty_will_record_residence_event_later(),
-            inner: once_cell::sync::OnceCell::new(),
+            inner: RwLock::new(DeltaLayerInner {
+                loaded: false,
+                file: None,
+                index_start_blk: 0,
+                index_root_blk: 0,
+            }),
         })
     }
 
@@ -757,7 +800,12 @@ impl DeltaLayerWriterInner {
                 metadata.len(),
             ),
             access_stats: LayerAccessStats::empty_will_record_residence_event_later(),
-            inner: once_cell::sync::OnceCell::new(),
+            inner: RwLock::new(DeltaLayerInner {
+                loaded: false,
+                file: None,
+                index_start_blk,
+                index_root_blk,
+            }),
         };
 
         // fsync the file
@@ -892,13 +940,13 @@ struct DeltaValueIter<'a> {
     reader: BlockCursor<Adapter<'a>>,
 }
 
-struct Adapter<'a>(&'a DeltaLayerInner);
+struct Adapter<'a>(RwLockReadGuard<'a, DeltaLayerInner>);
 
 impl<'a> BlockReader for Adapter<'a> {
     type BlockLease = PageReadGuard<'static>;
 
     fn read_blk(&self, blknum: u32) -> Result<Self::BlockLease, std::io::Error> {
-        self.0.file.read_blk(blknum)
+        self.0.file.as_ref().unwrap().read_blk(blknum)
     }
 }
 
@@ -911,8 +959,8 @@ impl<'a> Iterator for DeltaValueIter<'a> {
 }
 
 impl<'a> DeltaValueIter<'a> {
-    fn new(inner: &'a DeltaLayerInner) -> Result<Self> {
-        let file = &inner.file;
+    fn new(inner: RwLockReadGuard<'a, DeltaLayerInner>) -> Result<Self> {
+        let file = inner.file.as_ref().unwrap();
         let tree_reader = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
             inner.index_start_blk,
             inner.index_root_blk,
@@ -985,8 +1033,8 @@ impl Iterator for DeltaKeyIter {
 }
 
 impl<'a> DeltaKeyIter {
-    fn new(inner: &'a DeltaLayerInner) -> Result<Self> {
-        let file = &inner.file;
+    fn new(inner: RwLockReadGuard<'a, DeltaLayerInner>) -> Result<Self> {
+        let file = inner.file.as_ref().unwrap();
         let tree_reader = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
             inner.index_start_blk,
             inner.index_root_blk,
@@ -1026,21 +1074,3 @@ impl<'a> DeltaKeyIter {
         Ok(iter)
     }
 }
-
-#[cfg(test)]
-mod test {
-    use super::DeltaKeyIter;
-    use super::DeltaLayer;
-    use super::DeltaValueIter;
-
-    // We will soon need the iters to be send in the compaction code.
-    // Cf https://github.com/neondatabase/neon/pull/4462#issuecomment-1587398883
-    // Cf https://github.com/neondatabase/neon/issues/4471
-    #[test]
-    fn is_send() {
-        fn assert_send<T: Send>() {}
-        assert_send::<DeltaLayer>();
-        assert_send::<DeltaValueIter>();
-        assert_send::<DeltaKeyIter>();
-    }
-}

pageserver/src/tenant/storage_layer/inmemory_layer.rs

@@ -304,7 +304,7 @@ impl InMemoryLayer {
         Ok(())
     }
 
-    pub async fn put_tombstone(&self, _key_range: Range<Key>, _lsn: Lsn) -> Result<()> {
+    pub fn put_tombstone(&self, _key_range: Range<Key>, _lsn: Lsn) -> Result<()> {
         // TODO: Currently, we just leak the storage for any deleted keys
 
         Ok(())

pageserver/src/tenant/storage_layer/layer_desc.rs

@@ -9,12 +9,10 @@ use crate::{context::RequestContext, repository::Key};
 
 use super::{DeltaFileName, ImageFileName, LayerFileName};
 
-use serde::{Deserialize, Serialize};
-
 /// A unique identifier of a persistent layer. This is different from `LayerDescriptor`, which is only used in the
 /// benchmarks. This struct contains all necessary information to find the image / delta layer. It also provides
 /// a unified way to generate layer information like file name.
-#[derive(Debug, PartialEq, Eq, Clone, Serialize, Deserialize)]
+#[derive(Debug, PartialEq, Eq, Clone)]
 pub struct PersistentLayerDesc {
     pub tenant_id: TenantId,
     pub timeline_id: TimelineId,
@@ -52,19 +50,6 @@ impl PersistentLayerDesc {
         self.filename().file_name()
     }
 
-    #[cfg(test)]
-    pub fn new_test(key_range: Range<Key>) -> Self {
-        Self {
-            tenant_id: TenantId::generate(),
-            timeline_id: TimelineId::generate(),
-            key_range,
-            lsn_range: Lsn(0)..Lsn(1),
-            is_delta: false,
-            is_incremental: false,
-            file_size: 0,
-        }
-    }
-
     pub fn new_img(
         tenant_id: TenantId,
         timeline_id: TimelineId,

pageserver/src/tenant/timeline.rs

@@ -15,7 +15,6 @@ use pageserver_api::models::{
     TimelineState,
 };
 use remote_storage::GenericRemoteStorage;
-use serde_with::serde_as;
 use storage_broker::BrokerClientChannel;
 use tokio::sync::{oneshot, watch, Semaphore, TryAcquireError};
 use tokio_util::sync::CancellationToken;
@@ -29,7 +28,7 @@ use std::ops::{Deref, Range};
 use std::path::{Path, PathBuf};
 use std::pin::pin;
 use std::sync::atomic::{AtomicI64, Ordering as AtomicOrdering};
-use std::sync::{Arc, Mutex, RwLock, Weak};
+use std::sync::{Arc, Mutex, MutexGuard, RwLock, Weak};
 use std::time::{Duration, Instant, SystemTime};
 
 use crate::context::{DownloadBehavior, RequestContext};
@@ -48,10 +47,7 @@ use crate::tenant::{
 
 use crate::config::PageServerConf;
 use crate::keyspace::{KeyPartitioning, KeySpace, KeySpaceRandomAccum};
-use crate::metrics::{
-    TimelineMetrics, MATERIALIZED_PAGE_CACHE_HIT, MATERIALIZED_PAGE_CACHE_HIT_DIRECT,
-    RECONSTRUCT_TIME, UNEXPECTED_ONDEMAND_DOWNLOADS,
-};
+use crate::metrics::{TimelineMetrics, UNEXPECTED_ONDEMAND_DOWNLOADS};
 use crate::pgdatadir_mapping::LsnForTimestamp;
 use crate::pgdatadir_mapping::{is_rel_fsm_block_key, is_rel_vm_block_key};
 use crate::pgdatadir_mapping::{BlockNumber, CalculateLogicalSizeError};
@@ -88,14 +84,9 @@ use super::remote_timeline_client::RemoteTimelineClient;
 use super::storage_layer::{DeltaLayer, ImageLayer, Layer, LayerAccessStatsReset};
 
 #[derive(Debug, PartialEq, Eq, Clone, Copy)]
-pub(super) enum FlushLoopState {
+enum FlushLoopState {
     NotStarted,
-    Running {
-        #[cfg(test)]
-        expect_initdb_optimization: bool,
-        #[cfg(test)]
-        initdb_optimization_count: usize,
-    },
+    Running,
     Exited,
 }
 
@@ -129,7 +120,7 @@ pub struct Timeline {
 
     pub pg_version: u32,
 
-    pub(crate) layers: tokio::sync::RwLock<LayerMap<dyn PersistentLayer>>,
+    pub(super) layers: RwLock<LayerMap<dyn PersistentLayer>>,
 
     /// Set of key ranges which should be covered by image layers to
     /// allow GC to remove old layers. This set is created by GC and its cutoff LSN is also stored.
@@ -189,10 +180,10 @@ pub struct Timeline {
     /// Locked automatically by [`TimelineWriter`] and checkpointer.
     /// Must always be acquired before the layer map/individual layer lock
     /// to avoid deadlock.
-    write_lock: tokio::sync::Mutex<()>,
+    write_lock: Mutex<()>,
 
     /// Used to avoid multiple `flush_loop` tasks running
-    pub(super) flush_loop_state: Mutex<FlushLoopState>,
+    flush_loop_state: Mutex<FlushLoopState>,
 
     /// layer_flush_start_tx can be used to wake up the layer-flushing task.
     /// The value is a counter, incremented every time a new flush cycle is requested.
@@ -224,6 +215,10 @@ pub struct Timeline {
     // though let's keep them both for better error visibility.
     pub initdb_lsn: Lsn,
 
+    // Compute nodes can set this field after successful application
+    // of a new spec, in order to avoid reapplying it on next restart.
+    pub compute_spec_id: tokio::sync::Mutex<Option<String>>,
+
     /// When did we last calculate the partitioning?
     partitioning: Mutex<(KeyPartitioning, Lsn)>,
 
@@ -543,7 +538,9 @@ impl Timeline {
                 match cached_lsn.cmp(&lsn) {
                     Ordering::Less => {} // there might be WAL between cached_lsn and lsn, we need to check
                     Ordering::Equal => {
-                        MATERIALIZED_PAGE_CACHE_HIT_DIRECT.inc();
+                        self.metrics
+                            .materialized_page_cache_hit_upon_request_counter
+                            .inc();
                         return Ok(cached_img); // exact LSN match, return the image
                     }
                     Ordering::Greater => {
@@ -565,7 +562,8 @@ impl Timeline {
             .await?;
         timer.stop_and_record();
 
-        RECONSTRUCT_TIME
+        self.metrics
+            .reconstruct_time_histo
             .observe_closure_duration(|| self.reconstruct_value(key, lsn, reconstruct_state))
     }
 
@@ -598,8 +596,8 @@ impl Timeline {
     /// The sum of the file size of all historic layers in the layer map.
     /// This method makes no distinction between local and remote layers.
     /// Hence, the result **does not represent local filesystem usage**.
-    pub async fn layer_size_sum(&self) -> u64 {
-        let layer_map = self.layers.read().await;
+    pub fn layer_size_sum(&self) -> u64 {
+        let layer_map = self.layers.read().unwrap();
         let mut size = 0;
         for l in layer_map.iter_historic_layers() {
             size += l.file_size();
@@ -690,7 +688,7 @@ impl Timeline {
     /// Flush to disk all data that was written with the put_* functions
     #[instrument(skip(self), fields(tenant_id=%self.tenant_id, timeline_id=%self.timeline_id))]
     pub async fn freeze_and_flush(&self) -> anyhow::Result<()> {
-        self.freeze_inmem_layer(false).await;
+        self.freeze_inmem_layer(false);
         self.flush_frozen_layers_and_wait().await
     }
 
@@ -869,10 +867,10 @@ impl Timeline {
     }
 
     /// Mutate the timeline with a [`TimelineWriter`].
-    pub async fn writer(&self) -> TimelineWriter<'_> {
+    pub fn writer(&self) -> TimelineWriter<'_> {
         TimelineWriter {
             tl: self,
-            _write_guard: self.write_lock.lock().await,
+            _write_guard: self.write_lock.lock().unwrap(),
         }
     }
 
@@ -906,39 +904,37 @@ impl Timeline {
     ///
     /// Also flush after a period of time without new data -- it helps
     /// safekeepers to regard pageserver as caught up and suspend activity.
-    pub async fn check_checkpoint_distance(self: &Arc<Timeline>) -> anyhow::Result<()> {
+    pub fn check_checkpoint_distance(self: &Arc<Timeline>) -> anyhow::Result<()> {
         let last_lsn = self.get_last_record_lsn();
-        let open_layer_size = {
-            let layers = self.layers.read().await;
-            let Some(open_layer) = layers.open_layer.as_ref() else {
-                return Ok(());
-            };
-            open_layer.size()?
-        };
-        let last_freeze_at = self.last_freeze_at.load();
-        let last_freeze_ts = *(self.last_freeze_ts.read().unwrap());
-        let distance = last_lsn.widening_sub(last_freeze_at);
-        // Checkpointing the open layer can be triggered by layer size or LSN range.
-        // S3 has a 5 GB limit on the size of one upload (without multi-part upload), and
-        // we want to stay below that with a big margin.  The LSN distance determines how
-        // much WAL the safekeepers need to store.
-        if distance >= self.get_checkpoint_distance().into()
-            || open_layer_size > self.get_checkpoint_distance()
-            || (distance > 0 && last_freeze_ts.elapsed() >= self.get_checkpoint_timeout())
-        {
-            info!(
-                "check_checkpoint_distance {}, layer size {}, elapsed since last flush {:?}",
-                distance,
-                open_layer_size,
-                last_freeze_ts.elapsed()
-            );
+        let layers = self.layers.read().unwrap();
+        if let Some(open_layer) = &layers.open_layer {
+            let open_layer_size = open_layer.size()?;
+            drop(layers);
+            let last_freeze_at = self.last_freeze_at.load();
+            let last_freeze_ts = *(self.last_freeze_ts.read().unwrap());
+            let distance = last_lsn.widening_sub(last_freeze_at);
+            // Checkpointing the open layer can be triggered by layer size or LSN range.
+            // S3 has a 5 GB limit on the size of one upload (without multi-part upload), and
+            // we want to stay below that with a big margin.  The LSN distance determines how
+            // much WAL the safekeepers need to store.
+            if distance >= self.get_checkpoint_distance().into()
+                || open_layer_size > self.get_checkpoint_distance()
+                || (distance > 0 && last_freeze_ts.elapsed() >= self.get_checkpoint_timeout())
+            {
+                info!(
+                    "check_checkpoint_distance {}, layer size {}, elapsed since last flush {:?}",
+                    distance,
+                    open_layer_size,
+                    last_freeze_ts.elapsed()
+                );
 
-            self.freeze_inmem_layer(true).await;
-            self.last_freeze_at.store(last_lsn);
-            *(self.last_freeze_ts.write().unwrap()) = Instant::now();
+                self.freeze_inmem_layer(true);
+                self.last_freeze_at.store(last_lsn);
+                *(self.last_freeze_ts.write().unwrap()) = Instant::now();
 
-            // Wake up the layer flusher
-            self.flush_frozen_layers();
+                // Wake up the layer flusher
+                self.flush_frozen_layers();
+            }
         }
         Ok(())
     }
@@ -1039,8 +1035,8 @@ impl Timeline {
         }
     }
 
-    pub async fn layer_map_info(&self, reset: LayerAccessStatsReset) -> LayerMapInfo {
-        let layer_map = self.layers.read().await;
+    pub fn layer_map_info(&self, reset: LayerAccessStatsReset) -> LayerMapInfo {
+        let layer_map = self.layers.read().unwrap();
         let mut in_memory_layers = Vec::with_capacity(layer_map.frozen_layers.len() + 1);
         if let Some(open_layer) = &layer_map.open_layer {
             in_memory_layers.push(open_layer.info());
@@ -1062,7 +1058,7 @@ impl Timeline {
 
     #[instrument(skip_all, fields(tenant = %self.tenant_id, timeline = %self.timeline_id))]
     pub async fn download_layer(&self, layer_file_name: &str) -> anyhow::Result<Option<bool>> {
-        let Some(layer) = self.find_layer(layer_file_name).await else { return Ok(None) };
+        let Some(layer) = self.find_layer(layer_file_name) else { return Ok(None) };
         let Some(remote_layer) = layer.downcast_remote_layer() else { return  Ok(Some(false)) };
         if self.remote_client.is_none() {
             return Ok(Some(false));
@@ -1075,7 +1071,7 @@ impl Timeline {
     /// Like [`evict_layer_batch`], but for just one layer.
     /// Additional case `Ok(None)` covers the case where the layer could not be found by its `layer_file_name`.
     pub async fn evict_layer(&self, layer_file_name: &str) -> anyhow::Result<Option<bool>> {
-        let Some(local_layer) = self.find_layer(layer_file_name).await else { return Ok(None) };
+        let Some(local_layer) = self.find_layer(layer_file_name) else { return Ok(None) };
         let remote_client = self
             .remote_client
             .as_ref()
@@ -1160,7 +1156,7 @@ impl Timeline {
         }
 
         // start the batch update
-        let mut layer_map = self.layers.write().await;
+        let mut layer_map = self.layers.write().unwrap();
         let mut batch_updates = layer_map.batch_update();
 
         let mut results = Vec::with_capacity(layers_to_evict.len());
@@ -1418,7 +1414,7 @@ impl Timeline {
                 timeline_id,
                 tenant_id,
                 pg_version,
-                layers: tokio::sync::RwLock::new(LayerMap::default()),
+                layers: RwLock::new(LayerMap::default()),
                 wanted_image_layers: Mutex::new(None),
 
                 walredo_mgr,
@@ -1453,7 +1449,7 @@ impl Timeline {
                 layer_flush_start_tx,
                 layer_flush_done_tx,
 
-                write_lock: tokio::sync::Mutex::new(()),
+                write_lock: Mutex::new(()),
                 layer_removal_cs: Default::default(),
 
                 gc_info: std::sync::RwLock::new(GcInfo {
@@ -1464,6 +1460,7 @@ impl Timeline {
 
                 latest_gc_cutoff_lsn: Rcu::new(metadata.latest_gc_cutoff_lsn()),
                 initdb_lsn: metadata.initdb_lsn(),
+                compute_spec_id: tokio::sync::Mutex::new(None),
 
                 current_logical_size: if disk_consistent_lsn.is_valid() {
                     // we're creating timeline data with some layer files existing locally,
@@ -1505,7 +1502,7 @@ impl Timeline {
         let mut flush_loop_state = self.flush_loop_state.lock().unwrap();
         match *flush_loop_state {
             FlushLoopState::NotStarted => (),
-            FlushLoopState::Running { .. } => {
+            FlushLoopState::Running => {
                 info!(
                     "skipping attempt to start flush_loop twice {}/{}",
                     self.tenant_id, self.timeline_id
@@ -1525,12 +1522,6 @@ impl Timeline {
         let self_clone = Arc::clone(self);
 
         info!("spawning flush loop");
-        *flush_loop_state = FlushLoopState::Running {
-            #[cfg(test)]
-            expect_initdb_optimization: false,
-            #[cfg(test)]
-            initdb_optimization_count: 0,
-        };
         task_mgr::spawn(
             task_mgr::BACKGROUND_RUNTIME.handle(),
             task_mgr::TaskKind::LayerFlushTask,
@@ -1542,12 +1533,14 @@ impl Timeline {
                 let background_ctx = RequestContext::todo_child(TaskKind::LayerFlushTask, DownloadBehavior::Error);
                 self_clone.flush_loop(layer_flush_start_rx, &background_ctx).await;
                 let mut flush_loop_state = self_clone.flush_loop_state.lock().unwrap();
-                assert!(matches!(*flush_loop_state, FlushLoopState::Running{ ..}));
+                assert_eq!(*flush_loop_state, FlushLoopState::Running);
                 *flush_loop_state  = FlushLoopState::Exited;
                 Ok(())
             }
             .instrument(info_span!(parent: None, "layer flush task", tenant = %self.tenant_id, timeline = %self.timeline_id))
         );
+
+        *flush_loop_state = FlushLoopState::Running;
     }
 
     /// Creates and starts the wal receiver.
@@ -1595,21 +1588,12 @@ impl Timeline {
         ));
     }
 
-    ///
-    /// Initialize with an empty layer map. Used when creating a new timeline.
-    ///
-    pub(super) fn init_empty_layer_map(&self, start_lsn: Lsn) {
-        let mut layers = self.layers.try_write().expect(
-            "in the context where we call this function, no other task has access to the object",
-        );
-        layers.next_open_layer_at = Some(Lsn(start_lsn.0));
-    }
-
     ///
     /// Scan the timeline directory to populate the layer map.
+    /// Returns all timeline-related files that were found and loaded.
     ///
-    pub(super) async fn load_layer_map(&self, disk_consistent_lsn: Lsn) -> anyhow::Result<()> {
-        let mut layers = self.layers.write().await;
+    pub(super) fn load_layer_map(&self, disk_consistent_lsn: Lsn) -> anyhow::Result<()> {
+        let mut layers = self.layers.write().unwrap();
         let mut updates = layers.batch_update();
         let mut num_layers = 0;
 
@@ -1738,7 +1722,7 @@ impl Timeline {
 
         // We're holding a layer map lock for a while but this
         // method is only called during init so it's fine.
-        let mut layer_map = self.layers.write().await;
+        let mut layer_map = self.layers.write().unwrap();
         let mut updates = layer_map.batch_update();
         for remote_layer_name in &index_part.timeline_layers {
             let local_layer = local_only_layers.remove(remote_layer_name);
@@ -1891,7 +1875,7 @@ impl Timeline {
         let local_layers = self
             .layers
             .read()
-            .await
+            .unwrap()
             .iter_historic_layers()
             .map(|l| (l.filename(), l))
             .collect::<HashMap<_, _>>();
@@ -2264,8 +2248,8 @@ impl Timeline {
         }
     }
 
-    async fn find_layer(&self, layer_file_name: &str) -> Option<Arc<dyn PersistentLayer>> {
-        for historic_layer in self.layers.read().await.iter_historic_layers() {
+    fn find_layer(&self, layer_file_name: &str) -> Option<Arc<dyn PersistentLayer>> {
+        for historic_layer in self.layers.read().unwrap().iter_historic_layers() {
             let historic_layer_name = historic_layer.filename().file_name();
             if layer_file_name == historic_layer_name {
                 return Some(historic_layer);
@@ -2388,7 +2372,7 @@ impl Timeline {
                 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);
+                        self.metrics.materialized_page_cache_hit_counter.inc_by(1);
                         return Ok(());
                     }
                     if prev_lsn <= cont_lsn {
@@ -2406,17 +2390,10 @@ impl Timeline {
                 }
                 ValueReconstructResult::Missing => {
                     return Err(layer_traversal_error(
-                        if cfg!(test) {
-                            format!(
-                                "could not find data for key {} at LSN {}, for request at LSN {}\n{}",
-                                key, cont_lsn, request_lsn, std::backtrace::Backtrace::force_capture(),
-                            )
-                        } else {
-                            format!(
-                                "could not find data for key {} at LSN {}, for request at LSN {}",
-                                key, cont_lsn, request_lsn
-                            )
-                        },
+                        format!(
+                            "could not find data for key {} at LSN {}, for request at LSN {}",
+                            key, cont_lsn, request_lsn
+                        ),
                         traversal_path,
                     ));
                 }
@@ -2482,7 +2459,7 @@ impl Timeline {
             #[allow(clippy::never_loop)] // see comment at bottom of this loop
             'layer_map_search: loop {
                 let remote_layer = {
-                    let layers = timeline.layers.read().await;
+                    let layers = timeline.layers.read().unwrap();
 
                     // Check the open and frozen in-memory layers first, in order from newest
                     // to oldest.
@@ -2664,29 +2641,24 @@ impl Timeline {
     ///
     /// Get a handle to the latest layer for appending.
     ///
-    async fn get_layer_for_write(&self, lsn: Lsn) -> anyhow::Result<Arc<InMemoryLayer>> {
-        let mut layers = self.layers.write().await;
+    fn get_layer_for_write(&self, lsn: Lsn) -> anyhow::Result<Arc<InMemoryLayer>> {
+        let mut layers = self.layers.write().unwrap();
 
         ensure!(lsn.is_aligned());
 
         let last_record_lsn = self.get_last_record_lsn();
         ensure!(
             lsn > last_record_lsn,
-            "cannot modify relation after advancing last_record_lsn (incoming_lsn={}, last_record_lsn={})\n{}",
+            "cannot modify relation after advancing last_record_lsn (incoming_lsn={}, last_record_lsn={})",
             lsn,
             last_record_lsn,
-            std::backtrace::Backtrace::force_capture(),
         );
 
         // Do we have a layer open for writing already?
         let layer;
         if let Some(open_layer) = &layers.open_layer {
             if open_layer.get_lsn_range().start > lsn {
-                bail!(
-                    "unexpected open layer in the future: open layers starts at {}, write lsn {}",
-                    open_layer.get_lsn_range().start,
-                    lsn
-                );
+                bail!("unexpected open layer in the future");
             }
 
             layer = Arc::clone(open_layer);
@@ -2714,16 +2686,17 @@ impl Timeline {
         Ok(layer)
     }
 
-    async fn put_value(&self, key: Key, lsn: Lsn, val: &Value) -> anyhow::Result<()> {
+    fn put_value(&self, key: Key, lsn: Lsn, val: &Value) -> anyhow::Result<()> {
         //info!("PUT: key {} at {}", key, lsn);
-        let layer = self.get_layer_for_write(lsn).await?;
+        let layer = self.get_layer_for_write(lsn)?;
         layer.put_value(key, lsn, val)?;
         Ok(())
     }
 
-    async fn put_tombstone(&self, key_range: Range<Key>, lsn: Lsn) -> anyhow::Result<()> {
-        let layer = self.get_layer_for_write(lsn).await?;
-        layer.put_tombstone(key_range, lsn).await?;
+    fn put_tombstone(&self, key_range: Range<Key>, lsn: Lsn) -> anyhow::Result<()> {
+        let layer = self.get_layer_for_write(lsn)?;
+        layer.put_tombstone(key_range, lsn)?;
+
         Ok(())
     }
 
@@ -2734,15 +2707,15 @@ impl Timeline {
         self.last_record_lsn.advance(new_lsn);
     }
 
-    async fn freeze_inmem_layer(&self, write_lock_held: bool) {
+    fn freeze_inmem_layer(&self, write_lock_held: bool) {
         // Freeze the current open in-memory layer. It will be written to disk on next
         // iteration.
         let _write_guard = if write_lock_held {
             None
         } else {
-            Some(self.write_lock.lock().await)
+            Some(self.write_lock.lock().unwrap())
         };
-        let mut layers = self.layers.write().await;
+        let mut layers = self.layers.write().unwrap();
         if let Some(open_layer) = &layers.open_layer {
             let open_layer_rc = Arc::clone(open_layer);
             // Does this layer need freezing?
@@ -2780,7 +2753,7 @@ impl Timeline {
             let flush_counter = *layer_flush_start_rx.borrow();
             let result = loop {
                 let layer_to_flush = {
-                    let layers = self.layers.read().await;
+                    let layers = self.layers.read().unwrap();
                     layers.frozen_layers.front().cloned()
                     // drop 'layers' lock to allow concurrent reads and writes
                 };
@@ -2813,7 +2786,7 @@ impl Timeline {
         let mut my_flush_request = 0;
 
         let flush_loop_state = { *self.flush_loop_state.lock().unwrap() };
-        if !matches!(flush_loop_state, FlushLoopState::Running { .. }) {
+        if flush_loop_state != FlushLoopState::Running {
             anyhow::bail!("cannot flush frozen layers when flush_loop is not running, state is {flush_loop_state:?}")
         }
 
@@ -2863,18 +2836,6 @@ impl Timeline {
         let lsn_range = frozen_layer.get_lsn_range();
         let layer_paths_to_upload =
             if lsn_range.start == self.initdb_lsn && lsn_range.end == Lsn(self.initdb_lsn.0 + 1) {
-                #[cfg(test)]
-                match &mut *self.flush_loop_state.lock().unwrap() {
-                    FlushLoopState::NotStarted | FlushLoopState::Exited => {
-                        panic!("flush loop not running")
-                    }
-                    FlushLoopState::Running {
-                        initdb_optimization_count,
-                        ..
-                    } => {
-                        *initdb_optimization_count += 1;
-                    }
-                }
                 // Note: The 'ctx' in use here has DownloadBehavior::Error. We should not
                 // require downloading anything during initial import.
                 let (partitioning, _lsn) = self
@@ -2883,20 +2844,17 @@ impl Timeline {
                 self.create_image_layers(&partitioning, self.initdb_lsn, true, ctx)
                     .await?
             } else {
-                #[cfg(test)]
-                match &mut *self.flush_loop_state.lock().unwrap() {
-                    FlushLoopState::NotStarted | FlushLoopState::Exited => {
-                        panic!("flush loop not running")
-                    }
-                    FlushLoopState::Running {
-                        expect_initdb_optimization,
-                        ..
-                    } => {
-                        assert!(!*expect_initdb_optimization, "expected initdb optimization");
-                    }
-                }
                 // normal case, write out a L0 delta layer file.
-                let (delta_path, metadata) = self.create_delta_layer(&frozen_layer).await?;
+                let this = self.clone();
+                let frozen_layer = frozen_layer.clone();
+                let span = tracing::info_span!("blocking");
+                let (delta_path, metadata) = tokio::task::spawn_blocking(move || {
+                    let _g = span.entered();
+                    this.create_delta_layer(&frozen_layer)
+                })
+                .await
+                .context("create_delta_layer spawn_blocking")
+                .and_then(|res| res)?;
                 HashMap::from([(delta_path, metadata)])
             };
 
@@ -2905,7 +2863,7 @@ impl Timeline {
         // The new on-disk layers are now in the layer map. We can remove the
         // in-memory layer from the map now.
         {
-            let mut layers = self.layers.write().await;
+            let mut layers = self.layers.write().unwrap();
             let l = layers.frozen_layers.pop_front();
 
             // Only one thread may call this function at a time (for this
@@ -2999,52 +2957,34 @@ impl Timeline {
     }
 
     // Write out the given frozen in-memory layer as a new L0 delta file
-    async fn create_delta_layer(
+    fn create_delta_layer(
         self: &Arc<Self>,
-        frozen_layer: &Arc<InMemoryLayer>,
+        frozen_layer: &InMemoryLayer,
     ) -> anyhow::Result<(LayerFileName, LayerFileMetadata)> {
-        let span = tracing::info_span!("blocking");
-        let new_delta: DeltaLayer = tokio::task::spawn_blocking({
-            let _g = span.entered();
-            let self_clone = Arc::clone(self);
-            let frozen_layer = Arc::clone(frozen_layer);
-            move || {
-                // Write it out
-                let new_delta = frozen_layer.write_to_disk()?;
-                let new_delta_path = new_delta.path();
+        // Write it out
+        let new_delta = frozen_layer.write_to_disk()?;
+        let new_delta_path = new_delta.path();
+        let new_delta_filename = new_delta.filename();
 
-                // Sync it to disk.
-                //
-                // We must also fsync the timeline dir to ensure the directory entries for
-                // new layer files are durable.
-                //
-                // NB: timeline dir must be synced _after_ the file contents are durable.
-                // So, two separate fsyncs are required, they mustn't be batched.
-                //
-                // TODO: If we're running inside 'flush_frozen_layers' and there are multiple
-                // files to flush, the fsync overhead can be reduces as follows:
-                // 1. write them all to temporary file names
-                // 2. fsync them
-                // 3. rename to the final name
-                // 4. fsync the parent directory.
-                // Note that (1),(2),(3) today happen inside write_to_disk().
-                par_fsync::par_fsync(&[new_delta_path]).context("fsync of delta layer")?;
-                par_fsync::par_fsync(&[self_clone
-                    .conf
-                    .timeline_path(&self_clone.timeline_id, &self_clone.tenant_id)])
-                .context("fsync of timeline dir")?;
+        // Sync it to disk.
+        //
+        // We must also fsync the timeline dir to ensure the directory entries for
+        // new layer files are durable
+        //
+        // TODO: If we're running inside 'flush_frozen_layers' and there are multiple
+        // files to flush, it might be better to first write them all, and then fsync
+        // them all in parallel.
 
-                anyhow::Ok(new_delta)
-            }
-        })
-        .await
-        .context("spawn_blocking")??;
-        let new_delta_name = new_delta.filename();
-        let sz = new_delta.desc.file_size;
+        // First sync the delta layer. We still use par_fsync here to keep everything consistent. Feel free to replace
+        // this with a single fsync in future refactors.
+        par_fsync::par_fsync(&[new_delta_path.clone()]).context("fsync of delta layer")?;
+        // Then sync the parent directory.
+        par_fsync::par_fsync(&[self.conf.timeline_path(&self.timeline_id, &self.tenant_id)])
+            .context("fsync of timeline dir")?;
 
         // Add it to the layer map
         let l = Arc::new(new_delta);
-        let mut layers = self.layers.write().await;
+        let mut layers = self.layers.write().unwrap();
         let mut batch_updates = layers.batch_update();
         l.access_stats().record_residence_event(
             &batch_updates,
@@ -3054,12 +2994,15 @@ impl Timeline {
         batch_updates.insert_historic(l.layer_desc().clone(), l);
         batch_updates.flush();
 
-        // update metrics
+        // update the timeline's physical size
+        let sz = new_delta_path.metadata()?.len();
+
         self.metrics.resident_physical_size_gauge.add(sz);
+        // update metrics
         self.metrics.num_persistent_files_created.inc_by(1);
         self.metrics.persistent_bytes_written.inc_by(sz);
 
-        Ok((new_delta_name, LayerFileMetadata::new(sz)))
+        Ok((new_delta_filename, LayerFileMetadata::new(sz)))
     }
 
     async fn repartition(
@@ -3093,14 +3036,10 @@ impl Timeline {
     }
 
     // Is it time to create a new image layer for the given partition?
-    async fn time_for_new_image_layer(
-        &self,
-        partition: &KeySpace,
-        lsn: Lsn,
-    ) -> anyhow::Result<bool> {
+    fn time_for_new_image_layer(&self, partition: &KeySpace, lsn: Lsn) -> anyhow::Result<bool> {
         let threshold = self.get_image_creation_threshold();
 
-        let layers = self.layers.read().await;
+        let layers = self.layers.read().unwrap();
 
         let mut max_deltas = 0;
         {
@@ -3195,7 +3134,7 @@ impl Timeline {
         for partition in partitioning.parts.iter() {
             let img_range = start..partition.ranges.last().unwrap().end;
             start = img_range.end;
-            if force || self.time_for_new_image_layer(partition, lsn).await? {
+            if force || self.time_for_new_image_layer(partition, lsn)? {
                 let mut image_layer_writer = ImageLayerWriter::new(
                     self.conf,
                     self.timeline_id,
@@ -3278,7 +3217,7 @@ impl Timeline {
 
         let mut layer_paths_to_upload = HashMap::with_capacity(image_layers.len());
 
-        let mut layers = self.layers.write().await;
+        let mut layers = self.layers.write().unwrap();
         let mut updates = layers.batch_update();
         let timeline_path = self.conf.timeline_path(&self.timeline_id, &self.tenant_id);
 
@@ -3334,159 +3273,21 @@ impl From<anyhow::Error> for CompactionError {
     }
 }
 
-#[serde_as]
-#[derive(serde::Serialize)]
-struct RecordedDuration(#[serde_as(as = "serde_with::DurationMicroSeconds")] Duration);
-
-#[derive(Default)]
-enum DurationRecorder {
-    #[default]
-    NotStarted,
-    Recorded(RecordedDuration, tokio::time::Instant),
-}
-
-impl DurationRecorder {
-    pub fn till_now(&self) -> DurationRecorder {
-        match self {
-            DurationRecorder::NotStarted => {
-                panic!("must only call on recorded measurements")
-            }
-            DurationRecorder::Recorded(_, ended) => {
-                let now = tokio::time::Instant::now();
-                DurationRecorder::Recorded(RecordedDuration(now - *ended), now)
-            }
-        }
-    }
-    pub fn into_recorded(self) -> Option<RecordedDuration> {
-        match self {
-            DurationRecorder::NotStarted => None,
-            DurationRecorder::Recorded(recorded, _) => Some(recorded),
-        }
-    }
-}
-
-#[derive(Default)]
-struct CompactLevel0Phase1StatsBuilder {
-    version: Option<u64>,
-    tenant_id: Option<TenantId>,
-    timeline_id: Option<TimelineId>,
-    first_read_lock_acquisition_micros: DurationRecorder,
-    get_level0_deltas_plus_drop_lock_micros: DurationRecorder,
-    level0_deltas_count: Option<usize>,
-    time_spent_between_locks: DurationRecorder,
-    second_read_lock_acquisition_micros: DurationRecorder,
-    second_read_lock_held_micros: DurationRecorder,
-    sort_holes_micros: DurationRecorder,
-    write_layer_files_micros: DurationRecorder,
-    new_deltas_count: Option<usize>,
-    new_deltas_size: Option<u64>,
-}
-
-#[serde_as]
-#[derive(serde::Serialize)]
-struct CompactLevel0Phase1Stats {
-    version: u64,
-    #[serde_as(as = "serde_with::DisplayFromStr")]
-    tenant_id: TenantId,
-    #[serde_as(as = "serde_with::DisplayFromStr")]
-    timeline_id: TimelineId,
-    first_read_lock_acquisition_micros: RecordedDuration,
-    get_level0_deltas_plus_drop_lock_micros: RecordedDuration,
-    level0_deltas_count: usize,
-    time_spent_between_locks: RecordedDuration,
-    second_read_lock_acquisition_micros: RecordedDuration,
-    second_read_lock_held_micros: RecordedDuration,
-    sort_holes_micros: RecordedDuration,
-    write_layer_files_micros: RecordedDuration,
-    new_deltas_count: usize,
-    new_deltas_size: u64,
-}
-
-impl TryFrom<CompactLevel0Phase1StatsBuilder> for CompactLevel0Phase1Stats {
-    type Error = anyhow::Error;
-
-    fn try_from(value: CompactLevel0Phase1StatsBuilder) -> Result<Self, Self::Error> {
-        let CompactLevel0Phase1StatsBuilder {
-            version,
-            tenant_id,
-            timeline_id,
-            first_read_lock_acquisition_micros,
-            get_level0_deltas_plus_drop_lock_micros,
-            level0_deltas_count,
-            time_spent_between_locks,
-            second_read_lock_acquisition_micros,
-            second_read_lock_held_micros,
-            sort_holes_micros,
-            write_layer_files_micros,
-            new_deltas_count,
-            new_deltas_size,
-        } = value;
-        Ok(CompactLevel0Phase1Stats {
-            version: version.ok_or_else(|| anyhow::anyhow!("version not set"))?,
-            tenant_id: tenant_id.ok_or_else(|| anyhow::anyhow!("tenant_id not set"))?,
-            timeline_id: timeline_id.ok_or_else(|| anyhow::anyhow!("timeline_id not set"))?,
-            first_read_lock_acquisition_micros: first_read_lock_acquisition_micros
-                .into_recorded()
-                .ok_or_else(|| anyhow::anyhow!("first_read_lock_acquisition_micros not set"))?,
-            get_level0_deltas_plus_drop_lock_micros: get_level0_deltas_plus_drop_lock_micros
-                .into_recorded()
-                .ok_or_else(|| {
-                    anyhow::anyhow!("get_level0_deltas_plus_drop_lock_micros not set")
-                })?,
-            level0_deltas_count: level0_deltas_count
-                .ok_or_else(|| anyhow::anyhow!("level0_deltas_count not set"))?,
-            time_spent_between_locks: time_spent_between_locks
-                .into_recorded()
-                .ok_or_else(|| anyhow::anyhow!("time_spent_between_locks not set"))?,
-            second_read_lock_acquisition_micros: second_read_lock_acquisition_micros
-                .into_recorded()
-                .ok_or_else(|| anyhow::anyhow!("second_read_lock_acquisition_micros not set"))?,
-            second_read_lock_held_micros: second_read_lock_held_micros
-                .into_recorded()
-                .ok_or_else(|| anyhow::anyhow!("second_read_lock_held_micros not set"))?,
-            sort_holes_micros: sort_holes_micros
-                .into_recorded()
-                .ok_or_else(|| anyhow::anyhow!("sort_holes_micros not set"))?,
-            write_layer_files_micros: write_layer_files_micros
-                .into_recorded()
-                .ok_or_else(|| anyhow::anyhow!("write_layer_files_micros not set"))?,
-            new_deltas_count: new_deltas_count
-                .ok_or_else(|| anyhow::anyhow!("new_deltas_count not set"))?,
-            new_deltas_size: new_deltas_size
-                .ok_or_else(|| anyhow::anyhow!("new_deltas_size not set"))?,
-        })
-    }
-}
-
 impl Timeline {
     /// Level0 files first phase of compaction, explained in the [`compact_inner`] comment.
     ///
     /// This method takes the `_layer_removal_cs` guard to highlight it required downloads are
     /// returned as an error. If the `layer_removal_cs` boundary is changed not to be taken in the
     /// start of level0 files compaction, the on-demand download should be revisited as well.
-    async fn compact_level0_phase1(
+    fn compact_level0_phase1(
         &self,
         _layer_removal_cs: Arc<tokio::sync::OwnedMutexGuard<()>>,
         target_file_size: u64,
         ctx: &RequestContext,
     ) -> Result<CompactLevel0Phase1Result, CompactionError> {
-        let mut stats = CompactLevel0Phase1StatsBuilder {
-            version: Some(1),
-            tenant_id: Some(self.tenant_id),
-            timeline_id: Some(self.timeline_id),
-            ..Default::default()
-        };
-
-        let begin = tokio::time::Instant::now();
-        let layers = self.layers.read().await;
-        let now = tokio::time::Instant::now();
-        stats.first_read_lock_acquisition_micros =
-            DurationRecorder::Recorded(RecordedDuration(now - begin), now);
+        let layers = self.layers.read().unwrap();
         let mut level0_deltas = layers.get_level0_deltas()?;
         drop(layers);
-        stats.level0_deltas_count = Some(level0_deltas.len());
-        stats.get_level0_deltas_plus_drop_lock_micros =
-            stats.first_read_lock_acquisition_micros.till_now();
 
         // Only compact if enough layers have accumulated.
         let threshold = self.get_compaction_threshold();
@@ -3607,9 +3408,7 @@ impl Timeline {
         // Determine N largest holes where N is number of compacted layers.
         let max_holes = deltas_to_compact.len();
         let last_record_lsn = self.get_last_record_lsn();
-        stats.time_spent_between_locks = stats.get_level0_deltas_plus_drop_lock_micros.till_now();
-        let layers = self.layers.read().await; // Is'n it better to hold original layers lock till here?
-        stats.second_read_lock_acquisition_micros = stats.time_spent_between_locks.till_now();
+        let layers = self.layers.read().unwrap(); // Is'n it better to hold original layers lock till here?
         let min_hole_range = (target_file_size / page_cache::PAGE_SZ as u64) as i128;
         let min_hole_coverage_size = 3; // TODO: something more flexible?
 
@@ -3643,11 +3442,9 @@ impl Timeline {
             prev = Some(next_key.next());
         }
         drop(layers);
-        stats.second_read_lock_held_micros = stats.second_read_lock_acquisition_micros.till_now();
         let mut holes = heap.into_vec();
         holes.sort_unstable_by_key(|hole| hole.key_range.start);
         let mut next_hole = 0; // index of next hole in holes vector
-        stats.sort_holes_micros = stats.second_read_lock_held_micros.till_now();
 
         // Merge the contents of all the input delta layers into a new set
         // of delta layers, based on the current partitioning.
@@ -3807,26 +3604,8 @@ impl Timeline {
             layer_paths.pop().unwrap();
         }
 
-        stats.write_layer_files_micros = stats.sort_holes_micros.till_now();
-        stats.new_deltas_count = Some(new_layers.len());
-        stats.new_deltas_size = Some(new_layers.iter().map(|l| l.desc.file_size).sum());
-
         drop(all_keys_iter); // So that deltas_to_compact is no longer borrowed
 
-        match TryInto::<CompactLevel0Phase1Stats>::try_into(stats)
-            .and_then(|stats| serde_json::to_string(&stats).context("serde_json::to_string"))
-        {
-            Ok(stats_json) => {
-                info!(
-                    stats_json = stats_json.as_str(),
-                    "compact_level0_phase1 stats available"
-                )
-            }
-            Err(e) => {
-                warn!("compact_level0_phase1 stats failed to serialize: {:#}", e);
-            }
-        }
-
         Ok(CompactLevel0Phase1Result {
             new_layers,
             deltas_to_compact,
@@ -3843,12 +3622,21 @@ impl Timeline {
         target_file_size: u64,
         ctx: &RequestContext,
     ) -> Result<(), CompactionError> {
+        let this = self.clone();
+        let ctx_inner = ctx.clone();
+        let layer_removal_cs_inner = layer_removal_cs.clone();
+        let span = tracing::info_span!("blocking");
         let CompactLevel0Phase1Result {
             new_layers,
             deltas_to_compact,
-        } = self
-            .compact_level0_phase1(layer_removal_cs.clone(), target_file_size, ctx)
-            .await?;
+        } = tokio::task::spawn_blocking(move || {
+            let _g = span.entered();
+            this.compact_level0_phase1(layer_removal_cs_inner, target_file_size, &ctx_inner)
+        })
+        .await
+        .context("compact_level0_phase1 spawn_blocking")
+        .map_err(CompactionError::Other)
+        .and_then(|res| res)?;
 
         if new_layers.is_empty() && deltas_to_compact.is_empty() {
             // nothing to do
@@ -3866,7 +3654,7 @@ impl Timeline {
                 .context("wait for layer upload ops to complete")?;
         }
 
-        let mut layers = self.layers.write().await;
+        let mut layers = self.layers.write().unwrap();
         let mut updates = layers.batch_update();
         let mut new_layer_paths = HashMap::with_capacity(new_layers.len());
         for l in new_layers {
@@ -3953,7 +3741,6 @@ impl Timeline {
     /// for example. The caller should hold `Tenant::gc_cs` lock to ensure
     /// that.
     ///
-    #[instrument(skip_all, fields(timline_id=%self.timeline_id))]
     pub(super) async fn update_gc_info(
         &self,
         retain_lsns: Vec<Lsn>,
@@ -4126,7 +3913,7 @@ impl Timeline {
         // 4. newer on-disk image layers cover the layer's whole key range
         //
         // TODO holding a write lock is too agressive and avoidable
-        let mut layers = self.layers.write().await;
+        let mut layers = self.layers.write().unwrap();
         'outer: for l in layers.iter_historic_layers() {
             result.layers_total += 1;
 
@@ -4426,7 +4213,7 @@ impl Timeline {
 
                     // Download complete. Replace the RemoteLayer with the corresponding
                     // Delta- or ImageLayer in the layer map.
-                    let mut layers = self_clone.layers.write().await;
+                    let mut layers = self_clone.layers.write().unwrap();
                     let mut updates = layers.batch_update();
                     let new_layer = remote_layer.create_downloaded_layer(&updates, self_clone.conf, *size);
                     {
@@ -4584,7 +4371,7 @@ impl Timeline {
     ) {
         let mut downloads = Vec::new();
         {
-            let layers = self.layers.read().await;
+            let layers = self.layers.read().unwrap();
             layers
                 .iter_historic_layers()
                 .filter_map(|l| l.downcast_remote_layer())
@@ -4686,8 +4473,8 @@ impl LocalLayerInfoForDiskUsageEviction {
 }
 
 impl Timeline {
-    pub(crate) async fn get_local_layers_for_disk_usage_eviction(&self) -> DiskUsageEvictionInfo {
-        let layers = self.layers.read().await;
+    pub(crate) fn get_local_layers_for_disk_usage_eviction(&self) -> DiskUsageEvictionInfo {
+        let layers = self.layers.read().unwrap();
 
         let mut max_layer_size: Option<u64> = None;
         let mut resident_layers = Vec::new();
@@ -4759,7 +4546,7 @@ fn layer_traversal_error(msg: String, path: Vec<TraversalPathItem>) -> PageRecon
 // but will cause large code changes.
 pub struct TimelineWriter<'a> {
     tl: &'a Timeline,
-    _write_guard: tokio::sync::MutexGuard<'a, ()>,
+    _write_guard: MutexGuard<'a, ()>,
 }
 
 impl Deref for TimelineWriter<'_> {
@@ -4775,12 +4562,12 @@ impl<'a> TimelineWriter<'a> {
     ///
     /// This will implicitly extend the relation, if the page is beyond the
     /// current end-of-file.
-    pub async fn put(&self, key: Key, lsn: Lsn, value: &Value) -> anyhow::Result<()> {
-        self.tl.put_value(key, lsn, value).await
+    pub fn put(&self, key: Key, lsn: Lsn, value: &Value) -> anyhow::Result<()> {
+        self.tl.put_value(key, lsn, value)
     }
 
-    pub async fn delete(&self, key_range: Range<Key>, lsn: Lsn) -> anyhow::Result<()> {
-        self.tl.put_tombstone(key_range, lsn).await
+    pub fn delete(&self, key_range: Range<Key>, lsn: Lsn) -> anyhow::Result<()> {
+        self.tl.put_tombstone(key_range, lsn)
     }
 
     /// Track the end of the latest digested WAL record.
@@ -4800,14 +4587,6 @@ impl<'a> TimelineWriter<'a> {
     }
 }
 
-// We need TimelineWriter to be send in upcoming conversion of
-// Timeline::layers to tokio::sync::RwLock.
-#[test]
-fn is_send() {
-    fn _assert_send<T: Send>() {}
-    _assert_send::<TimelineWriter<'_>>();
-}
-
 /// Add a suffix to a layer file's name: .{num}.old
 /// Uses the first available num (starts at 0)
 fn rename_to_backup(path: &Path) -> anyhow::Result<()> {

pageserver/src/tenant/timeline/eviction_task.rs

@@ -197,7 +197,7 @@ impl Timeline {
         // We don't want to hold the layer map lock during eviction.
         // So, we just need to deal with this.
         let candidates: Vec<Arc<dyn PersistentLayer>> = {
-            let layers = self.layers.read().await;
+            let layers = self.layers.read().unwrap();
             let mut candidates = Vec::new();
             for hist_layer in layers.iter_historic_layers() {
                 if hist_layer.is_remote_layer() {

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

@@ -1324,8 +1324,7 @@ mod tests {
     async fn dummy_state(harness: &TenantHarness<'_>) -> ConnectionManagerState {
         let (tenant, ctx) = harness.load().await;
         let timeline = tenant
-            .create_test_timeline(TIMELINE_ID, Lsn(0x8), crate::DEFAULT_PG_VERSION, &ctx)
-            .await
+            .create_test_timeline(TIMELINE_ID, Lsn(0), crate::DEFAULT_PG_VERSION, &ctx)
             .expect("Failed to create an empty timeline for dummy wal connection manager");
 
         ConnectionManagerState {

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

@@ -304,15 +304,12 @@ pub(super) async fn handle_walreceiver_connection(
             }
         }
 
-        timeline
-            .check_checkpoint_distance()
-            .await
-            .with_context(|| {
-                format!(
-                    "Failed to check checkpoint distance for timeline {}",
-                    timeline.timeline_id
-                )
-            })?;
+        timeline.check_checkpoint_distance().with_context(|| {
+            format!(
+                "Failed to check checkpoint distance for timeline {}",
+                timeline.timeline_id
+            )
+        })?;
 
         if let Some(last_lsn) = status_update {
             let timeline_remote_consistent_lsn =

pageserver/src/walingest.rs

@@ -25,7 +25,7 @@ use postgres_ffi::v14::nonrelfile_utils::clogpage_precedes;
 use postgres_ffi::v14::nonrelfile_utils::slru_may_delete_clogsegment;
 use postgres_ffi::{fsm_logical_to_physical, page_is_new, page_set_lsn};
 
-use anyhow::{Context, Result};
+use anyhow::Result;
 use bytes::{Buf, Bytes, BytesMut};
 use tracing::*;
 
@@ -333,7 +333,7 @@ impl<'a> WalIngest<'a> {
 
         // Now that this record has been fully handled, including updating the
         // checkpoint data, let the repository know that it is up-to-date to this LSN
-        modification.commit().await?;
+        modification.commit()?;
 
         Ok(())
     }
@@ -1082,10 +1082,7 @@ impl<'a> WalIngest<'a> {
             .await?
         {
             // create it with 0 size initially, the logic below will extend it
-            modification
-                .put_rel_creation(rel, 0, ctx)
-                .await
-                .context("Relation Error")?;
+            modification.put_rel_creation(rel, 0, ctx).await?;
             0
         } else {
             self.timeline.get_rel_size(rel, last_lsn, true, ctx).await?
@@ -1174,6 +1171,7 @@ impl<'a> WalIngest<'a> {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use crate::pgdatadir_mapping::create_test_timeline;
     use crate::tenant::harness::*;
     use crate::tenant::Timeline;
     use postgres_ffi::v14::xlog_utils::SIZEOF_CHECKPOINT;
@@ -1202,7 +1200,7 @@ mod tests {
         let mut m = tline.begin_modification(Lsn(0x10));
         m.put_checkpoint(ZERO_CHECKPOINT.clone())?;
         m.put_relmap_file(0, 111, Bytes::from(""), ctx).await?; // dummy relmapper file
-        m.commit().await?;
+        m.commit()?;
         let walingest = WalIngest::new(tline, Lsn(0x10), ctx).await?;
 
         Ok(walingest)
@@ -1211,9 +1209,7 @@ mod tests {
     #[tokio::test]
     async fn test_relsize() -> Result<()> {
         let (tenant, ctx) = TenantHarness::create("test_relsize")?.load().await;
-        let tline = tenant
-            .create_test_timeline(TIMELINE_ID, Lsn(8), DEFAULT_PG_VERSION, &ctx)
-            .await?;
+        let tline = create_test_timeline(&tenant, TIMELINE_ID, DEFAULT_PG_VERSION, &ctx)?;
         let mut walingest = init_walingest_test(&tline, &ctx).await?;
 
         let mut m = tline.begin_modification(Lsn(0x20));
@@ -1221,22 +1217,22 @@ mod tests {
         walingest
             .put_rel_page_image(&mut m, TESTREL_A, 0, TEST_IMG("foo blk 0 at 2"), &ctx)
             .await?;
-        m.commit().await?;
+        m.commit()?;
         let mut m = tline.begin_modification(Lsn(0x30));
         walingest
             .put_rel_page_image(&mut m, TESTREL_A, 0, TEST_IMG("foo blk 0 at 3"), &ctx)
             .await?;
-        m.commit().await?;
+        m.commit()?;
         let mut m = tline.begin_modification(Lsn(0x40));
         walingest
             .put_rel_page_image(&mut m, TESTREL_A, 1, TEST_IMG("foo blk 1 at 4"), &ctx)
             .await?;
-        m.commit().await?;
+        m.commit()?;
         let mut m = tline.begin_modification(Lsn(0x50));
         walingest
             .put_rel_page_image(&mut m, TESTREL_A, 2, TEST_IMG("foo blk 2 at 5"), &ctx)
             .await?;
-        m.commit().await?;
+        m.commit()?;
 
         assert_current_logical_size(&tline, Lsn(0x50));
 
@@ -1322,7 +1318,7 @@ mod tests {
         walingest
             .put_rel_truncation(&mut m, TESTREL_A, 2, &ctx)
             .await?;
-        m.commit().await?;
+        m.commit()?;
         assert_current_logical_size(&tline, Lsn(0x60));
 
         // Check reported size and contents after truncation
@@ -1364,7 +1360,7 @@ mod tests {
         walingest
             .put_rel_truncation(&mut m, TESTREL_A, 0, &ctx)
             .await?;
-        m.commit().await?;
+        m.commit()?;
         assert_eq!(
             tline
                 .get_rel_size(TESTREL_A, Lsn(0x68), false, &ctx)
@@ -1377,7 +1373,7 @@ mod tests {
         walingest
             .put_rel_page_image(&mut m, TESTREL_A, 1, TEST_IMG("foo blk 1"), &ctx)
             .await?;
-        m.commit().await?;
+        m.commit()?;
         assert_eq!(
             tline
                 .get_rel_size(TESTREL_A, Lsn(0x70), false, &ctx)
@@ -1402,7 +1398,7 @@ mod tests {
         walingest
             .put_rel_page_image(&mut m, TESTREL_A, 1500, TEST_IMG("foo blk 1500"), &ctx)
             .await?;
-        m.commit().await?;
+        m.commit()?;
         assert_eq!(
             tline
                 .get_rel_size(TESTREL_A, Lsn(0x80), false, &ctx)
@@ -1432,16 +1428,14 @@ mod tests {
     #[tokio::test]
     async fn test_drop_extend() -> Result<()> {
         let (tenant, ctx) = TenantHarness::create("test_drop_extend")?.load().await;
-        let tline = tenant
-            .create_test_timeline(TIMELINE_ID, Lsn(8), DEFAULT_PG_VERSION, &ctx)
-            .await?;
+        let tline = create_test_timeline(&tenant, TIMELINE_ID, DEFAULT_PG_VERSION, &ctx)?;
         let mut walingest = init_walingest_test(&tline, &ctx).await?;
 
         let mut m = tline.begin_modification(Lsn(0x20));
         walingest
             .put_rel_page_image(&mut m, TESTREL_A, 0, TEST_IMG("foo blk 0 at 2"), &ctx)
             .await?;
-        m.commit().await?;
+        m.commit()?;
 
         // Check that rel exists and size is correct
         assert_eq!(
@@ -1460,7 +1454,7 @@ mod tests {
         // Drop rel
         let mut m = tline.begin_modification(Lsn(0x30));
         walingest.put_rel_drop(&mut m, TESTREL_A, &ctx).await?;
-        m.commit().await?;
+        m.commit()?;
 
         // Check that rel is not visible anymore
         assert_eq!(
@@ -1478,7 +1472,7 @@ mod tests {
         walingest
             .put_rel_page_image(&mut m, TESTREL_A, 0, TEST_IMG("foo blk 0 at 4"), &ctx)
             .await?;
-        m.commit().await?;
+        m.commit()?;
 
         // Check that rel exists and size is correct
         assert_eq!(
@@ -1503,9 +1497,7 @@ mod tests {
     #[tokio::test]
     async fn test_truncate_extend() -> Result<()> {
         let (tenant, ctx) = TenantHarness::create("test_truncate_extend")?.load().await;
-        let tline = tenant
-            .create_test_timeline(TIMELINE_ID, Lsn(8), DEFAULT_PG_VERSION, &ctx)
-            .await?;
+        let tline = create_test_timeline(&tenant, TIMELINE_ID, DEFAULT_PG_VERSION, &ctx)?;
         let mut walingest = init_walingest_test(&tline, &ctx).await?;
 
         // Create a 20 MB relation (the size is arbitrary)
@@ -1517,7 +1509,7 @@ mod tests {
                 .put_rel_page_image(&mut m, TESTREL_A, blkno, TEST_IMG(&data), &ctx)
                 .await?;
         }
-        m.commit().await?;
+        m.commit()?;
 
         // The relation was created at LSN 20, not visible at LSN 1 yet.
         assert_eq!(
@@ -1562,7 +1554,7 @@ mod tests {
         walingest
             .put_rel_truncation(&mut m, TESTREL_A, 1, &ctx)
             .await?;
-        m.commit().await?;
+        m.commit()?;
 
         // Check reported size and contents after truncation
         assert_eq!(
@@ -1611,7 +1603,7 @@ mod tests {
                 .put_rel_page_image(&mut m, TESTREL_A, blkno, TEST_IMG(&data), &ctx)
                 .await?;
         }
-        m.commit().await?;
+        m.commit()?;
 
         assert_eq!(
             tline
@@ -1645,9 +1637,7 @@ mod tests {
     #[tokio::test]
     async fn test_large_rel() -> Result<()> {
         let (tenant, ctx) = TenantHarness::create("test_large_rel")?.load().await;
-        let tline = tenant
-            .create_test_timeline(TIMELINE_ID, Lsn(8), DEFAULT_PG_VERSION, &ctx)
-            .await?;
+        let tline = create_test_timeline(&tenant, TIMELINE_ID, DEFAULT_PG_VERSION, &ctx)?;
         let mut walingest = init_walingest_test(&tline, &ctx).await?;
 
         let mut lsn = 0x10;
@@ -1658,7 +1648,7 @@ mod tests {
             walingest
                 .put_rel_page_image(&mut m, TESTREL_A, blknum as BlockNumber, img, &ctx)
                 .await?;
-            m.commit().await?;
+            m.commit()?;
         }
 
         assert_current_logical_size(&tline, Lsn(lsn));
@@ -1674,7 +1664,7 @@ mod tests {
         walingest
             .put_rel_truncation(&mut m, TESTREL_A, RELSEG_SIZE, &ctx)
             .await?;
-        m.commit().await?;
+        m.commit()?;
         assert_eq!(
             tline.get_rel_size(TESTREL_A, Lsn(lsn), false, &ctx).await?,
             RELSEG_SIZE
@@ -1687,7 +1677,7 @@ mod tests {
         walingest
             .put_rel_truncation(&mut m, TESTREL_A, RELSEG_SIZE - 1, &ctx)
             .await?;
-        m.commit().await?;
+        m.commit()?;
         assert_eq!(
             tline.get_rel_size(TESTREL_A, Lsn(lsn), false, &ctx).await?,
             RELSEG_SIZE - 1
@@ -1703,7 +1693,7 @@ mod tests {
             walingest
                 .put_rel_truncation(&mut m, TESTREL_A, size as BlockNumber, &ctx)
                 .await?;
-            m.commit().await?;
+            m.commit()?;
             assert_eq!(
                 tline.get_rel_size(TESTREL_A, Lsn(lsn), false, &ctx).await?,
                 size as BlockNumber

pgxn/neon/Makefile

@@ -8,7 +8,6 @@ OBJS = \
 	libpagestore.o \
 	libpqwalproposer.o \
 	neon.o \
-	access_stat.o \
 	pagestore_smgr.o \
 	relsize_cache.o \
 	walproposer.o \

pgxn/neon/access_stat.c

@@ -1,274 +0,0 @@
-
-/*
- * We want this statistic to rpresent current access patern mthis is why when
- * (n_seq_accesses + n_rnd_accesses) > MAX_ACCESS_COUNTER then we divide both counters by two,
- * so decreasng weight of historical data
- */
-#include "postgres.h"
-#include "funcapi.h"
-#include "miscadmin.h"
-#include "common/hashfn.h"
-#include "pagestore_client.h"
-#include "storage/relfilenode.h"
-#include "utils/guc.h"
-
-/* Structure used to predict sequential access */
-
-typedef struct AccessStatEntry {
-	RelFileNode relnode;
-	BlockNumber blkno; /* last accessed black number */
-	uint32      n_seq_accesses; /* number of sequential accesses (when block N+1 is accessed after block N) */
-	uint32      n_rnd_accesses; /* number of random accesses */
-	uint32      hash;
-	uint32      status;
-	uint64      access_count;  /* total number of relation accesses since backend start */
-	dlist_node	lru_node; /* LRU list node */
-}  AccessStatEntry;
-
-#define SH_PREFIX			as
-#define SH_ELEMENT_TYPE		AccessStatEntry
-#define SH_KEY_TYPE			RelFileNode
-#define SH_KEY				relnode
-#define SH_STORE_HASH
-#define SH_GET_HASH(tb, a)	((a)->hash)
-#define SH_HASH_KEY(tb, key) hash_bytes(		\
-		((const unsigned char *) &(key)),		\
-		sizeof(RelFileNode)					\
-)
-
-#define SH_EQUAL(tb, a, b)  RelFileNodeEquals((a), (b))
-#define SH_SCOPE			static inline
-#define SH_DEFINE
-#define SH_DECLARE
-#include "lib/simplehash.h"
-
-static as_hash *hash;
-static dlist_head lru;
-static int max_access_stat_size;
-static int max_access_stat_count;
-static double min_seq_access_ratio;
-static int min_seq_access_count;
-
-void access_stat_init(void)
-{
-	MemoryContext memctx = AllocSetContextCreate(TopMemoryContext,
-												 "NeonSMGR/access_stat",
-												 ALLOCSET_DEFAULT_SIZES);
-	DefineCustomIntVariable("neon.max_access_stat_size",
-							"Maximal size of Neon relation access statistic hash",
-							NULL,
-							&max_access_stat_size,
-							1024,
-							0,
-							INT_MAX,
-							PGC_USERSET,
-							0,
-							NULL,
-							NULL,
-							NULL);
-	DefineCustomIntVariable("neon.max_access_stat_count",
-							"Maximal value of relation access counter after which counters are divided by 2",
-							NULL,
-							&max_access_stat_count,
-							1024,
-							0,
-							INT_MAX,
-							PGC_USERSET,
-							0,
-							NULL,
-							NULL,
-							NULL);
-	DefineCustomRealVariable("neon.min_seq_access_ratio",
-							 "Minimal seq/(rnd+seq) ratio to determine sequential access",
-							 NULL,
-							 &min_seq_access_ratio,
-							 0.9,
-							 0,
-							 INT_MAX,
-							 PGC_USERSET,
-							 0,
-							 NULL,
-							 NULL,
-							 NULL);
-	DefineCustomIntVariable("neon.min_seq_access_count",
-							"Minimal access count to determine sequetial access",
-							NULL,
-							&min_seq_access_count,
-							10,
-							0,
-							INT_MAX,
-							PGC_USERSET,
-							0,
-							NULL,
-							NULL,
-							NULL);
-	hash = as_create(memctx, max_access_stat_size, NULL);
-	dlist_init(&lru);
-}
-
-
-bool is_sequential_access(RelFileNode rnode, ForkNumber forkNum, BlockNumber blkno)
-{
-	bool is_seq_access = false;
-	if (forkNum == MAIN_FORKNUM /* prefetch makes sense only for main fork */
-		&& max_access_stat_size != 0)
-	{
-		AccessStatEntry* entry = as_lookup(hash, rnode);
-		if (entry == NULL)
-		{
-			bool found;
-			/* New item */
-			while (hash->members >= max_access_stat_size)
-			{
-				/* Hash overflow: find candidate for replacement */
-				AccessStatEntry* victim = dlist_container(AccessStatEntry, lru_node, dlist_pop_head_node(&lru));
-				as_delete_item(hash, victim);
-			}
-			entry = as_insert(hash, rnode, &found);
-			Assert(!found);
-			/* Set both counter to zero because we don't know whethr first access is sequential or random */
-			entry->n_seq_accesses = 0;
-			entry->n_rnd_accesses = 0;
-			entry->access_count = 1;
-		}
-		else
-		{
-			uint32 access_count = entry->n_seq_accesses + entry->n_rnd_accesses;
-			/*
-			 * We want this function to represent most recent access pattern,
-			 * so when number of accesses exceed threashold value `max_access_stat_count`
-			 * we divide bother coutners by two devaluing old data
-			 */
-			if (access_count >= max_access_stat_count)
-			{
-				entry->n_seq_accesses >>= 1;
-				entry->n_rnd_accesses >>= 1;
-			}
-			if (entry->blkno+1 == blkno)
-				entry->n_seq_accesses += 1;
-			else
-				entry->n_rnd_accesses += 1;
-			entry->access_count += 1;
-			access_count = entry->n_seq_accesses + entry->n_rnd_accesses;
-
-			is_seq_access = access_count >= min_seq_access_count
-				&& (double)entry->n_seq_accesses / access_count >= min_seq_access_ratio;
-
-
-			/* Remove entry from LRU list tobe able to insert it to the end of this list */
-			dlist_delete(&entry->lru_node);
-		}
-		/* Place entry to the tail of LRU list */
-		dlist_push_tail(&lru, &entry->lru_node);
-		entry->blkno = blkno;
-	}
-	return is_seq_access;
-}
-
-/*
- * Get relation access pattern
- */
-PG_FUNCTION_INFO_V1(get_relation_access_statistics);
-
-
-typedef struct
-{
-	TupleDesc	tupdesc;
-	dlist_node* curr;
-} AccessStatContext;
-
-#define NUM_ACCESS_STAT_COLUMNS 6
-
-Datum
-get_relation_access_statistics(PG_FUNCTION_ARGS)
-{
-	FuncCallContext *funcctx;
-	Datum		result;
-	MemoryContext oldcontext;
-	AccessStatContext *fctx;	/* User function context. */
-	TupleDesc	tupledesc;
-	TupleDesc	expected_tupledesc;
-	HeapTuple	tuple;
-
-	if (SRF_IS_FIRSTCALL())
-	{
-		funcctx = SRF_FIRSTCALL_INIT();
-
-		/* Switch context when allocating stuff to be used in later calls */
-		oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
-
-		/* Create a user function context for cross-call persistence */
-		fctx = (AccessStatContext *) palloc(sizeof(AccessStatContext));
-
-		/*
-		 * To smoothly support upgrades from version 1.0 of this extension
-		 * transparently handle the (non-)existence of the pinning_backends
-		 * column. We unfortunately have to get the result type for that... -
-		 * we can't use the result type determined by the function definition
-		 * without potentially crashing when somebody uses the old (or even
-		 * wrong) function definition though.
-		 */
-		if (get_call_result_type(fcinfo, NULL, &expected_tupledesc) != TYPEFUNC_COMPOSITE)
-			elog(ERROR, "return type must be a row type");
-
-		if (expected_tupledesc->natts != NUM_ACCESS_STAT_COLUMNS)
-			elog(ERROR, "incorrect number of output arguments");
-
-		/* Construct a tuple descriptor for the result rows. */
-		tupledesc = CreateTemplateTupleDesc(expected_tupledesc->natts);
-		TupleDescInitEntry(tupledesc, (AttrNumber) 1, "relfilenode",
-						   OIDOID, -1, 0);
-		TupleDescInitEntry(tupledesc, (AttrNumber) 2, "reltablespace",
-						   OIDOID, -1, 0);
-		TupleDescInitEntry(tupledesc, (AttrNumber) 3, "reldatabase",
-						   OIDOID, -1, 0);
-		TupleDescInitEntry(tupledesc, (AttrNumber) 4, "seqaccess",
-						   INT4OID, -1, 0);
-		TupleDescInitEntry(tupledesc, (AttrNumber) 5, "rndaccess",
-						   INT4OID, -1, 0);
-		TupleDescInitEntry(tupledesc, (AttrNumber) 6, "accesscnt",
-						   INT8OID, -1, 0);
-
-		fctx->tupdesc = BlessTupleDesc(tupledesc);
-		fctx->curr = dlist_is_empty(&lru) ? NULL : dlist_tail_node(&lru);
-
-
-		/* Set max calls and remember the user function context. */
-		funcctx->max_calls = hash->members;
-		funcctx->user_fctx = fctx;
-
-		/* Return to original context when allocating transient memory */
-		MemoryContextSwitchTo(oldcontext);
-	}
-
-	funcctx = SRF_PERCALL_SETUP();
-
-	/* Get the saved state */
-	fctx = funcctx->user_fctx;
-	if (fctx->curr)
-	{
-		AccessStatEntry* entry = dlist_container(AccessStatEntry, lru_node, fctx->curr);
-		Datum		values[NUM_ACCESS_STAT_COLUMNS];
-		bool		nulls[NUM_ACCESS_STAT_COLUMNS] = {
-			false, false, false, false, false, false
-		};
-
-		values[0] = ObjectIdGetDatum(entry->relnode.relNode);
-		values[1] = ObjectIdGetDatum(entry->relnode.spcNode);
-		values[2] = ObjectIdGetDatum(entry->relnode.dbNode);
-		values[3] = Int32GetDatum(entry->n_seq_accesses);
-		values[4] = Int32GetDatum(entry->n_rnd_accesses);
-		values[5] = Int64GetDatum(entry->access_count);
-
-		/* Build and return the tuple. */
-		tuple = heap_form_tuple(fctx->tupdesc, values, nulls);
-		result = HeapTupleGetDatum(tuple);
-
-		fctx->curr = dlist_has_prev(&lru, fctx->curr) ? dlist_prev_node(&lru, fctx->curr) : NULL;
-
-		SRF_RETURN_NEXT(funcctx, result);
-	}
-	else
-		SRF_RETURN_DONE(funcctx);
-}
-

pgxn/neon/libpagestore.c

@@ -58,7 +58,6 @@ char	   *neon_auth_token;
 int			n_unflushed_requests = 0;
 int			flush_every_n_requests = 8;
 int			readahead_buffer_size = 128;
-int         readahead_distance = 10;
 
 bool	(*old_redo_read_buffer_filter) (XLogReaderState *record, uint8 block_id) = NULL;
 
@@ -453,18 +452,6 @@ pg_init_libpagestore(void)
 							PGC_USERSET,
 							0,	/* no flags required */
 							NULL, (GucIntAssignHook) &readahead_buffer_resize, NULL);
-	DefineCustomIntVariable("neon.readahead_distance",
-							"Number of read-ahead blocks",
-							NULL,
-							&readahead_distance,
-							10,
-							0,
-							INT_MAX,
-							PGC_USERSET,
-							0,
-							NULL,
-							NULL,
-							NULL);
 
 	relsize_hash_init();
 
@@ -489,5 +476,4 @@ pg_init_libpagestore(void)
 		redo_read_buffer_filter = neon_redo_read_buffer_filter;
 	}
 	lfc_init();
-	access_stat_init();
 }

pgxn/neon/neon--1.0.sql

@@ -27,18 +27,8 @@ RETURNS SETOF RECORD
 AS 'MODULE_PATHNAME', 'local_cache_pages'
 LANGUAGE C PARALLEL SAFE;
 
-CREATE FUNCTION get_relation_access_statistics()
-RETURNS SETOF RECORD
-AS 'MODULE_PATHNAME', 'get_relation_access_statistics'
-LANGUAGE C PARALLEL SAFE;
-
 -- Create a view for convenient access.
 CREATE VIEW local_cache AS
-	SELECT relname,P.* FROM local_cache_pages() AS P
+	SELECT P.* FROM local_cache_pages() AS P
 	(pageoffs int8, relfilenode oid, reltablespace oid, reldatabase oid,
-	 relforknumber int2, relblocknumber int8, accesscount int4) JOIN pg_class pc ON (P.relfilenode = pc.relfilenode);
-
-CREATE VIEW relation_access_statistics AS
-	SELECT relname,P.* FROM get_relation_access_statistics() AS P
-	(relfilenode oid, reltablespace oid, reldatabase oid,
-	 seqaccess int4, rndaccess int4, access_count int8) JOIN pg_class pc ON (P.relfilenode = pc.relfilenode);
+	 relforknumber int2, relblocknumber int8, accesscount int4);

pgxn/neon/pagestore_client.h

@@ -157,7 +157,6 @@ extern page_server_api * page_server;
 extern char *page_server_connstring;
 extern int flush_every_n_requests;
 extern int readahead_buffer_size;
-extern int readahead_distance;
 extern bool seqscan_prefetch_enabled;
 extern int seqscan_prefetch_distance;
 extern char *neon_timeline;
@@ -211,8 +210,5 @@ extern bool lfc_cache_contains(RelFileNode rnode, ForkNumber forkNum, BlockNumbe
 extern void lfc_evict(RelFileNode rnode, ForkNumber forkNum, BlockNumber blkno);
 extern void lfc_init(void);
 
-/* Access statistic */
-extern void access_stat_init(void);
-extern bool is_sequential_access(RelFileNode rnode, ForkNumber forkNum, BlockNumber blkno);
 
 #endif

pgxn/neon/pagestore_smgr.c

@@ -1881,7 +1881,6 @@ neon_read_at_lsn(RelFileNode rnode, ForkNumber forkNum, BlockNumber blkno,
 	if (RecoveryInProgress() && !(MyBackendType == B_STARTUP))
 		XLogWaitForReplayOf(request_lsn);
 
-
 	/*
 	 * Try to find prefetched page in the list of received pages.
 	 */
@@ -2004,10 +2003,6 @@ neon_read(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno,
 			elog(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
 	}
 
-	/* If it is expected to be sequential access then initiate prefetch of next block */
-	if (is_sequential_access(reln->smgr_rnode.node, forkNum, blkno))
-		neon_prefetch(reln, forkNum, blkno + readahead_distance);
-
 	/* Try to read from local file cache */
 	if (lfc_read(reln->smgr_rnode.node, forkNum, blkno, buffer))
 	{

pgxn/neon/walproposer.c

@@ -257,7 +257,7 @@ nwp_register_gucs(void)
 							"Walproposer reconnects to offline safekeepers once in this interval.",
 							NULL,
 							&wal_acceptor_reconnect_timeout,
-							1000, 0, INT_MAX,	/* default, min, max */
+							5000, 0, INT_MAX,	/* default, min, max */
 							PGC_SIGHUP, /* context */
 							GUC_UNIT_MS,	/* flags */
 							NULL, NULL, NULL);

poetry.lock

@@ -1,4 +1,4 @@
-# This file is automatically @generated by Poetry and should not be changed by hand.
+# This file is automatically @generated by Poetry 1.4.2 and should not be changed by hand.
 
 [[package]]
 name = "aiohttp"
@@ -855,31 +855,35 @@ files = [
 
 [[package]]
 name = "cryptography"
-version = "41.0.0"
+version = "39.0.1"
 description = "cryptography is a package which provides cryptographic recipes and primitives to Python developers."
 category = "main"
 optional = false
-python-versions = ">=3.7"
+python-versions = ">=3.6"
 files = [
-    {file = "cryptography-41.0.0-cp37-abi3-macosx_10_12_universal2.whl", hash = "sha256:3c5ef25d060c80d6d9f7f9892e1d41bb1c79b78ce74805b8cb4aa373cb7d5ec8"},
-    {file = "cryptography-41.0.0-cp37-abi3-macosx_10_12_x86_64.whl", hash = "sha256:8362565b3835ceacf4dc8f3b56471a2289cf51ac80946f9087e66dc283a810e0"},
-    {file = "cryptography-41.0.0-cp37-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:3680248309d340fda9611498a5319b0193a8dbdb73586a1acf8109d06f25b92d"},
-    {file = "cryptography-41.0.0-cp37-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:84a165379cb9d411d58ed739e4af3396e544eac190805a54ba2e0322feb55c46"},
-    {file = "cryptography-41.0.0-cp37-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:4ab14d567f7bbe7f1cdff1c53d5324ed4d3fc8bd17c481b395db224fb405c237"},
-    {file = "cryptography-41.0.0-cp37-abi3-manylinux_2_28_x86_64.whl", hash = "sha256:9f65e842cb02550fac96536edb1d17f24c0a338fd84eaf582be25926e993dde4"},
-    {file = "cryptography-41.0.0-cp37-abi3-musllinux_1_1_aarch64.whl", hash = "sha256:b7f2f5c525a642cecad24ee8670443ba27ac1fab81bba4cc24c7b6b41f2d0c75"},
-    {file = "cryptography-41.0.0-cp37-abi3-musllinux_1_1_x86_64.whl", hash = "sha256:7d92f0248d38faa411d17f4107fc0bce0c42cae0b0ba5415505df72d751bf62d"},
-    {file = "cryptography-41.0.0-cp37-abi3-win32.whl", hash = "sha256:34d405ea69a8b34566ba3dfb0521379b210ea5d560fafedf9f800a9a94a41928"},
-    {file = "cryptography-41.0.0-cp37-abi3-win_amd64.whl", hash = "sha256:344c6de9f8bda3c425b3a41b319522ba3208551b70c2ae00099c205f0d9fd3be"},
-    {file = "cryptography-41.0.0-pp38-pypy38_pp73-macosx_10_12_x86_64.whl", hash = "sha256:88ff107f211ea696455ea8d911389f6d2b276aabf3231bf72c8853d22db755c5"},
-    {file = "cryptography-41.0.0-pp38-pypy38_pp73-manylinux_2_28_aarch64.whl", hash = "sha256:b846d59a8d5a9ba87e2c3d757ca019fa576793e8758174d3868aecb88d6fc8eb"},
-    {file = "cryptography-41.0.0-pp38-pypy38_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:f5d0bf9b252f30a31664b6f64432b4730bb7038339bd18b1fafe129cfc2be9be"},
-    {file = "cryptography-41.0.0-pp38-pypy38_pp73-win_amd64.whl", hash = "sha256:5c1f7293c31ebc72163a9a0df246f890d65f66b4a40d9ec80081969ba8c78cc9"},
-    {file = "cryptography-41.0.0-pp39-pypy39_pp73-macosx_10_12_x86_64.whl", hash = "sha256:bf8fc66012ca857d62f6a347007e166ed59c0bc150cefa49f28376ebe7d992a2"},
-    {file = "cryptography-41.0.0-pp39-pypy39_pp73-manylinux_2_28_aarch64.whl", hash = "sha256:a4fc68d1c5b951cfb72dfd54702afdbbf0fb7acdc9b7dc4301bbf2225a27714d"},
-    {file = "cryptography-41.0.0-pp39-pypy39_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:14754bcdae909d66ff24b7b5f166d69340ccc6cb15731670435efd5719294895"},
-    {file = "cryptography-41.0.0-pp39-pypy39_pp73-win_amd64.whl", hash = "sha256:0ddaee209d1cf1f180f1efa338a68c4621154de0afaef92b89486f5f96047c55"},
-    {file = "cryptography-41.0.0.tar.gz", hash = "sha256:6b71f64beeea341c9b4f963b48ee3b62d62d57ba93eb120e1196b31dc1025e78"},
+    {file = "cryptography-39.0.1-cp36-abi3-macosx_10_12_universal2.whl", hash = "sha256:6687ef6d0a6497e2b58e7c5b852b53f62142cfa7cd1555795758934da363a965"},
+    {file = "cryptography-39.0.1-cp36-abi3-macosx_10_12_x86_64.whl", hash = "sha256:706843b48f9a3f9b9911979761c91541e3d90db1ca905fd63fee540a217698bc"},
+    {file = "cryptography-39.0.1-cp36-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.manylinux_2_24_aarch64.whl", hash = "sha256:5d2d8b87a490bfcd407ed9d49093793d0f75198a35e6eb1a923ce1ee86c62b41"},
+    {file = "cryptography-39.0.1-cp36-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:83e17b26de248c33f3acffb922748151d71827d6021d98c70e6c1a25ddd78505"},
+    {file = "cryptography-39.0.1-cp36-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:e124352fd3db36a9d4a21c1aa27fd5d051e621845cb87fb851c08f4f75ce8be6"},
+    {file = "cryptography-39.0.1-cp36-abi3-manylinux_2_24_x86_64.whl", hash = "sha256:5aa67414fcdfa22cf052e640cb5ddc461924a045cacf325cd164e65312d99502"},
+    {file = "cryptography-39.0.1-cp36-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:35f7c7d015d474f4011e859e93e789c87d21f6f4880ebdc29896a60403328f1f"},
+    {file = "cryptography-39.0.1-cp36-abi3-manylinux_2_28_x86_64.whl", hash = "sha256:f24077a3b5298a5a06a8e0536e3ea9ec60e4c7ac486755e5fb6e6ea9b3500106"},
+    {file = "cryptography-39.0.1-cp36-abi3-musllinux_1_1_aarch64.whl", hash = "sha256:f0c64d1bd842ca2633e74a1a28033d139368ad959872533b1bab8c80e8240a0c"},
+    {file = "cryptography-39.0.1-cp36-abi3-musllinux_1_1_x86_64.whl", hash = "sha256:0f8da300b5c8af9f98111ffd512910bc792b4c77392a9523624680f7956a99d4"},
+    {file = "cryptography-39.0.1-cp36-abi3-win32.whl", hash = "sha256:fe913f20024eb2cb2f323e42a64bdf2911bb9738a15dba7d3cce48151034e3a8"},
+    {file = "cryptography-39.0.1-cp36-abi3-win_amd64.whl", hash = "sha256:ced4e447ae29ca194449a3f1ce132ded8fcab06971ef5f618605aacaa612beac"},
+    {file = "cryptography-39.0.1-pp38-pypy38_pp73-macosx_10_12_x86_64.whl", hash = "sha256:807ce09d4434881ca3a7594733669bd834f5b2c6d5c7e36f8c00f691887042ad"},
+    {file = "cryptography-39.0.1-pp38-pypy38_pp73-manylinux_2_28_aarch64.whl", hash = "sha256:c5caeb8188c24888c90b5108a441c106f7faa4c4c075a2bcae438c6e8ca73cef"},
+    {file = "cryptography-39.0.1-pp38-pypy38_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:4789d1e3e257965e960232345002262ede4d094d1a19f4d3b52e48d4d8f3b885"},
+    {file = "cryptography-39.0.1-pp38-pypy38_pp73-win_amd64.whl", hash = "sha256:96f1157a7c08b5b189b16b47bc9db2332269d6680a196341bf30046330d15388"},
+    {file = "cryptography-39.0.1-pp39-pypy39_pp73-macosx_10_12_x86_64.whl", hash = "sha256:e422abdec8b5fa8462aa016786680720d78bdce7a30c652b7fadf83a4ba35336"},
+    {file = "cryptography-39.0.1-pp39-pypy39_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.manylinux_2_24_aarch64.whl", hash = "sha256:b0afd054cd42f3d213bf82c629efb1ee5f22eba35bf0eec88ea9ea7304f511a2"},
+    {file = "cryptography-39.0.1-pp39-pypy39_pp73-manylinux_2_24_x86_64.whl", hash = "sha256:6f8ba7f0328b79f08bdacc3e4e66fb4d7aab0c3584e0bd41328dce5262e26b2e"},
+    {file = "cryptography-39.0.1-pp39-pypy39_pp73-manylinux_2_28_aarch64.whl", hash = "sha256:ef8b72fa70b348724ff1218267e7f7375b8de4e8194d1636ee60510aae104cd0"},
+    {file = "cryptography-39.0.1-pp39-pypy39_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:aec5a6c9864be7df2240c382740fcf3b96928c46604eaa7f3091f58b878c0bb6"},
+    {file = "cryptography-39.0.1-pp39-pypy39_pp73-win_amd64.whl", hash = "sha256:fdd188c8a6ef8769f148f88f859884507b954cc64db6b52f66ef199bb9ad660a"},
+    {file = "cryptography-39.0.1.tar.gz", hash = "sha256:d1f6198ee6d9148405e49887803907fe8962a23e6c6f83ea7d98f1c0de375695"},
 ]
 
 [package.dependencies]
@@ -888,12 +892,12 @@ cffi = ">=1.12"
 [package.extras]
 docs = ["sphinx (>=5.3.0)", "sphinx-rtd-theme (>=1.1.1)"]
 docstest = ["pyenchant (>=1.6.11)", "sphinxcontrib-spelling (>=4.0.1)", "twine (>=1.12.0)"]
-nox = ["nox"]
-pep8test = ["black", "check-sdist", "mypy", "ruff"]
-sdist = ["build"]
+pep8test = ["black", "check-manifest", "mypy", "ruff", "types-pytz", "types-requests"]
+sdist = ["setuptools-rust (>=0.11.4)"]
 ssh = ["bcrypt (>=3.1.5)"]
-test = ["pretend", "pytest (>=6.2.0)", "pytest-benchmark", "pytest-cov", "pytest-xdist"]
+test = ["hypothesis (>=1.11.4,!=3.79.2)", "iso8601", "pretend", "pytest (>=6.2.0)", "pytest-benchmark", "pytest-cov", "pytest-shard (>=0.1.2)", "pytest-subtests", "pytest-xdist", "pytz"]
 test-randomorder = ["pytest-randomly"]
+tox = ["tox"]
 
 [[package]]
 name = "docker"

proxy/src/http/sql_over_http.rs

@@ -1,6 +1,5 @@
 use futures::pin_mut;
 use futures::StreamExt;
-use futures::TryFutureExt;
 use hyper::body::HttpBody;
 use hyper::http::HeaderName;
 use hyper::http::HeaderValue;
@@ -12,13 +11,8 @@ use serde_json::Value;
 use tokio_postgres::types::Kind;
 use tokio_postgres::types::Type;
 use tokio_postgres::Row;
-use tracing::error;
-use tracing::info;
-use tracing::instrument;
 use url::Url;
 
-use crate::proxy::invalidate_cache;
-use crate::proxy::NUM_RETRIES_WAKE_COMPUTE;
 use crate::{auth, config::ProxyConfig, console};
 
 #[derive(serde::Deserialize)]
@@ -96,17 +90,10 @@ fn json_array_to_pg_array(value: &Value) -> Result<Option<String>, serde_json::E
     }
 }
 
-struct ConnInfo {
-    username: String,
-    dbname: String,
-    hostname: String,
-    password: String,
-}
-
 fn get_conn_info(
     headers: &HeaderMap,
     sni_hostname: Option<String>,
-) -> Result<ConnInfo, anyhow::Error> {
+) -> Result<(String, String, String, String), anyhow::Error> {
     let connection_string = headers
         .get("Neon-Connection-String")
         .ok_or(anyhow::anyhow!("missing connection string"))?
@@ -159,12 +146,12 @@ fn get_conn_info(
         }
     }
 
-    Ok(ConnInfo {
-        username: username.to_owned(),
-        dbname: dbname.to_owned(),
-        hostname: hostname.to_owned(),
-        password: password.to_owned(),
-    })
+    Ok((
+        username.to_owned(),
+        dbname.to_owned(),
+        hostname.to_owned(),
+        password.to_owned(),
+    ))
 }
 
 // TODO: return different http error codes
@@ -177,10 +164,10 @@ pub async fn handle(
     // Determine the destination and connection params
     //
     let headers = request.headers();
-    let conn_info = get_conn_info(headers, sni_hostname)?;
+    let (username, dbname, hostname, password) = get_conn_info(headers, sni_hostname)?;
     let credential_params = StartupMessageParams::new([
-        ("user", &conn_info.username),
-        ("database", &conn_info.dbname),
+        ("user", &username),
+        ("database", &dbname),
         ("application_name", APP_NAME),
     ]);
 
@@ -199,20 +186,21 @@ pub async fn handle(
     let creds = config
         .auth_backend
         .as_ref()
-        .map(|_| {
-            auth::ClientCredentials::parse(
-                &credential_params,
-                Some(&conn_info.hostname),
-                common_names,
-            )
-        })
+        .map(|_| auth::ClientCredentials::parse(&credential_params, Some(&hostname), common_names))
         .transpose()?;
     let extra = console::ConsoleReqExtra {
         session_id: uuid::Uuid::new_v4(),
         application_name: Some(APP_NAME),
     };
-
-    let mut node_info = creds.wake_compute(&extra).await?.expect("msg");
+    let node = creds.wake_compute(&extra).await?.expect("msg");
+    let conf = node.value.config;
+    let port = *conf.get_ports().first().expect("no port");
+    let host = match conf.get_hosts().first().expect("no host") {
+        tokio_postgres::config::Host::Tcp(host) => host,
+        tokio_postgres::config::Host::Unix(_) => {
+            return Err(anyhow::anyhow!("unix socket is not supported"));
+        }
+    };
 
     let request_content_length = match request.body().size_hint().upper() {
         Some(v) => v,
@@ -232,10 +220,28 @@ pub async fn handle(
     let QueryData { query, params } = serde_json::from_slice(&body)?;
     let query_params = json_to_pg_text(params)?;
 
+    //
+    // Connenct to the destination
+    //
+    let (client, connection) = tokio_postgres::Config::new()
+        .host(host)
+        .port(port)
+        .user(&username)
+        .password(&password)
+        .dbname(&dbname)
+        .max_backend_message_size(MAX_RESPONSE_SIZE)
+        .connect(tokio_postgres::NoTls)
+        .await?;
+
+    tokio::spawn(async move {
+        if let Err(e) = connection.await {
+            eprintln!("connection error: {}", e);
+        }
+    });
+
     //
     // Now execute the query and return the result
     //
-    let client = connect_to_compute(&mut node_info, &extra, &creds, &conn_info).await?;
     let row_stream = client.query_raw_txt(query, query_params).await?;
 
     // Manually drain the stream into a vector to leave row_stream hanging
@@ -274,11 +280,6 @@ pub async fn handle(
                 json!({
                     "name": Value::String(c.name().to_owned()),
                     "dataTypeID": Value::Number(c.type_().oid().into()),
-                    "tableID": c.table_oid(),
-                    "columnID": c.column_id(),
-                    "dataTypeSize": c.type_size(),
-                    "dataTypeModifier": c.type_modifier(),
-                    "format": "text",
                 })
             })
             .collect::<Vec<_>>()
@@ -302,70 +303,6 @@ pub async fn handle(
     }))
 }
 
-/// This function is a copy of `connect_to_compute` from `src/proxy.rs` with
-/// the difference that it uses `tokio_postgres` for the connection.
-#[instrument(skip_all)]
-async fn connect_to_compute(
-    node_info: &mut console::CachedNodeInfo,
-    extra: &console::ConsoleReqExtra<'_>,
-    creds: &auth::BackendType<'_, auth::ClientCredentials<'_>>,
-    conn_info: &ConnInfo,
-) -> anyhow::Result<tokio_postgres::Client> {
-    let mut num_retries: usize = NUM_RETRIES_WAKE_COMPUTE;
-
-    loop {
-        match connect_to_compute_once(node_info, conn_info).await {
-            Err(e) if num_retries > 0 => {
-                info!("compute node's state has changed; requesting a wake-up");
-                match creds.wake_compute(extra).await? {
-                    // Update `node_info` and try one more time.
-                    Some(new) => {
-                        *node_info = new;
-                    }
-                    // Link auth doesn't work that way, so we just exit.
-                    None => return Err(e),
-                }
-            }
-            other => return other,
-        }
-
-        num_retries -= 1;
-        info!("retrying after wake-up ({num_retries} attempts left)");
-    }
-}
-
-async fn connect_to_compute_once(
-    node_info: &console::CachedNodeInfo,
-    conn_info: &ConnInfo,
-) -> anyhow::Result<tokio_postgres::Client> {
-    let mut config = (*node_info.config).clone();
-
-    let (client, connection) = config
-        .user(&conn_info.username)
-        .password(&conn_info.password)
-        .dbname(&conn_info.dbname)
-        .max_backend_message_size(MAX_RESPONSE_SIZE)
-        .connect(tokio_postgres::NoTls)
-        .inspect_err(|e: &tokio_postgres::Error| {
-            error!(
-                "failed to connect to compute node hosts={:?} ports={:?}: {}",
-                node_info.config.get_hosts(),
-                node_info.config.get_ports(),
-                e
-            );
-            invalidate_cache(node_info)
-        })
-        .await?;
-
-    tokio::spawn(async move {
-        if let Err(e) = connection.await {
-            error!("connection error: {}", e);
-        }
-    });
-
-    Ok(client)
-}
-
 //
 // Convert postgres row with text-encoded values to JSON object
 //

proxy/src/http/websocket.rs

@@ -26,6 +26,7 @@ use tls_listener::TlsListener;
 use tokio::{
     io::{self, AsyncBufRead, AsyncRead, AsyncWrite, ReadBuf},
     net::TcpListener,
+    select,
 };
 use tokio_util::sync::CancellationToken;
 use tracing::{error, info, info_span, warn, Instrument};
@@ -192,9 +193,14 @@ async fn ws_handler(
     // TODO: that deserves a refactor as now this function also handles http json client besides websockets.
     // Right now I don't want to blow up sql-over-http patch with file renames and do that as a follow up instead.
     } else if request.uri().path() == "/sql" && request.method() == Method::POST {
-        let result = sql_over_http::handle(config, request, sni_hostname)
-            .instrument(info_span!("sql-over-http"))
-            .await;
+        let result = select! {
+            _ = tokio::time::sleep(std::time::Duration::from_secs(10)) => {
+                Err(anyhow::anyhow!("Query timed out"))
+            }
+            response = sql_over_http::handle(config, request, sni_hostname) => {
+                response
+            }
+        };
         let status_code = match result {
             Ok(_) => StatusCode::OK,
             Err(_) => StatusCode::BAD_REQUEST,

proxy/src/proxy.rs

@@ -22,7 +22,7 @@ use tracing::{error, info, warn};
 use utils::measured_stream::MeasuredStream;
 
 /// Number of times we should retry the `/proxy_wake_compute` http request.
-pub const NUM_RETRIES_WAKE_COMPUTE: usize = 1;
+const NUM_RETRIES_WAKE_COMPUTE: usize = 1;
 
 const ERR_INSECURE_CONNECTION: &str = "connection is insecure (try using `sslmode=require`)";
 const ERR_PROTO_VIOLATION: &str = "protocol violation";
@@ -283,35 +283,34 @@ async fn handshake<S: AsyncRead + AsyncWrite + Unpin>(
     }
 }
 
-/// If we couldn't connect, a cached connection info might be to blame
-/// (e.g. the compute node's address might've changed at the wrong time).
-/// Invalidate the cache entry (if any) to prevent subsequent errors.
-#[tracing::instrument(name = "invalidate_cache", skip_all)]
-pub fn invalidate_cache(node_info: &console::CachedNodeInfo) {
-    let is_cached = node_info.cached();
-    if is_cached {
-        warn!("invalidating stalled compute node info cache entry");
-        node_info.invalidate();
-    }
-
-    let label = match is_cached {
-        true => "compute_cached",
-        false => "compute_uncached",
-    };
-    NUM_CONNECTION_FAILURES.with_label_values(&[label]).inc();
-}
-
 /// Try to connect to the compute node once.
 #[tracing::instrument(name = "connect_once", skip_all)]
 async fn connect_to_compute_once(
     node_info: &console::CachedNodeInfo,
 ) -> Result<PostgresConnection, compute::ConnectionError> {
+    // If we couldn't connect, a cached connection info might be to blame
+    // (e.g. the compute node's address might've changed at the wrong time).
+    // Invalidate the cache entry (if any) to prevent subsequent errors.
+    let invalidate_cache = |_: &compute::ConnectionError| {
+        let is_cached = node_info.cached();
+        if is_cached {
+            warn!("invalidating stalled compute node info cache entry");
+            node_info.invalidate();
+        }
+
+        let label = match is_cached {
+            true => "compute_cached",
+            false => "compute_uncached",
+        };
+        NUM_CONNECTION_FAILURES.with_label_values(&[label]).inc();
+    };
+
     let allow_self_signed_compute = node_info.allow_self_signed_compute;
 
     node_info
         .config
         .connect(allow_self_signed_compute)
-        .inspect_err(|_: &compute::ConnectionError| invalidate_cache(node_info))
+        .inspect_err(invalidate_cache)
         .await
 }
 

safekeeper/src/bin/safekeeper.rs

@@ -3,19 +3,15 @@
 //
 use anyhow::{bail, Context, Result};
 use clap::Parser;
-use futures::future::BoxFuture;
-use futures::stream::FuturesUnordered;
-use futures::{FutureExt, StreamExt};
 use remote_storage::RemoteStorageConfig;
-use tokio::runtime::Handle;
-use tokio::signal::unix::{signal, SignalKind};
-use tokio::task::JoinError;
 use toml_edit::Document;
+use utils::signals::ShutdownSignals;
 
 use std::fs::{self, File};
 use std::io::{ErrorKind, Write};
 use std::path::{Path, PathBuf};
 use std::sync::Arc;
+use std::thread;
 use std::time::Duration;
 use storage_broker::Uri;
 use tokio::sync::mpsc;
@@ -24,21 +20,22 @@ use tracing::*;
 use utils::pid_file;
 
 use metrics::set_build_info_metric;
+use safekeeper::broker;
+use safekeeper::control_file;
 use safekeeper::defaults::{
     DEFAULT_HEARTBEAT_TIMEOUT, DEFAULT_HTTP_LISTEN_ADDR, DEFAULT_MAX_OFFLOADER_LAG_BYTES,
     DEFAULT_PG_LISTEN_ADDR,
 };
+use safekeeper::http;
+use safekeeper::remove_wal;
+use safekeeper::wal_backup;
 use safekeeper::wal_service;
 use safekeeper::GlobalTimelines;
 use safekeeper::SafeKeeperConf;
-use safekeeper::{broker, WAL_SERVICE_RUNTIME};
-use safekeeper::{control_file, BROKER_RUNTIME};
-use safekeeper::{http, WAL_REMOVER_RUNTIME};
-use safekeeper::{remove_wal, WAL_BACKUP_RUNTIME};
-use safekeeper::{wal_backup, HTTP_RUNTIME};
 use storage_broker::DEFAULT_ENDPOINT;
 use utils::auth::JwtAuth;
 use utils::{
+    http::endpoint,
     id::NodeId,
     logging::{self, LogFormat},
     project_git_version,
@@ -107,6 +104,10 @@ struct Args {
     /// Safekeeper won't be elected for WAL offloading if it is lagging for more than this value in bytes
     #[arg(long, default_value_t = DEFAULT_MAX_OFFLOADER_LAG_BYTES)]
     max_offloader_lag: u64,
+    /// Number of threads for wal backup runtime, by default number of cores
+    /// available to the system.
+    #[arg(long)]
+    wal_backup_threads: Option<usize>,
     /// Number of max parallel WAL segments to be offloaded to remote storage.
     #[arg(long, default_value = "5")]
     wal_backup_parallel_jobs: usize,
@@ -120,14 +121,9 @@ struct Args {
     /// Format for logging, either 'plain' or 'json'.
     #[arg(long, default_value = "plain")]
     log_format: String,
-    /// Run everything in single threaded current thread runtime, might be
-    /// useful for debugging.
-    #[arg(long)]
-    current_thread_runtime: bool,
 }
 
-#[tokio::main(flavor = "current_thread")]
-async fn main() -> anyhow::Result<()> {
+fn main() -> anyhow::Result<()> {
     let args = Args::parse();
 
     if let Some(addr) = args.dump_control_file {
@@ -187,10 +183,10 @@ async fn main() -> anyhow::Result<()> {
         heartbeat_timeout: args.heartbeat_timeout,
         remote_storage: args.remote_storage,
         max_offloader_lag_bytes: args.max_offloader_lag,
+        backup_runtime_threads: args.wal_backup_threads,
         wal_backup_enabled: !args.disable_wal_backup,
         backup_parallel_jobs: args.wal_backup_parallel_jobs,
         auth,
-        current_thread_runtime: args.current_thread_runtime,
     };
 
     // initialize sentry if SENTRY_DSN is provided
@@ -198,14 +194,10 @@ async fn main() -> anyhow::Result<()> {
         Some(GIT_VERSION.into()),
         &[("node_id", &conf.my_id.to_string())],
     );
-    start_safekeeper(conf).await
+    start_safekeeper(conf)
 }
 
-/// Result of joining any of main tasks: upper error means task failed to
-/// complete, e.g. panicked, inner is error produced by task itself.
-type JoinTaskRes = Result<anyhow::Result<()>, JoinError>;
-
-async fn start_safekeeper(conf: SafeKeeperConf) -> Result<()> {
+fn start_safekeeper(conf: SafeKeeperConf) -> Result<()> {
     // Prevent running multiple safekeepers on the same directory
     let lock_file_path = conf.workdir.join(PID_FILE_NAME);
     let lock_file =
@@ -216,18 +208,14 @@ async fn start_safekeeper(conf: SafeKeeperConf) -> Result<()> {
     // we need to release the lock file only when the current process is gone
     std::mem::forget(lock_file);
 
-    info!("starting safekeeper WAL service on {}", conf.listen_pg_addr);
-    let pg_listener = tcp_listener::bind(conf.listen_pg_addr.clone()).map_err(|e| {
-        error!("failed to bind to address {}: {}", conf.listen_pg_addr, e);
+    let http_listener = tcp_listener::bind(conf.listen_http_addr.clone()).map_err(|e| {
+        error!("failed to bind to address {}: {}", conf.listen_http_addr, e);
         e
     })?;
 
-    info!(
-        "starting safekeeper HTTP service on {}",
-        conf.listen_http_addr
-    );
-    let http_listener = tcp_listener::bind(conf.listen_http_addr.clone()).map_err(|e| {
-        error!("failed to bind to address {}: {}", conf.listen_http_addr, e);
+    info!("starting safekeeper on {}", conf.listen_pg_addr);
+    let pg_listener = tcp_listener::bind(conf.listen_pg_addr.clone()).map_err(|e| {
+        error!("failed to bind to address {}: {}", conf.listen_pg_addr, e);
         e
     })?;
 
@@ -236,88 +224,71 @@ async fn start_safekeeper(conf: SafeKeeperConf) -> Result<()> {
     let timeline_collector = safekeeper::metrics::TimelineCollector::new();
     metrics::register_internal(Box::new(timeline_collector))?;
 
+    let mut threads = vec![];
     let (wal_backup_launcher_tx, wal_backup_launcher_rx) = mpsc::channel(100);
 
     // Load all timelines from disk to memory.
     GlobalTimelines::init(conf.clone(), wal_backup_launcher_tx)?;
 
-    // Keep handles to main tasks to die if any of them disappears.
-    let mut tasks_handles: FuturesUnordered<BoxFuture<(String, JoinTaskRes)>> =
-        FuturesUnordered::new();
+    let conf_ = conf.clone();
+    threads.push(
+        thread::Builder::new()
+            .name("http_endpoint_thread".into())
+            .spawn(|| {
+                let router = http::make_router(conf_);
+                endpoint::serve_thread_main(
+                    router,
+                    http_listener,
+                    std::future::pending(), // never shut down
+                )
+                .unwrap();
+            })?,
+    );
+
+    let conf_cloned = conf.clone();
+    let safekeeper_thread = thread::Builder::new()
+        .name("WAL service thread".into())
+        .spawn(|| wal_service::thread_main(conf_cloned, pg_listener))
+        .unwrap();
+
+    threads.push(safekeeper_thread);
 
     let conf_ = conf.clone();
-    // Run everything in current thread rt, if asked.
-    if conf.current_thread_runtime {
-        info!("running in current thread runtime");
-    }
-    let current_thread_rt = conf
-        .current_thread_runtime
-        .then(|| Handle::try_current().expect("no runtime in main"));
-    let wal_service_handle = current_thread_rt
-        .as_ref()
-        .unwrap_or_else(|| WAL_SERVICE_RUNTIME.handle())
-        .spawn(wal_service::task_main(conf_, pg_listener))
-        // wrap with task name for error reporting
-        .map(|res| ("WAL service main".to_owned(), res));
-    tasks_handles.push(Box::pin(wal_service_handle));
+    threads.push(
+        thread::Builder::new()
+            .name("broker thread".into())
+            .spawn(|| {
+                broker::thread_main(conf_);
+            })?,
+    );
 
     let conf_ = conf.clone();
-    let http_handle = current_thread_rt
-        .as_ref()
-        .unwrap_or_else(|| HTTP_RUNTIME.handle())
-        .spawn(http::task_main(conf_, http_listener))
-        .map(|res| ("HTTP service main".to_owned(), res));
-    tasks_handles.push(Box::pin(http_handle));
+    threads.push(
+        thread::Builder::new()
+            .name("WAL removal thread".into())
+            .spawn(|| {
+                remove_wal::thread_main(conf_);
+            })?,
+    );
 
-    let conf_ = conf.clone();
-    let broker_task_handle = current_thread_rt
-        .as_ref()
-        .unwrap_or_else(|| BROKER_RUNTIME.handle())
-        .spawn(broker::task_main(conf_).instrument(info_span!("broker")))
-        .map(|res| ("broker main".to_owned(), res));
-    tasks_handles.push(Box::pin(broker_task_handle));
-
-    let conf_ = conf.clone();
-    let wal_remover_handle = current_thread_rt
-        .as_ref()
-        .unwrap_or_else(|| WAL_REMOVER_RUNTIME.handle())
-        .spawn(remove_wal::task_main(conf_))
-        .map(|res| ("WAL remover".to_owned(), res));
-    tasks_handles.push(Box::pin(wal_remover_handle));
-
-    let conf_ = conf.clone();
-    let wal_backup_handle = current_thread_rt
-        .as_ref()
-        .unwrap_or_else(|| WAL_BACKUP_RUNTIME.handle())
-        .spawn(wal_backup::wal_backup_launcher_task_main(
-            conf_,
-            wal_backup_launcher_rx,
-        ))
-        .map(|res| ("WAL backup launcher".to_owned(), res));
-    tasks_handles.push(Box::pin(wal_backup_handle));
+    threads.push(
+        thread::Builder::new()
+            .name("WAL backup launcher thread".into())
+            .spawn(move || {
+                wal_backup::wal_backup_launcher_thread_main(conf, wal_backup_launcher_rx);
+            })?,
+    );
 
     set_build_info_metric(GIT_VERSION);
+    // TODO: put more thoughts into handling of failed threads
+    // We should catch & die if they are in trouble.
 
-    // TODO: update tokio-stream, convert to real async Stream with
-    // SignalStream, map it to obtain missing signal name, combine streams into
-    // single stream we can easily sit on.
-    let mut sigquit_stream = signal(SignalKind::quit())?;
-    let mut sigint_stream = signal(SignalKind::interrupt())?;
-    let mut sigterm_stream = signal(SignalKind::terminate())?;
-
-    tokio::select! {
-        Some((task_name, res)) = tasks_handles.next()=> {
-            error!("{} task failed: {:?}, exiting", task_name, res);
-            std::process::exit(1);
-        }
-        // On any shutdown signal, log receival and exit. Additionally, handling
-        // SIGQUIT prevents coredump.
-        _ = sigquit_stream.recv() => info!("received SIGQUIT, terminating"),
-        _ = sigint_stream.recv() => info!("received SIGINT, terminating"),
-        _ = sigterm_stream.recv() => info!("received SIGTERM, terminating")
-
-    };
-    std::process::exit(0);
+    // On any shutdown signal, log receival and exit. Additionally, handling
+    // SIGQUIT prevents coredump.
+    ShutdownSignals::handle(|signal| {
+        info!("received {}, terminating", signal.name());
+        std::process::exit(0);
+    })
 }
 
 /// Determine safekeeper id.

safekeeper/src/broker.rs

@@ -8,7 +8,7 @@ use anyhow::Error;
 use anyhow::Result;
 
 use storage_broker::parse_proto_ttid;
-
+use storage_broker::proto::broker_service_client::BrokerServiceClient;
 use storage_broker::proto::subscribe_safekeeper_info_request::SubscriptionKey as ProtoSubscriptionKey;
 use storage_broker::proto::SubscribeSafekeeperInfoRequest;
 use storage_broker::Request;
@@ -16,7 +16,7 @@ use storage_broker::Request;
 use std::time::Duration;
 use std::time::Instant;
 use tokio::task::JoinHandle;
-use tokio::time::sleep;
+use tokio::{runtime, time::sleep};
 use tracing::*;
 
 use crate::metrics::BROKER_ITERATION_TIMELINES;
@@ -29,10 +29,23 @@ use crate::SafeKeeperConf;
 const RETRY_INTERVAL_MSEC: u64 = 1000;
 const PUSH_INTERVAL_MSEC: u64 = 1000;
 
+pub fn thread_main(conf: SafeKeeperConf) {
+    let runtime = runtime::Builder::new_current_thread()
+        .enable_all()
+        .build()
+        .unwrap();
+
+    let _enter = info_span!("broker").entered();
+    info!("started, broker endpoint {:?}", conf.broker_endpoint);
+
+    runtime.block_on(async {
+        main_loop(conf).await;
+    });
+}
+
 /// Push once in a while data about all active timelines to the broker.
 async fn push_loop(conf: SafeKeeperConf) -> anyhow::Result<()> {
-    let mut client =
-        storage_broker::connect(conf.broker_endpoint.clone(), conf.broker_keepalive_interval)?;
+    let mut client = BrokerServiceClient::connect(conf.broker_endpoint.clone()).await?;
     let push_interval = Duration::from_millis(PUSH_INTERVAL_MSEC);
 
     let outbound = async_stream::stream! {
@@ -42,27 +55,20 @@ async fn push_loop(conf: SafeKeeperConf) -> anyhow::Result<()> {
             // sensitive and there is no risk of deadlock as we don't await while
             // lock is held.
             let now = Instant::now();
-            let all_tlis = GlobalTimelines::get_all();
-            let mut n_pushed_tlis = 0;
-            for tli in &all_tlis {
-                // filtering alternative futures::stream::iter(all_tlis)
-                //   .filter(|tli| {let tli = tli.clone(); async move { tli.is_active().await}}).collect::<Vec<_>>().await;
-                // doesn't look better, and I'm not sure how to do that without collect.
-                if !tli.is_active().await {
-                    continue;
-                }
-                let sk_info = tli.get_safekeeper_info(&conf).await;
+            let mut active_tlis = GlobalTimelines::get_all();
+            active_tlis.retain(|tli| tli.is_active());
+            for tli in &active_tlis {
+                let sk_info = tli.get_safekeeper_info(&conf);
                 yield sk_info;
                 BROKER_PUSHED_UPDATES.inc();
-                n_pushed_tlis += 1;
             }
             let elapsed = now.elapsed();
 
             BROKER_PUSH_ALL_UPDATES_SECONDS.observe(elapsed.as_secs_f64());
-            BROKER_ITERATION_TIMELINES.observe(n_pushed_tlis as f64);
+            BROKER_ITERATION_TIMELINES.observe(active_tlis.len() as f64);
 
             if elapsed > push_interval / 2 {
-                info!("broker push is too long, pushed {} timeline updates to broker in {:?}", n_pushed_tlis, elapsed);
+                info!("broker push is too long, pushed {} timeline updates to broker in {:?}", active_tlis.len(), elapsed);
             }
 
             sleep(push_interval).await;
@@ -119,13 +125,10 @@ async fn pull_loop(conf: SafeKeeperConf) -> Result<()> {
     bail!("end of stream");
 }
 
-pub async fn task_main(conf: SafeKeeperConf) -> anyhow::Result<()> {
-    info!("started, broker endpoint {:?}", conf.broker_endpoint);
-
+async fn main_loop(conf: SafeKeeperConf) {
     let mut ticker = tokio::time::interval(Duration::from_millis(RETRY_INTERVAL_MSEC));
     let mut push_handle: Option<JoinHandle<Result<(), Error>>> = None;
     let mut pull_handle: Option<JoinHandle<Result<(), Error>>> = None;
-
     // Selecting on JoinHandles requires some squats; is there a better way to
     // reap tasks individually?
 

safekeeper/src/control_file.rs

@@ -2,10 +2,9 @@
 
 use anyhow::{bail, ensure, Context, Result};
 use byteorder::{LittleEndian, ReadBytesExt, WriteBytesExt};
-use tokio::fs::{self, File};
-use tokio::io::AsyncWriteExt;
 
-use std::io::Read;
+use std::fs::{self, File, OpenOptions};
+use std::io::{Read, Write};
 use std::ops::Deref;
 use std::path::{Path, PathBuf};
 use std::time::Instant;
@@ -27,10 +26,9 @@ pub const CHECKSUM_SIZE: usize = std::mem::size_of::<u32>();
 
 /// Storage should keep actual state inside of it. It should implement Deref
 /// trait to access state fields and have persist method for updating that state.
-#[async_trait::async_trait]
 pub trait Storage: Deref<Target = SafeKeeperState> {
     /// Persist safekeeper state on disk and update internal state.
-    async fn persist(&mut self, s: &SafeKeeperState) -> Result<()>;
+    fn persist(&mut self, s: &SafeKeeperState) -> Result<()>;
 
     /// Timestamp of last persist.
     fn last_persist_at(&self) -> Instant;
@@ -84,7 +82,7 @@ impl FileStorage {
     /// Check the magic/version in the on-disk data and deserialize it, if possible.
     fn deser_sk_state(buf: &mut &[u8]) -> Result<SafeKeeperState> {
         // Read the version independent part
-        let magic = ReadBytesExt::read_u32::<LittleEndian>(buf)?;
+        let magic = buf.read_u32::<LittleEndian>()?;
         if magic != SK_MAGIC {
             bail!(
                 "bad control file magic: {:X}, expected {:X}",
@@ -92,7 +90,7 @@ impl FileStorage {
                 SK_MAGIC
             );
         }
-        let version = ReadBytesExt::read_u32::<LittleEndian>(buf)?;
+        let version = buf.read_u32::<LittleEndian>()?;
         if version == SK_FORMAT_VERSION {
             let res = SafeKeeperState::des(buf)?;
             return Ok(res);
@@ -112,7 +110,7 @@ impl FileStorage {
 
     /// Read in the control file.
     pub fn load_control_file<P: AsRef<Path>>(control_file_path: P) -> Result<SafeKeeperState> {
-        let mut control_file = std::fs::OpenOptions::new()
+        let mut control_file = OpenOptions::new()
             .read(true)
             .write(true)
             .open(&control_file_path)
@@ -161,31 +159,30 @@ impl Deref for FileStorage {
     }
 }
 
-#[async_trait::async_trait]
 impl Storage for FileStorage {
     /// persists state durably to underlying storage
     /// for description see https://lwn.net/Articles/457667/
-    async fn persist(&mut self, s: &SafeKeeperState) -> Result<()> {
+    fn persist(&mut self, s: &SafeKeeperState) -> Result<()> {
         let _timer = PERSIST_CONTROL_FILE_SECONDS.start_timer();
 
         // write data to safekeeper.control.partial
         let control_partial_path = self.timeline_dir.join(CONTROL_FILE_NAME_PARTIAL);
-        let mut control_partial = File::create(&control_partial_path).await.with_context(|| {
+        let mut control_partial = File::create(&control_partial_path).with_context(|| {
             format!(
                 "failed to create partial control file at: {}",
                 &control_partial_path.display()
             )
         })?;
         let mut buf: Vec<u8> = Vec::new();
-        WriteBytesExt::write_u32::<LittleEndian>(&mut buf, SK_MAGIC)?;
-        WriteBytesExt::write_u32::<LittleEndian>(&mut buf, SK_FORMAT_VERSION)?;
+        buf.write_u32::<LittleEndian>(SK_MAGIC)?;
+        buf.write_u32::<LittleEndian>(SK_FORMAT_VERSION)?;
         s.ser_into(&mut buf)?;
 
         // calculate checksum before resize
         let checksum = crc32c::crc32c(&buf);
         buf.extend_from_slice(&checksum.to_le_bytes());
 
-        control_partial.write_all(&buf).await.with_context(|| {
+        control_partial.write_all(&buf).with_context(|| {
             format!(
                 "failed to write safekeeper state into control file at: {}",
                 control_partial_path.display()
@@ -194,7 +191,7 @@ impl Storage for FileStorage {
 
         // fsync the file
         if !self.conf.no_sync {
-            control_partial.sync_all().await.with_context(|| {
+            control_partial.sync_all().with_context(|| {
                 format!(
                     "failed to sync partial control file at {}",
                     control_partial_path.display()
@@ -205,22 +202,21 @@ impl Storage for FileStorage {
         let control_path = self.timeline_dir.join(CONTROL_FILE_NAME);
 
         // rename should be atomic
-        fs::rename(&control_partial_path, &control_path).await?;
+        fs::rename(&control_partial_path, &control_path)?;
         // this sync is not required by any standard but postgres does this (see durable_rename)
         if !self.conf.no_sync {
-            let new_f = File::open(&control_path).await?;
-            new_f.sync_all().await.with_context(|| {
-                format!(
-                    "failed to sync control file at: {}",
-                    &control_path.display()
-                )
-            })?;
+            File::open(&control_path)
+                .and_then(|f| f.sync_all())
+                .with_context(|| {
+                    format!(
+                        "failed to sync control file at: {}",
+                        &control_path.display()
+                    )
+                })?;
 
             // fsync the directory (linux specific)
-            let tli_dir = File::open(&self.timeline_dir).await?;
-            tli_dir
-                .sync_all()
-                .await
+            File::open(&self.timeline_dir)
+                .and_then(|f| f.sync_all())
                 .context("failed to sync control file directory")?;
         }
 
@@ -240,6 +236,7 @@ mod test {
     use super::*;
     use crate::{safekeeper::SafeKeeperState, SafeKeeperConf};
     use anyhow::Result;
+    use std::fs;
     use utils::{id::TenantTimelineId, lsn::Lsn};
 
     fn stub_conf() -> SafeKeeperConf {
@@ -250,75 +247,59 @@ mod test {
         }
     }
 
-    async fn load_from_control_file(
+    fn load_from_control_file(
         conf: &SafeKeeperConf,
         ttid: &TenantTimelineId,
     ) -> Result<(FileStorage, SafeKeeperState)> {
-        fs::create_dir_all(conf.timeline_dir(ttid))
-            .await
-            .expect("failed to create timeline dir");
+        fs::create_dir_all(conf.timeline_dir(ttid)).expect("failed to create timeline dir");
         Ok((
             FileStorage::restore_new(ttid, conf)?,
             FileStorage::load_control_file_conf(conf, ttid)?,
         ))
     }
 
-    async fn create(
+    fn create(
         conf: &SafeKeeperConf,
         ttid: &TenantTimelineId,
     ) -> Result<(FileStorage, SafeKeeperState)> {
-        fs::create_dir_all(conf.timeline_dir(ttid))
-            .await
-            .expect("failed to create timeline dir");
+        fs::create_dir_all(conf.timeline_dir(ttid)).expect("failed to create timeline dir");
         let state = SafeKeeperState::empty();
         let storage = FileStorage::create_new(ttid, conf, state.clone())?;
         Ok((storage, state))
     }
 
-    #[tokio::test]
-    async fn test_read_write_safekeeper_state() {
+    #[test]
+    fn test_read_write_safekeeper_state() {
         let conf = stub_conf();
         let ttid = TenantTimelineId::generate();
         {
-            let (mut storage, mut state) =
-                create(&conf, &ttid).await.expect("failed to create state");
+            let (mut storage, mut state) = create(&conf, &ttid).expect("failed to create state");
             // change something
             state.commit_lsn = Lsn(42);
-            storage
-                .persist(&state)
-                .await
-                .expect("failed to persist state");
+            storage.persist(&state).expect("failed to persist state");
         }
 
-        let (_, state) = load_from_control_file(&conf, &ttid)
-            .await
-            .expect("failed to read state");
+        let (_, state) = load_from_control_file(&conf, &ttid).expect("failed to read state");
         assert_eq!(state.commit_lsn, Lsn(42));
     }
 
-    #[tokio::test]
-    async fn test_safekeeper_state_checksum_mismatch() {
+    #[test]
+    fn test_safekeeper_state_checksum_mismatch() {
         let conf = stub_conf();
         let ttid = TenantTimelineId::generate();
         {
-            let (mut storage, mut state) =
-                create(&conf, &ttid).await.expect("failed to read state");
+            let (mut storage, mut state) = create(&conf, &ttid).expect("failed to read state");
 
             // change something
             state.commit_lsn = Lsn(42);
-            storage
-                .persist(&state)
-                .await
-                .expect("failed to persist state");
+            storage.persist(&state).expect("failed to persist state");
         }
         let control_path = conf.timeline_dir(&ttid).join(CONTROL_FILE_NAME);
-        let mut data = fs::read(&control_path).await.unwrap();
+        let mut data = fs::read(&control_path).unwrap();
         data[0] += 1; // change the first byte of the file to fail checksum validation
-        fs::write(&control_path, &data)
-            .await
-            .expect("failed to write control file");
+        fs::write(&control_path, &data).expect("failed to write control file");
 
-        match load_from_control_file(&conf, &ttid).await {
+        match load_from_control_file(&conf, &ttid) {
             Err(err) => assert!(err
                 .to_string()
                 .contains("safekeeper control file checksum mismatch")),

safekeeper/src/debug_dump.rs

@@ -121,7 +121,7 @@ pub struct FileInfo {
 }
 
 /// Build debug dump response, using the provided [`Args`] filters.
-pub async fn build(args: Args) -> Result<Response> {
+pub fn build(args: Args) -> Result<Response> {
     let start_time = Utc::now();
     let timelines_count = GlobalTimelines::timelines_count();
 
@@ -155,7 +155,7 @@ pub async fn build(args: Args) -> Result<Response> {
         }
 
         let control_file = if args.dump_control_file {
-            let mut state = tli.get_state().await.1;
+            let mut state = tli.get_state().1;
             if !args.dump_term_history {
                 state.acceptor_state.term_history = TermHistory(vec![]);
             }
@@ -165,7 +165,7 @@ pub async fn build(args: Args) -> Result<Response> {
         };
 
         let memory = if args.dump_memory {
-            Some(tli.memory_dump().await)
+            Some(tli.memory_dump())
         } else {
             None
         };

safekeeper/src/handler.rs

@@ -256,14 +256,14 @@ impl SafekeeperPostgresHandler {
 
         let lsn = if self.is_walproposer_recovery() {
             // walproposer should get all local WAL until flush_lsn
-            tli.get_flush_lsn().await
+            tli.get_flush_lsn()
         } else {
             // other clients shouldn't get any uncommitted WAL
-            tli.get_state().await.0.commit_lsn
+            tli.get_state().0.commit_lsn
         }
         .to_string();
 
-        let sysid = tli.get_state().await.1.server.system_id.to_string();
+        let sysid = tli.get_state().1.server.system_id.to_string();
         let lsn_bytes = lsn.as_bytes();
         let tli = PG_TLI.to_string();
         let tli_bytes = tli.as_bytes();

safekeeper/src/http/mod.rs

@@ -2,18 +2,3 @@ pub mod routes;
 pub use routes::make_router;
 
 pub use safekeeper_api::models;
-
-use crate::SafeKeeperConf;
-
-pub async fn task_main(
-    conf: SafeKeeperConf,
-    http_listener: std::net::TcpListener,
-) -> anyhow::Result<()> {
-    let router = make_router(conf)
-        .build()
-        .map_err(|err| anyhow::anyhow!(err))?;
-    let service = utils::http::RouterService::new(router).unwrap();
-    let server = hyper::Server::from_tcp(http_listener)?;
-    server.serve(service).await?;
-    Ok(()) // unreachable
-}

safekeeper/src/http/routes.rs

@@ -13,7 +13,7 @@ use storage_broker::proto::SafekeeperTimelineInfo;
 use storage_broker::proto::TenantTimelineId as ProtoTenantTimelineId;
 use tokio::fs::File;
 use tokio::io::AsyncReadExt;
-use utils::http::endpoint::request_span;
+use tokio::task::JoinError;
 
 use crate::safekeeper::ServerInfo;
 use crate::safekeeper::Term;
@@ -116,8 +116,8 @@ async fn timeline_status_handler(request: Request<Body>) -> Result<Response<Body
     check_permission(&request, Some(ttid.tenant_id))?;
 
     let tli = GlobalTimelines::get(ttid).map_err(ApiError::from)?;
-    let (inmem, state) = tli.get_state().await;
-    let flush_lsn = tli.get_flush_lsn().await;
+    let (inmem, state) = tli.get_state();
+    let flush_lsn = tli.get_flush_lsn();
 
     let epoch = state.acceptor_state.get_epoch(flush_lsn);
     let term_history = state
@@ -232,11 +232,13 @@ async fn timeline_delete_force_handler(
     );
     check_permission(&request, Some(ttid.tenant_id))?;
     ensure_no_body(&mut request).await?;
-    // FIXME: `delete_force` can fail from both internal errors and bad requests. Add better
-    // error handling here when we're able to.
-    let resp = GlobalTimelines::delete_force(&ttid)
-        .await
-        .map_err(ApiError::InternalServerError)?;
+    let resp = tokio::task::spawn_blocking(move || {
+        // FIXME: `delete_force` can fail from both internal errors and bad requests. Add better
+        // error handling here when we're able to.
+        GlobalTimelines::delete_force(&ttid).map_err(ApiError::InternalServerError)
+    })
+    .await
+    .map_err(|e: JoinError| ApiError::InternalServerError(e.into()))??;
     json_response(StatusCode::OK, resp)
 }
 
@@ -248,11 +250,14 @@ async fn tenant_delete_force_handler(
     let tenant_id = parse_request_param(&request, "tenant_id")?;
     check_permission(&request, Some(tenant_id))?;
     ensure_no_body(&mut request).await?;
-    // FIXME: `delete_force_all_for_tenant` can return an error for multiple different reasons;
-    // Using an `InternalServerError` should be fixed when the types support it
-    let delete_info = GlobalTimelines::delete_force_all_for_tenant(&tenant_id)
-        .await
-        .map_err(ApiError::InternalServerError)?;
+    let delete_info = tokio::task::spawn_blocking(move || {
+        // FIXME: `delete_force_all_for_tenant` can return an error for multiple different reasons;
+        // Using an `InternalServerError` should be fixed when the types support it
+        GlobalTimelines::delete_force_all_for_tenant(&tenant_id)
+            .map_err(ApiError::InternalServerError)
+    })
+    .await
+    .map_err(|e: JoinError| ApiError::InternalServerError(e.into()))??;
     json_response(
         StatusCode::OK,
         delete_info
@@ -348,9 +353,11 @@ async fn dump_debug_handler(mut request: Request<Body>) -> Result<Response<Body>
         timeline_id,
     };
 
-    let resp = debug_dump::build(args)
-        .await
-        .map_err(ApiError::InternalServerError)?;
+    let resp = tokio::task::spawn_blocking(move || {
+        debug_dump::build(args).map_err(ApiError::InternalServerError)
+    })
+    .await
+    .map_err(|e: JoinError| ApiError::InternalServerError(e.into()))??;
 
     // TODO: use streaming response
     json_response(StatusCode::OK, resp)
@@ -379,32 +386,29 @@ pub fn make_router(conf: SafeKeeperConf) -> RouterBuilder<hyper::Body, ApiError>
     router
         .data(Arc::new(conf))
         .data(auth)
-        .get("/v1/status", |r| request_span(r, status_handler))
+        .get("/v1/status", status_handler)
         // Will be used in the future instead of implicit timeline creation
-        .post("/v1/tenant/timeline", |r| {
-            request_span(r, timeline_create_handler)
-        })
-        .get("/v1/tenant/:tenant_id/timeline/:timeline_id", |r| {
-            request_span(r, timeline_status_handler)
-        })
-        .delete("/v1/tenant/:tenant_id/timeline/:timeline_id", |r| {
-            request_span(r, timeline_delete_force_handler)
-        })
-        .delete("/v1/tenant/:tenant_id", |r| {
-            request_span(r, tenant_delete_force_handler)
-        })
-        .post("/v1/pull_timeline", |r| {
-            request_span(r, timeline_pull_handler)
-        })
+        .post("/v1/tenant/timeline", timeline_create_handler)
+        .get(
+            "/v1/tenant/:tenant_id/timeline/:timeline_id",
+            timeline_status_handler,
+        )
+        .delete(
+            "/v1/tenant/:tenant_id/timeline/:timeline_id",
+            timeline_delete_force_handler,
+        )
+        .delete("/v1/tenant/:tenant_id", tenant_delete_force_handler)
+        .post("/v1/pull_timeline", timeline_pull_handler)
         .get(
             "/v1/tenant/:tenant_id/timeline/:timeline_id/file/:filename",
-            |r| request_span(r, timeline_files_handler),
+            timeline_files_handler,
         )
         // for tests
-        .post("/v1/record_safekeeper_info/:tenant_id/:timeline_id", |r| {
-            request_span(r, record_safekeeper_info)
-        })
-        .get("/v1/debug_dump", |r| request_span(r, dump_debug_handler))
+        .post(
+            "/v1/record_safekeeper_info/:tenant_id/:timeline_id",
+            record_safekeeper_info,
+        )
+        .get("/v1/debug_dump", dump_debug_handler)
 }
 
 #[cfg(test)]

safekeeper/src/json_ctrl.rs

@@ -73,12 +73,12 @@ pub async fn handle_json_ctrl<IO: AsyncRead + AsyncWrite + Unpin>(
 
     // if send_proposer_elected is true, we need to update local history
     if append_request.send_proposer_elected {
-        send_proposer_elected(&tli, append_request.term, append_request.epoch_start_lsn).await?;
+        send_proposer_elected(&tli, append_request.term, append_request.epoch_start_lsn)?;
     }
 
-    let inserted_wal = append_logical_message(&tli, append_request).await?;
+    let inserted_wal = append_logical_message(&tli, append_request)?;
     let response = AppendResult {
-        state: tli.get_state().await.1,
+        state: tli.get_state().1,
         inserted_wal,
     };
     let response_data = serde_json::to_vec(&response)
@@ -114,9 +114,9 @@ async fn prepare_safekeeper(
     .await
 }
 
-async fn send_proposer_elected(tli: &Arc<Timeline>, term: Term, lsn: Lsn) -> anyhow::Result<()> {
+fn send_proposer_elected(tli: &Arc<Timeline>, term: Term, lsn: Lsn) -> anyhow::Result<()> {
     // add new term to existing history
-    let history = tli.get_state().await.1.acceptor_state.term_history;
+    let history = tli.get_state().1.acceptor_state.term_history;
     let history = history.up_to(lsn.checked_sub(1u64).unwrap());
     let mut history_entries = history.0;
     history_entries.push(TermSwitchEntry { term, lsn });
@@ -129,7 +129,7 @@ async fn send_proposer_elected(tli: &Arc<Timeline>, term: Term, lsn: Lsn) -> any
         timeline_start_lsn: lsn,
     });
 
-    tli.process_msg(&proposer_elected_request).await?;
+    tli.process_msg(&proposer_elected_request)?;
     Ok(())
 }
 
@@ -142,12 +142,12 @@ pub struct InsertedWAL {
 
 /// Extend local WAL with new LogicalMessage record. To do that,
 /// create AppendRequest with new WAL and pass it to safekeeper.
-pub async fn append_logical_message(
+pub fn append_logical_message(
     tli: &Arc<Timeline>,
     msg: &AppendLogicalMessage,
 ) -> anyhow::Result<InsertedWAL> {
     let wal_data = encode_logical_message(&msg.lm_prefix, &msg.lm_message);
-    let sk_state = tli.get_state().await.1;
+    let sk_state = tli.get_state().1;
 
     let begin_lsn = msg.begin_lsn;
     let end_lsn = begin_lsn + wal_data.len() as u64;
@@ -171,7 +171,7 @@ pub async fn append_logical_message(
         wal_data: Bytes::from(wal_data),
     });
 
-    let response = tli.process_msg(&append_request).await?;
+    let response = tli.process_msg(&append_request)?;
 
     let append_response = match response {
         Some(AcceptorProposerMessage::AppendResponse(resp)) => resp,

safekeeper/src/lib.rs

@@ -1,6 +1,4 @@
-use once_cell::sync::Lazy;
 use remote_storage::RemoteStorageConfig;
-use tokio::runtime::Runtime;
 
 use std::path::PathBuf;
 use std::time::Duration;
@@ -38,6 +36,7 @@ pub mod defaults {
         DEFAULT_PG_LISTEN_PORT,
     };
 
+    pub const DEFAULT_WAL_BACKUP_RUNTIME_THREADS: usize = 8;
     pub const DEFAULT_HEARTBEAT_TIMEOUT: &str = "5000ms";
     pub const DEFAULT_MAX_OFFLOADER_LAG_BYTES: u64 = 128 * (1 << 20);
 }
@@ -61,10 +60,10 @@ pub struct SafeKeeperConf {
     pub heartbeat_timeout: Duration,
     pub remote_storage: Option<RemoteStorageConfig>,
     pub max_offloader_lag_bytes: u64,
+    pub backup_runtime_threads: Option<usize>,
     pub backup_parallel_jobs: usize,
     pub wal_backup_enabled: bool,
     pub auth: Option<Arc<JwtAuth>>,
-    pub current_thread_runtime: bool,
 }
 
 impl SafeKeeperConf {
@@ -93,64 +92,12 @@ impl SafeKeeperConf {
                 .parse()
                 .expect("failed to parse default broker endpoint"),
             broker_keepalive_interval: Duration::from_secs(5),
+            backup_runtime_threads: None,
             wal_backup_enabled: true,
             backup_parallel_jobs: 1,
             auth: None,
             heartbeat_timeout: Duration::new(5, 0),
             max_offloader_lag_bytes: defaults::DEFAULT_MAX_OFFLOADER_LAG_BYTES,
-            current_thread_runtime: false,
         }
     }
 }
-
-// Tokio runtimes.
-pub static WAL_SERVICE_RUNTIME: Lazy<Runtime> = Lazy::new(|| {
-    tokio::runtime::Builder::new_multi_thread()
-        .thread_name("WAL service worker")
-        .enable_all()
-        .build()
-        .expect("Failed to create WAL service runtime")
-});
-
-pub static HTTP_RUNTIME: Lazy<Runtime> = Lazy::new(|| {
-    tokio::runtime::Builder::new_multi_thread()
-        .thread_name("HTTP worker")
-        .enable_all()
-        .build()
-        .expect("Failed to create WAL service runtime")
-});
-
-pub static BROKER_RUNTIME: Lazy<Runtime> = Lazy::new(|| {
-    tokio::runtime::Builder::new_multi_thread()
-        .thread_name("broker worker")
-        .worker_threads(2) // there are only 2 tasks, having more threads doesn't make sense
-        .enable_all()
-        .build()
-        .expect("Failed to create broker runtime")
-});
-
-pub static WAL_REMOVER_RUNTIME: Lazy<Runtime> = Lazy::new(|| {
-    tokio::runtime::Builder::new_multi_thread()
-        .thread_name("WAL remover")
-        .worker_threads(1)
-        .enable_all()
-        .build()
-        .expect("Failed to create broker runtime")
-});
-
-pub static WAL_BACKUP_RUNTIME: Lazy<Runtime> = Lazy::new(|| {
-    tokio::runtime::Builder::new_multi_thread()
-        .thread_name("WAL backup worker")
-        .enable_all()
-        .build()
-        .expect("Failed to create WAL backup runtime")
-});
-
-pub static METRICS_SHIFTER_RUNTIME: Lazy<Runtime> = Lazy::new(|| {
-    tokio::runtime::Builder::new_multi_thread()
-        .thread_name("metric shifter")
-        .worker_threads(1)
-        .enable_all()
-        .build()
-        .expect("Failed to create broker runtime")
-});

safekeeper/src/metrics.rs

@@ -7,7 +7,6 @@ use std::{
 
 use ::metrics::{register_histogram, GaugeVec, Histogram, IntGauge, DISK_WRITE_SECONDS_BUCKETS};
 use anyhow::Result;
-use futures::Future;
 use metrics::{
     core::{AtomicU64, Collector, Desc, GenericCounter, GenericGaugeVec, Opts},
     proto::MetricFamily,
@@ -293,17 +292,14 @@ impl WalStorageMetrics {
     }
 }
 
-/// Accepts async function that returns empty anyhow result, and returns the duration of its execution.
-pub async fn time_io_closure<E: Into<anyhow::Error>>(
-    closure: impl Future<Output = Result<(), E>>,
-) -> Result<f64> {
+/// Accepts a closure that returns a result, and returns the duration of the closure.
+pub fn time_io_closure(closure: impl FnOnce() -> Result<()>) -> Result<f64> {
     let start = std::time::Instant::now();
-    closure.await.map_err(|e| e.into())?;
+    closure()?;
     Ok(start.elapsed().as_secs_f64())
 }
 
 /// Metrics for a single timeline.
-#[derive(Clone)]
 pub struct FullTimelineInfo {
     pub ttid: TenantTimelineId,
     pub ps_feedback: PageserverFeedback,
@@ -579,19 +575,13 @@ impl Collector for TimelineCollector {
         let timelines = GlobalTimelines::get_all();
         let timelines_count = timelines.len();
 
-        // Prometheus Collector is sync, and data is stored under async lock. To
-        // bridge the gap with a crutch, collect data in spawned thread with
-        // local tokio runtime.
-        let infos = std::thread::spawn(|| {
-            let rt = tokio::runtime::Builder::new_current_thread()
-                .build()
-                .expect("failed to create rt");
-            rt.block_on(collect_timeline_metrics())
-        })
-        .join()
-        .expect("collect_timeline_metrics thread panicked");
+        for arc_tli in timelines {
+            let tli = arc_tli.info_for_metrics();
+            if tli.is_none() {
+                continue;
+            }
+            let tli = tli.unwrap();
 
-        for tli in &infos {
             let tenant_id = tli.ttid.tenant_id.to_string();
             let timeline_id = tli.ttid.timeline_id.to_string();
             let labels = &[tenant_id.as_str(), timeline_id.as_str()];
@@ -692,15 +682,3 @@ impl Collector for TimelineCollector {
         mfs
     }
 }
-
-async fn collect_timeline_metrics() -> Vec<FullTimelineInfo> {
-    let mut res = vec![];
-    let timelines = GlobalTimelines::get_all();
-
-    for tli in timelines {
-        if let Some(info) = tli.info_for_metrics().await {
-            res.push(info);
-        }
-    }
-    res
-}

safekeeper/src/pull_timeline.rs

@@ -231,7 +231,7 @@ async fn pull_timeline(status: TimelineStatus, host: String) -> Result<Response>
     info!(
         "Loaded timeline {}, flush_lsn={}",
         ttid,
-        tli.get_flush_lsn().await
+        tli.get_flush_lsn()
     );
 
     Ok(Response {

safekeeper/src/receive_wal.rs

@@ -18,14 +18,15 @@ use postgres_backend::QueryError;
 use pq_proto::BeMessage;
 use std::net::SocketAddr;
 use std::sync::Arc;
+use std::thread;
+use std::thread::JoinHandle;
 use tokio::io::AsyncRead;
 use tokio::io::AsyncWrite;
 use tokio::sync::mpsc::channel;
 use tokio::sync::mpsc::error::TryRecvError;
 use tokio::sync::mpsc::Receiver;
 use tokio::sync::mpsc::Sender;
-use tokio::task;
-use tokio::task::JoinHandle;
+use tokio::task::spawn_blocking;
 use tokio::time::Duration;
 use tokio::time::Instant;
 use tracing::*;
@@ -96,7 +97,7 @@ impl SafekeeperPostgresHandler {
                 Err(res.expect_err("no error with WalAcceptor not spawn"))
             }
             Some(handle) => {
-                let wal_acceptor_res = handle.await;
+                let wal_acceptor_res = handle.join();
 
                 // If there was any network error, return it.
                 res?;
@@ -106,7 +107,7 @@ impl SafekeeperPostgresHandler {
                     Ok(Ok(_)) => Ok(()), // can't happen currently; would be if we add graceful termination
                     Ok(Err(e)) => Err(CopyStreamHandlerEnd::Other(e.context("WAL acceptor"))),
                     Err(_) => Err(CopyStreamHandlerEnd::Other(anyhow!(
-                        "WalAcceptor task panicked",
+                        "WalAcceptor thread panicked",
                     ))),
                 }
             }
@@ -153,12 +154,10 @@ impl<'a, IO: AsyncRead + AsyncWrite + Unpin> NetworkReader<'a, IO> {
             }
         };
 
-        *self.acceptor_handle = Some(WalAcceptor::spawn(
-            tli.clone(),
-            msg_rx,
-            reply_tx,
-            self.conn_id,
-        ));
+        *self.acceptor_handle = Some(
+            WalAcceptor::spawn(tli.clone(), msg_rx, reply_tx, self.conn_id)
+                .context("spawn WalAcceptor thread")?,
+        );
 
         // Forward all messages to WalAcceptor
         read_network_loop(self.pgb_reader, msg_tx, next_msg).await
@@ -227,19 +226,28 @@ impl WalAcceptor {
         msg_rx: Receiver<ProposerAcceptorMessage>,
         reply_tx: Sender<AcceptorProposerMessage>,
         conn_id: ConnectionId,
-    ) -> JoinHandle<anyhow::Result<()>> {
-        task::spawn(async move {
-            let mut wa = WalAcceptor {
-                tli,
-                msg_rx,
-                reply_tx,
-            };
+    ) -> anyhow::Result<JoinHandle<anyhow::Result<()>>> {
+        let thread_name = format!("WAL acceptor {}", tli.ttid);
+        thread::Builder::new()
+            .name(thread_name)
+            .spawn(move || -> anyhow::Result<()> {
+                let mut wa = WalAcceptor {
+                    tli,
+                    msg_rx,
+                    reply_tx,
+                };
 
-            let span_ttid = wa.tli.ttid; // satisfy borrow checker
-            wa.run()
-                .instrument(info_span!("WAL acceptor", cid = %conn_id, ttid = %span_ttid))
-                .await
-        })
+                let runtime = tokio::runtime::Builder::new_current_thread()
+                    .enable_all()
+                    .build()?;
+
+                let span_ttid = wa.tli.ttid; // satisfy borrow checker
+                runtime.block_on(
+                    wa.run()
+                        .instrument(info_span!("WAL acceptor", cid = %conn_id, ttid = %span_ttid)),
+                )
+            })
+            .map_err(anyhow::Error::from)
     }
 
     /// The main loop. Returns Ok(()) if either msg_rx or reply_tx got closed;
@@ -273,7 +281,7 @@ impl WalAcceptor {
                 while let ProposerAcceptorMessage::AppendRequest(append_request) = next_msg {
                     let noflush_msg = ProposerAcceptorMessage::NoFlushAppendRequest(append_request);
 
-                    if let Some(reply) = self.tli.process_msg(&noflush_msg).await? {
+                    if let Some(reply) = self.tli.process_msg(&noflush_msg)? {
                         if self.reply_tx.send(reply).await.is_err() {
                             return Ok(()); // chan closed, streaming terminated
                         }
@@ -292,12 +300,10 @@ impl WalAcceptor {
                 }
 
                 // flush all written WAL to the disk
-                self.tli
-                    .process_msg(&ProposerAcceptorMessage::FlushWAL)
-                    .await?
+                self.tli.process_msg(&ProposerAcceptorMessage::FlushWAL)?
             } else {
                 // process message other than AppendRequest
-                self.tli.process_msg(&next_msg).await?
+                self.tli.process_msg(&next_msg)?
             };
 
             if let Some(reply) = reply_msg {
@@ -320,8 +326,8 @@ impl Drop for ComputeConnectionGuard {
         let tli = self.timeline.clone();
         // tokio forbids to call blocking_send inside the runtime, and see
         // comments in on_compute_disconnect why we call blocking_send.
-        tokio::spawn(async move {
-            if let Err(e) = tli.on_compute_disconnect().await {
+        spawn_blocking(move || {
+            if let Err(e) = tli.on_compute_disconnect() {
                 error!("failed to unregister compute connection: {}", e);
             }
         });

safekeeper/src/remove_wal.rs

@@ -1,36 +1,29 @@
 //! Thread removing old WAL.
 
-use std::time::Duration;
+use std::{thread, time::Duration};
 
-use tokio::time::sleep;
 use tracing::*;
 
 use crate::{GlobalTimelines, SafeKeeperConf};
 
-pub async fn task_main(conf: SafeKeeperConf) -> anyhow::Result<()> {
+pub fn thread_main(conf: SafeKeeperConf) {
     let wal_removal_interval = Duration::from_millis(5000);
     loop {
         let tlis = GlobalTimelines::get_all();
         for tli in &tlis {
-            if !tli.is_active().await {
+            if !tli.is_active() {
                 continue;
             }
             let ttid = tli.ttid;
-            if let Err(e) = tli
-                .maybe_persist_control_file()
-                .instrument(info_span!("", tenant = %ttid.tenant_id, timeline = %ttid.timeline_id))
-                .await
-            {
+            let _enter =
+                info_span!("", tenant = %ttid.tenant_id, timeline = %ttid.timeline_id).entered();
+            if let Err(e) = tli.maybe_pesist_control_file() {
                 warn!("failed to persist control file: {e}");
             }
-            if let Err(e) = tli
-                .remove_old_wal(conf.wal_backup_enabled)
-                .instrument(info_span!("", tenant = %ttid.tenant_id, timeline = %ttid.timeline_id))
-                .await
-            {
-                error!("failed to remove WAL: {}", e);
+            if let Err(e) = tli.remove_old_wal(conf.wal_backup_enabled) {
+                warn!("failed to remove WAL: {}", e);
             }
         }
-        sleep(wal_removal_interval).await;
+        thread::sleep(wal_removal_interval)
     }
 }

safekeeper/src/safekeeper.rs

@@ -568,27 +568,25 @@ where
 
     /// Process message from proposer and possibly form reply. Concurrent
     /// callers must exclude each other.
-    pub async fn process_msg(
+    pub fn process_msg(
         &mut self,
         msg: &ProposerAcceptorMessage,
     ) -> Result<Option<AcceptorProposerMessage>> {
         match msg {
-            ProposerAcceptorMessage::Greeting(msg) => self.handle_greeting(msg).await,
-            ProposerAcceptorMessage::VoteRequest(msg) => self.handle_vote_request(msg).await,
-            ProposerAcceptorMessage::Elected(msg) => self.handle_elected(msg).await,
-            ProposerAcceptorMessage::AppendRequest(msg) => {
-                self.handle_append_request(msg, true).await
-            }
+            ProposerAcceptorMessage::Greeting(msg) => self.handle_greeting(msg),
+            ProposerAcceptorMessage::VoteRequest(msg) => self.handle_vote_request(msg),
+            ProposerAcceptorMessage::Elected(msg) => self.handle_elected(msg),
+            ProposerAcceptorMessage::AppendRequest(msg) => self.handle_append_request(msg, true),
             ProposerAcceptorMessage::NoFlushAppendRequest(msg) => {
-                self.handle_append_request(msg, false).await
+                self.handle_append_request(msg, false)
             }
-            ProposerAcceptorMessage::FlushWAL => self.handle_flush().await,
+            ProposerAcceptorMessage::FlushWAL => self.handle_flush(),
         }
     }
 
     /// Handle initial message from proposer: check its sanity and send my
     /// current term.
-    async fn handle_greeting(
+    fn handle_greeting(
         &mut self,
         msg: &ProposerGreeting,
     ) -> Result<Option<AcceptorProposerMessage>> {
@@ -651,7 +649,7 @@ where
             if msg.pg_version != UNKNOWN_SERVER_VERSION {
                 state.server.pg_version = msg.pg_version;
             }
-            self.state.persist(&state).await?;
+            self.state.persist(&state)?;
         }
 
         info!(
@@ -666,7 +664,7 @@ where
     }
 
     /// Give vote for the given term, if we haven't done that previously.
-    async fn handle_vote_request(
+    fn handle_vote_request(
         &mut self,
         msg: &VoteRequest,
     ) -> Result<Option<AcceptorProposerMessage>> {
@@ -680,7 +678,7 @@ where
         // handle_elected instead. Currently not a big deal, as proposer is the
         // only source of WAL; with peer2peer recovery it would be more
         // important.
-        self.wal_store.flush_wal().await?;
+        self.wal_store.flush_wal()?;
         // initialize with refusal
         let mut resp = VoteResponse {
             term: self.state.acceptor_state.term,
@@ -694,7 +692,7 @@ where
             let mut state = self.state.clone();
             state.acceptor_state.term = msg.term;
             // persist vote before sending it out
-            self.state.persist(&state).await?;
+            self.state.persist(&state)?;
 
             resp.term = self.state.acceptor_state.term;
             resp.vote_given = true as u64;
@@ -717,15 +715,12 @@ where
         ar
     }
 
-    async fn handle_elected(
-        &mut self,
-        msg: &ProposerElected,
-    ) -> Result<Option<AcceptorProposerMessage>> {
+    fn handle_elected(&mut self, msg: &ProposerElected) -> Result<Option<AcceptorProposerMessage>> {
         info!("received ProposerElected {:?}", msg);
         if self.state.acceptor_state.term < msg.term {
             let mut state = self.state.clone();
             state.acceptor_state.term = msg.term;
-            self.state.persist(&state).await?;
+            self.state.persist(&state)?;
         }
 
         // If our term is higher, ignore the message (next feedback will inform the compute)
@@ -755,7 +750,7 @@ where
         // intersection of our history and history from msg
 
         // truncate wal, update the LSNs
-        self.wal_store.truncate_wal(msg.start_streaming_at).await?;
+        self.wal_store.truncate_wal(msg.start_streaming_at)?;
 
         // and now adopt term history from proposer
         {
@@ -789,7 +784,7 @@ where
             self.inmem.backup_lsn = max(self.inmem.backup_lsn, state.timeline_start_lsn);
 
             state.acceptor_state.term_history = msg.term_history.clone();
-            self.persist_control_file(state).await?;
+            self.persist_control_file(state)?;
         }
 
         info!("start receiving WAL since {:?}", msg.start_streaming_at);
@@ -801,7 +796,7 @@ where
     ///
     /// Note: it is assumed that 'WAL we have is from the right term' check has
     /// already been done outside.
-    async fn update_commit_lsn(&mut self, mut candidate: Lsn) -> Result<()> {
+    fn update_commit_lsn(&mut self, mut candidate: Lsn) -> Result<()> {
         // Both peers and walproposer communicate this value, we might already
         // have a fresher (higher) version.
         candidate = max(candidate, self.inmem.commit_lsn);
@@ -823,32 +818,29 @@ where
         // that we receive new epoch_start_lsn, and we still need to sync
         // control file in this case.
         if commit_lsn == self.epoch_start_lsn && self.state.commit_lsn != commit_lsn {
-            self.persist_control_file(self.state.clone()).await?;
+            self.persist_control_file(self.state.clone())?;
         }
 
         Ok(())
     }
 
     /// Persist control file to disk, called only after timeline creation (bootstrap).
-    pub async fn persist(&mut self) -> Result<()> {
-        self.persist_control_file(self.state.clone()).await
+    pub fn persist(&mut self) -> Result<()> {
+        self.persist_control_file(self.state.clone())
     }
 
     /// Persist in-memory state to the disk, taking other data from state.
-    async fn persist_control_file(&mut self, mut state: SafeKeeperState) -> Result<()> {
+    fn persist_control_file(&mut self, mut state: SafeKeeperState) -> Result<()> {
         state.commit_lsn = self.inmem.commit_lsn;
         state.backup_lsn = self.inmem.backup_lsn;
         state.peer_horizon_lsn = self.inmem.peer_horizon_lsn;
         state.proposer_uuid = self.inmem.proposer_uuid;
-        self.state.persist(&state).await
+        self.state.persist(&state)
     }
 
     /// Persist control file if there is something to save and enough time
     /// passed after the last save.
-    pub async fn maybe_persist_control_file(
-        &mut self,
-        inmem_remote_consistent_lsn: Lsn,
-    ) -> Result<()> {
+    pub fn maybe_persist_control_file(&mut self, inmem_remote_consistent_lsn: Lsn) -> Result<()> {
         const CF_SAVE_INTERVAL: Duration = Duration::from_secs(300);
         if self.state.last_persist_at().elapsed() < CF_SAVE_INTERVAL {
             return Ok(());
@@ -860,7 +852,7 @@ where
         if need_persist {
             let mut state = self.state.clone();
             state.remote_consistent_lsn = inmem_remote_consistent_lsn;
-            self.persist_control_file(state).await?;
+            self.persist_control_file(state)?;
             trace!("saved control file: {CF_SAVE_INTERVAL:?} passed");
         }
         Ok(())
@@ -868,7 +860,7 @@ where
 
     /// Handle request to append WAL.
     #[allow(clippy::comparison_chain)]
-    async fn handle_append_request(
+    fn handle_append_request(
         &mut self,
         msg: &AppendRequest,
         require_flush: bool,
@@ -891,19 +883,17 @@ where
 
         // do the job
         if !msg.wal_data.is_empty() {
-            self.wal_store
-                .write_wal(msg.h.begin_lsn, &msg.wal_data)
-                .await?;
+            self.wal_store.write_wal(msg.h.begin_lsn, &msg.wal_data)?;
         }
 
         // flush wal to the disk, if required
         if require_flush {
-            self.wal_store.flush_wal().await?;
+            self.wal_store.flush_wal()?;
         }
 
         // Update commit_lsn.
         if msg.h.commit_lsn != Lsn(0) {
-            self.update_commit_lsn(msg.h.commit_lsn).await?;
+            self.update_commit_lsn(msg.h.commit_lsn)?;
         }
         // Value calculated by walproposer can always lag:
         // - safekeepers can forget inmem value and send to proposer lower
@@ -919,7 +909,7 @@ where
         if self.state.peer_horizon_lsn + (self.state.server.wal_seg_size as u64)
             < self.inmem.peer_horizon_lsn
         {
-            self.persist_control_file(self.state.clone()).await?;
+            self.persist_control_file(self.state.clone())?;
         }
 
         trace!(
@@ -941,15 +931,15 @@ where
     }
 
     /// Flush WAL to disk. Return AppendResponse with latest LSNs.
-    async fn handle_flush(&mut self) -> Result<Option<AcceptorProposerMessage>> {
-        self.wal_store.flush_wal().await?;
+    fn handle_flush(&mut self) -> Result<Option<AcceptorProposerMessage>> {
+        self.wal_store.flush_wal()?;
         Ok(Some(AcceptorProposerMessage::AppendResponse(
             self.append_response(),
         )))
     }
 
     /// Update timeline state with peer safekeeper data.
-    pub async fn record_safekeeper_info(&mut self, sk_info: &SafekeeperTimelineInfo) -> Result<()> {
+    pub fn record_safekeeper_info(&mut self, sk_info: &SafekeeperTimelineInfo) -> Result<()> {
         let mut sync_control_file = false;
 
         if (Lsn(sk_info.commit_lsn) != Lsn::INVALID) && (sk_info.last_log_term != INVALID_TERM) {
@@ -957,7 +947,7 @@ where
             // commit_lsn if our history matches (is part of) history of advanced
             // commit_lsn provider.
             if sk_info.last_log_term == self.get_epoch() {
-                self.update_commit_lsn(Lsn(sk_info.commit_lsn)).await?;
+                self.update_commit_lsn(Lsn(sk_info.commit_lsn))?;
             }
         }
 
@@ -983,7 +973,7 @@ where
             // Note: we could make remote_consistent_lsn update in cf common by
             // storing Arc to walsenders in Safekeeper.
             state.remote_consistent_lsn = new_remote_consistent_lsn;
-            self.persist_control_file(state).await?;
+            self.persist_control_file(state)?;
         }
         Ok(())
     }
@@ -1007,7 +997,6 @@ where
 
 #[cfg(test)]
 mod tests {
-    use futures::future::BoxFuture;
     use postgres_ffi::WAL_SEGMENT_SIZE;
 
     use super::*;
@@ -1019,9 +1008,8 @@ mod tests {
         persisted_state: SafeKeeperState,
     }
 
-    #[async_trait::async_trait]
     impl control_file::Storage for InMemoryState {
-        async fn persist(&mut self, s: &SafeKeeperState) -> Result<()> {
+        fn persist(&mut self, s: &SafeKeeperState) -> Result<()> {
             self.persisted_state = s.clone();
             Ok(())
         }
@@ -1051,28 +1039,27 @@ mod tests {
         lsn: Lsn,
     }
 
-    #[async_trait::async_trait]
     impl wal_storage::Storage for DummyWalStore {
         fn flush_lsn(&self) -> Lsn {
             self.lsn
         }
 
-        async fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()> {
+        fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()> {
             self.lsn = startpos + buf.len() as u64;
             Ok(())
         }
 
-        async fn truncate_wal(&mut self, end_pos: Lsn) -> Result<()> {
+        fn truncate_wal(&mut self, end_pos: Lsn) -> Result<()> {
             self.lsn = end_pos;
             Ok(())
         }
 
-        async fn flush_wal(&mut self) -> Result<()> {
+        fn flush_wal(&mut self) -> Result<()> {
             Ok(())
         }
 
-        fn remove_up_to(&self, _segno_up_to: XLogSegNo) -> BoxFuture<'static, anyhow::Result<()>> {
-            Box::pin(async { Ok(()) })
+        fn remove_up_to(&self) -> Box<dyn Fn(XLogSegNo) -> Result<()>> {
+            Box::new(move |_segno_up_to: XLogSegNo| Ok(()))
         }
 
         fn get_metrics(&self) -> crate::metrics::WalStorageMetrics {
@@ -1080,8 +1067,8 @@ mod tests {
         }
     }
 
-    #[tokio::test]
-    async fn test_voting() {
+    #[test]
+    fn test_voting() {
         let storage = InMemoryState {
             persisted_state: test_sk_state(),
         };
@@ -1090,7 +1077,7 @@ mod tests {
 
         // check voting for 1 is ok
         let vote_request = ProposerAcceptorMessage::VoteRequest(VoteRequest { term: 1 });
-        let mut vote_resp = sk.process_msg(&vote_request).await;
+        let mut vote_resp = sk.process_msg(&vote_request);
         match vote_resp.unwrap() {
             Some(AcceptorProposerMessage::VoteResponse(resp)) => assert!(resp.vote_given != 0),
             r => panic!("unexpected response: {:?}", r),
@@ -1105,15 +1092,15 @@ mod tests {
         sk = SafeKeeper::new(storage, sk.wal_store, NodeId(0)).unwrap();
 
         // and ensure voting second time for 1 is not ok
-        vote_resp = sk.process_msg(&vote_request).await;
+        vote_resp = sk.process_msg(&vote_request);
         match vote_resp.unwrap() {
             Some(AcceptorProposerMessage::VoteResponse(resp)) => assert!(resp.vote_given == 0),
             r => panic!("unexpected response: {:?}", r),
         }
     }
 
-    #[tokio::test]
-    async fn test_epoch_switch() {
+    #[test]
+    fn test_epoch_switch() {
         let storage = InMemoryState {
             persisted_state: test_sk_state(),
         };
@@ -1145,13 +1132,10 @@ mod tests {
             timeline_start_lsn: Lsn(0),
         };
         sk.process_msg(&ProposerAcceptorMessage::Elected(pem))
-            .await
             .unwrap();
 
         // check that AppendRequest before epochStartLsn doesn't switch epoch
-        let resp = sk
-            .process_msg(&ProposerAcceptorMessage::AppendRequest(append_request))
-            .await;
+        let resp = sk.process_msg(&ProposerAcceptorMessage::AppendRequest(append_request));
         assert!(resp.is_ok());
         assert_eq!(sk.get_epoch(), 0);
 
@@ -1162,11 +1146,9 @@ mod tests {
             h: ar_hdr,
             wal_data: Bytes::from_static(b"b"),
         };
-        let resp = sk
-            .process_msg(&ProposerAcceptorMessage::AppendRequest(append_request))
-            .await;
+        let resp = sk.process_msg(&ProposerAcceptorMessage::AppendRequest(append_request));
         assert!(resp.is_ok());
-        sk.wal_store.truncate_wal(Lsn(3)).await.unwrap(); // imitate the complete record at 3 %)
+        sk.wal_store.truncate_wal(Lsn(3)).unwrap(); // imitate the complete record at 3 %)
         assert_eq!(sk.get_epoch(), 1);
     }
 }

safekeeper/src/send_wal.rs

@@ -396,7 +396,7 @@ impl SafekeeperPostgresHandler {
         // on this safekeeper itself. That's ok as (old) proposer will never be
         // able to commit such WAL.
         let stop_pos: Option<Lsn> = if self.is_walproposer_recovery() {
-            let wal_end = tli.get_flush_lsn().await;
+            let wal_end = tli.get_flush_lsn();
             Some(wal_end)
         } else {
             None
@@ -418,7 +418,7 @@ impl SafekeeperPostgresHandler {
         // switch to copy
         pgb.write_message(&BeMessage::CopyBothResponse).await?;
 
-        let (_, persisted_state) = tli.get_state().await;
+        let (_, persisted_state) = tli.get_state();
         let wal_reader = WalReader::new(
             self.conf.workdir.clone(),
             self.conf.timeline_dir(&tli.ttid),
@@ -562,7 +562,7 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> WalSender<'_, IO> {
                 .walsenders
                 .get_ws_remote_consistent_lsn(self.ws_guard.id)
             {
-                if self.tli.should_walsender_stop(remote_consistent_lsn).await {
+                if self.tli.should_walsender_stop(remote_consistent_lsn) {
                     // Terminate if there is nothing more to send.
                     return Err(CopyStreamHandlerEnd::ServerInitiated(format!(
                         "ending streaming to {:?} at {}, receiver is caughtup and there is no computes",

safekeeper/src/timeline.rs

@@ -2,13 +2,12 @@
 //! to glue together SafeKeeper and all other background services.
 
 use anyhow::{anyhow, bail, Result};
+use parking_lot::{Mutex, MutexGuard};
 use postgres_ffi::XLogSegNo;
-use tokio::fs;
 
 use std::cmp::max;
 use std::path::PathBuf;
 use std::sync::Arc;
-use tokio::sync::{Mutex, MutexGuard};
 use tokio::{
     sync::{mpsc::Sender, watch},
     time::Instant,
@@ -287,9 +286,8 @@ pub struct Timeline {
     commit_lsn_watch_tx: watch::Sender<Lsn>,
     commit_lsn_watch_rx: watch::Receiver<Lsn>,
 
-    /// Safekeeper and other state, that should remain consistent and
-    /// synchronized with the disk. This is tokio mutex as we write WAL to disk
-    /// while holding it, ensuring that consensus checks are in order.
+    /// Safekeeper and other state, that should remain consistent and synchronized
+    /// with the disk.
     mutex: Mutex<SharedState>,
     walsenders: Arc<WalSenders>,
 
@@ -363,8 +361,8 @@ impl Timeline {
     ///
     /// Bootstrap is transactional, so if it fails, created files will be deleted,
     /// and state on disk should remain unchanged.
-    pub async fn bootstrap(&self, shared_state: &mut MutexGuard<'_, SharedState>) -> Result<()> {
-        match fs::metadata(&self.timeline_dir).await {
+    pub fn bootstrap(&self, shared_state: &mut MutexGuard<SharedState>) -> Result<()> {
+        match std::fs::metadata(&self.timeline_dir) {
             Ok(_) => {
                 // Timeline directory exists on disk, we should leave state unchanged
                 // and return error.
@@ -377,51 +375,53 @@ impl Timeline {
         }
 
         // Create timeline directory.
-        fs::create_dir_all(&self.timeline_dir).await?;
+        std::fs::create_dir_all(&self.timeline_dir)?;
 
         // Write timeline to disk and TODO: start background tasks.
-        if let Err(e) = shared_state.sk.persist().await {
-            // Bootstrap failed, cancel timeline and remove timeline directory.
-            self.cancel(shared_state);
+        match || -> Result<()> {
+            shared_state.sk.persist()?;
+            // TODO: add more initialization steps here
+            self.update_status(shared_state);
+            Ok(())
+        }() {
+            Ok(_) => Ok(()),
+            Err(e) => {
+                // Bootstrap failed, cancel timeline and remove timeline directory.
+                self.cancel(shared_state);
 
-            if let Err(fs_err) = fs::remove_dir_all(&self.timeline_dir).await {
-                warn!(
-                    "failed to remove timeline {} directory after bootstrap failure: {}",
-                    self.ttid, fs_err
-                );
+                if let Err(fs_err) = std::fs::remove_dir_all(&self.timeline_dir) {
+                    warn!(
+                        "failed to remove timeline {} directory after bootstrap failure: {}",
+                        self.ttid, fs_err
+                    );
+                }
+
+                Err(e)
             }
-
-            return Err(e);
         }
-
-        // TODO: add more initialization steps here
-        self.update_status(shared_state);
-        Ok(())
     }
 
     /// Delete timeline from disk completely, by removing timeline directory. Background
     /// timeline activities will stop eventually.
-    pub async fn delete_from_disk(
+    pub fn delete_from_disk(
         &self,
-        shared_state: &mut MutexGuard<'_, SharedState>,
+        shared_state: &mut MutexGuard<SharedState>,
     ) -> Result<(bool, bool)> {
         let was_active = shared_state.active;
         self.cancel(shared_state);
-        let dir_existed = delete_dir(&self.timeline_dir).await?;
+        let dir_existed = delete_dir(&self.timeline_dir)?;
         Ok((dir_existed, was_active))
     }
 
     /// Cancel timeline to prevent further usage. Background tasks will stop
     /// eventually after receiving cancellation signal.
-    ///
-    /// Note that we can't notify backup launcher here while holding
-    /// shared_state lock, as this is a potential deadlock: caller is
-    /// responsible for that. Generally we should probably make WAL backup tasks
-    /// to shut down on their own, checking once in a while whether it is the
-    /// time.
-    fn cancel(&self, shared_state: &mut MutexGuard<'_, SharedState>) {
+    fn cancel(&self, shared_state: &mut MutexGuard<SharedState>) {
         info!("timeline {} is cancelled", self.ttid);
         let _ = self.cancellation_tx.send(true);
+        let res = self.wal_backup_launcher_tx.blocking_send(self.ttid);
+        if let Err(e) = res {
+            error!("Failed to send stop signal to wal_backup_launcher: {}", e);
+        }
         // Close associated FDs. Nobody will be able to touch timeline data once
         // it is cancelled, so WAL storage won't be opened again.
         shared_state.sk.wal_store.close();
@@ -433,8 +433,8 @@ impl Timeline {
     }
 
     /// Take a writing mutual exclusive lock on timeline shared_state.
-    pub async fn write_shared_state(&self) -> MutexGuard<SharedState> {
-        self.mutex.lock().await
+    pub fn write_shared_state(&self) -> MutexGuard<SharedState> {
+        self.mutex.lock()
     }
 
     fn update_status(&self, shared_state: &mut SharedState) -> bool {
@@ -450,7 +450,7 @@ impl Timeline {
 
         let is_wal_backup_action_pending: bool;
         {
-            let mut shared_state = self.write_shared_state().await;
+            let mut shared_state = self.write_shared_state();
             shared_state.num_computes += 1;
             is_wal_backup_action_pending = self.update_status(&mut shared_state);
         }
@@ -464,17 +464,22 @@ impl Timeline {
 
     /// De-register compute connection, shutting down timeline activity if
     /// pageserver doesn't need catchup.
-    pub async fn on_compute_disconnect(&self) -> Result<()> {
+    pub fn on_compute_disconnect(&self) -> Result<()> {
         let is_wal_backup_action_pending: bool;
         {
-            let mut shared_state = self.write_shared_state().await;
+            let mut shared_state = self.write_shared_state();
             shared_state.num_computes -= 1;
             is_wal_backup_action_pending = self.update_status(&mut shared_state);
         }
         // Wake up wal backup launcher, if it is time to stop the offloading.
         if is_wal_backup_action_pending {
             // Can fail only if channel to a static thread got closed, which is not normal at all.
-            self.wal_backup_launcher_tx.send(self.ttid).await?;
+            //
+            // Note: this is blocking_send because on_compute_disconnect is called in Drop, there is
+            // no async Drop and we use current thread runtimes. With current thread rt spawning
+            // task in drop impl is racy, as thread along with runtime might finish before the task.
+            // This should be switched send.await when/if we go to full async.
+            self.wal_backup_launcher_tx.blocking_send(self.ttid)?;
         }
         Ok(())
     }
@@ -484,11 +489,11 @@ impl Timeline {
     /// computes. While there might be nothing to stream already, we learn about
     /// remote_consistent_lsn update through replication feedback, and we want
     /// to stop pushing to the broker if pageserver is fully caughtup.
-    pub async fn should_walsender_stop(&self, reported_remote_consistent_lsn: Lsn) -> bool {
+    pub fn should_walsender_stop(&self, reported_remote_consistent_lsn: Lsn) -> bool {
         if self.is_cancelled() {
             return true;
         }
-        let shared_state = self.write_shared_state().await;
+        let shared_state = self.write_shared_state();
         if shared_state.num_computes == 0 {
             return shared_state.sk.inmem.commit_lsn == Lsn(0) || // no data at all yet
             reported_remote_consistent_lsn >= shared_state.sk.inmem.commit_lsn;
@@ -498,12 +503,12 @@ impl Timeline {
 
     /// Returns whether s3 offloading is required and sets current status as
     /// matching it.
-    pub async fn wal_backup_attend(&self) -> bool {
+    pub fn wal_backup_attend(&self) -> bool {
         if self.is_cancelled() {
             return false;
         }
 
-        self.write_shared_state().await.wal_backup_attend()
+        self.write_shared_state().wal_backup_attend()
     }
 
     /// Returns commit_lsn watch channel.
@@ -512,7 +517,7 @@ impl Timeline {
     }
 
     /// Pass arrived message to the safekeeper.
-    pub async fn process_msg(
+    pub fn process_msg(
         &self,
         msg: &ProposerAcceptorMessage,
     ) -> Result<Option<AcceptorProposerMessage>> {
@@ -523,8 +528,8 @@ impl Timeline {
         let mut rmsg: Option<AcceptorProposerMessage>;
         let commit_lsn: Lsn;
         {
-            let mut shared_state = self.write_shared_state().await;
-            rmsg = shared_state.sk.process_msg(msg).await?;
+            let mut shared_state = self.write_shared_state();
+            rmsg = shared_state.sk.process_msg(msg)?;
 
             // if this is AppendResponse, fill in proper pageserver and hot
             // standby feedback.
@@ -541,37 +546,37 @@ impl Timeline {
     }
 
     /// Returns wal_seg_size.
-    pub async fn get_wal_seg_size(&self) -> usize {
-        self.write_shared_state().await.get_wal_seg_size()
+    pub fn get_wal_seg_size(&self) -> usize {
+        self.write_shared_state().get_wal_seg_size()
     }
 
     /// Returns true only if the timeline is loaded and active.
-    pub async fn is_active(&self) -> bool {
+    pub fn is_active(&self) -> bool {
         if self.is_cancelled() {
             return false;
         }
 
-        self.write_shared_state().await.active
+        self.write_shared_state().active
     }
 
     /// Returns state of the timeline.
-    pub async fn get_state(&self) -> (SafekeeperMemState, SafeKeeperState) {
-        let state = self.write_shared_state().await;
+    pub fn get_state(&self) -> (SafekeeperMemState, SafeKeeperState) {
+        let state = self.write_shared_state();
         (state.sk.inmem.clone(), state.sk.state.clone())
     }
 
     /// Returns latest backup_lsn.
-    pub async fn get_wal_backup_lsn(&self) -> Lsn {
-        self.write_shared_state().await.sk.inmem.backup_lsn
+    pub fn get_wal_backup_lsn(&self) -> Lsn {
+        self.write_shared_state().sk.inmem.backup_lsn
     }
 
     /// Sets backup_lsn to the given value.
-    pub async fn set_wal_backup_lsn(&self, backup_lsn: Lsn) -> Result<()> {
+    pub fn set_wal_backup_lsn(&self, backup_lsn: Lsn) -> Result<()> {
         if self.is_cancelled() {
             bail!(TimelineError::Cancelled(self.ttid));
         }
 
-        let mut state = self.write_shared_state().await;
+        let mut state = self.write_shared_state();
         state.sk.inmem.backup_lsn = max(state.sk.inmem.backup_lsn, backup_lsn);
         // we should check whether to shut down offloader, but this will be done
         // soon by peer communication anyway.
@@ -579,8 +584,8 @@ impl Timeline {
     }
 
     /// Get safekeeper info for broadcasting to broker and other peers.
-    pub async fn get_safekeeper_info(&self, conf: &SafeKeeperConf) -> SafekeeperTimelineInfo {
-        let shared_state = self.write_shared_state().await;
+    pub fn get_safekeeper_info(&self, conf: &SafeKeeperConf) -> SafekeeperTimelineInfo {
+        let shared_state = self.write_shared_state();
         shared_state.get_safekeeper_info(
             &self.ttid,
             conf,
@@ -599,8 +604,8 @@ impl Timeline {
         let is_wal_backup_action_pending: bool;
         let commit_lsn: Lsn;
         {
-            let mut shared_state = self.write_shared_state().await;
-            shared_state.sk.record_safekeeper_info(&sk_info).await?;
+            let mut shared_state = self.write_shared_state();
+            shared_state.sk.record_safekeeper_info(&sk_info)?;
             let peer_info = PeerInfo::from_sk_info(&sk_info, Instant::now());
             shared_state.peers_info.upsert(&peer_info);
             is_wal_backup_action_pending = self.update_status(&mut shared_state);
@@ -617,8 +622,8 @@ impl Timeline {
     /// Get our latest view of alive peers status on the timeline.
     /// We pass our own info through the broker as well, so when we don't have connection
     /// to the broker returned vec is empty.
-    pub async fn get_peers(&self, conf: &SafeKeeperConf) -> Vec<PeerInfo> {
-        let shared_state = self.write_shared_state().await;
+    pub fn get_peers(&self, conf: &SafeKeeperConf) -> Vec<PeerInfo> {
+        let shared_state = self.write_shared_state();
         let now = Instant::now();
         shared_state
             .peers_info
@@ -635,34 +640,34 @@ impl Timeline {
     }
 
     /// Returns flush_lsn.
-    pub async fn get_flush_lsn(&self) -> Lsn {
-        self.write_shared_state().await.sk.wal_store.flush_lsn()
+    pub fn get_flush_lsn(&self) -> Lsn {
+        self.write_shared_state().sk.wal_store.flush_lsn()
     }
 
     /// Delete WAL segments from disk that are no longer needed. This is determined
     /// based on pageserver's remote_consistent_lsn and local backup_lsn/peer_lsn.
-    pub async fn remove_old_wal(&self, wal_backup_enabled: bool) -> Result<()> {
+    pub fn remove_old_wal(&self, wal_backup_enabled: bool) -> Result<()> {
         if self.is_cancelled() {
             bail!(TimelineError::Cancelled(self.ttid));
         }
 
         let horizon_segno: XLogSegNo;
-        let remover = {
-            let shared_state = self.write_shared_state().await;
+        let remover: Box<dyn Fn(u64) -> Result<(), anyhow::Error>>;
+        {
+            let shared_state = self.write_shared_state();
             horizon_segno = shared_state.sk.get_horizon_segno(wal_backup_enabled);
+            remover = shared_state.sk.wal_store.remove_up_to();
             if horizon_segno <= 1 || horizon_segno <= shared_state.last_removed_segno {
-                return Ok(()); // nothing to do
+                return Ok(());
             }
-            let remover = shared_state.sk.wal_store.remove_up_to(horizon_segno - 1);
             // release the lock before removing
-            remover
-        };
+        }
 
         // delete old WAL files
-        remover.await?;
+        remover(horizon_segno - 1)?;
 
         // update last_removed_segno
-        let mut shared_state = self.write_shared_state().await;
+        let mut shared_state = self.write_shared_state();
         shared_state.last_removed_segno = horizon_segno;
         Ok(())
     }
@@ -671,24 +676,22 @@ impl Timeline {
     /// passed after the last save. This helps to keep remote_consistent_lsn up
     /// to date so that storage nodes restart doesn't cause many pageserver ->
     /// safekeeper reconnections.
-    pub async fn maybe_persist_control_file(&self) -> Result<()> {
+    pub fn maybe_pesist_control_file(&self) -> Result<()> {
         let remote_consistent_lsn = self.walsenders.get_remote_consistent_lsn();
         self.write_shared_state()
-            .await
             .sk
             .maybe_persist_control_file(remote_consistent_lsn)
-            .await
     }
 
-    /// Gather timeline data for metrics. If the timeline is not active, returns
-    /// None, we do not collect these.
-    pub async fn info_for_metrics(&self) -> Option<FullTimelineInfo> {
+    /// Returns full timeline info, required for the metrics. If the timeline is
+    /// not active, returns None instead.
+    pub fn info_for_metrics(&self) -> Option<FullTimelineInfo> {
         if self.is_cancelled() {
             return None;
         }
 
         let ps_feedback = self.walsenders.get_ps_feedback();
-        let state = self.write_shared_state().await;
+        let state = self.write_shared_state();
         if state.active {
             Some(FullTimelineInfo {
                 ttid: self.ttid,
@@ -710,8 +713,8 @@ impl Timeline {
     }
 
     /// Returns in-memory timeline state to build a full debug dump.
-    pub async fn memory_dump(&self) -> debug_dump::Memory {
-        let state = self.write_shared_state().await;
+    pub fn memory_dump(&self) -> debug_dump::Memory {
+        let state = self.write_shared_state();
 
         let (write_lsn, write_record_lsn, flush_lsn, file_open) =
             state.sk.wal_store.internal_state();
@@ -735,8 +738,8 @@ impl Timeline {
 }
 
 /// Deletes directory and it's contents. Returns false if directory does not exist.
-async fn delete_dir(path: &PathBuf) -> Result<bool> {
-    match fs::remove_dir_all(path).await {
+fn delete_dir(path: &PathBuf) -> Result<bool> {
+    match std::fs::remove_dir_all(path) {
         Ok(_) => Ok(true),
         Err(e) if e.kind() == std::io::ErrorKind::NotFound => Ok(false),
         Err(e) => Err(e.into()),

safekeeper/src/timelines_global_map.rs

@@ -113,17 +113,9 @@ impl GlobalTimelines {
         Ok(())
     }
 
-    /// Loads all timelines for the given tenant to memory. Returns fs::read_dir
-    /// errors if any.
-    ///
-    /// Note: This function (and all reading/loading below) is sync because
-    /// timelines are loaded while holding GlobalTimelinesState lock. Which is
-    /// fine as this is called only from single threaded main runtime on boot,
-    /// but clippy complains anyway, and suppressing that isn't trivial as async
-    /// is the keyword, ha. That only other user is pull_timeline.rs for which
-    /// being blocked is not that bad, and we can do spawn_blocking.
+    /// Loads all timelines for the given tenant to memory. Returns fs::read_dir errors if any.
     fn load_tenant_timelines(
-        state: &mut MutexGuard<'_, GlobalTimelinesState>,
+        state: &mut MutexGuard<GlobalTimelinesState>,
         tenant_id: TenantId,
     ) -> Result<()> {
         let timelines_dir = state.get_conf().tenant_dir(&tenant_id);
@@ -228,7 +220,7 @@ impl GlobalTimelines {
         // Take a lock and finish the initialization holding this mutex. No other threads
         // can interfere with creation after we will insert timeline into the map.
         {
-            let mut shared_state = timeline.write_shared_state().await;
+            let mut shared_state = timeline.write_shared_state();
 
             // We can get a race condition here in case of concurrent create calls, but only
             // in theory. create() will return valid timeline on the next try.
@@ -240,7 +232,7 @@ impl GlobalTimelines {
             // Write the new timeline to the disk and start background workers.
             // Bootstrap is transactional, so if it fails, the timeline will be deleted,
             // and the state on disk should remain unchanged.
-            if let Err(e) = timeline.bootstrap(&mut shared_state).await {
+            if let Err(e) = timeline.bootstrap(&mut shared_state) {
                 // Note: the most likely reason for bootstrap failure is that the timeline
                 // directory already exists on disk. This happens when timeline is corrupted
                 // and wasn't loaded from disk on startup because of that. We want to preserve
@@ -302,16 +294,15 @@ impl GlobalTimelines {
     }
 
     /// Cancels timeline, then deletes the corresponding data directory.
-    pub async fn delete_force(ttid: &TenantTimelineId) -> Result<TimelineDeleteForceResult> {
+    pub fn delete_force(ttid: &TenantTimelineId) -> Result<TimelineDeleteForceResult> {
         let tli_res = TIMELINES_STATE.lock().unwrap().get(ttid);
         match tli_res {
             Ok(timeline) => {
                 // Take a lock and finish the deletion holding this mutex.
-                let mut shared_state = timeline.write_shared_state().await;
+                let mut shared_state = timeline.write_shared_state();
 
                 info!("deleting timeline {}", ttid);
-                let (dir_existed, was_active) =
-                    timeline.delete_from_disk(&mut shared_state).await?;
+                let (dir_existed, was_active) = timeline.delete_from_disk(&mut shared_state)?;
 
                 // Remove timeline from the map.
                 // FIXME: re-enable it once we fix the issue with recreation of deleted timelines
@@ -344,7 +335,7 @@ impl GlobalTimelines {
     /// the tenant had, `true` if a timeline was active. There may be a race if new timelines are
     /// created simultaneously. In that case the function will return error and the caller should
     /// retry tenant deletion again later.
-    pub async fn delete_force_all_for_tenant(
+    pub fn delete_force_all_for_tenant(
         tenant_id: &TenantId,
     ) -> Result<HashMap<TenantTimelineId, TimelineDeleteForceResult>> {
         info!("deleting all timelines for tenant {}", tenant_id);
@@ -354,7 +345,7 @@ impl GlobalTimelines {
 
         let mut deleted = HashMap::new();
         for tli in &to_delete {
-            match Self::delete_force(&tli.ttid).await {
+            match Self::delete_force(&tli.ttid) {
                 Ok(result) => {
                     deleted.insert(tli.ttid, result);
                 }

safekeeper/src/wal_backup.rs

@@ -17,6 +17,7 @@ use postgres_ffi::XLogFileName;
 use postgres_ffi::{XLogSegNo, PG_TLI};
 use remote_storage::{GenericRemoteStorage, RemotePath};
 use tokio::fs::File;
+use tokio::runtime::Builder;
 
 use tokio::select;
 use tokio::sync::mpsc::{self, Receiver, Sender};
@@ -35,16 +36,30 @@ use once_cell::sync::OnceCell;
 const UPLOAD_FAILURE_RETRY_MIN_MS: u64 = 10;
 const UPLOAD_FAILURE_RETRY_MAX_MS: u64 = 5000;
 
+pub fn wal_backup_launcher_thread_main(
+    conf: SafeKeeperConf,
+    wal_backup_launcher_rx: Receiver<TenantTimelineId>,
+) {
+    let mut builder = Builder::new_multi_thread();
+    if let Some(num_threads) = conf.backup_runtime_threads {
+        builder.worker_threads(num_threads);
+    }
+    let rt = builder
+        .enable_all()
+        .build()
+        .expect("failed to create wal backup runtime");
+
+    rt.block_on(async {
+        wal_backup_launcher_main_loop(conf, wal_backup_launcher_rx).await;
+    });
+}
+
 /// Check whether wal backup is required for timeline. If yes, mark that launcher is
 /// aware of current status and return the timeline.
-async fn is_wal_backup_required(ttid: TenantTimelineId) -> Option<Arc<Timeline>> {
-    match GlobalTimelines::get(ttid).ok() {
-        Some(tli) => {
-            tli.wal_backup_attend().await;
-            Some(tli)
-        }
-        None => None,
-    }
+fn is_wal_backup_required(ttid: TenantTimelineId) -> Option<Arc<Timeline>> {
+    GlobalTimelines::get(ttid)
+        .ok()
+        .filter(|tli| tli.wal_backup_attend())
 }
 
 struct WalBackupTaskHandle {
@@ -128,8 +143,8 @@ async fn update_task(
     ttid: TenantTimelineId,
     entry: &mut WalBackupTimelineEntry,
 ) {
-    let alive_peers = entry.timeline.get_peers(conf).await;
-    let wal_backup_lsn = entry.timeline.get_wal_backup_lsn().await;
+    let alive_peers = entry.timeline.get_peers(conf);
+    let wal_backup_lsn = entry.timeline.get_wal_backup_lsn();
     let (offloader, election_dbg_str) =
         determine_offloader(&alive_peers, wal_backup_lsn, ttid, conf);
     let elected_me = Some(conf.my_id) == offloader;
@@ -168,10 +183,10 @@ const CHECK_TASKS_INTERVAL_MSEC: u64 = 1000;
 /// Sits on wal_backup_launcher_rx and starts/stops per timeline wal backup
 /// tasks. Having this in separate task simplifies locking, allows to reap
 /// panics and separate elections from offloading itself.
-pub async fn wal_backup_launcher_task_main(
+async fn wal_backup_launcher_main_loop(
     conf: SafeKeeperConf,
     mut wal_backup_launcher_rx: Receiver<TenantTimelineId>,
-) -> anyhow::Result<()> {
+) {
     info!(
         "WAL backup launcher started, remote config {:?}",
         conf.remote_storage
@@ -199,7 +214,7 @@ pub async fn wal_backup_launcher_task_main(
                 if conf.remote_storage.is_none() || !conf.wal_backup_enabled {
                     continue; /* just drain the channel and do nothing */
                 }
-                let timeline = is_wal_backup_required(ttid).await;
+                let timeline = is_wal_backup_required(ttid);
                 // do we need to do anything at all?
                 if timeline.is_some() != tasks.contains_key(&ttid) {
                     if let Some(timeline) = timeline {
@@ -254,7 +269,7 @@ async fn backup_task_main(
     let tli = res.unwrap();
 
     let mut wb = WalBackupTask {
-        wal_seg_size: tli.get_wal_seg_size().await,
+        wal_seg_size: tli.get_wal_seg_size(),
         commit_lsn_watch_rx: tli.get_commit_lsn_watch_rx(),
         timeline: tli,
         timeline_dir,
@@ -311,7 +326,7 @@ impl WalBackupTask {
                 continue; /* nothing to do, common case as we wake up on every commit_lsn bump */
             }
             // Perhaps peers advanced the position, check shmem value.
-            backup_lsn = self.timeline.get_wal_backup_lsn().await;
+            backup_lsn = self.timeline.get_wal_backup_lsn();
             if backup_lsn.segment_number(self.wal_seg_size)
                 >= commit_lsn.segment_number(self.wal_seg_size)
             {
@@ -387,7 +402,6 @@ pub async fn backup_lsn_range(
                 let new_backup_lsn = segment.end_lsn;
                 timeline
                     .set_wal_backup_lsn(new_backup_lsn)
-                    .await
                     .context("setting wal_backup_lsn")?;
                 *backup_lsn = new_backup_lsn;
             } else {

safekeeper/src/wal_service.rs

@@ -4,7 +4,7 @@
 //!
 use anyhow::{Context, Result};
 use postgres_backend::QueryError;
-use std::{future, time::Duration};
+use std::{future, thread, time::Duration};
 use tokio::net::TcpStream;
 use tokio_io_timeout::TimeoutReader;
 use tracing::*;
@@ -16,82 +16,104 @@ use crate::SafeKeeperConf;
 use postgres_backend::{AuthType, PostgresBackend};
 
 /// Accept incoming TCP connections and spawn them into a background thread.
-pub async fn task_main(
-    conf: SafeKeeperConf,
-    pg_listener: std::net::TcpListener,
-) -> anyhow::Result<()> {
-    // Tokio's from_std won't do this for us, per its comment.
-    pg_listener.set_nonblocking(true)?;
+pub fn thread_main(conf: SafeKeeperConf, pg_listener: std::net::TcpListener) {
+    let runtime = tokio::runtime::Builder::new_current_thread()
+        .enable_all()
+        .build()
+        .context("create runtime")
+        // todo catch error in main thread
+        .expect("failed to create runtime");
 
-    let listener = tokio::net::TcpListener::from_std(pg_listener)?;
-    let mut connection_count: ConnectionCount = 0;
+    runtime
+        .block_on(async move {
+            // Tokio's from_std won't do this for us, per its comment.
+            pg_listener.set_nonblocking(true)?;
+            let listener = tokio::net::TcpListener::from_std(pg_listener)?;
+            let mut connection_count: ConnectionCount = 0;
 
-    loop {
-        let (socket, peer_addr) = listener.accept().await.context("accept")?;
-        debug!("accepted connection from {}", peer_addr);
-        let conf = conf.clone();
-        let conn_id = issue_connection_id(&mut connection_count);
+            loop {
+                match listener.accept().await {
+                    Ok((socket, peer_addr)) => {
+                        debug!("accepted connection from {}", peer_addr);
+                        let conf = conf.clone();
+                        let conn_id = issue_connection_id(&mut connection_count);
 
-        tokio::spawn(async move {
-            if let Err(err) = handle_socket(socket, conf, conn_id)
-                .instrument(info_span!("", cid = %conn_id))
-                .await
-            {
-                error!("connection handler exited: {}", err);
+                        let _ = thread::Builder::new()
+                            .name("WAL service thread".into())
+                            .spawn(move || {
+                                if let Err(err) = handle_socket(socket, conf, conn_id) {
+                                    error!("connection handler exited: {}", err);
+                                }
+                            })
+                            .unwrap();
+                    }
+                    Err(e) => error!("Failed to accept connection: {}", e),
+                }
             }
-        });
-    }
+            #[allow(unreachable_code)] // hint compiler the closure return type
+            Ok::<(), anyhow::Error>(())
+        })
+        .expect("listener failed")
 }
 
-/// This is run by `task_main` above, inside a background thread.
+/// This is run by `thread_main` above, inside a background thread.
 ///
-async fn handle_socket(
+fn handle_socket(
     socket: TcpStream,
     conf: SafeKeeperConf,
     conn_id: ConnectionId,
 ) -> Result<(), QueryError> {
+    let _enter = info_span!("", cid = %conn_id).entered();
+
+    let runtime = tokio::runtime::Builder::new_current_thread()
+        .enable_all()
+        .build()?;
+
     socket.set_nodelay(true)?;
     let peer_addr = socket.peer_addr()?;
 
-    // Set timeout on reading from the socket. It prevents hanged up connection
-    // if client suddenly disappears. Note that TCP_KEEPALIVE is not enabled by
-    // default, and tokio doesn't provide ability to set it out of the box.
-    let mut socket = TimeoutReader::new(socket);
-    let wal_service_timeout = Duration::from_secs(60 * 10);
-    socket.set_timeout(Some(wal_service_timeout));
-    // pin! is here because TimeoutReader (due to storing sleep future inside)
-    // is not Unpin, and all pgbackend/framed/tokio dependencies require stream
-    // to be Unpin. Which is reasonable, as indeed something like TimeoutReader
-    // shouldn't be moved.
-    tokio::pin!(socket);
+    // TimeoutReader wants async runtime during creation.
+    runtime.block_on(async move {
+        // Set timeout on reading from the socket. It prevents hanged up connection
+        // if client suddenly disappears. Note that TCP_KEEPALIVE is not enabled by
+        // default, and tokio doesn't provide ability to set it out of the box.
+        let mut socket = TimeoutReader::new(socket);
+        let wal_service_timeout = Duration::from_secs(60 * 10);
+        socket.set_timeout(Some(wal_service_timeout));
+        // pin! is here because TimeoutReader (due to storing sleep future inside)
+        // is not Unpin, and all pgbackend/framed/tokio dependencies require stream
+        // to be Unpin. Which is reasonable, as indeed something like TimeoutReader
+        // shouldn't be moved.
+        tokio::pin!(socket);
 
-    let traffic_metrics = TrafficMetrics::new();
-    if let Some(current_az) = conf.availability_zone.as_deref() {
-        traffic_metrics.set_sk_az(current_az);
-    }
+        let traffic_metrics = TrafficMetrics::new();
+        if let Some(current_az) = conf.availability_zone.as_deref() {
+            traffic_metrics.set_sk_az(current_az);
+        }
 
-    let socket = MeasuredStream::new(
-        socket,
-        |cnt| {
-            traffic_metrics.observe_read(cnt);
-        },
-        |cnt| {
-            traffic_metrics.observe_write(cnt);
-        },
-    );
+        let socket = MeasuredStream::new(
+            socket,
+            |cnt| {
+                traffic_metrics.observe_read(cnt);
+            },
+            |cnt| {
+                traffic_metrics.observe_write(cnt);
+            },
+        );
 
-    let auth_type = match conf.auth {
-        None => AuthType::Trust,
-        Some(_) => AuthType::NeonJWT,
-    };
-    let mut conn_handler =
-        SafekeeperPostgresHandler::new(conf, conn_id, Some(traffic_metrics.clone()));
-    let pgbackend = PostgresBackend::new_from_io(socket, peer_addr, auth_type, None)?;
-    // libpq protocol between safekeeper and walproposer / pageserver
-    // We don't use shutdown.
-    pgbackend
-        .run(&mut conn_handler, future::pending::<()>)
-        .await
+        let auth_type = match conf.auth {
+            None => AuthType::Trust,
+            Some(_) => AuthType::NeonJWT,
+        };
+        let mut conn_handler =
+            SafekeeperPostgresHandler::new(conf, conn_id, Some(traffic_metrics.clone()));
+        let pgbackend = PostgresBackend::new_from_io(socket, peer_addr, auth_type, None)?;
+        // libpq protocol between safekeeper and walproposer / pageserver
+        // We don't use shutdown.
+        pgbackend
+            .run(&mut conn_handler, future::pending::<()>)
+            .await
+    })
 }
 
 /// Unique WAL service connection ids are logged in spans for observability.

safekeeper/src/wal_storage.rs

@@ -8,47 +8,54 @@
 //! Note that last file has `.partial` suffix, that's different from postgres.
 
 use anyhow::{bail, Context, Result};
-use bytes::Bytes;
-use futures::future::BoxFuture;
+use remote_storage::RemotePath;
+
+use std::io::{self, Seek, SeekFrom};
+use std::pin::Pin;
+use tokio::io::AsyncRead;
+
 use postgres_ffi::v14::xlog_utils::{IsPartialXLogFileName, IsXLogFileName, XLogFromFileName};
 use postgres_ffi::{XLogSegNo, PG_TLI};
-use remote_storage::RemotePath;
 use std::cmp::{max, min};
-use std::io::{self, SeekFrom};
+
+use bytes::Bytes;
+use std::fs::{self, remove_file, File, OpenOptions};
+use std::io::Write;
 use std::path::{Path, PathBuf};
-use std::pin::Pin;
-use tokio::fs::{self, remove_file, File, OpenOptions};
-use tokio::io::{AsyncRead, AsyncWriteExt};
-use tokio::io::{AsyncReadExt, AsyncSeekExt};
+
 use tracing::*;
 
+use utils::{id::TenantTimelineId, lsn::Lsn};
+
 use crate::metrics::{time_io_closure, WalStorageMetrics, REMOVED_WAL_SEGMENTS};
 use crate::safekeeper::SafeKeeperState;
+
 use crate::wal_backup::read_object;
 use crate::SafeKeeperConf;
-use postgres_ffi::waldecoder::WalStreamDecoder;
 use postgres_ffi::XLogFileName;
 use postgres_ffi::XLOG_BLCKSZ;
-use pq_proto::SystemId;
-use utils::{id::TenantTimelineId, lsn::Lsn};
 
-#[async_trait::async_trait]
+use postgres_ffi::waldecoder::WalStreamDecoder;
+
+use pq_proto::SystemId;
+use tokio::io::{AsyncReadExt, AsyncSeekExt};
+
 pub trait Storage {
     /// LSN of last durably stored WAL record.
     fn flush_lsn(&self) -> Lsn;
 
     /// Write piece of WAL from buf to disk, but not necessarily sync it.
-    async fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()>;
+    fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()>;
 
     /// Truncate WAL at specified LSN, which must be the end of WAL record.
-    async fn truncate_wal(&mut self, end_pos: Lsn) -> Result<()>;
+    fn truncate_wal(&mut self, end_pos: Lsn) -> Result<()>;
 
     /// Durably store WAL on disk, up to the last written WAL record.
-    async fn flush_wal(&mut self) -> Result<()>;
+    fn flush_wal(&mut self) -> Result<()>;
 
-    /// Remove all segments <= given segno. Returns function doing that as we
-    /// want to perform it without timeline lock.
-    fn remove_up_to(&self, segno_up_to: XLogSegNo) -> BoxFuture<'static, anyhow::Result<()>>;
+    /// Remove all segments <= given segno. Returns closure as we want to do
+    /// that without timeline lock.
+    fn remove_up_to(&self) -> Box<dyn Fn(XLogSegNo) -> Result<()>>;
 
     /// Release resources associated with the storage -- technically, close FDs.
     /// Currently we don't remove timelines until restart (#3146), so need to
@@ -98,22 +105,6 @@ pub struct PhysicalStorage {
     /// - points to write_lsn, so no seek is needed for writing
     /// - doesn't point to the end of the segment
     file: Option<File>,
-
-    /// When false, we have just initialized storage using the LSN from find_end_of_wal().
-    /// In this case, [`write_lsn`] can be less than actually written WAL on disk. In particular,
-    /// there can be a case with unexpected .partial file.
-    ///
-    /// Imagine the following:
-    /// - 000000010000000000000001
-    ///   - it was fully written, but the last record is split between 2 segments
-    ///   - after restart, find_end_of_wal() returned 0/1FFFFF0, which is in the end of this segment
-    ///   - write_lsn, write_record_lsn and flush_record_lsn were initialized to 0/1FFFFF0
-    /// - 000000010000000000000002.partial
-    ///   - it has only 1 byte written, which is not enough to make a full WAL record
-    ///
-    /// Partial segment 002 has no WAL records, and it will be removed by the next truncate_wal().
-    /// This flag will be set to true after the first truncate_wal() call.
-    is_truncated_after_restart: bool,
 }
 
 impl PhysicalStorage {
@@ -173,7 +164,6 @@ impl PhysicalStorage {
             flush_record_lsn: flush_lsn,
             decoder: WalStreamDecoder::new(write_lsn, state.server.pg_version / 10000),
             file: None,
-            is_truncated_after_restart: false,
         })
     }
 
@@ -188,37 +178,33 @@ impl PhysicalStorage {
     }
 
     /// Call fdatasync if config requires so.
-    async fn fdatasync_file(&mut self, file: &mut File) -> Result<()> {
+    fn fdatasync_file(&mut self, file: &mut File) -> Result<()> {
         if !self.conf.no_sync {
             self.metrics
-                .observe_flush_seconds(time_io_closure(file.sync_data()).await?);
+                .observe_flush_seconds(time_io_closure(|| Ok(file.sync_data()?))?);
         }
         Ok(())
     }
 
     /// Call fsync if config requires so.
-    async fn fsync_file(&mut self, file: &mut File) -> Result<()> {
+    fn fsync_file(&mut self, file: &mut File) -> Result<()> {
         if !self.conf.no_sync {
             self.metrics
-                .observe_flush_seconds(time_io_closure(file.sync_all()).await?);
+                .observe_flush_seconds(time_io_closure(|| Ok(file.sync_all()?))?);
         }
         Ok(())
     }
 
     /// Open or create WAL segment file. Caller must call seek to the wanted position.
     /// Returns `file` and `is_partial`.
-    async fn open_or_create(&mut self, segno: XLogSegNo) -> Result<(File, bool)> {
+    fn open_or_create(&mut self, segno: XLogSegNo) -> Result<(File, bool)> {
         let (wal_file_path, wal_file_partial_path) =
             wal_file_paths(&self.timeline_dir, segno, self.wal_seg_size)?;
 
         // Try to open already completed segment
-        if let Ok(file) = OpenOptions::new().write(true).open(&wal_file_path).await {
+        if let Ok(file) = OpenOptions::new().write(true).open(&wal_file_path) {
             Ok((file, false))
-        } else if let Ok(file) = OpenOptions::new()
-            .write(true)
-            .open(&wal_file_partial_path)
-            .await
-        {
+        } else if let Ok(file) = OpenOptions::new().write(true).open(&wal_file_partial_path) {
             // Try to open existing partial file
             Ok((file, true))
         } else {
@@ -227,36 +213,35 @@ impl PhysicalStorage {
                 .create(true)
                 .write(true)
                 .open(&wal_file_partial_path)
-                .await
                 .with_context(|| format!("Failed to open log file {:?}", &wal_file_path))?;
 
-            write_zeroes(&mut file, self.wal_seg_size).await?;
-            self.fsync_file(&mut file).await?;
+            write_zeroes(&mut file, self.wal_seg_size)?;
+            self.fsync_file(&mut file)?;
             Ok((file, true))
         }
     }
 
     /// Write WAL bytes, which are known to be located in a single WAL segment.
-    async fn write_in_segment(&mut self, segno: u64, xlogoff: usize, buf: &[u8]) -> Result<()> {
+    fn write_in_segment(&mut self, segno: u64, xlogoff: usize, buf: &[u8]) -> Result<()> {
         let mut file = if let Some(file) = self.file.take() {
             file
         } else {
-            let (mut file, is_partial) = self.open_or_create(segno).await?;
+            let (mut file, is_partial) = self.open_or_create(segno)?;
             assert!(is_partial, "unexpected write into non-partial segment file");
-            file.seek(SeekFrom::Start(xlogoff as u64)).await?;
+            file.seek(SeekFrom::Start(xlogoff as u64))?;
             file
         };
 
-        file.write_all(buf).await?;
+        file.write_all(buf)?;
 
         if xlogoff + buf.len() == self.wal_seg_size {
             // If we reached the end of a WAL segment, flush and close it.
-            self.fdatasync_file(&mut file).await?;
+            self.fdatasync_file(&mut file)?;
 
             // Rename partial file to completed file
             let (wal_file_path, wal_file_partial_path) =
                 wal_file_paths(&self.timeline_dir, segno, self.wal_seg_size)?;
-            fs::rename(wal_file_partial_path, wal_file_path).await?;
+            fs::rename(wal_file_partial_path, wal_file_path)?;
         } else {
             // otherwise, file can be reused later
             self.file = Some(file);
@@ -270,11 +255,11 @@ impl PhysicalStorage {
     /// be flushed separately later.
     ///
     /// Updates `write_lsn`.
-    async fn write_exact(&mut self, pos: Lsn, mut buf: &[u8]) -> Result<()> {
+    fn write_exact(&mut self, pos: Lsn, mut buf: &[u8]) -> Result<()> {
         if self.write_lsn != pos {
             // need to flush the file before discarding it
             if let Some(mut file) = self.file.take() {
-                self.fdatasync_file(&mut file).await?;
+                self.fdatasync_file(&mut file)?;
             }
 
             self.write_lsn = pos;
@@ -292,8 +277,7 @@ impl PhysicalStorage {
                 buf.len()
             };
 
-            self.write_in_segment(segno, xlogoff, &buf[..bytes_write])
-                .await?;
+            self.write_in_segment(segno, xlogoff, &buf[..bytes_write])?;
             self.write_lsn += bytes_write as u64;
             buf = &buf[bytes_write..];
         }
@@ -302,7 +286,6 @@ impl PhysicalStorage {
     }
 }
 
-#[async_trait::async_trait]
 impl Storage for PhysicalStorage {
     /// flush_lsn returns LSN of last durably stored WAL record.
     fn flush_lsn(&self) -> Lsn {
@@ -310,7 +293,7 @@ impl Storage for PhysicalStorage {
     }
 
     /// Write WAL to disk.
-    async fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()> {
+    fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()> {
         // Disallow any non-sequential writes, which can result in gaps or overwrites.
         // If we need to move the pointer, use truncate_wal() instead.
         if self.write_lsn > startpos {
@@ -328,7 +311,7 @@ impl Storage for PhysicalStorage {
             );
         }
 
-        let write_seconds = time_io_closure(self.write_exact(startpos, buf)).await?;
+        let write_seconds = time_io_closure(|| self.write_exact(startpos, buf))?;
         // WAL is written, updating write metrics
         self.metrics.observe_write_seconds(write_seconds);
         self.metrics.observe_write_bytes(buf.len());
@@ -357,14 +340,14 @@ impl Storage for PhysicalStorage {
         Ok(())
     }
 
-    async fn flush_wal(&mut self) -> Result<()> {
+    fn flush_wal(&mut self) -> Result<()> {
         if self.flush_record_lsn == self.write_record_lsn {
             // no need to do extra flush
             return Ok(());
         }
 
         if let Some(mut unflushed_file) = self.file.take() {
-            self.fdatasync_file(&mut unflushed_file).await?;
+            self.fdatasync_file(&mut unflushed_file)?;
             self.file = Some(unflushed_file);
         } else {
             // We have unflushed data (write_lsn != flush_lsn), but no file.
@@ -386,7 +369,7 @@ impl Storage for PhysicalStorage {
 
     /// Truncate written WAL by removing all WAL segments after the given LSN.
     /// end_pos must point to the end of the WAL record.
-    async fn truncate_wal(&mut self, end_pos: Lsn) -> Result<()> {
+    fn truncate_wal(&mut self, end_pos: Lsn) -> Result<()> {
         // Streaming must not create a hole, so truncate cannot be called on non-written lsn
         if self.write_lsn != Lsn(0) && end_pos > self.write_lsn {
             bail!(
@@ -398,51 +381,47 @@ impl Storage for PhysicalStorage {
 
         // Quick exit if nothing to do to avoid writing up to 16 MiB of zeros on
         // disk (this happens on each connect).
-        if self.is_truncated_after_restart
-            && end_pos == self.write_lsn
-            && end_pos == self.flush_record_lsn
-        {
+        if end_pos == self.write_lsn {
             return Ok(());
         }
 
         // Close previously opened file, if any
         if let Some(mut unflushed_file) = self.file.take() {
-            self.fdatasync_file(&mut unflushed_file).await?;
+            self.fdatasync_file(&mut unflushed_file)?;
         }
 
         let xlogoff = end_pos.segment_offset(self.wal_seg_size);
         let segno = end_pos.segment_number(self.wal_seg_size);
 
         // Remove all segments after the given LSN.
-        remove_segments_from_disk(&self.timeline_dir, self.wal_seg_size, |x| x > segno).await?;
+        remove_segments_from_disk(&self.timeline_dir, self.wal_seg_size, |x| x > segno)?;
 
-        let (mut file, is_partial) = self.open_or_create(segno).await?;
+        let (mut file, is_partial) = self.open_or_create(segno)?;
 
         // Fill end with zeroes
-        file.seek(SeekFrom::Start(xlogoff as u64)).await?;
-        write_zeroes(&mut file, self.wal_seg_size - xlogoff).await?;
-        self.fdatasync_file(&mut file).await?;
+        file.seek(SeekFrom::Start(xlogoff as u64))?;
+        write_zeroes(&mut file, self.wal_seg_size - xlogoff)?;
+        self.fdatasync_file(&mut file)?;
 
         if !is_partial {
             // Make segment partial once again
             let (wal_file_path, wal_file_partial_path) =
                 wal_file_paths(&self.timeline_dir, segno, self.wal_seg_size)?;
-            fs::rename(wal_file_path, wal_file_partial_path).await?;
+            fs::rename(wal_file_path, wal_file_partial_path)?;
         }
 
         // Update LSNs
         self.write_lsn = end_pos;
         self.write_record_lsn = end_pos;
         self.flush_record_lsn = end_pos;
-        self.is_truncated_after_restart = true;
         Ok(())
     }
 
-    fn remove_up_to(&self, segno_up_to: XLogSegNo) -> BoxFuture<'static, anyhow::Result<()>> {
+    fn remove_up_to(&self) -> Box<dyn Fn(XLogSegNo) -> Result<()>> {
         let timeline_dir = self.timeline_dir.clone();
         let wal_seg_size = self.wal_seg_size;
-        Box::pin(async move {
-            remove_segments_from_disk(&timeline_dir, wal_seg_size, |x| x <= segno_up_to).await
+        Box::new(move |segno_up_to: XLogSegNo| {
+            remove_segments_from_disk(&timeline_dir, wal_seg_size, |x| x <= segno_up_to)
         })
     }
 
@@ -457,7 +436,7 @@ impl Storage for PhysicalStorage {
 }
 
 /// Remove all WAL segments in timeline_dir that match the given predicate.
-async fn remove_segments_from_disk(
+fn remove_segments_from_disk(
     timeline_dir: &Path,
     wal_seg_size: usize,
     remove_predicate: impl Fn(XLogSegNo) -> bool,
@@ -466,8 +445,8 @@ async fn remove_segments_from_disk(
     let mut min_removed = u64::MAX;
     let mut max_removed = u64::MIN;
 
-    let mut entries = fs::read_dir(timeline_dir).await?;
-    while let Some(entry) = entries.next_entry().await? {
+    for entry in fs::read_dir(timeline_dir)? {
+        let entry = entry?;
         let entry_path = entry.path();
         let fname = entry_path.file_name().unwrap();
 
@@ -478,7 +457,7 @@ async fn remove_segments_from_disk(
             }
             let (segno, _) = XLogFromFileName(fname_str, wal_seg_size);
             if remove_predicate(segno) {
-                remove_file(entry_path).await?;
+                remove_file(entry_path)?;
                 n_removed += 1;
                 min_removed = min(min_removed, segno);
                 max_removed = max(max_removed, segno);
@@ -710,12 +689,12 @@ impl WalReader {
 const ZERO_BLOCK: &[u8] = &[0u8; XLOG_BLCKSZ];
 
 /// Helper for filling file with zeroes.
-async fn write_zeroes(file: &mut File, mut count: usize) -> Result<()> {
+fn write_zeroes(file: &mut File, mut count: usize) -> Result<()> {
     while count >= XLOG_BLCKSZ {
-        file.write_all(ZERO_BLOCK).await?;
+        file.write_all(ZERO_BLOCK)?;
         count -= XLOG_BLCKSZ;
     }
-    file.write_all(&ZERO_BLOCK[0..count]).await?;
+    file.write_all(&ZERO_BLOCK[0..count])?;
     Ok(())
 }
 

scripts/comment-test-report.js

@@ -1,5 +1,3 @@
-#! /usr/bin/env node
-
 //
 // The script parses Allure reports and posts a comment with a summary of the test results to the PR or to the latest commit in the branch.
 //
@@ -21,7 +19,7 @@
 //       })
 //
 
-// Equivalent of Python's defaultdict.
+// Analog of Python's defaultdict.
 //
 // const dm = new DefaultMap(() => new DefaultMap(() => []))
 // dm["firstKey"]["secondKey"].push("value")
@@ -34,7 +32,34 @@ class DefaultMap extends Map {
     }
 }
 
-const parseReportJson = async ({ reportJsonUrl, fetch }) => {
+module.exports = async ({ github, context, fetch, report }) => {
+    // Marker to find the comment in the subsequent runs
+    const startMarker = `<!--AUTOMATIC COMMENT START #${context.payload.number}-->`
+    // If we run the script in the PR or in the branch (main/release/...)
+    const isPullRequest = !!context.payload.pull_request
+    // Latest commit in PR or in the branch
+    const commitSha = isPullRequest ? context.payload.pull_request.head.sha : context.sha
+    // Let users know that the comment is updated automatically
+    const autoupdateNotice = `<div align="right"><sub>The comment gets automatically updated with the latest test results<br>${commitSha} at ${new Date().toISOString()} :recycle:</sub></div>`
+    // GitHub bot id taken from (https://api.github.com/users/github-actions[bot])
+    const githubActionsBotId = 41898282
+    // Commend body itself
+    let commentBody = `${startMarker}\n`
+
+    // Common parameters for GitHub API requests
+    const ownerRepoParams = {
+        owner: context.repo.owner,
+        repo: context.repo.repo,
+    }
+
+    const {reportUrl, reportJsonUrl} = report
+
+    if (!reportUrl || !reportJsonUrl) {
+        commentBody += `#### No tests were run or test report is not available\n`
+        commentBody += autoupdateNotice
+        return
+    }
+
     const suites = await (await fetch(reportJsonUrl)).json()
 
     // Allure distinguishes "failed" (with an assertion error) and "broken" (with any other error) tests.
@@ -58,7 +83,7 @@ const parseReportJson = async ({ reportJsonUrl, fetch }) => {
                 let buildType, pgVersion
                 const match = test.name.match(/[\[-](?<buildType>debug|release)-pg(?<pgVersion>\d+)[-\]]/)?.groups
                 if (match) {
-                    ({ buildType, pgVersion } = match)
+                    ({buildType, pgVersion} = match)
                 } else {
                     // It's ok, we embed BUILD_TYPE and Postgres Version into the test name only for regress suite and do not for other suites (like performance).
                     console.info(`Cannot get BUILD_TYPE and Postgres Version from test name: "${test.name}", defaulting to "release" and "14"`)
@@ -98,68 +123,37 @@ const parseReportJson = async ({ reportJsonUrl, fetch }) => {
         }
     }
 
-    return {
-        failedTests,
-        failedTestsCount,
-        passedTests,
-        passedTestsCount,
-        skippedTests,
-        skippedTestsCount,
-        flakyTests,
-        flakyTestsCount,
-        retriedTests,
-        pgVersions,
-    }
-}
-
-const reportSummary = async (params) => {
-    const {
-        failedTests,
-        failedTestsCount,
-        passedTests,
-        passedTestsCount,
-        skippedTests,
-        skippedTestsCount,
-        flakyTests,
-        flakyTestsCount,
-        retriedTests,
-        pgVersions,
-        reportUrl,
-    } = params
-
-    let summary = ""
-
     const totalTestsCount = failedTestsCount + passedTestsCount + skippedTestsCount
-    summary += `### ${totalTestsCount} tests run: ${passedTestsCount} passed, ${failedTestsCount} failed, ${skippedTestsCount} skipped ([full report](${reportUrl}))\n___\n`
+    commentBody += `### ${totalTestsCount} tests run: ${passedTestsCount} passed, ${failedTestsCount} failed, ${skippedTestsCount} skipped ([full report](${reportUrl}))\n___\n`
 
     // Print test resuls from the newest to the oldest Postgres version for release and debug builds.
     for (const pgVersion of Array.from(pgVersions).sort().reverse()) {
         if (Object.keys(failedTests[pgVersion]).length > 0) {
-            summary += `#### Failures on Posgres ${pgVersion}\n\n`
+            commentBody += `#### Failures on Posgres ${pgVersion}\n\n`
             for (const [testName, tests] of Object.entries(failedTests[pgVersion])) {
                 const links = []
                 for (const test of tests) {
                     const allureLink = `${reportUrl}#suites/${test.parentUid}/${test.uid}`
                     links.push(`[${test.buildType}](${allureLink})`)
                 }
-                summary += `- \`${testName}\`: ${links.join(", ")}\n`
+                commentBody += `- \`${testName}\`: ${links.join(", ")}\n`
             }
 
             const testsToRerun = Object.values(failedTests[pgVersion]).map(x => x[0].name)
             const command = `DEFAULT_PG_VERSION=${pgVersion} scripts/pytest -k "${testsToRerun.join(" or ")}"`
 
-            summary += "```\n"
-            summary += `# Run failed on Postgres ${pgVersion} tests locally:\n`
-            summary += `${command}\n`
-            summary += "```\n"
+            commentBody += "```\n"
+            commentBody += `# Run failed on Postgres ${pgVersion} tests locally:\n`
+            commentBody += `${command}\n`
+            commentBody += "```\n"
         }
     }
 
     if (flakyTestsCount > 0) {
-        summary += `<details>\n<summary>Flaky tests (${flakyTestsCount})</summary>\n\n`
+        commentBody += `<details>\n<summary>Flaky tests (${flakyTestsCount})</summary>\n\n`
         for (const pgVersion of Array.from(pgVersions).sort().reverse()) {
             if (Object.keys(flakyTests[pgVersion]).length > 0) {
-                summary += `#### Postgres ${pgVersion}\n\n`
+                commentBody += `#### Postgres ${pgVersion}\n\n`
                 for (const [testName, tests] of Object.entries(flakyTests[pgVersion])) {
                     const links = []
                     for (const test of tests) {
@@ -167,57 +161,11 @@ const reportSummary = async (params) => {
                         const status = test.status === "passed" ? ":white_check_mark:" : ":x:"
                         links.push(`[${status} ${test.buildType}](${allureLink})`)
                     }
-                    summary += `- \`${testName}\`: ${links.join(", ")}\n`
+                    commentBody += `- \`${testName}\`: ${links.join(", ")}\n`
                 }
             }
         }
-        summary += "\n</details>\n"
-    }
-
-    return summary
-}
-
-module.exports = async ({ github, context, fetch, report }) => {
-    // Marker to find the comment in the subsequent runs
-    const startMarker = `<!--AUTOMATIC COMMENT START #${context.payload.number}-->`
-    // If we run the script in the PR or in the branch (main/release/...)
-    const isPullRequest = !!context.payload.pull_request
-    // Latest commit in PR or in the branch
-    const commitSha = isPullRequest ? context.payload.pull_request.head.sha : context.sha
-    // Let users know that the comment is updated automatically
-    const autoupdateNotice = `<div align="right"><sub>The comment gets automatically updated with the latest test results<br>${commitSha} at ${new Date().toISOString()} :recycle:</sub></div>`
-    // GitHub bot id taken from (https://api.github.com/users/github-actions[bot])
-    const githubActionsBotId = 41898282
-    // Commend body itself
-    let commentBody = `${startMarker}\n`
-
-    // Common parameters for GitHub API requests
-    const ownerRepoParams = {
-        owner: context.repo.owner,
-        repo: context.repo.repo,
-    }
-
-    const {reportUrl, reportJsonUrl} = report
-
-    if (!reportUrl || !reportJsonUrl) {
-        commentBody += `#### No tests were run or test report is not available\n`
-        commentBody += autoupdateNotice
-        return
-    }
-
-    try {
-        const parsed = await parseReportJson({ reportJsonUrl, fetch })
-        commentBody += await reportSummary({ ...parsed, reportUrl })
-    } catch (error) {
-        commentBody += `### [full report](${reportUrl})\n___\n`
-        commentBody += `#### Failed to create a summary for the test run: \n`
-        commentBody += "```\n"
-        commentBody += `${error.stack}\n`
-        commentBody += "```\n"
-        commentBody += "\nTo reproduce and debug the error locally run:\n"
-        commentBody += "```\n"
-        commentBody += `scripts/comment-test-report.js ${reportJsonUrl}`
-        commentBody += "\n```\n"
+        commentBody += "\n</details>\n"
     }
 
     commentBody += autoupdateNotice
@@ -259,60 +207,3 @@ module.exports = async ({ github, context, fetch, report }) => {
         })
     }
 }
-
-// Equivalent of Python's `if __name__ == "__main__":`
-// https://nodejs.org/docs/latest/api/modules.html#accessing-the-main-module
-if (require.main === module) {
-    // Poor man's argument parsing: we expect the third argument is a JSON URL (0: node binary, 1: this script, 2: JSON url)
-    if (process.argv.length !== 3) {
-        console.error(`Unexpected number of arguments\nUsage: node ${process.argv[1]} <jsonUrl>`)
-        process.exit(1)
-    }
-    const jsonUrl = process.argv[2]
-
-    try {
-        new URL(jsonUrl)
-    } catch (error) {
-        console.error(`Invalid URL: ${jsonUrl}\nUsage: node ${process.argv[1]} <jsonUrl>`)
-        process.exit(1)
-    }
-
-    const htmlUrl = jsonUrl.replace("/data/suites.json", "/index.html")
-
-    const githubMock = {
-        rest: {
-            issues: {
-                createComment: console.log,
-                listComments: async () => ({ data: [] }),
-                updateComment: console.log
-            },
-            repos: {
-                createCommitComment: console.log,
-                listCommentsForCommit: async () => ({ data: [] }),
-                updateCommitComment: console.log
-            }
-        }
-    }
-
-    const contextMock = {
-        repo: {
-            owner: 'testOwner',
-            repo: 'testRepo'
-        },
-        payload: {
-            number: 42,
-            pull_request: null,
-        },
-        sha: '0000000000000000000000000000000000000000',
-    }
-
-    module.exports({
-        github: githubMock,
-        context: contextMock,
-        fetch: fetch,
-        report: {
-            reportUrl: htmlUrl,
-            reportJsonUrl: jsonUrl,
-        }
-    })
-}

scripts/ingest_perf_test_result.py

@@ -1,14 +1,12 @@
 #!/usr/bin/env python3
 import argparse
 import json
-import logging
 import os
 import sys
 from contextlib import contextmanager
 from datetime import datetime
 from pathlib import Path
 
-import backoff
 import psycopg2
 import psycopg2.extras
 
@@ -37,20 +35,9 @@ def get_connection_cursor():
     connstr = os.getenv("DATABASE_URL")
     if not connstr:
         err("DATABASE_URL environment variable is not set")
-
-    @backoff.on_exception(backoff.expo, psycopg2.OperationalError, max_time=150)
-    def connect(connstr):
-        conn = psycopg2.connect(connstr, connect_timeout=30)
-        conn.autocommit = True
-        return conn
-
-    conn = connect(connstr)
-    try:
+    with psycopg2.connect(connstr, connect_timeout=30) as conn:
         with conn.cursor() as cur:
             yield cur
-    finally:
-        if conn is not None:
-            conn.close()
 
 
 def create_table(cur):
@@ -128,7 +115,6 @@ def main():
     parser.add_argument(
         "--ingest",
         type=Path,
-        required=True,
         help="Path to perf test result file, or directory with perf test result files",
     )
     parser.add_argument("--initdb", action="store_true", help="Initialuze database")
@@ -154,5 +140,4 @@ def main():
 
 
 if __name__ == "__main__":
-    logging.getLogger("backoff").addHandler(logging.StreamHandler())
     main()

scripts/ingest_regress_test_result.py

@@ -1,13 +1,11 @@
 #!/usr/bin/env python3
 import argparse
-import logging
 import os
 import re
 import sys
 from contextlib import contextmanager
 from pathlib import Path
 
-import backoff
 import psycopg2
 
 CREATE_TABLE = """
@@ -31,20 +29,9 @@ def get_connection_cursor():
     connstr = os.getenv("DATABASE_URL")
     if not connstr:
         err("DATABASE_URL environment variable is not set")
-
-    @backoff.on_exception(backoff.expo, psycopg2.OperationalError, max_time=150)
-    def connect(connstr):
-        conn = psycopg2.connect(connstr, connect_timeout=30)
-        conn.autocommit = True
-        return conn
-
-    conn = connect(connstr)
-    try:
+    with psycopg2.connect(connstr, connect_timeout=30) as conn:
         with conn.cursor() as cur:
             yield cur
-    finally:
-        if conn is not None:
-            conn.close()
 
 
 def create_table(cur):
@@ -114,5 +101,4 @@ def main():
 
 
 if __name__ == "__main__":
-    logging.getLogger("backoff").addHandler(logging.StreamHandler())
     main()

storage_broker/src/lib.rs

@@ -32,7 +32,6 @@ pub const DEFAULT_LISTEN_ADDR: &str = "127.0.0.1:50051";
 pub const DEFAULT_ENDPOINT: &str = const_format::formatcp!("http://{DEFAULT_LISTEN_ADDR}");
 
 pub const DEFAULT_KEEPALIVE_INTERVAL: &str = "5000 ms";
-pub const DEFAULT_CONNECT_TIMEOUT: Duration = Duration::from_millis(5000);
 
 // BrokerServiceClient charged with tonic provided Channel transport; helps to
 // avoid depending on tonic directly in user crates.
@@ -59,8 +58,7 @@ where
     }
     tonic_endpoint = tonic_endpoint
         .http2_keep_alive_interval(keepalive_interval)
-        .keep_alive_while_idle(true)
-        .connect_timeout(DEFAULT_CONNECT_TIMEOUT);
+        .keep_alive_while_idle(true);
     //  keep_alive_timeout is 20s by default on both client and server side
     let channel = tonic_endpoint.connect_lazy();
     Ok(BrokerClientChannel::new(channel))

test_runner/fixtures/flaky.py

@@ -1,6 +1,6 @@
 import json
 from pathlib import Path
-from typing import Any, List, MutableMapping, cast
+from typing import List
 
 import pytest
 from _pytest.config import Config
@@ -56,15 +56,3 @@ def pytest_collection_modifyitems(config: Config, items: List[pytest.Item]):
             # Rerun 3 times = 1 original run + 2 reruns
             log.info(f"Marking {item.nodeid} as flaky. It will be rerun up to 3 times")
             item.add_marker(pytest.mark.flaky(reruns=2))
-
-            # pytest-rerunfailures is not compatible with pytest-timeout (timeout is not set for reruns),
-            #   we can workaround it by setting `timeout_func_only` to True[1].
-            # Unfortunately, setting `timeout_func_only = True` globally in pytest.ini is broken[2],
-            #   but we still can do it using pytest marker.
-            #
-            # - [1] https://github.com/pytest-dev/pytest-rerunfailures/issues/99
-            # - [2] https://github.com/pytest-dev/pytest-timeout/issues/142
-            timeout_marker = item.get_closest_marker("timeout")
-            if timeout_marker is not None:
-                kwargs = cast(MutableMapping[str, Any], timeout_marker.kwargs)
-                kwargs["func_only"] = True

test_runner/fixtures/metrics.py

@@ -57,16 +57,14 @@ PAGESERVER_GLOBAL_METRICS: Tuple[str, ...] = (
     "libmetrics_launch_timestamp",
     "libmetrics_build_info",
     "libmetrics_tracing_event_count_total",
-    "pageserver_materialized_cache_hits_total",
-    "pageserver_materialized_cache_hits_direct_total",
-    "pageserver_getpage_reconstruct_seconds_bucket",
-    "pageserver_getpage_reconstruct_seconds_count",
-    "pageserver_getpage_reconstruct_seconds_sum",
 )
 
 PAGESERVER_PER_TENANT_METRICS: Tuple[str, ...] = (
     "pageserver_current_logical_size",
     "pageserver_resident_physical_size",
+    "pageserver_getpage_reconstruct_seconds_bucket",
+    "pageserver_getpage_reconstruct_seconds_count",
+    "pageserver_getpage_reconstruct_seconds_sum",
     "pageserver_getpage_get_reconstruct_data_seconds_bucket",
     "pageserver_getpage_get_reconstruct_data_seconds_count",
     "pageserver_getpage_get_reconstruct_data_seconds_sum",
@@ -75,6 +73,8 @@ PAGESERVER_PER_TENANT_METRICS: Tuple[str, ...] = (
     "pageserver_io_operations_seconds_count",
     "pageserver_io_operations_seconds_sum",
     "pageserver_last_record_lsn",
+    "pageserver_materialized_cache_hits_total",
+    "pageserver_materialized_cache_hits_direct_total",
     "pageserver_read_num_fs_layers_bucket",
     "pageserver_read_num_fs_layers_count",
     "pageserver_read_num_fs_layers_sum",

test_runner/fixtures/neon_fixtures.py

@@ -1631,8 +1631,6 @@ class NeonPageserver(PgProtocol):
             r".*ERROR.*ancestor timeline \S+ is being stopped",
             # this is expected given our collaborative shutdown approach for the UploadQueue
             ".*Compaction failed, retrying in .*: queue is in state Stopped.*",
-            # Pageserver timeline deletion should be polled until it gets 404, so ignore it globally
-            ".*Error processing HTTP request: NotFound: Timeline .* was not found",
         ]
 
     def start(
@@ -2914,6 +2912,7 @@ SKIP_FILES = frozenset(
         "pg_internal.init",
         "pg.log",
         "zenith.signal",
+        "neon_compute_spec_id.txt",
         "pg_hba.conf",
         "postgresql.conf",
         "postmaster.opts",

test_runner/fixtures/pageserver/http.py

@@ -342,11 +342,6 @@ class PageserverHttpClient(requests.Session):
         return res_json
 
     def timeline_delete(self, tenant_id: TenantId, timeline_id: TimelineId, **kwargs):
-        """
-        Note that deletion is not instant, it is scheduled and performed mostly in the background.
-        So if you need to wait for it to complete use `timeline_delete_wait_completed`.
-        For longer description consult with pageserver openapi spec.
-        """
         res = self.delete(
             f"http://localhost:{self.port}/v1/tenant/{tenant_id}/timeline/{timeline_id}", **kwargs
         )

test_runner/fixtures/pageserver/utils.py

@@ -193,30 +193,19 @@ def wait_for_upload_queue_empty(
         time.sleep(0.2)
 
 
-def wait_timeline_detail_404(
-    pageserver_http: PageserverHttpClient, tenant_id: TenantId, timeline_id: TimelineId
-):
-    last_exc = None
-    for _ in range(2):
-        time.sleep(0.250)
-        try:
-            data = pageserver_http.timeline_detail(tenant_id, timeline_id)
-            log.error(f"detail {data}")
-        except PageserverApiException as e:
-            log.debug(e)
-            if e.status_code == 404:
-                return
-
-            last_exc = e
-
-    raise last_exc or RuntimeError(f"Timeline wasnt deleted in time, state: {data['state']}")
-
-
-def timeline_delete_wait_completed(
+def assert_timeline_detail_404(
     pageserver_http: PageserverHttpClient,
     tenant_id: TenantId,
     timeline_id: TimelineId,
-    **delete_args,
 ):
-    pageserver_http.timeline_delete(tenant_id=tenant_id, timeline_id=timeline_id, **delete_args)
-    wait_timeline_detail_404(pageserver_http, tenant_id, timeline_id)
+    """Asserts that timeline_detail returns 404, or dumps the detail."""
+    try:
+        data = pageserver_http.timeline_detail(tenant_id, timeline_id)
+        log.error(f"detail {data}")
+    except PageserverApiException as e:
+        log.error(e)
+        if e.status_code == 404:
+            return
+        else:
+            raise
+    raise Exception("detail succeeded (it should return 404)")

test_runner/regress/test_compatibility.py

@@ -15,11 +15,7 @@ from fixtures.neon_fixtures import (
     PortDistributor,
 )
 from fixtures.pageserver.http import PageserverHttpClient
-from fixtures.pageserver.utils import (
-    timeline_delete_wait_completed,
-    wait_for_last_record_lsn,
-    wait_for_upload,
-)
+from fixtures.pageserver.utils import wait_for_last_record_lsn, wait_for_upload
 from fixtures.pg_version import PgVersion
 from fixtures.types import Lsn
 from pytest import FixtureRequest
@@ -421,7 +417,7 @@ def check_neon_works(
     )
 
     shutil.rmtree(repo_dir / "local_fs_remote_storage")
-    timeline_delete_wait_completed(pageserver_http, tenant_id, timeline_id)
+    pageserver_http.timeline_delete(tenant_id, timeline_id)
     pageserver_http.timeline_create(pg_version, tenant_id, timeline_id)
     pg_bin.run(
         ["pg_dumpall", f"--dbname={connstr}", f"--file={test_output_dir / 'dump-from-wal.sql'}"]

test_runner/regress/test_import.py

@@ -14,11 +14,7 @@ from fixtures.neon_fixtures import (
     NeonEnvBuilder,
     PgBin,
 )
-from fixtures.pageserver.utils import (
-    timeline_delete_wait_completed,
-    wait_for_last_record_lsn,
-    wait_for_upload,
-)
+from fixtures.pageserver.utils import wait_for_last_record_lsn, wait_for_upload
 from fixtures.types import Lsn, TenantId, TimelineId
 from fixtures.utils import subprocess_capture
 
@@ -155,7 +151,7 @@ def test_import_from_vanilla(test_output_dir, pg_bin, vanilla_pg, neon_env_build
         ".*files not bound to index_file.json, proceeding with their deletion.*"
     )
 
-    timeline_delete_wait_completed(client, tenant, timeline)
+    client.timeline_delete(tenant, timeline)
 
     # Importing correct backup works
     import_tar(base_tar, wal_tar)

test_runner/regress/test_layer_eviction.py

@@ -24,13 +24,7 @@ def test_basic_eviction(
         test_name="test_download_remote_layers_api",
     )
 
-    env = neon_env_builder.init_start(
-        initial_tenant_conf={
-            # disable gc and compaction background loops because they perform on-demand downloads
-            "gc_period": "0s",
-            "compaction_period": "0s",
-        }
-    )
+    env = neon_env_builder.init_start()
     client = env.pageserver.http_client()
     endpoint = env.endpoints.create_start("main")
 
@@ -53,11 +47,6 @@ def test_basic_eviction(
     client.timeline_checkpoint(tenant_id, timeline_id)
     wait_for_upload(client, tenant_id, timeline_id, current_lsn)
 
-    # disable compute & sks to avoid on-demand downloads by walreceiver / getpage
-    endpoint.stop()
-    for sk in env.safekeepers:
-        sk.stop()
-
     timeline_path = env.repo_dir / "tenants" / str(tenant_id) / "timelines" / str(timeline_id)
     initial_local_layers = sorted(
         list(filter(lambda path: path.name != "metadata", timeline_path.glob("*")))

test_runner/regress/test_proxy.py

@@ -163,6 +163,7 @@ def test_forward_params_to_client(static_proxy: NeonProxy):
                 assert conn.get_parameter_status(name) == value
 
 
+@pytest.mark.timeout(5)
 def test_close_on_connections_exit(static_proxy: NeonProxy):
     # Open two connections, send SIGTERM, then ensure that proxy doesn't exit
     # until after connections close.

test_runner/regress/test_remote_storage.py

@@ -20,7 +20,7 @@ from fixtures.neon_fixtures import (
 )
 from fixtures.pageserver.http import PageserverApiException, PageserverHttpClient
 from fixtures.pageserver.utils import (
-    timeline_delete_wait_completed,
+    assert_timeline_detail_404,
     wait_for_last_record_lsn,
     wait_for_upload,
     wait_until_tenant_active,
@@ -597,11 +597,14 @@ def test_timeline_deletion_with_files_stuck_in_upload_queue(
     env.pageserver.allowed_errors.append(
         ".* ERROR .*Error processing HTTP request: InternalServerError\\(timeline is Stopping"
     )
+    client.timeline_delete(tenant_id, timeline_id)
 
+    env.pageserver.allowed_errors.append(f".*Timeline {tenant_id}/{timeline_id} was not found.*")
     env.pageserver.allowed_errors.append(
         ".*files not bound to index_file.json, proceeding with their deletion.*"
     )
-    timeline_delete_wait_completed(client, tenant_id, timeline_id)
+
+    wait_until(2, 0.5, lambda: assert_timeline_detail_404(client, tenant_id, timeline_id))
 
     assert not timeline_path.exists()
 

test_runner/regress/test_tenant_size.py

@@ -11,7 +11,6 @@ from fixtures.neon_fixtures import (
     wait_for_wal_insert_lsn,
 )
 from fixtures.pageserver.http import PageserverHttpClient
-from fixtures.pageserver.utils import timeline_delete_wait_completed
 from fixtures.pg_version import PgVersion, xfail_on_postgres
 from fixtures.types import Lsn, TenantId, TimelineId
 
@@ -629,12 +628,12 @@ def test_get_tenant_size_with_multiple_branches(
     size_debug_file_before.write(size_debug)
 
     # teardown, delete branches, and the size should be going down
-    timeline_delete_wait_completed(http_client, tenant_id, first_branch_timeline_id)
+    http_client.timeline_delete(tenant_id, first_branch_timeline_id)
 
     size_after_deleting_first = http_client.tenant_size(tenant_id)
     assert size_after_deleting_first < size_after_thinning_branch
 
-    timeline_delete_wait_completed(http_client, tenant_id, second_branch_timeline_id)
+    http_client.timeline_delete(tenant_id, second_branch_timeline_id)
     size_after_deleting_second = http_client.tenant_size(tenant_id)
     assert size_after_deleting_second < size_after_deleting_first
 

test_runner/regress/test_tenant_tasks.py

@@ -1,10 +1,6 @@
 from fixtures.log_helper import log
 from fixtures.neon_fixtures import NeonEnvBuilder
-from fixtures.pageserver.utils import (
-    assert_tenant_state,
-    timeline_delete_wait_completed,
-    wait_until_tenant_active,
-)
+from fixtures.pageserver.utils import assert_tenant_state, wait_until_tenant_active
 from fixtures.types import TenantId, TimelineId
 from fixtures.utils import wait_until
 
@@ -28,7 +24,7 @@ def test_tenant_tasks(neon_env_builder: NeonEnvBuilder):
     def delete_all_timelines(tenant: TenantId):
         timelines = [TimelineId(t["timeline_id"]) for t in client.timeline_list(tenant)]
         for t in timelines:
-            timeline_delete_wait_completed(client, tenant, t)
+            client.timeline_delete(tenant, t)
 
     # Create tenant, start compute
     tenant, _ = env.neon_cli.create_tenant()

test_runner/regress/test_tenants.py

@@ -21,7 +21,6 @@ from fixtures.neon_fixtures import (
     RemoteStorageKind,
     available_remote_storages,
 )
-from fixtures.pageserver.utils import timeline_delete_wait_completed
 from fixtures.types import Lsn, TenantId, TimelineId
 from fixtures.utils import wait_until
 from prometheus_client.samples import Sample
@@ -214,7 +213,7 @@ def test_metrics_normal_work(neon_env_builder: NeonEnvBuilder):
     # Test (a subset of) pageserver global metrics
     for metric in PAGESERVER_GLOBAL_METRICS:
         ps_samples = ps_metrics.query_all(metric, {})
-        assert len(ps_samples) > 0, f"expected at least one sample for {metric}"
+        assert len(ps_samples) > 0
         for sample in ps_samples:
             labels = ",".join([f'{key}="{value}"' for key, value in sample.labels.items()])
             log.info(f"{sample.name}{{{labels}}} {sample.value}")
@@ -319,10 +318,9 @@ def test_pageserver_with_empty_tenants(
     client.tenant_create(tenant_with_empty_timelines)
     temp_timelines = client.timeline_list(tenant_with_empty_timelines)
     for temp_timeline in temp_timelines:
-        timeline_delete_wait_completed(
-            client, tenant_with_empty_timelines, TimelineId(temp_timeline["timeline_id"])
+        client.timeline_delete(
+            tenant_with_empty_timelines, TimelineId(temp_timeline["timeline_id"])
         )
-
     files_in_timelines_dir = sum(
         1
         for _p in Path.iterdir(

test_runner/regress/test_timeline_delete.py

@@ -17,10 +17,9 @@ from fixtures.neon_fixtures import (
 )
 from fixtures.pageserver.http import PageserverApiException
 from fixtures.pageserver.utils import (
-    timeline_delete_wait_completed,
+    assert_timeline_detail_404,
     wait_for_last_record_lsn,
     wait_for_upload,
-    wait_timeline_detail_404,
     wait_until_tenant_active,
     wait_until_timeline_state,
 )
@@ -84,7 +83,7 @@ def test_timeline_delete(neon_simple_env: NeonEnv):
     wait_until(
         number_of_iterations=3,
         interval=0.2,
-        func=lambda: timeline_delete_wait_completed(ps_http, env.initial_tenant, leaf_timeline_id),
+        func=lambda: ps_http.timeline_delete(env.initial_tenant, leaf_timeline_id),
     )
 
     assert not timeline_path.exists()
@@ -95,15 +94,15 @@ def test_timeline_delete(neon_simple_env: NeonEnv):
         match=f"Timeline {env.initial_tenant}/{leaf_timeline_id} was not found",
     ) as exc:
         ps_http.timeline_detail(env.initial_tenant, leaf_timeline_id)
-    assert exc.value.status_code == 404
 
-    wait_until(
-        number_of_iterations=3,
-        interval=0.2,
-        func=lambda: timeline_delete_wait_completed(
-            ps_http, env.initial_tenant, parent_timeline_id
-        ),
-    )
+        # FIXME leaves tenant without timelines, should we prevent deletion of root timeline?
+        wait_until(
+            number_of_iterations=3,
+            interval=0.2,
+            func=lambda: ps_http.timeline_delete(env.initial_tenant, parent_timeline_id),
+        )
+
+    assert exc.value.status_code == 404
 
     # Check that we didn't pick up the timeline again after restart.
     # See https://github.com/neondatabase/neon/issues/3560
@@ -144,6 +143,7 @@ def test_delete_timeline_post_rm_failure(
     ps_http.configure_failpoints((failpoint_name, "return"))
 
     ps_http.timeline_delete(env.initial_tenant, env.initial_timeline)
+
     timeline_info = wait_until_timeline_state(
         pageserver_http=ps_http,
         tenant_id=env.initial_tenant,
@@ -165,7 +165,13 @@ def test_delete_timeline_post_rm_failure(
 
     # this should succeed
     # this also checks that delete can be retried even when timeline is in Broken state
-    timeline_delete_wait_completed(ps_http, env.initial_tenant, env.initial_timeline)
+    ps_http.timeline_delete(env.initial_tenant, env.initial_timeline, timeout=2)
+    with pytest.raises(PageserverApiException) as e:
+        ps_http.timeline_detail(env.initial_tenant, env.initial_timeline)
+
+    assert e.value.status_code == 404
+
+    env.pageserver.allowed_errors.append(f".*NotFound: Timeline.*{env.initial_timeline}.*")
     env.pageserver.allowed_errors.append(
         f".*{env.initial_timeline}.*timeline directory not found, proceeding anyway.*"
     )
@@ -241,7 +247,13 @@ def test_timeline_resurrection_on_attach(
         pass
 
     # delete new timeline
-    timeline_delete_wait_completed(ps_http, tenant_id=tenant_id, timeline_id=branch_timeline_id)
+    ps_http.timeline_delete(tenant_id=tenant_id, timeline_id=branch_timeline_id)
+
+    env.pageserver.allowed_errors.append(
+        f".*Timeline {tenant_id}/{branch_timeline_id} was not found.*"
+    )
+
+    wait_until(2, 0.5, lambda: assert_timeline_detail_404(ps_http, tenant_id, branch_timeline_id))
 
     ##### Stop the pageserver instance, erase all its data
     env.endpoints.stop_all()
@@ -326,6 +338,7 @@ def test_timeline_delete_fail_before_local_delete(neon_env_builder: NeonEnvBuild
     )
 
     ps_http.timeline_delete(env.initial_tenant, leaf_timeline_id)
+
     timeline_info = wait_until_timeline_state(
         pageserver_http=ps_http,
         tenant_id=env.initial_tenant,
@@ -344,15 +357,12 @@ def test_timeline_delete_fail_before_local_delete(neon_env_builder: NeonEnvBuild
     # Wait for tenant to finish loading.
     wait_until_tenant_active(ps_http, tenant_id=env.initial_tenant, iterations=10, period=1)
 
-    try:
-        data = ps_http.timeline_detail(env.initial_tenant, leaf_timeline_id)
-        log.debug(f"detail {data}")
-    except PageserverApiException as e:
-        log.debug(e)
-        if e.status_code != 404:
-            raise
-    else:
-        raise Exception("detail succeeded (it should return 404)")
+    env.pageserver.allowed_errors.append(
+        f".*Timeline {env.initial_tenant}/{leaf_timeline_id} was not found.*"
+    )
+    wait_until(
+        2, 0.5, lambda: assert_timeline_detail_404(ps_http, env.initial_tenant, leaf_timeline_id)
+    )
 
     assert (
         not leaf_timeline_path.exists()
@@ -379,8 +389,13 @@ def test_timeline_delete_fail_before_local_delete(neon_env_builder: NeonEnvBuild
     assert env.initial_timeline is not None
 
     for timeline_id in (intermediate_timeline_id, env.initial_timeline):
-        timeline_delete_wait_completed(
-            ps_http, tenant_id=env.initial_tenant, timeline_id=timeline_id
+        ps_http.timeline_delete(env.initial_tenant, timeline_id)
+
+        env.pageserver.allowed_errors.append(
+            f".*Timeline {env.initial_tenant}/{timeline_id} was not found.*"
+        )
+        wait_until(
+            2, 0.5, lambda: assert_timeline_detail_404(ps_http, env.initial_tenant, timeline_id)
         )
 
         assert_prefix_empty(
@@ -404,27 +419,23 @@ def test_timeline_delete_fail_before_local_delete(neon_env_builder: NeonEnvBuild
     )
 
 
-@pytest.mark.parametrize(
-    "stuck_failpoint",
-    ["persist_deleted_index_part", "in_progress_delete"],
-)
-def test_concurrent_timeline_delete_stuck_on(
-    neon_env_builder: NeonEnvBuilder, stuck_failpoint: str
+def test_concurrent_timeline_delete_if_first_stuck_at_index_upload(
+    neon_env_builder: NeonEnvBuilder,
 ):
     """
-    If delete is stuck console will eventually retry deletion.
-    So we need to be sure that these requests wont interleave with each other.
-    In this tests we check two places where we can spend a lot of time.
-    This is a regression test because there was a bug when DeletionGuard wasnt propagated
-    to the background task.
-
-    Ensure that when retry comes if we're still stuck request will get an immediate error response,
-    signalling to console that it should retry later.
+    If we're stuck uploading the index file with the is_delete flag,
+    eventually console will hand up and retry.
+    If we're still stuck at the retry time, ensure that the retry
+    fails with status 500, signalling to console that it should retry
+    later.
+    Ideally, timeline_delete should return 202 Accepted and require
+    console to poll for completion, but, that would require changing
+    the API contract.
     """
 
     neon_env_builder.enable_remote_storage(
         remote_storage_kind=RemoteStorageKind.MOCK_S3,
-        test_name=f"concurrent_timeline_delete_stuck_on_{stuck_failpoint}",
+        test_name="test_concurrent_timeline_delete_if_first_stuck_at_index_upload",
     )
 
     env = neon_env_builder.init_start()
@@ -434,14 +445,13 @@ def test_concurrent_timeline_delete_stuck_on(
     ps_http = env.pageserver.http_client()
 
     # make the first call sleep practically forever
-    ps_http.configure_failpoints((stuck_failpoint, "pause"))
+    failpoint_name = "persist_index_part_with_deleted_flag_after_set_before_upload_pause"
+    ps_http.configure_failpoints((failpoint_name, "pause"))
 
     def first_call(result_queue):
         try:
             log.info("first call start")
-            timeline_delete_wait_completed(
-                ps_http, env.initial_tenant, child_timeline_id, timeout=10
-            )
+            ps_http.timeline_delete(env.initial_tenant, child_timeline_id, timeout=10)
             log.info("first call success")
             result_queue.put("success")
         except Exception:
@@ -456,7 +466,7 @@ def test_concurrent_timeline_delete_stuck_on(
 
         def first_call_hit_failpoint():
             assert env.pageserver.log_contains(
-                f".*{child_timeline_id}.*at failpoint {stuck_failpoint}"
+                f".*{child_timeline_id}.*at failpoint {failpoint_name}"
             )
 
         wait_until(50, 0.1, first_call_hit_failpoint)
@@ -474,12 +484,8 @@ def test_concurrent_timeline_delete_stuck_on(
         )
         log.info("second call failed as expected")
 
-        # ensure it is not 404 and stopping
-        detail = ps_http.timeline_detail(env.initial_tenant, child_timeline_id)
-        assert detail["state"] == "Stopping"
-
         # by now we know that the second call failed, let's ensure the first call will finish
-        ps_http.configure_failpoints((stuck_failpoint, "off"))
+        ps_http.configure_failpoints((failpoint_name, "off"))
 
         result = first_call_result.get()
         assert result == "success"
@@ -492,10 +498,8 @@ def test_concurrent_timeline_delete_stuck_on(
 
 def test_delete_timeline_client_hangup(neon_env_builder: NeonEnvBuilder):
     """
-    If the client hangs up before we start the index part upload but after deletion is scheduled
-    we mark it
+    If the client hangs up before we start the index part upload but after we mark it
     deleted in local memory, a subsequent delete_timeline call should be able to do
-
     another delete timeline operation.
 
     This tests cancel safety up to the given failpoint.
@@ -511,18 +515,12 @@ def test_delete_timeline_client_hangup(neon_env_builder: NeonEnvBuilder):
 
     ps_http = env.pageserver.http_client()
 
-    failpoint_name = "persist_deleted_index_part"
+    failpoint_name = "persist_index_part_with_deleted_flag_after_set_before_upload_pause"
     ps_http.configure_failpoints((failpoint_name, "pause"))
 
     with pytest.raises(requests.exceptions.Timeout):
         ps_http.timeline_delete(env.initial_tenant, child_timeline_id, timeout=2)
 
-    env.pageserver.allowed_errors.append(
-        f".*{child_timeline_id}.*timeline deletion is already in progress.*"
-    )
-    with pytest.raises(PageserverApiException, match="timeline deletion is already in progress"):
-        ps_http.timeline_delete(env.initial_tenant, child_timeline_id, timeout=2)
-
     # make sure the timeout was due to the failpoint
     at_failpoint_log_message = f".*{child_timeline_id}.*at failpoint {failpoint_name}.*"
 
@@ -554,7 +552,12 @@ def test_delete_timeline_client_hangup(neon_env_builder: NeonEnvBuilder):
     wait_until(50, 0.1, first_request_finished)
 
     # check that the timeline is gone
-    wait_timeline_detail_404(ps_http, env.initial_tenant, child_timeline_id)
+    notfound_message = f"Timeline {env.initial_tenant}/{child_timeline_id} was not found"
+    env.pageserver.allowed_errors.append(".*" + notfound_message)
+    with pytest.raises(PageserverApiException, match=notfound_message) as exc:
+        ps_http.timeline_detail(env.initial_tenant, child_timeline_id)
+
+    assert exc.value.status_code == 404
 
 
 @pytest.mark.parametrize(
@@ -613,7 +616,12 @@ def test_timeline_delete_works_for_remote_smoke(
     for timeline_id in reversed(timeline_ids):
         # note that we need to finish previous deletion before scheduling next one
         # otherwise we can get an "HasChildren" error if deletion is not fast enough (real_s3)
-        timeline_delete_wait_completed(ps_http, tenant_id=tenant_id, timeline_id=timeline_id)
+        ps_http.timeline_delete(tenant_id=tenant_id, timeline_id=timeline_id)
+
+        env.pageserver.allowed_errors.append(
+            f".*Timeline {env.initial_tenant}/{timeline_id} was not found.*"
+        )
+        wait_until(2, 0.5, lambda: assert_timeline_detail_404(ps_http, tenant_id, timeline_id))
 
         assert_prefix_empty(
             neon_env_builder,

test_runner/regress/test_timeline_size.py

@@ -24,7 +24,6 @@ from fixtures.neon_fixtures import (
 from fixtures.pageserver.http import PageserverApiException, PageserverHttpClient
 from fixtures.pageserver.utils import (
     assert_tenant_state,
-    timeline_delete_wait_completed,
     wait_for_upload_queue_empty,
     wait_until_tenant_active,
 )
@@ -273,7 +272,7 @@ def test_timeline_initial_logical_size_calculation_cancellation(
             if deletion_method == "tenant_detach":
                 client.tenant_detach(tenant_id)
             elif deletion_method == "timeline_delete":
-                timeline_delete_wait_completed(client, tenant_id, timeline_id)
+                client.timeline_delete(tenant_id, timeline_id)
             delete_timeline_success.put(True)
         except PageserverApiException:
             delete_timeline_success.put(False)

test_runner/regress/test_wal_acceptor.py

@@ -31,11 +31,7 @@ from fixtures.neon_fixtures import (
     SafekeeperPort,
     available_remote_storages,
 )
-from fixtures.pageserver.utils import (
-    timeline_delete_wait_completed,
-    wait_for_last_record_lsn,
-    wait_for_upload,
-)
+from fixtures.pageserver.utils import wait_for_last_record_lsn, wait_for_upload
 from fixtures.pg_version import PgVersion
 from fixtures.types import Lsn, TenantId, TimelineId
 from fixtures.utils import get_dir_size, query_scalar, start_in_background
@@ -552,15 +548,15 @@ def test_s3_wal_replay(neon_env_builder: NeonEnvBuilder, remote_storage_kind: Re
                     f"sk_id={sk.id} to flush {last_lsn}",
                 )
 
-    ps_http = env.pageserver.http_client()
-    pageserver_lsn = Lsn(ps_http.timeline_detail(tenant_id, timeline_id)["last_record_lsn"])
+    ps_cli = env.pageserver.http_client()
+    pageserver_lsn = Lsn(ps_cli.timeline_detail(tenant_id, timeline_id)["last_record_lsn"])
     lag = last_lsn - pageserver_lsn
     log.info(
         f"Pageserver last_record_lsn={pageserver_lsn}; flush_lsn={last_lsn}; lag before replay is {lag / 1024}kb"
     )
 
     endpoint.stop_and_destroy()
-    timeline_delete_wait_completed(ps_http, tenant_id, timeline_id)
+    ps_cli.timeline_delete(tenant_id, timeline_id)
 
     # Also delete and manually create timeline on safekeepers -- this tests
     # scenario of manual recovery on different set of safekeepers.
@@ -575,21 +571,11 @@ def test_s3_wal_replay(neon_env_builder: NeonEnvBuilder, remote_storage_kind: Re
 
     pg_version = sk.http_client().timeline_status(tenant_id, timeline_id).pg_version
 
-    # Terminate first all safekeepers to prevent communication unexpectantly
-    # advancing peer_horizon_lsn.
     for sk in env.safekeepers:
         cli = sk.http_client()
         cli.timeline_delete_force(tenant_id, timeline_id)
         # restart safekeeper to clear its in-memory state
-        sk.stop()
-    # wait all potenital in flight pushes to broker arrive before starting
-    # safekeepers (even without sleep, it is very unlikely they are not
-    # delivered yet).
-    time.sleep(1)
-
-    for sk in env.safekeepers:
-        sk.start()
-        cli = sk.http_client()
+        sk.stop().start()
         cli.timeline_create(tenant_id, timeline_id, pg_version, last_lsn)
         f_partial_path = (
             Path(sk.data_dir()) / str(tenant_id) / str(timeline_id) / f_partial_saved.name
@@ -597,7 +583,7 @@ def test_s3_wal_replay(neon_env_builder: NeonEnvBuilder, remote_storage_kind: Re
         shutil.copy(f_partial_saved, f_partial_path)
 
     # recreate timeline on pageserver from scratch
-    ps_http.timeline_create(
+    ps_cli.timeline_create(
         pg_version=PgVersion(pg_version),
         tenant_id=tenant_id,
         new_timeline_id=timeline_id,
@@ -612,7 +598,7 @@ def test_s3_wal_replay(neon_env_builder: NeonEnvBuilder, remote_storage_kind: Re
         if elapsed > wait_lsn_timeout:
             raise RuntimeError("Timed out waiting for WAL redo")
 
-        tenant_status = ps_http.tenant_status(tenant_id)
+        tenant_status = ps_cli.tenant_status(tenant_id)
         if tenant_status["state"]["slug"] == "Loading":
             log.debug(f"Tenant {tenant_id} is still loading, retrying")
         else:

test_runner/regress/test_wal_receiver.py

@@ -1,5 +1,3 @@
-import time
-
 from fixtures.log_helper import log
 from fixtures.neon_fixtures import NeonEnv, NeonEnvBuilder
 from fixtures.types import Lsn, TenantId
@@ -42,10 +40,7 @@ def test_pageserver_lsn_wait_error_start(neon_env_builder: NeonEnvBuilder):
 # Kills one of the safekeepers and ensures that only the active ones are printed in the state.
 def test_pageserver_lsn_wait_error_safekeeper_stop(neon_env_builder: NeonEnvBuilder):
     # Trigger WAL wait timeout faster
-    neon_env_builder.pageserver_config_override = """
-        wait_lsn_timeout = "1s"
-        tenant_config={walreceiver_connect_timeout = "2s", lagging_wal_timeout = "2s"}
-    """
+    neon_env_builder.pageserver_config_override = "wait_lsn_timeout = '1s'"
     # Have notable SK ids to ensure we check logs for their presence, not some other random numbers
     neon_env_builder.safekeepers_id_start = 12345
     neon_env_builder.num_safekeepers = 3
@@ -75,8 +70,6 @@ def test_pageserver_lsn_wait_error_safekeeper_stop(neon_env_builder: NeonEnvBuil
     stopped_safekeeper_id = stopped_safekeeper.id
     log.info(f"Stopping safekeeper {stopped_safekeeper.id}")
     stopped_safekeeper.stop()
-    # sleep until stopped safekeeper is removed from candidates
-    time.sleep(2)
 
     # Spend some more time inserting, to ensure SKs report updated statuses and walreceiver in PS have time to update its connection stats.
     insert_test_elements(env, tenant_id, start=elements_to_insert + 1, count=elements_to_insert)