don't require shared_preload_libraries in spec in local tests -

we don't pass settings this way in local setup

.github/actionlint.yml

@@ -33,9 +33,14 @@ config-variables:
   - REMOTE_STORAGE_AZURE_CONTAINER
   - REMOTE_STORAGE_AZURE_REGION
   - SLACK_CICD_CHANNEL_ID
+  - SLACK_COMPUTE_CHANNEL_ID
   - SLACK_ON_CALL_DEVPROD_STREAM
   - SLACK_ON_CALL_QA_STAGING_STREAM
   - SLACK_ON_CALL_STORAGE_STAGING_STREAM
+  - SLACK_ONCALL_COMPUTE_GROUP
+  - SLACK_ONCALL_PROXY_GROUP
+  - SLACK_ONCALL_STORAGE_GROUP
+  - SLACK_PROXY_CHANNEL_ID
   - SLACK_RUST_CHANNEL_ID
   - SLACK_STORAGE_CHANNEL_ID
   - SLACK_UPCOMING_RELEASE_CHANNEL_ID

.github/workflows/build_and_test.yml

@@ -824,7 +824,7 @@ jobs:
           - pg: v17
             debian: bookworm
     env:
-      VM_BUILDER_VERSION: v0.42.2
+      VM_BUILDER_VERSION: v0.46.0
 
     steps:
       - name: Harden the runner (Audit all outbound calls)
@@ -1434,10 +1434,10 @@ jobs:
             ;;
           esac
 
-  notify-storage-release-deploy-failure:
-    needs: [ deploy ]
+  notify-release-deploy-failure:
+    needs: [ meta, deploy ]
     # We want this to run even if (transitive) dependencies are skipped, because deploy should really be successful on release branch workflow runs.
-    if: github.ref_name == 'release' && needs.deploy.result != 'success' && always()
+    if: contains(fromJSON('["storage-release", "compute-release", "proxy-release"]'), needs.meta.outputs.run-kind) && needs.deploy.result != 'success' && always()
     runs-on: ubuntu-22.04
     steps:
       - name: Harden the runner (Audit all outbound calls)
@@ -1445,15 +1445,40 @@ jobs:
         with:
           egress-policy: audit
 
-      - name: Post release-deploy failure to team-storage slack channel
+      - name: Post release-deploy failure to team slack channel
         uses: slackapi/slack-github-action@485a9d42d3a73031f12ec201c457e2162c45d02d # v2.0.0
+        env:
+          TEAM_ONCALL: >-
+            ${{
+              fromJSON(format('{
+                "storage-release": "<!subteam^{0}|@oncall-storage>",
+                "compute-release": "<!subteam^{1}|@oncall-compute>",
+                "proxy-release":   "<!subteam^{2}|@oncall-proxy>"
+              }',
+                vars.SLACK_ONCALL_STORAGE_GROUP,
+                vars.SLACK_ONCALL_COMPUTE_GROUP,
+                vars.SLACK_ONCALL_PROXY_GROUP
+              ))[needs.meta.outputs.run-kind]
+            }}
+          CHANNEL: >-
+            ${{
+              fromJSON(format('{
+                "storage-release": "{0}",
+                "compute-release": "{1}",
+                "proxy-release":   "{2}"
+              }',
+                vars.SLACK_STORAGE_CHANNEL_ID,
+                vars.SLACK_COMPUTE_CHANNEL_ID,
+                vars.SLACK_PROXY_CHANNEL_ID
+              ))[needs.meta.outputs.run-kind]
+            }}
         with:
           method: chat.postMessage
           token: ${{ secrets.SLACK_BOT_TOKEN }}
           payload: |
-            channel: ${{ vars.SLACK_STORAGE_CHANNEL_ID }}
+            channel: ${{ env.CHANNEL }}
             text: |
-              🔴 <!subteam^S06CJ87UMNY|@oncall-storage>: deploy job on release branch had unexpected status "${{ needs.deploy.result }}" <${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}|GitHub Run>.
+              🔴 ${{ env.TEAM_ONCALL }}: deploy job on release branch had unexpected status "${{ needs.deploy.result }}" <${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}|GitHub Run>.
 
   # The job runs on `release` branch and copies compatibility data and Neon artifact from the last *release PR* to the latest directory
   promote-compatibility-data:

compute/compute-node.Dockerfile

@@ -1083,6 +1083,34 @@ ARG PG_VERSION
 RUN cargo install --locked --version 0.12.9 cargo-pgrx && \
     /bin/bash -c 'cargo pgrx init --pg${PG_VERSION:1}=/usr/local/pgsql/bin/pg_config'
 
+USER root
+#########################################################################################
+#
+# Layer "rust extensions pgrx14"
+#
+#########################################################################################
+FROM pg-build-nonroot-with-cargo AS rust-extensions-build-pgrx14
+ARG PG_VERSION
+
+RUN cargo install --locked --version 0.14.1 cargo-pgrx && \
+    /bin/bash -c 'cargo pgrx init --pg${PG_VERSION:1}=/usr/local/pgsql/bin/pg_config'
+
+USER root
+#########################################################################################
+#
+# Layer "rust extensions pgrx14"
+#
+# Version 14 is now required by a few 
+# This layer should be used as a base for new pgrx extensions,
+# and eventually get merged with `rust-extensions-build`
+#
+#########################################################################################
+FROM pg-build-nonroot-with-cargo AS rust-extensions-build-pgrx14
+ARG PG_VERSION
+
+RUN cargo install --locked --version 0.14.1 cargo-pgrx && \
+    /bin/bash -c 'cargo pgrx init --pg${PG_VERSION:1}=/usr/local/pgsql/bin/pg_config'
+
 USER root
 
 #########################################################################################
@@ -1100,11 +1128,11 @@ RUN wget https://github.com/microsoft/onnxruntime/archive/refs/tags/v1.18.1.tar.
     mkdir onnxruntime-src && cd onnxruntime-src && tar xzf ../onnxruntime.tar.gz --strip-components=1 -C . && \
     echo "#nothing to test here" > neon-test.sh
 
-RUN wget https://github.com/neondatabase-labs/pgrag/archive/refs/tags/v0.0.0.tar.gz -O pgrag.tar.gz &&  \
-    echo "2cbe394c1e74fc8bcad9b52d5fbbfb783aef834ca3ce44626cfd770573700bb4 pgrag.tar.gz" | sha256sum --check && \
+RUN wget https://github.com/neondatabase-labs/pgrag/archive/refs/tags/v0.1.1.tar.gz -O pgrag.tar.gz &&  \
+    echo "087b2ecd11ba307dc968042ef2e9e43dc04d9ba60e8306e882c407bbe1350a50 pgrag.tar.gz" | sha256sum --check && \
     mkdir pgrag-src && cd pgrag-src && tar xzf ../pgrag.tar.gz --strip-components=1 -C .
 
-FROM rust-extensions-build-pgrx12 AS pgrag-build
+FROM rust-extensions-build-pgrx14 AS pgrag-build
 COPY --from=pgrag-src /ext-src/ /ext-src/
 
 # Install build-time dependencies
@@ -1124,19 +1152,19 @@ RUN . venv/bin/activate && \
 
 WORKDIR /ext-src/pgrag-src
 RUN cd exts/rag && \
-    sed -i 's/pgrx = "0.12.6"/pgrx = { version = "0.12.9", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    sed -i 's/pgrx = "0.14.1"/pgrx = { version = "0.14.1", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     cargo pgrx install --release && \
     echo "trusted = true" >> /usr/local/pgsql/share/extension/rag.control
 
 RUN cd exts/rag_bge_small_en_v15 && \
-    sed -i 's/pgrx = "0.12.6"/pgrx = { version = "0.12.9", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    sed -i 's/pgrx = "0.14.1"/pgrx = { version = "0.14.1", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     ORT_LIB_LOCATION=/ext-src/onnxruntime-src/build/Linux \
         REMOTE_ONNX_URL=http://pg-ext-s3-gateway/pgrag-data/bge_small_en_v15.onnx \
         cargo pgrx install --release --features remote_onnx && \
     echo "trusted = true" >> /usr/local/pgsql/share/extension/rag_bge_small_en_v15.control
 
 RUN cd exts/rag_jina_reranker_v1_tiny_en && \
-    sed -i 's/pgrx = "0.12.6"/pgrx = { version = "0.12.9", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    sed -i 's/pgrx = "0.14.1"/pgrx = { version = "0.14.1", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     ORT_LIB_LOCATION=/ext-src/onnxruntime-src/build/Linux \
         REMOTE_ONNX_URL=http://pg-ext-s3-gateway/pgrag-data/jina_reranker_v1_tiny_en.onnx \
         cargo pgrx install --release --features remote_onnx && \
@@ -1305,8 +1333,8 @@ ARG PG_VERSION
 # Do not update without approve from proxy team
 # Make sure the version is reflected in proxy/src/serverless/local_conn_pool.rs
 WORKDIR /ext-src
-RUN wget https://github.com/neondatabase/pg_session_jwt/archive/refs/heads/alexk/tmp-disable-audit.tar.gz -O pg_session_jwt.tar.gz && \
-    echo "b868227950973df5186af54dfa03617536d21e96dddcb7b25e7ce199b4956bdb pg_session_jwt.tar.gz" | sha256sum --check && \
+RUN wget https://github.com/neondatabase/pg_session_jwt/archive/refs/tags/v0.3.0.tar.gz -O pg_session_jwt.tar.gz && \
+    echo "19be2dc0b3834d643706ed430af998bb4c2cdf24b3c45e7b102bb3a550e8660c pg_session_jwt.tar.gz" | sha256sum --check && \
     mkdir pg_session_jwt-src && cd pg_session_jwt-src && tar xzf ../pg_session_jwt.tar.gz --strip-components=1 -C . && \
     sed -i 's/pgrx = "0.12.6"/pgrx = { version = "0.12.9", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     sed -i 's/version = "0.12.6"/version = "0.12.9"/g' pgrx-tests/Cargo.toml && \
@@ -1319,6 +1347,39 @@ COPY --from=pg_session_jwt-src /ext-src/ /ext-src/
 WORKDIR /ext-src/pg_session_jwt-src
 RUN cargo pgrx install --release
 
+#########################################################################################
+#
+# Layer "pg-anon-pg-build"
+# compile anon extension
+#
+#########################################################################################
+FROM pg-build AS pg_anon-src
+ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+WORKDIR /ext-src
+COPY compute/patches/anon_v2.patch .
+
+# This is an experimental extension, never got to real production.
+# !Do not remove! It can be present in shared_preload_libraries and compute will fail to start if library is not found.
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
+RUN wget https://gitlab.com/dalibo/postgresql_anonymizer/-/archive/latest/postgresql_anonymizer-latest.tar.gz -O pg_anon.tar.gz && \
+    mkdir pg_anon-src && cd pg_anon-src && tar xzf ../pg_anon.tar.gz --strip-components=1 -C . && \
+    find /usr/local/pgsql -type f | sed 's|^/usr/local/pgsql/||' > /before.txt && \
+    sed -i 's/pgrx = "0.14.1"/pgrx = { version = "=0.14.1", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    patch -p1 < /ext-src/anon_v2.patch
+
+FROM rust-extensions-build-pgrx14 AS pg-anon-pg-build
+ARG PG_VERSION
+COPY --from=pg_anon-src /ext-src/ /ext-src/
+WORKDIR /ext-src
+RUN cd pg_anon-src && \
+    make -j $(getconf _NPROCESSORS_ONLN) extension PG_CONFIG=/usr/local/pgsql/bin/pg_config PGVER=pg$(echo "$PG_VERSION" | sed 's/^v//') && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config PGVER=pg$(echo "$PG_VERSION" | sed 's/^v//') && \
+    chmod -R a+r ../pg_anon-src && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/anon.control;
+
+########################################################################################
+
 #########################################################################################
 #
 # Layer "wal2json-build"
@@ -1615,6 +1676,7 @@ COPY --from=pg_uuidv7-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg_roaringbitmap-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg_semver-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=wal2json-build /usr/local/pgsql /usr/local/pgsql
+COPY --from=pg-anon-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg_ivm-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg_partman-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg_mooncake-build /usr/local/pgsql/ /usr/local/pgsql/

compute/etc/neon_collector.jsonnet

@@ -23,6 +23,8 @@
     import 'sql_exporter/getpage_prefetch_requests_total.libsonnet',
     import 'sql_exporter/getpage_prefetches_buffered.libsonnet',
     import 'sql_exporter/getpage_sync_requests_total.libsonnet',
+    import 'sql_exporter/compute_getpage_stuck_requests_total.libsonnet',
+    import 'sql_exporter/compute_getpage_max_inflight_stuck_time_ms.libsonnet',
     import 'sql_exporter/getpage_wait_seconds_bucket.libsonnet',
     import 'sql_exporter/getpage_wait_seconds_count.libsonnet',
     import 'sql_exporter/getpage_wait_seconds_sum.libsonnet',

compute/etc/sql_exporter/compute_getpage_max_inflight_stuck_time_ms.libsonnet

@@ -0,0 +1,9 @@
+{
+  metric_name: 'compute_getpage_max_inflight_stuck_time_ms',
+  type: 'gauge',
+  help: 'Max wait time for stuck requests among all backends. Includes only active stuck requests, terminated or disconnected ones are not accounted for',
+  values: [
+    'compute_getpage_max_inflight_stuck_time_ms',
+  ],
+  query_ref: 'neon_perf_counters',
+}

compute/etc/sql_exporter/compute_getpage_stuck_requests_total.libsonnet

@@ -0,0 +1,9 @@
+{
+  metric_name: 'compute_getpage_stuck_requests_total',
+  type: 'counter',
+  help: 'Total number of Getpage requests left without an answer for more than pageserver_response_log_timeout but less than pageserver_response_disconnect_timeout',
+  values: [
+    'compute_getpage_stuck_requests_total',
+  ],
+  query_ref: 'neon_perf_counters',
+}

compute/etc/sql_exporter/neon_perf_counters.sql

@@ -9,6 +9,8 @@ SELECT d.* FROM pg_catalog.jsonb_to_record((SELECT jb FROM c)) AS d(
   getpage_wait_seconds_sum numeric,
   getpage_prefetch_requests_total numeric,
   getpage_sync_requests_total numeric,
+  compute_getpage_stuck_requests_total numeric,
+  compute_getpage_max_inflight_stuck_time_ms numeric,
   getpage_prefetch_misses_total numeric,
   getpage_prefetch_discards_total numeric,
   getpage_prefetches_buffered numeric,

compute/patches/anon_v2.patch

@@ -0,0 +1,129 @@
+diff --git a/sql/anon.sql b/sql/anon.sql
+index 0cdc769..f6cc950 100644
+--- a/sql/anon.sql
++++ b/sql/anon.sql
+@@ -1141,3 +1141,8 @@ $$
+ -- TODO : https://en.wikipedia.org/wiki/L-diversity
+ 
+ -- TODO : https://en.wikipedia.org/wiki/T-closeness
++
++-- NEON Patches
++
++GRANT ALL ON SCHEMA anon to neon_superuser;
++GRANT ALL ON ALL TABLES IN SCHEMA anon TO neon_superuser;
+diff --git a/sql/init.sql b/sql/init.sql
+index 7da6553..9b6164b 100644
+--- a/sql/init.sql
++++ b/sql/init.sql
+@@ -74,50 +74,49 @@ $$
+ 
+ SECURITY LABEL FOR anon ON FUNCTION anon.load_csv IS 'UNTRUSTED';
+ 
+--- load fake data from a given path
+-CREATE OR REPLACE FUNCTION anon.init(
+-  datapath TEXT
+-)
++CREATE OR REPLACE FUNCTION anon.load_fake_data()
+ RETURNS BOOLEAN
+ AS $$
+ DECLARE
+-  datapath_check TEXT;
+   success BOOLEAN;
++  sharedir TEXT;
++  datapath TEXT;
+ BEGIN
+ 
+-  IF anon.is_initialized() THEN
+-    RAISE NOTICE 'The anon extension is already initialized.';
+-    RETURN TRUE;
+-  END IF;
++  datapath := '/extension/anon/';
++  -- find the local extension directory
++  SELECT setting INTO sharedir
++  FROM pg_catalog.pg_config
++  WHERE name = 'SHAREDIR';
+ 
+   SELECT bool_or(results) INTO success
+   FROM unnest(array[
+-    anon.load_csv('anon.identifiers_category',datapath||'/identifiers_category.csv'),
+-    anon.load_csv('anon.identifier',datapath ||'/identifier.csv'),
+-    anon.load_csv('anon.address',datapath ||'/address.csv'),
+-    anon.load_csv('anon.city',datapath ||'/city.csv'),
+-    anon.load_csv('anon.company',datapath ||'/company.csv'),
+-    anon.load_csv('anon.country',datapath ||'/country.csv'),
+-    anon.load_csv('anon.email', datapath ||'/email.csv'),
+-    anon.load_csv('anon.first_name',datapath ||'/first_name.csv'),
+-    anon.load_csv('anon.iban',datapath ||'/iban.csv'),
+-    anon.load_csv('anon.last_name',datapath ||'/last_name.csv'),
+-    anon.load_csv('anon.postcode',datapath ||'/postcode.csv'),
+-    anon.load_csv('anon.siret',datapath ||'/siret.csv'),
+-    anon.load_csv('anon.lorem_ipsum',datapath ||'/lorem_ipsum.csv')
++    anon.load_csv('anon.identifiers_category',sharedir || datapath || '/identifiers_category.csv'),
++    anon.load_csv('anon.identifier',sharedir || datapath || '/identifier.csv'),
++    anon.load_csv('anon.address',sharedir || datapath || '/address.csv'),
++    anon.load_csv('anon.city',sharedir || datapath || '/city.csv'),
++    anon.load_csv('anon.company',sharedir || datapath || '/company.csv'),
++    anon.load_csv('anon.country',sharedir || datapath || '/country.csv'),
++    anon.load_csv('anon.email', sharedir || datapath || '/email.csv'),
++    anon.load_csv('anon.first_name',sharedir || datapath || '/first_name.csv'),
++    anon.load_csv('anon.iban',sharedir || datapath || '/iban.csv'),
++    anon.load_csv('anon.last_name',sharedir || datapath || '/last_name.csv'),
++    anon.load_csv('anon.postcode',sharedir || datapath || '/postcode.csv'),
++    anon.load_csv('anon.siret',sharedir || datapath || '/siret.csv'),
++    anon.load_csv('anon.lorem_ipsum',sharedir || datapath || '/lorem_ipsum.csv')
+   ]) results;
+   RETURN success;
+-
+ END;
+ $$
+-  LANGUAGE PLPGSQL
++  LANGUAGE plpgsql
+   VOLATILE
+   RETURNS NULL ON NULL INPUT
+-  PARALLEL UNSAFE -- because load_csv is unsafe
+-  SECURITY INVOKER
++  PARALLEL UNSAFE -- because of the EXCEPTION
++  SECURITY DEFINER
+   SET search_path=''
+ ;
+-SECURITY LABEL FOR anon ON FUNCTION anon.init(TEXT) IS 'UNTRUSTED';
++
++SECURITY LABEL FOR anon ON FUNCTION anon.load_fake_data IS 'UNTRUSTED';
+ 
+ -- People tend to forget the anon.init() step
+ -- This is a friendly notice for them
+@@ -144,7 +143,7 @@ SECURITY LABEL FOR anon ON FUNCTION anon.notice_if_not_init IS 'UNTRUSTED';
+ CREATE OR REPLACE FUNCTION anon.load(TEXT)
+ RETURNS BOOLEAN AS
+ $$
+-  SELECT anon.init($1);
++  SELECT anon.init();
+ $$
+   LANGUAGE SQL
+   VOLATILE
+@@ -159,16 +158,16 @@ SECURITY LABEL FOR anon ON FUNCTION anon.load(TEXT) IS 'UNTRUSTED';
+ CREATE OR REPLACE FUNCTION anon.init()
+ RETURNS BOOLEAN
+ AS $$
+-  WITH conf AS (
+-        -- find the local extension directory
+-        SELECT setting AS sharedir
+-        FROM pg_catalog.pg_config
+-        WHERE name = 'SHAREDIR'
+-    )
+-  SELECT anon.init(conf.sharedir || '/extension/anon/')
+-  FROM conf;
++BEGIN
++  IF anon.is_initialized() THEN
++    RAISE NOTICE 'The anon extension is already initialized.';
++    RETURN TRUE;
++  END IF;
++
++  RETURN anon.load_fake_data();
++END;
+ $$
+-  LANGUAGE SQL
++  LANGUAGE plpgsql
+   VOLATILE
+   PARALLEL UNSAFE -- because init is unsafe
+   SECURITY INVOKER

compute_tools/src/config.rs

@@ -168,6 +168,35 @@ pub fn write_postgres_conf(
         writeln!(file, "# Managed by compute_ctl: end")?;
     }
 
+    // Always add pgaudit to shared_preload_libraries.
+    //
+    // This is needed to handle the downgrade scenario.
+    // pgaudit extension creates event triggers that require library to be loaded.
+    // so, once extension was installed it must always be present in shared_preload_libraries.
+    let mut extra_shared_preload_libraries = String::new();
+
+    let libs = {
+        // We don't distribute pgaudit in the testing image,
+        // and don't pass shared_preload_libraries via spec,
+        // so disable this logic there.
+        #[cfg(feature = "testing")]
+        {
+            String::new()
+        }
+        #[cfg(not(feature = "testing"))]
+        {
+            spec.cluster
+                .settings
+                .find("shared_preload_libraries")
+                .expect("shared_preload_libraries setting is missing in the spec")
+        }
+    };
+
+    #[cfg(not(feature = "testing"))]
+    if !libs.contains("pgaudit") {
+        extra_shared_preload_libraries.push_str(",pgaudit");
+    };
+
     // If base audit logging is enabled, configure it.
     // In this setup, the audit log will be written to the standard postgresql log.
     //
@@ -177,29 +206,22 @@ pub fn write_postgres_conf(
     // This way we always override the settings from the spec
     // and don't allow the user or the control plane admin to change them.
     match spec.audit_log_level {
-        ComputeAudit::Disabled => {}
+        ComputeAudit::Disabled => {
+            // this is the default, but let's be explicit
+            writeln!(file, "pgaudit.log='none'")?;
+        }
         ComputeAudit::Log | ComputeAudit::Base => {
             writeln!(file, "# Managed by compute_ctl base audit settings: start")?;
             writeln!(file, "pgaudit.log='ddl,role'")?;
             // Disable logging of catalog queries to reduce the noise
             writeln!(file, "pgaudit.log_catalog=off")?;
 
-            if let Some(libs) = spec.cluster.settings.find("shared_preload_libraries") {
-                let mut extra_shared_preload_libraries = String::new();
-                if !libs.contains("pgaudit") {
-                    extra_shared_preload_libraries.push_str(",pgaudit");
-                }
-                writeln!(
-                    file,
-                    "shared_preload_libraries='{}{}'",
-                    libs, extra_shared_preload_libraries
-                )?;
-            } else {
-                // Typically, this should be unreacheable,
-                // because we always set at least some shared_preload_libraries in the spec
-                // but let's handle it explicitly anyway.
-                writeln!(file, "shared_preload_libraries='neon,pgaudit'")?;
-            }
+            writeln!(
+                file,
+                "shared_preload_libraries='{}{}'",
+                libs, extra_shared_preload_libraries
+            )?;
+
             writeln!(file, "# Managed by compute_ctl base audit settings: end")?;
         }
         ComputeAudit::Hipaa | ComputeAudit::Extended | ComputeAudit::Full => {
@@ -228,28 +250,15 @@ pub fn write_postgres_conf(
             // The caller who sets the flag is responsible for ensuring that the necessary
             // shared_preload_libraries are present in the compute image,
             // otherwise the compute start will fail.
-            if let Some(libs) = spec.cluster.settings.find("shared_preload_libraries") {
-                let mut extra_shared_preload_libraries = String::new();
-                if !libs.contains("pgaudit") {
-                    extra_shared_preload_libraries.push_str(",pgaudit");
-                }
-                if !libs.contains("pgauditlogtofile") {
-                    extra_shared_preload_libraries.push_str(",pgauditlogtofile");
-                }
-                writeln!(
-                    file,
-                    "shared_preload_libraries='{}{}'",
-                    libs, extra_shared_preload_libraries
-                )?;
-            } else {
-                // Typically, this should be unreacheable,
-                // because we always set at least some shared_preload_libraries in the spec
-                // but let's handle it explicitly anyway.
-                writeln!(
-                    file,
-                    "shared_preload_libraries='neon,pgaudit,pgauditlogtofile'"
-                )?;
+            if !libs.contains("pgauditlogtofile") {
+                extra_shared_preload_libraries.push_str(",pgauditlogtofile");
             }
+            writeln!(
+                file,
+                "shared_preload_libraries='{}{}'",
+                libs, extra_shared_preload_libraries
+            )?;
+
             writeln!(
                 file,
                 "# Managed by compute_ctl compliance audit settings: end"

compute_tools/src/monitor.rs

@@ -424,10 +424,10 @@ pub fn launch_monitor(compute: &Arc<ComputeNode>) -> thread::JoinHandle<()> {
         experimental,
     };
 
-    let span = span!(Level::INFO, "compute_monitor");
     thread::Builder::new()
         .name("compute-monitor".into())
         .spawn(move || {
+            let span = span!(Level::INFO, "compute_monitor");
             let _enter = span.enter();
             monitor.run();
         })

docs/consumption_metrics.md

@@ -38,11 +38,6 @@ Currently, the following metrics are collected:
 Amount of WAL produced , by a timeline, i.e. last_record_lsn
 This is an absolute, per-timeline metric.
 
-- `resident_size`
-
-Size of all the layer files in the tenant's directory on disk on the pageserver.
-This is an absolute, per-tenant metric.
-
 - `remote_storage_size`
 
 Size of the remote storage (S3) directory.

libs/postgres_backend/src/lib.rs

@@ -841,6 +841,10 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> PostgresBackend<IO> {
 
         let expected_end = match &end {
             ServerInitiated(_) | CopyDone | CopyFail | Terminate | EOF | Cancelled => true,
+            // The timeline doesn't exist and we have been requested to not auto-create it.
+            // Compute requests for timelines that haven't been created yet
+            // might reach us before the storcon request to create those timelines.
+            TimelineNoCreate => true,
             CopyStreamHandlerEnd::Disconnected(ConnectionError::Io(io_error))
                 if is_expected_io_error(io_error) =>
             {
@@ -1059,6 +1063,8 @@ pub enum CopyStreamHandlerEnd {
     Terminate,
     #[error("EOF on COPY stream")]
     EOF,
+    #[error("timeline not found, and allow_timeline_creation is false")]
+    TimelineNoCreate,
     /// The connection was lost
     #[error("connection error: {0}")]
     Disconnected(#[from] ConnectionError),

libs/safekeeper_api/src/models.rs

@@ -303,7 +303,8 @@ pub struct PullTimelineRequest {
 
 #[derive(Debug, Serialize, Deserialize)]
 pub struct PullTimelineResponse {
-    // Donor safekeeper host
-    pub safekeeper_host: String,
+    /// Donor safekeeper host.
+    /// None if no pull happened because the timeline already exists.
+    pub safekeeper_host: Option<String>,
     // TODO: add more fields?
 }

pageserver/ctl/src/layers.rs

@@ -10,6 +10,7 @@ use pageserver::tenant::storage_layer::{DeltaLayer, ImageLayer, delta_layer, ima
 use pageserver::tenant::{TENANTS_SEGMENT_NAME, TIMELINES_SEGMENT_NAME};
 use pageserver::virtual_file::api::IoMode;
 use pageserver::{page_cache, virtual_file};
+use pageserver_api::key::Key;
 use utils::id::{TenantId, TimelineId};
 
 use crate::layer_map_analyzer::parse_filename;
@@ -27,6 +28,7 @@ pub(crate) enum LayerCmd {
         path: PathBuf,
         tenant: String,
         timeline: String,
+        key: Option<Key>,
     },
     /// Dump all information of a layer file
     DumpLayer {
@@ -100,6 +102,7 @@ pub(crate) async fn main(cmd: &LayerCmd) -> Result<()> {
             path,
             tenant,
             timeline,
+            key,
         } => {
             let timeline_path = path
                 .join(TENANTS_SEGMENT_NAME)
@@ -107,21 +110,37 @@ pub(crate) async fn main(cmd: &LayerCmd) -> Result<()> {
                 .join(TIMELINES_SEGMENT_NAME)
                 .join(timeline);
             let mut idx = 0;
+            let mut to_print = Vec::default();
             for layer in fs::read_dir(timeline_path)? {
                 let layer = layer?;
                 if let Ok(layer_file) = parse_filename(&layer.file_name().into_string().unwrap()) {
-                    println!(
-                        "[{:3}]  key:{}-{}\n       lsn:{}-{}\n       delta:{}",
-                        idx,
-                        layer_file.key_range.start,
-                        layer_file.key_range.end,
-                        layer_file.lsn_range.start,
-                        layer_file.lsn_range.end,
-                        layer_file.is_delta,
-                    );
+                    if let Some(key) = key {
+                        if layer_file.key_range.start <= *key && *key < layer_file.key_range.end {
+                            to_print.push((idx, layer_file));
+                        }
+                    } else {
+                        to_print.push((idx, layer_file));
+                    }
                     idx += 1;
                 }
             }
+
+            if key.is_some() {
+                to_print
+                    .sort_by_key(|(_idx, layer_file)| std::cmp::Reverse(layer_file.lsn_range.end));
+            }
+
+            for (idx, layer_file) in to_print {
+                println!(
+                    "[{:3}]  key:{}-{}\n       lsn:{}-{}\n       delta:{}",
+                    idx,
+                    layer_file.key_range.start,
+                    layer_file.key_range.end,
+                    layer_file.lsn_range.start,
+                    layer_file.lsn_range.end,
+                    layer_file.is_delta,
+                );
+            }
             Ok(())
         }
         LayerCmd::DumpLayer {

pageserver/src/consumption_metrics/metrics.rs

@@ -30,9 +30,6 @@ pub(super) enum Name {
     /// Tenant remote size
     #[serde(rename = "remote_storage_size")]
     RemoteSize,
-    /// Tenant resident size
-    #[serde(rename = "resident_size")]
-    ResidentSize,
     /// Tenant synthetic size
     #[serde(rename = "synthetic_storage_size")]
     SyntheticSize,
@@ -187,18 +184,6 @@ impl MetricsKey {
         .absolute_values()
     }
 
-    /// Sum of [`Timeline::resident_physical_size`] for each `Tenant`.
-    ///
-    /// [`Timeline::resident_physical_size`]: crate::tenant::Timeline::resident_physical_size
-    const fn resident_size(tenant_id: TenantId) -> AbsoluteValueFactory {
-        MetricsKey {
-            tenant_id,
-            timeline_id: None,
-            metric: Name::ResidentSize,
-        }
-        .absolute_values()
-    }
-
     /// [`TenantShard::cached_synthetic_size`] as refreshed by [`calculate_synthetic_size_worker`].
     ///
     /// [`TenantShard::cached_synthetic_size`]: crate::tenant::TenantShard::cached_synthetic_size
@@ -261,10 +246,7 @@ where
     let mut tenants = std::pin::pin!(tenants);
 
     while let Some((tenant_id, tenant)) = tenants.next().await {
-        let mut tenant_resident_size = 0;
-
         let timelines = tenant.list_timelines();
-        let timelines_len = timelines.len();
         for timeline in timelines {
             let timeline_id = timeline.timeline_id;
 
@@ -287,16 +269,9 @@ where
                     continue;
                 }
             }
-
-            tenant_resident_size += timeline.resident_physical_size();
         }
 
-        if timelines_len == 0 {
-            // Force set it to 1 byte to avoid not being reported -- all timelines are offloaded.
-            tenant_resident_size = 1;
-        }
-
-        let snap = TenantSnapshot::collect(&tenant, tenant_resident_size);
+        let snap = TenantSnapshot::collect(&tenant);
         snap.to_metrics(tenant_id, Utc::now(), cache, &mut current_metrics);
     }
 
@@ -305,19 +280,14 @@ where
 
 /// In-between abstraction to allow testing metrics without actual Tenants.
 struct TenantSnapshot {
-    resident_size: u64,
     remote_size: u64,
     synthetic_size: u64,
 }
 
 impl TenantSnapshot {
     /// Collect tenant status to have metrics created out of it.
-    ///
-    /// `resident_size` is calculated of the timelines we had access to for other metrics, so we
-    /// cannot just list timelines here.
-    fn collect(t: &Arc<crate::tenant::TenantShard>, resident_size: u64) -> Self {
+    fn collect(t: &Arc<crate::tenant::TenantShard>) -> Self {
         TenantSnapshot {
-            resident_size,
             remote_size: t.remote_size(),
             // Note that this metric is calculated in a separate bgworker
             // Here we only use cached value, which may lag behind the real latest one
@@ -334,8 +304,6 @@ impl TenantSnapshot {
     ) {
         let remote_size = MetricsKey::remote_storage_size(tenant_id).at(now, self.remote_size);
 
-        let resident_size = MetricsKey::resident_size(tenant_id).at(now, self.resident_size);
-
         let synthetic_size = {
             let factory = MetricsKey::synthetic_size(tenant_id);
             let mut synthetic_size = self.synthetic_size;
@@ -355,11 +323,7 @@ impl TenantSnapshot {
             }
         };
 
-        metrics.extend(
-            [Some(remote_size), Some(resident_size), synthetic_size]
-                .into_iter()
-                .flatten(),
-        );
+        metrics.extend([Some(remote_size), synthetic_size].into_iter().flatten());
     }
 }
 

pageserver/src/consumption_metrics/metrics/tests.rs

@@ -224,7 +224,6 @@ fn post_restart_synthetic_size_uses_cached_if_available() {
     let tenant_id = TenantId::generate();
 
     let ts = TenantSnapshot {
-        resident_size: 1000,
         remote_size: 1000,
         // not yet calculated
         synthetic_size: 0,
@@ -245,7 +244,6 @@ fn post_restart_synthetic_size_uses_cached_if_available() {
         metrics,
         &[
             MetricsKey::remote_storage_size(tenant_id).at(now, 1000),
-            MetricsKey::resident_size(tenant_id).at(now, 1000),
             MetricsKey::synthetic_size(tenant_id).at(now, 1000),
         ]
     );
@@ -256,7 +254,6 @@ fn post_restart_synthetic_size_is_not_sent_when_not_cached() {
     let tenant_id = TenantId::generate();
 
     let ts = TenantSnapshot {
-        resident_size: 1000,
         remote_size: 1000,
         // not yet calculated
         synthetic_size: 0,
@@ -274,7 +271,6 @@ fn post_restart_synthetic_size_is_not_sent_when_not_cached() {
         metrics,
         &[
             MetricsKey::remote_storage_size(tenant_id).at(now, 1000),
-            MetricsKey::resident_size(tenant_id).at(now, 1000),
             // no synthetic size here
         ]
     );
@@ -295,14 +291,13 @@ pub(crate) const fn metric_examples_old(
     timeline_id: TimelineId,
     now: DateTime<Utc>,
     before: DateTime<Utc>,
-) -> [RawMetric; 6] {
+) -> [RawMetric; 5] {
     [
         MetricsKey::written_size(tenant_id, timeline_id).at_old_format(now, 0),
         MetricsKey::written_size_delta(tenant_id, timeline_id)
             .from_until_old_format(before, now, 0),
         MetricsKey::timeline_logical_size(tenant_id, timeline_id).at_old_format(now, 0),
         MetricsKey::remote_storage_size(tenant_id).at_old_format(now, 0),
-        MetricsKey::resident_size(tenant_id).at_old_format(now, 0),
         MetricsKey::synthetic_size(tenant_id).at_old_format(now, 1),
     ]
 }
@@ -312,13 +307,12 @@ pub(crate) const fn metric_examples(
     timeline_id: TimelineId,
     now: DateTime<Utc>,
     before: DateTime<Utc>,
-) -> [NewRawMetric; 6] {
+) -> [NewRawMetric; 5] {
     [
         MetricsKey::written_size(tenant_id, timeline_id).at(now, 0),
         MetricsKey::written_size_delta(tenant_id, timeline_id).from_until(before, now, 0),
         MetricsKey::timeline_logical_size(tenant_id, timeline_id).at(now, 0),
         MetricsKey::remote_storage_size(tenant_id).at(now, 0),
-        MetricsKey::resident_size(tenant_id).at(now, 0),
         MetricsKey::synthetic_size(tenant_id).at(now, 1),
     ]
 }

pageserver/src/consumption_metrics/upload.rs

@@ -521,10 +521,6 @@ mod tests {
                 line!(),
                 r#"{"type":"absolute","time":"2023-09-15T00:00:00.123456789Z","metric":"remote_storage_size","idempotency_key":"2023-09-15 00:00:00.123456789 UTC-1-0000","value":0,"tenant_id":"00000000000000000000000000000000"}"#,
             ),
-            (
-                line!(),
-                r#"{"type":"absolute","time":"2023-09-15T00:00:00.123456789Z","metric":"resident_size","idempotency_key":"2023-09-15 00:00:00.123456789 UTC-1-0000","value":0,"tenant_id":"00000000000000000000000000000000"}"#,
-            ),
             (
                 line!(),
                 r#"{"type":"absolute","time":"2023-09-15T00:00:00.123456789Z","metric":"synthetic_storage_size","idempotency_key":"2023-09-15 00:00:00.123456789 UTC-1-0000","value":1,"tenant_id":"00000000000000000000000000000000"}"#,
@@ -564,7 +560,7 @@ mod tests {
         assert_eq!(upgraded_samples, new_samples);
     }
 
-    fn metric_samples_old() -> [RawMetric; 6] {
+    fn metric_samples_old() -> [RawMetric; 5] {
         let tenant_id = TenantId::from_array([0; 16]);
         let timeline_id = TimelineId::from_array([0xff; 16]);
 
@@ -576,7 +572,7 @@ mod tests {
         super::super::metrics::metric_examples_old(tenant_id, timeline_id, now, before)
     }
 
-    fn metric_samples() -> [NewRawMetric; 6] {
+    fn metric_samples() -> [NewRawMetric; 5] {
         let tenant_id = TenantId::from_array([0; 16]);
         let timeline_id = TimelineId::from_array([0xff; 16]);
 

pageserver/src/metrics.rs

@@ -497,6 +497,24 @@ pub(crate) static WAIT_LSN_IN_PROGRESS_GLOBAL_MICROS: Lazy<IntCounter> = Lazy::n
     .expect("failed to define a metric")
 });
 
+pub(crate) static ONDEMAND_DOWNLOAD_BYTES: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "pageserver_ondemand_download_bytes_total",
+        "Total bytes of layers on-demand downloaded",
+        &["task_kind"]
+    )
+    .expect("failed to define a metric")
+});
+
+pub(crate) static ONDEMAND_DOWNLOAD_COUNT: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "pageserver_ondemand_download_count",
+        "Total count of layers on-demand downloaded",
+        &["task_kind"]
+    )
+    .expect("failed to define a metric")
+});
+
 pub(crate) mod wait_ondemand_download_time {
     use super::*;
     const WAIT_ONDEMAND_DOWNLOAD_TIME_BUCKETS: &[f64] = &[
@@ -2180,6 +2198,10 @@ impl BasebackupQueryTimeOngoingRecording<'_> {
         // If you want to change categorize of a specific error, also change it in `log_query_error`.
         let metric = match res {
             Ok(_) => &self.parent.ok,
+            Err(QueryError::Shutdown) => {
+                // Do not observe ok/err for shutdown
+                return;
+            }
             Err(QueryError::Disconnected(ConnectionError::Io(io_error)))
                 if is_expected_io_error(io_error) =>
             {

pageserver/src/page_service.rs

@@ -1035,10 +1035,25 @@ impl PageServerHandler {
                 // avoid a somewhat costly Span::record() by constructing the entire span in one go.
                 macro_rules! mkspan {
                     (before shard routing) => {{
-                        tracing::info_span!(parent: &parent_span, "handle_get_page_request", rel = %req.rel, blkno = %req.blkno, req_lsn = %req.hdr.request_lsn)
+                        tracing::info_span!(
+                            parent: &parent_span,
+                            "handle_get_page_request",
+                            rel = %req.rel,
+                            blkno = %req.blkno,
+                            req_lsn = %req.hdr.request_lsn,
+                            not_modified_since_lsn = %req.hdr.not_modified_since
+                        )
                     }};
                     ($shard_id:expr) => {{
-                        tracing::info_span!(parent: &parent_span, "handle_get_page_request", rel = %req.rel, blkno = %req.blkno, req_lsn = %req.hdr.request_lsn, shard_id = %$shard_id)
+                        tracing::info_span!(
+                            parent: &parent_span,
+                            "handle_get_page_request",
+                            rel = %req.rel,
+                            blkno = %req.blkno,
+                            req_lsn = %req.hdr.request_lsn,
+                            not_modified_since_lsn = %req.hdr.not_modified_since,
+                            shard_id = %$shard_id
+                        )
                     }};
                 }
 
@@ -1102,6 +1117,7 @@ impl PageServerHandler {
                             shard_id = %shard.get_shard_identity().shard_slug(),
                             timeline_id = %timeline_id,
                             lsn = %req.hdr.request_lsn,
+                            not_modified_since_lsn = %req.hdr.not_modified_since,
                             request_id = %req.hdr.reqid,
                             key = %key,
                             )

pageserver/src/tenant/storage_layer/layer.rs

@@ -4,6 +4,7 @@ use std::sync::{Arc, Weak};
 use std::time::{Duration, SystemTime};
 
 use crate::PERF_TRACE_TARGET;
+use crate::metrics::{ONDEMAND_DOWNLOAD_BYTES, ONDEMAND_DOWNLOAD_COUNT};
 use anyhow::Context;
 use camino::{Utf8Path, Utf8PathBuf};
 use pageserver_api::keyspace::KeySpace;
@@ -1255,6 +1256,14 @@ impl LayerInner {
 
                 self.access_stats.record_residence_event();
 
+                let task_kind: &'static str = ctx.task_kind().into();
+                ONDEMAND_DOWNLOAD_BYTES
+                    .with_label_values(&[task_kind])
+                    .inc_by(self.desc.file_size);
+                ONDEMAND_DOWNLOAD_COUNT
+                    .with_label_values(&[task_kind])
+                    .inc();
+
                 Ok(self.initialize_after_layer_is_on_disk(permit))
             }
             Err(e) => {

pageserver/src/virtual_file.rs

@@ -14,8 +14,6 @@
 use std::fs::File;
 use std::io::{Error, ErrorKind};
 use std::os::fd::{AsRawFd, FromRawFd, IntoRawFd, OwnedFd, RawFd};
-#[cfg(target_os = "linux")]
-use std::os::unix::fs::OpenOptionsExt;
 use std::sync::LazyLock;
 use std::sync::atomic::{AtomicBool, AtomicU8, AtomicUsize, Ordering};
 
@@ -99,7 +97,7 @@ impl VirtualFile {
 
     pub async fn open_with_options_v2<P: AsRef<Utf8Path>>(
         path: P,
-        open_options: &OpenOptions,
+        #[cfg_attr(not(target_os = "linux"), allow(unused_mut))] mut open_options: OpenOptions,
         ctx: &RequestContext,
     ) -> Result<Self, std::io::Error> {
         let mode = get_io_mode();
@@ -112,21 +110,16 @@ impl VirtualFile {
             #[cfg(target_os = "linux")]
             (IoMode::DirectRw, _) => true,
         };
-        let open_options = open_options.clone();
-        let open_options = if set_o_direct {
+        if set_o_direct {
             #[cfg(target_os = "linux")]
             {
-                let mut open_options = open_options;
-                open_options.custom_flags(nix::libc::O_DIRECT);
-                open_options
+                open_options = open_options.custom_flags(nix::libc::O_DIRECT);
             }
             #[cfg(not(target_os = "linux"))]
             unreachable!(
                 "O_DIRECT is not supported on this platform, IoMode's that result in set_o_direct=true shouldn't even be defined"
             );
-        } else {
-            open_options
-        };
+        }
         let inner = VirtualFileInner::open_with_options(path, open_options, ctx).await?;
         Ok(VirtualFile { inner, _mode: mode })
     }
@@ -530,7 +523,7 @@ impl VirtualFileInner {
         path: P,
         ctx: &RequestContext,
     ) -> Result<VirtualFileInner, std::io::Error> {
-        Self::open_with_options(path.as_ref(), OpenOptions::new().read(true).clone(), ctx).await
+        Self::open_with_options(path.as_ref(), OpenOptions::new().read(true), ctx).await
     }
 
     /// Open a file with given options.
@@ -558,10 +551,11 @@ impl VirtualFileInner {
         // It would perhaps be nicer to check just for the read and write flags
         // explicitly, but OpenOptions doesn't contain any functions to read flags,
         // only to set them.
-        let mut reopen_options = open_options.clone();
-        reopen_options.create(false);
-        reopen_options.create_new(false);
-        reopen_options.truncate(false);
+        let reopen_options = open_options
+            .clone()
+            .create(false)
+            .create_new(false)
+            .truncate(false);
 
         let vfile = VirtualFileInner {
             handle: RwLock::new(handle),
@@ -1307,7 +1301,7 @@ mod tests {
                 opts: OpenOptions,
                 ctx: &RequestContext,
             ) -> Result<MaybeVirtualFile, anyhow::Error> {
-                let vf = VirtualFile::open_with_options_v2(&path, &opts, ctx).await?;
+                let vf = VirtualFile::open_with_options_v2(&path, opts, ctx).await?;
                 Ok(MaybeVirtualFile::VirtualFile(vf))
             }
         }
@@ -1374,7 +1368,7 @@ mod tests {
         let _ = file_a.read_string_at(0, 1, &ctx).await.unwrap_err();
 
         // Close the file and re-open for reading
-        let mut file_a = A::open(path_a, OpenOptions::new().read(true).to_owned(), &ctx).await?;
+        let mut file_a = A::open(path_a, OpenOptions::new().read(true), &ctx).await?;
 
         // cannot write to a file opened in read-only mode
         let _ = file_a
@@ -1393,8 +1387,7 @@ mod tests {
                 .read(true)
                 .write(true)
                 .create(true)
-                .truncate(true)
-                .to_owned(),
+                .truncate(true),
             &ctx,
         )
         .await?;
@@ -1412,12 +1405,7 @@ mod tests {
 
         let mut vfiles = Vec::new();
         for _ in 0..100 {
-            let mut vfile = A::open(
-                path_b.clone(),
-                OpenOptions::new().read(true).to_owned(),
-                &ctx,
-            )
-            .await?;
+            let mut vfile = A::open(path_b.clone(), OpenOptions::new().read(true), &ctx).await?;
             assert_eq!("FOOBAR", vfile.read_string_at(0, 6, &ctx).await?);
             vfiles.push(vfile);
         }
@@ -1466,7 +1454,7 @@ mod tests {
         for _ in 0..VIRTUAL_FILES {
             let f = VirtualFileInner::open_with_options(
                 &test_file_path,
-                OpenOptions::new().read(true).clone(),
+                OpenOptions::new().read(true),
                 &ctx,
             )
             .await?;

pageserver/src/virtual_file/open_options.rs

@@ -1,6 +1,7 @@
 //! Enum-dispatch to the `OpenOptions` type of the respective [`super::IoEngineKind`];
 
 use std::os::fd::OwnedFd;
+use std::os::unix::fs::OpenOptionsExt;
 use std::path::Path;
 
 use super::io_engine::IoEngine;
@@ -43,7 +44,7 @@ impl OpenOptions {
         self.write
     }
 
-    pub fn read(&mut self, read: bool) -> &mut OpenOptions {
+    pub fn read(mut self, read: bool) -> Self {
         match &mut self.inner {
             Inner::StdFs(x) => {
                 let _ = x.read(read);
@@ -56,7 +57,7 @@ impl OpenOptions {
         self
     }
 
-    pub fn write(&mut self, write: bool) -> &mut OpenOptions {
+    pub fn write(mut self, write: bool) -> Self {
         self.write = write;
         match &mut self.inner {
             Inner::StdFs(x) => {
@@ -70,7 +71,7 @@ impl OpenOptions {
         self
     }
 
-    pub fn create(&mut self, create: bool) -> &mut OpenOptions {
+    pub fn create(mut self, create: bool) -> Self {
         match &mut self.inner {
             Inner::StdFs(x) => {
                 let _ = x.create(create);
@@ -83,7 +84,7 @@ impl OpenOptions {
         self
     }
 
-    pub fn create_new(&mut self, create_new: bool) -> &mut OpenOptions {
+    pub fn create_new(mut self, create_new: bool) -> Self {
         match &mut self.inner {
             Inner::StdFs(x) => {
                 let _ = x.create_new(create_new);
@@ -96,7 +97,7 @@ impl OpenOptions {
         self
     }
 
-    pub fn truncate(&mut self, truncate: bool) -> &mut OpenOptions {
+    pub fn truncate(mut self, truncate: bool) -> Self {
         match &mut self.inner {
             Inner::StdFs(x) => {
                 let _ = x.truncate(truncate);
@@ -124,10 +125,8 @@ impl OpenOptions {
             }
         }
     }
-}
 
-impl std::os::unix::prelude::OpenOptionsExt for OpenOptions {
-    fn mode(&mut self, mode: u32) -> &mut OpenOptions {
+    pub fn mode(mut self, mode: u32) -> Self {
         match &mut self.inner {
             Inner::StdFs(x) => {
                 let _ = x.mode(mode);
@@ -140,7 +139,7 @@ impl std::os::unix::prelude::OpenOptionsExt for OpenOptions {
         self
     }
 
-    fn custom_flags(&mut self, flags: i32) -> &mut OpenOptions {
+    pub fn custom_flags(mut self, flags: i32) -> Self {
         match &mut self.inner {
             Inner::StdFs(x) => {
                 let _ = x.custom_flags(flags);

pgxn/neon/communicator.c

@@ -687,8 +687,14 @@ prefetch_wait_for(uint64 ring_index)
 		END_PREFETCH_RECEIVE_WORK();
 		CHECK_FOR_INTERRUPTS();
 	}
-
-	return result;
+	if (result)
+	{
+		/* Check that slot is actually received (srver can be disconnected in prefetch_pump_state called from CHECK_FOR_INTERRUPTS */
+		PrefetchRequest *slot = GetPrfSlot(ring_index);
+		return slot->status == PRFS_RECEIVED;
+	}
+	return false;
+;
 }
 
 /*

pgxn/neon/libpagestore.c

@@ -26,6 +26,7 @@
 #include "portability/instr_time.h"
 #include "postmaster/interrupt.h"
 #include "storage/buf_internals.h"
+#include "storage/fd.h"
 #include "storage/ipc.h"
 #include "storage/lwlock.h"
 #include "storage/pg_shmem.h"
@@ -79,6 +80,7 @@ int         neon_protocol_version = 3;
 static int	neon_compute_mode = 0;
 static int	max_reconnect_attempts = 60;
 static int	stripe_size;
+static int	max_sockets;
 
 static int pageserver_response_log_timeout = 10000;
 /* 2.5 minutes. A bit higher than highest default TCP retransmission timeout */
@@ -336,6 +338,13 @@ load_shard_map(shardno_t shard_no, char *connstr_p, shardno_t *num_shards_p)
 				pageserver_disconnect(i);
 		}
 		pagestore_local_counter = end_update_counter;
+
+        /* Reserve file descriptors for sockets */
+		while (max_sockets < num_shards)
+		{
+			max_sockets += 1;
+			ReserveExternalFD();
+		}
 	}
 
 	if (num_shards_p)
@@ -877,6 +886,7 @@ retry:
 			int			port;
 			int			sndbuf;
 			int			recvbuf;
+			uint64*		max_wait;
 
 			get_local_port(PQsocket(pageserver_conn), &port);
 			get_socket_stats(PQsocket(pageserver_conn), &sndbuf, &recvbuf);
@@ -887,7 +897,10 @@ retry:
 						   shard->nrequests_sent, shard->nresponses_received, port, sndbuf, recvbuf,
 				           pageserver_conn->inStart, pageserver_conn->inEnd);
 			shard->receive_last_log_time = now;
+			MyNeonCounters->compute_getpage_stuck_requests_total += !shard->receive_logged;
 			shard->receive_logged = true;
+			max_wait = &MyNeonCounters->compute_getpage_max_inflight_stuck_time_ms;
+			*max_wait = Max(*max_wait, INSTR_TIME_GET_MILLISEC(since_start));
 		}
 
 		/*
@@ -910,6 +923,7 @@ retry:
 			get_local_port(PQsocket(pageserver_conn), &port);
 			neon_shard_log(shard_no, LOG, "no response from pageserver for %0.3f s, disconnecting (socket port=%d)",
 					   INSTR_TIME_GET_DOUBLE(since_start), port);
+			MyNeonCounters->compute_getpage_max_inflight_stuck_time_ms = 0;
 			pageserver_disconnect(shard_no);
 			return -1;
 		}
@@ -933,6 +947,7 @@ retry:
 	INSTR_TIME_SET_ZERO(shard->receive_start_time);
 	INSTR_TIME_SET_ZERO(shard->receive_last_log_time);
 	shard->receive_logged = false;
+	MyNeonCounters->compute_getpage_max_inflight_stuck_time_ms = 0;
 
 	return ret;
 }

pgxn/neon/neon_lwlsncache.c

@@ -4,6 +4,7 @@
 
 #include "miscadmin.h"
 #include "access/xlog.h"
+#include "access/xlog_internal.h"
 #include "storage/ipc.h"
 #include "storage/shmem.h"
 #include "storage/buf_internals.h"
@@ -396,9 +397,10 @@ SetLastWrittenLSNForBlockRangeInternal(XLogRecPtr lsn,
 XLogRecPtr
 neon_set_lwlsn_block_range(XLogRecPtr lsn, NRelFileInfo rlocator, ForkNumber forknum, BlockNumber from, BlockNumber n_blocks)
 {
-	if (lsn < FirstNormalUnloggedLSN || n_blocks == 0 || LwLsnCache->lastWrittenLsnCacheSize == 0)
+	if (lsn == InvalidXLogRecPtr || n_blocks == 0 || LwLsnCache->lastWrittenLsnCacheSize == 0)
 		return lsn;
 
+	Assert(lsn >= WalSegMinSize);
 	LWLockAcquire(LastWrittenLsnLock, LW_EXCLUSIVE);
 	lsn = SetLastWrittenLSNForBlockRangeInternal(lsn, rlocator, forknum, from, n_blocks);
 	LWLockRelease(LastWrittenLsnLock);
@@ -435,7 +437,6 @@ neon_set_lwlsn_block_v(const XLogRecPtr *lsns, NRelFileInfo relfilenode,
 		NInfoGetRelNumber(relfilenode) == InvalidOid)
 		return InvalidXLogRecPtr;
 
-	
 	BufTagInit(key,  relNumber, forknum, blockno, spcOid, dbOid);
 
 	LWLockAcquire(LastWrittenLsnLock, LW_EXCLUSIVE);
@@ -444,6 +445,10 @@ neon_set_lwlsn_block_v(const XLogRecPtr *lsns, NRelFileInfo relfilenode,
 	{
 		XLogRecPtr	lsn = lsns[i];
 
+		if (lsn == InvalidXLogRecPtr)
+			continue;
+
+		Assert(lsn >= WalSegMinSize);
 		key.blockNum = blockno + i;
 		entry = hash_search(lastWrittenLsnCache, &key, HASH_ENTER, &found);
 		if (found)

pgxn/neon/neon_perf_counters.c

@@ -148,7 +148,7 @@ histogram_to_metrics(IOHistogram histogram,
 static metric_t *
 neon_perf_counters_to_metrics(neon_per_backend_counters *counters)
 {
-#define NUM_METRICS ((2 + NUM_IO_WAIT_BUCKETS) * 3 + 10)
+#define NUM_METRICS ((2 + NUM_IO_WAIT_BUCKETS) * 3 + 12)
 	metric_t   *metrics = palloc((NUM_METRICS + 1) * sizeof(metric_t));
 	int			i = 0;
 
@@ -166,6 +166,8 @@ neon_perf_counters_to_metrics(neon_per_backend_counters *counters)
 
 	APPEND_METRIC(getpage_prefetch_requests_total);
 	APPEND_METRIC(getpage_sync_requests_total);
+	APPEND_METRIC(compute_getpage_stuck_requests_total);
+	APPEND_METRIC(compute_getpage_max_inflight_stuck_time_ms);
 	APPEND_METRIC(getpage_prefetch_misses_total);
 	APPEND_METRIC(getpage_prefetch_discards_total);
 	APPEND_METRIC(pageserver_requests_sent_total);
@@ -294,6 +296,11 @@ neon_get_perf_counters(PG_FUNCTION_ARGS)
 		totals.file_cache_hits_total += counters->file_cache_hits_total;
 		histogram_merge_into(&totals.file_cache_read_hist, &counters->file_cache_read_hist);
 		histogram_merge_into(&totals.file_cache_write_hist, &counters->file_cache_write_hist);
+
+		totals.compute_getpage_stuck_requests_total += counters->compute_getpage_stuck_requests_total;
+		totals.compute_getpage_max_inflight_stuck_time_ms = Max(
+			totals.compute_getpage_max_inflight_stuck_time_ms,
+			counters->compute_getpage_max_inflight_stuck_time_ms);
 	}
 
 	metrics = neon_perf_counters_to_metrics(&totals);

pgxn/neon/neon_perf_counters.h

@@ -57,6 +57,18 @@ typedef struct
 	uint64		getpage_prefetch_requests_total;
 	uint64		getpage_sync_requests_total;
 
+	/* 
+	 * Total number of Getpage requests left without an answer for more than
+	 * pageserver_response_log_timeout but less than pageserver_response_disconnect_timeout
+	 */
+	uint64 compute_getpage_stuck_requests_total;
+
+	/* 
+	 * Longest waiting time for active stuck requests. If a stuck request gets a
+	 * response or disconnects, this metric is updated
+	 */
+	uint64 compute_getpage_max_inflight_stuck_time_ms;
+
 	/*
 	 * Total number of readahead misses; consisting of either prefetches that
 	 * don't satisfy the LSN bounds, or cases where no readahead was issued

pgxn/neon/walproposer.c

@@ -836,7 +836,7 @@ TermsCollectedMset(WalProposer *wp, MemberSet *mset, Safekeeper **msk, StringInf
 {
 	uint32		n_greeted = 0;
 
-	for (uint32 i = 0; i < wp->mconf.members.len; i++)
+	for (uint32 i = 0; i < mset->len; i++)
 	{
 		Safekeeper *sk = msk[i];
 
@@ -1106,7 +1106,7 @@ VotesCollectedMset(WalProposer *wp, MemberSet *mset, Safekeeper **msk, StringInf
 {
 	uint32		n_votes = 0;
 
-	for (uint32 i = 0; i < wp->mconf.members.len; i++)
+	for (uint32 i = 0; i < mset->len; i++)
 	{
 		Safekeeper *sk = msk[i];
 

pgxn/neon/walproposer_pg.c

@@ -63,7 +63,7 @@
 char	   *wal_acceptors_list = "";
 int			wal_acceptor_reconnect_timeout = 1000;
 int			wal_acceptor_connection_timeout = 10000;
-int			safekeeper_proto_version = 2;
+int			safekeeper_proto_version = 3;
 
 /* Set to true in the walproposer bgw. */
 static bool am_walproposer;
@@ -228,7 +228,7 @@ nwp_register_gucs(void)
 							"Version of compute <-> safekeeper protocol.",
 							"Used while migrating from 2 to 3.",
 							&safekeeper_proto_version,
-							2, 0, INT_MAX,
+							3, 0, INT_MAX,
 							PGC_POSTMASTER,
 							0,
 							NULL, NULL, NULL);

proxy/README.md

@@ -32,7 +32,7 @@ To play with it locally one may start proxy over a local postgres installation
 (see end of this page on how to generate certs with openssl):
 
 ```
-./target/debug/proxy -c server.crt -k server.key --auth-backend=postgres --auth-endpoint=postgres://stas@127.0.0.1:5432/stas --wss 0.0.0.0:4444
+LOGFMT=text ./target/debug/proxy -c server.crt -k server.key --auth-backend=postgres --auth-endpoint=postgres://stas@127.0.0.1:5432/stas --wss 0.0.0.0:4444
 ```
 
 If both postgres and proxy are running you may send a SQL query:
@@ -130,7 +130,7 @@ openssl req -new -x509 -days 365 -nodes -text -out server.crt -keyout server.key
 
 Then we need to build proxy with 'testing' feature and run, e.g.:
 ```sh
-RUST_LOG=proxy cargo run -p proxy --bin proxy --features testing -- --auth-backend postgres --auth-endpoint 'postgresql://postgres:proxy-postgres@127.0.0.1:5432/postgres' -c server.crt -k server.key
+RUST_LOG=proxy LOGFMT=text cargo run -p proxy --bin proxy --features testing -- --auth-backend postgres --auth-endpoint 'postgresql://postgres:proxy-postgres@127.0.0.1:5432/postgres' -c server.crt -k server.key
 ```
 
 Now from client you can start a new session:

proxy/src/auth/credentials.rs

@@ -32,12 +32,6 @@ pub(crate) enum ComputeUserInfoParseError {
         option: EndpointId,
     },
 
-    #[error(
-        "Common name inferred from SNI ('{}') is not known",
-        .cn,
-    )]
-    UnknownCommonName { cn: String },
-
     #[error("Project name ('{0}') must contain only alphanumeric characters and hyphen.")]
     MalformedProjectName(EndpointId),
 }
@@ -66,22 +60,15 @@ impl ComputeUserInfoMaybeEndpoint {
     }
 }
 
-pub(crate) fn endpoint_sni(
-    sni: &str,
-    common_names: &HashSet<String>,
-) -> Result<Option<EndpointId>, ComputeUserInfoParseError> {
-    let Some((subdomain, common_name)) = sni.split_once('.') else {
-        return Err(ComputeUserInfoParseError::UnknownCommonName { cn: sni.into() });
-    };
+pub(crate) fn endpoint_sni(sni: &str, common_names: &HashSet<String>) -> Option<EndpointId> {
+    let (subdomain, common_name) = sni.split_once('.')?;
     if !common_names.contains(common_name) {
-        return Err(ComputeUserInfoParseError::UnknownCommonName {
-            cn: common_name.into(),
-        });
+        return None;
     }
     if subdomain == SERVERLESS_DRIVER_SNI {
-        return Ok(None);
+        return None;
     }
-    Ok(Some(EndpointId::from(subdomain)))
+    Some(EndpointId::from(subdomain))
 }
 
 impl ComputeUserInfoMaybeEndpoint {
@@ -113,15 +100,8 @@ impl ComputeUserInfoMaybeEndpoint {
             })
             .map(|name| name.into());
 
-        let endpoint_from_domain = if let Some(sni_str) = sni {
-            if let Some(cn) = common_names {
-                endpoint_sni(sni_str, cn)?
-            } else {
-                None
-            }
-        } else {
-            None
-        };
+        let endpoint_from_domain =
+            sni.and_then(|sni_str| common_names.and_then(|cn| endpoint_sni(sni_str, cn)));
 
         let endpoint = match (endpoint_option, endpoint_from_domain) {
             // Invariant: if we have both project name variants, they should match.
@@ -424,21 +404,34 @@ mod tests {
     }
 
     #[test]
-    fn parse_inconsistent_sni() {
+    fn parse_unknown_sni() {
         let options = StartupMessageParams::new([("user", "john_doe")]);
 
         let sni = Some("project.localhost");
         let common_names = Some(["example.com".into()].into());
 
         let ctx = RequestContext::test();
-        let err = ComputeUserInfoMaybeEndpoint::parse(&ctx, &options, sni, common_names.as_ref())
-            .expect_err("should fail");
-        match err {
-            UnknownCommonName { cn } => {
-                assert_eq!(cn, "localhost");
-            }
-            _ => panic!("bad error: {err:?}"),
-        }
+        let info = ComputeUserInfoMaybeEndpoint::parse(&ctx, &options, sni, common_names.as_ref())
+            .unwrap();
+
+        assert!(info.endpoint_id.is_none());
+    }
+
+    #[test]
+    fn parse_unknown_sni_with_options() {
+        let options = StartupMessageParams::new([
+            ("user", "john_doe"),
+            ("options", "endpoint=foo-bar-baz-1234"),
+        ]);
+
+        let sni = Some("project.localhost");
+        let common_names = Some(["example.com".into()].into());
+
+        let ctx = RequestContext::test();
+        let info = ComputeUserInfoMaybeEndpoint::parse(&ctx, &options, sni, common_names.as_ref())
+            .unwrap();
+
+        assert_eq!(info.endpoint_id.as_deref(), Some("foo-bar-baz-1234"));
     }
 
     #[test]

proxy/src/logging.rs

@@ -132,11 +132,10 @@ impl Drop for LoggingGuard {
     }
 }
 
-// TODO: make JSON the default
 #[derive(Copy, Clone, PartialEq, Eq, Default, Debug)]
 enum LogFormat {
+    Text,
     #[default]
-    Text = 1,
     Json,
 }
 

proxy/src/proxy/handshake.rs

@@ -24,9 +24,6 @@ pub(crate) enum HandshakeError {
     #[error("protocol violation")]
     ProtocolViolation,
 
-    #[error("missing certificate")]
-    MissingCertificate,
-
     #[error("{0}")]
     StreamUpgradeError(#[from] StreamUpgradeError),
 
@@ -42,10 +39,6 @@ impl ReportableError for HandshakeError {
         match self {
             HandshakeError::EarlyData => crate::error::ErrorKind::User,
             HandshakeError::ProtocolViolation => crate::error::ErrorKind::User,
-            // This error should not happen, but will if we have no default certificate and
-            // the client sends no SNI extension.
-            // If they provide SNI then we can be sure there is a certificate that matches.
-            HandshakeError::MissingCertificate => crate::error::ErrorKind::Service,
             HandshakeError::StreamUpgradeError(upgrade) => match upgrade {
                 StreamUpgradeError::AlreadyTls => crate::error::ErrorKind::Service,
                 StreamUpgradeError::Io(_) => crate::error::ErrorKind::ClientDisconnect,
@@ -146,7 +139,7 @@ pub(crate) async fn handshake<S: AsyncRead + AsyncWrite + Unpin>(
                         // try parse endpoint
                         let ep = conn_info
                             .server_name()
-                            .and_then(|sni| endpoint_sni(sni, &tls.common_names).ok().flatten());
+                            .and_then(|sni| endpoint_sni(sni, &tls.common_names));
                         if let Some(ep) = ep {
                             ctx.set_endpoint_id(ep);
                         }
@@ -161,10 +154,8 @@ pub(crate) async fn handshake<S: AsyncRead + AsyncWrite + Unpin>(
                             }
                         }
 
-                        let (_, tls_server_end_point) = tls
-                            .cert_resolver
-                            .resolve(conn_info.server_name())
-                            .ok_or(HandshakeError::MissingCertificate)?;
+                        let (_, tls_server_end_point) =
+                            tls.cert_resolver.resolve(conn_info.server_name());
 
                         stream = PqStream {
                             framed: Framed {

proxy/src/proxy/tests/mod.rs

@@ -98,8 +98,7 @@ fn generate_tls_config<'a>(
                 .with_no_client_auth()
                 .with_single_cert(vec![cert.clone()], key.clone_key())?;
 
-        let mut cert_resolver = CertResolver::new();
-        cert_resolver.add_cert(key, vec![cert], true)?;
+        let cert_resolver = CertResolver::new(key, vec![cert])?;
 
         let common_names = cert_resolver.get_common_names();
 

proxy/src/serverless/sql_over_http.rs

@@ -199,8 +199,7 @@ fn get_conn_info(
     let endpoint = match connection_url.host() {
         Some(url::Host::Domain(hostname)) => {
             if let Some(tls) = tls {
-                endpoint_sni(hostname, &tls.common_names)?
-                    .ok_or(ConnInfoError::MalformedEndpoint)?
+                endpoint_sni(hostname, &tls.common_names).ok_or(ConnInfoError::MalformedEndpoint)?
             } else {
                 hostname
                     .split_once('.')

proxy/src/tls/server_config.rs

@@ -5,6 +5,7 @@ use anyhow::{Context, bail};
 use itertools::Itertools;
 use rustls::crypto::ring::{self, sign};
 use rustls::pki_types::{CertificateDer, PrivateKeyDer};
+use rustls::sign::CertifiedKey;
 use x509_cert::der::{Reader, SliceReader};
 
 use super::{PG_ALPN_PROTOCOL, TlsServerEndPoint};
@@ -25,10 +26,8 @@ pub fn configure_tls(
     certs_dir: Option<&String>,
     allow_tls_keylogfile: bool,
 ) -> anyhow::Result<TlsConfig> {
-    let mut cert_resolver = CertResolver::new();
-
     // add default certificate
-    cert_resolver.add_cert_path(key_path, cert_path, true)?;
+    let mut cert_resolver = CertResolver::parse_new(key_path, cert_path)?;
 
     // add extra certificates
     if let Some(certs_dir) = certs_dir {
@@ -40,11 +39,8 @@ pub fn configure_tls(
                 let key_path = path.join("tls.key");
                 let cert_path = path.join("tls.crt");
                 if key_path.exists() && cert_path.exists() {
-                    cert_resolver.add_cert_path(
-                        &key_path.to_string_lossy(),
-                        &cert_path.to_string_lossy(),
-                        false,
-                    )?;
+                    cert_resolver
+                        .add_cert_path(&key_path.to_string_lossy(), &cert_path.to_string_lossy())?;
                 }
             }
         }
@@ -83,92 +79,42 @@ pub fn configure_tls(
     })
 }
 
-#[derive(Default, Debug)]
+#[derive(Debug)]
 pub struct CertResolver {
     certs: HashMap<String, (Arc<rustls::sign::CertifiedKey>, TlsServerEndPoint)>,
-    default: Option<(Arc<rustls::sign::CertifiedKey>, TlsServerEndPoint)>,
+    default: (Arc<rustls::sign::CertifiedKey>, TlsServerEndPoint),
 }
 
 impl CertResolver {
-    pub fn new() -> Self {
-        Self::default()
+    fn parse_new(key_path: &str, cert_path: &str) -> anyhow::Result<Self> {
+        let (priv_key, cert_chain) = parse_key_cert(key_path, cert_path)?;
+        Self::new(priv_key, cert_chain)
     }
 
-    fn add_cert_path(
-        &mut self,
-        key_path: &str,
-        cert_path: &str,
-        is_default: bool,
-    ) -> anyhow::Result<()> {
-        let priv_key = {
-            let key_bytes = std::fs::read(key_path)
-                .with_context(|| format!("Failed to read TLS keys at '{key_path}'"))?;
-            rustls_pemfile::private_key(&mut &key_bytes[..])
-                .with_context(|| format!("Failed to parse TLS keys at '{key_path}'"))?
-                .with_context(|| format!("Failed to parse TLS keys at '{key_path}'"))?
-        };
+    pub fn new(
+        priv_key: PrivateKeyDer<'static>,
+        cert_chain: Vec<CertificateDer<'static>>,
+    ) -> anyhow::Result<Self> {
+        let (common_name, cert, tls_server_end_point) = process_key_cert(priv_key, cert_chain)?;
 
-        let cert_chain_bytes = std::fs::read(cert_path)
-            .context(format!("Failed to read TLS cert file at '{cert_path}.'"))?;
-
-        let cert_chain = {
-            rustls_pemfile::certs(&mut &cert_chain_bytes[..])
-                .try_collect()
-                .with_context(|| {
-                    format!("Failed to read TLS certificate chain from bytes from file at '{cert_path}'.")
-                })?
-        };
-
-        self.add_cert(priv_key, cert_chain, is_default)
+        let mut certs = HashMap::new();
+        let default = (cert.clone(), tls_server_end_point);
+        certs.insert(common_name, (cert, tls_server_end_point));
+        Ok(Self { certs, default })
     }
 
-    pub fn add_cert(
+    fn add_cert_path(&mut self, key_path: &str, cert_path: &str) -> anyhow::Result<()> {
+        let (priv_key, cert_chain) = parse_key_cert(key_path, cert_path)?;
+        self.add_cert(priv_key, cert_chain)
+    }
+
+    fn add_cert(
         &mut self,
         priv_key: PrivateKeyDer<'static>,
         cert_chain: Vec<CertificateDer<'static>>,
-        is_default: bool,
     ) -> anyhow::Result<()> {
-        let key = sign::any_supported_type(&priv_key).context("invalid private key")?;
-
-        let first_cert = &cert_chain[0];
-        let tls_server_end_point = TlsServerEndPoint::new(first_cert)?;
-
-        let certificate = SliceReader::new(first_cert)
-            .context("Failed to parse cerficiate")?
-            .decode::<x509_cert::Certificate>()
-            .context("Failed to parse cerficiate")?;
-
-        let common_name = certificate.tbs_certificate.subject.to_string();
-
-        // We need to get the canonical name for this certificate so we can match them against any domain names
-        // seen within the proxy codebase.
-        //
-        // In scram-proxy we use wildcard certificates only, with the database endpoint as the wildcard subdomain, taken from SNI.
-        // We need to remove the wildcard prefix for the purposes of certificate selection.
-        //
-        // auth-broker does not use SNI and instead uses the Neon-Connection-String header.
-        // Auth broker has the subdomain `apiauth` we need to remove for the purposes of validating the Neon-Connection-String.
-        //
-        // Console Redirect proxy does not use any wildcard domains and does not need any certificate selection or conn string
-        // validation, so let's we can continue with any common-name
-        let common_name = if let Some(s) = common_name.strip_prefix("CN=*.") {
-            s.to_string()
-        } else if let Some(s) = common_name.strip_prefix("CN=apiauth.") {
-            s.to_string()
-        } else if let Some(s) = common_name.strip_prefix("CN=") {
-            s.to_string()
-        } else {
-            bail!("Failed to parse common name from certificate")
-        };
-
-        let cert = Arc::new(rustls::sign::CertifiedKey::new(cert_chain, key));
-
-        if is_default {
-            self.default = Some((cert.clone(), tls_server_end_point));
-        }
-
+        let (common_name, cert, tls_server_end_point) = process_key_cert(priv_key, cert_chain)?;
         self.certs.insert(common_name, (cert, tls_server_end_point));
-
         Ok(())
     }
 
@@ -177,12 +123,82 @@ impl CertResolver {
     }
 }
 
+fn parse_key_cert(
+    key_path: &str,
+    cert_path: &str,
+) -> anyhow::Result<(PrivateKeyDer<'static>, Vec<CertificateDer<'static>>)> {
+    let priv_key = {
+        let key_bytes = std::fs::read(key_path)
+            .with_context(|| format!("Failed to read TLS keys at '{key_path}'"))?;
+        rustls_pemfile::private_key(&mut &key_bytes[..])
+            .with_context(|| format!("Failed to parse TLS keys at '{key_path}'"))?
+            .with_context(|| format!("Failed to parse TLS keys at '{key_path}'"))?
+    };
+
+    let cert_chain_bytes = std::fs::read(cert_path)
+        .context(format!("Failed to read TLS cert file at '{cert_path}.'"))?;
+
+    let cert_chain = {
+        rustls_pemfile::certs(&mut &cert_chain_bytes[..])
+            .try_collect()
+            .with_context(|| {
+                format!(
+                    "Failed to read TLS certificate chain from bytes from file at '{cert_path}'."
+                )
+            })?
+    };
+
+    Ok((priv_key, cert_chain))
+}
+
+fn process_key_cert(
+    priv_key: PrivateKeyDer<'static>,
+    cert_chain: Vec<CertificateDer<'static>>,
+) -> anyhow::Result<(String, Arc<CertifiedKey>, TlsServerEndPoint)> {
+    let key = sign::any_supported_type(&priv_key).context("invalid private key")?;
+
+    let first_cert = &cert_chain[0];
+    let tls_server_end_point = TlsServerEndPoint::new(first_cert)?;
+
+    let certificate = SliceReader::new(first_cert)
+        .context("Failed to parse cerficiate")?
+        .decode::<x509_cert::Certificate>()
+        .context("Failed to parse cerficiate")?;
+
+    let common_name = certificate.tbs_certificate.subject.to_string();
+
+    // We need to get the canonical name for this certificate so we can match them against any domain names
+    // seen within the proxy codebase.
+    //
+    // In scram-proxy we use wildcard certificates only, with the database endpoint as the wildcard subdomain, taken from SNI.
+    // We need to remove the wildcard prefix for the purposes of certificate selection.
+    //
+    // auth-broker does not use SNI and instead uses the Neon-Connection-String header.
+    // Auth broker has the subdomain `apiauth` we need to remove for the purposes of validating the Neon-Connection-String.
+    //
+    // Console Redirect proxy does not use any wildcard domains and does not need any certificate selection or conn string
+    // validation, so let's we can continue with any common-name
+    let common_name = if let Some(s) = common_name.strip_prefix("CN=*.") {
+        s.to_string()
+    } else if let Some(s) = common_name.strip_prefix("CN=apiauth.") {
+        s.to_string()
+    } else if let Some(s) = common_name.strip_prefix("CN=") {
+        s.to_string()
+    } else {
+        bail!("Failed to parse common name from certificate")
+    };
+
+    let cert = Arc::new(rustls::sign::CertifiedKey::new(cert_chain, key));
+
+    Ok((common_name, cert, tls_server_end_point))
+}
+
 impl rustls::server::ResolvesServerCert for CertResolver {
     fn resolve(
         &self,
         client_hello: rustls::server::ClientHello<'_>,
     ) -> Option<Arc<rustls::sign::CertifiedKey>> {
-        self.resolve(client_hello.server_name()).map(|x| x.0)
+        Some(self.resolve(client_hello.server_name()).0)
     }
 }
 
@@ -190,7 +206,7 @@ impl CertResolver {
     pub fn resolve(
         &self,
         server_name: Option<&str>,
-    ) -> Option<(Arc<rustls::sign::CertifiedKey>, TlsServerEndPoint)> {
+    ) -> (Arc<rustls::sign::CertifiedKey>, TlsServerEndPoint) {
         // loop here and cut off more and more subdomains until we find
         // a match to get a proper wildcard support. OTOH, we now do not
         // use nested domains, so keep this simple for now.
@@ -200,12 +216,17 @@ impl CertResolver {
         if let Some(mut sni_name) = server_name {
             loop {
                 if let Some(cert) = self.certs.get(sni_name) {
-                    return Some(cert.clone());
+                    return cert.clone();
                 }
                 if let Some((_, rest)) = sni_name.split_once('.') {
                     sni_name = rest;
                 } else {
-                    return None;
+                    // The customer has some custom DNS mapping - just return
+                    // a default certificate.
+                    //
+                    // This will error if the customer uses anything stronger
+                    // than sslmode=require. That's a choice they can make.
+                    return self.default.clone();
                 }
             }
         } else {

safekeeper/src/pull_timeline.rs

@@ -401,7 +401,10 @@ pub async fn handle_request(
         request.timeline_id,
     ));
     if existing_tli.is_ok() {
-        bail!("Timeline {} already exists", request.timeline_id);
+        info!("Timeline {} already exists", request.timeline_id);
+        return Ok(PullTimelineResponse {
+            safekeeper_host: None,
+        });
     }
 
     let mut http_client = reqwest::Client::builder();
@@ -425,8 +428,25 @@ pub async fn handle_request(
 
     let mut statuses = Vec::new();
     for (i, response) in responses.into_iter().enumerate() {
-        let status = response.context(format!("fetching status from {}", http_hosts[i]))?;
-        statuses.push((status, i));
+        match response {
+            Ok(status) => {
+                statuses.push((status, i));
+            }
+            Err(e) => {
+                info!("error fetching status from {}: {e}", http_hosts[i]);
+            }
+        }
+    }
+
+    // Allow missing responses from up to one safekeeper (say due to downtime)
+    // e.g. if we created a timeline on PS A and B, with C being offline. Then B goes
+    // offline and C comes online. Then we want a pull on C with A and B as hosts to work.
+    let min_required_successful = (http_hosts.len() - 1).max(1);
+    if statuses.len() < min_required_successful {
+        bail!(
+            "only got {} successful status responses. required: {min_required_successful}",
+            statuses.len()
+        )
     }
 
     // Find the most advanced safekeeper
@@ -536,6 +556,6 @@ async fn pull_timeline(
         .await?;
 
     Ok(PullTimelineResponse {
-        safekeeper_host: host,
+        safekeeper_host: Some(host),
     })
 }

safekeeper/src/receive_wal.rs

@@ -32,7 +32,7 @@ use crate::metrics::{
     WAL_RECEIVERS,
 };
 use crate::safekeeper::{AcceptorProposerMessage, ProposerAcceptorMessage};
-use crate::timeline::WalResidentTimeline;
+use crate::timeline::{TimelineError, WalResidentTimeline};
 
 const DEFAULT_FEEDBACK_CAPACITY: usize = 8;
 
@@ -357,9 +357,14 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> NetworkReader<'_, IO> {
                         .await
                         .context("create timeline")?
                 } else {
-                    self.global_timelines
-                        .get(self.ttid)
-                        .context("get timeline")?
+                    let timeline_res = self.global_timelines.get(self.ttid);
+                    match timeline_res {
+                        Ok(tl) => tl,
+                        Err(TimelineError::NotFound(_)) => {
+                            return Err(CopyStreamHandlerEnd::TimelineNoCreate);
+                        }
+                        other => other.context("get_timeline")?,
+                    }
                 };
                 tli.wal_residence_guard().await?
             }

storage_controller/src/main.rs

@@ -19,7 +19,8 @@ use storage_controller::service::chaos_injector::ChaosInjector;
 use storage_controller::service::{
     Config, HEARTBEAT_INTERVAL_DEFAULT, LONG_RECONCILE_THRESHOLD_DEFAULT,
     MAX_OFFLINE_INTERVAL_DEFAULT, MAX_WARMING_UP_INTERVAL_DEFAULT,
-    PRIORITY_RECONCILER_CONCURRENCY_DEFAULT, RECONCILER_CONCURRENCY_DEFAULT, Service,
+    PRIORITY_RECONCILER_CONCURRENCY_DEFAULT, RECONCILER_CONCURRENCY_DEFAULT,
+    SAFEKEEPER_RECONCILER_CONCURRENCY_DEFAULT, Service,
 };
 use tokio::signal::unix::SignalKind;
 use tokio_util::sync::CancellationToken;
@@ -132,6 +133,10 @@ struct Cli {
     #[arg(long)]
     priority_reconciler_concurrency: Option<usize>,
 
+    /// Maximum number of safekeeper reconciliations that may run in parallel (per safekeeper)
+    #[arg(long)]
+    safekeeper_reconciler_concurrency: Option<usize>,
+
     /// Tenant API rate limit, as requests per second per tenant.
     #[arg(long, default_value = "10")]
     tenant_rate_limit: NonZeroU32,
@@ -403,6 +408,9 @@ async fn async_main() -> anyhow::Result<()> {
         priority_reconciler_concurrency: args
             .priority_reconciler_concurrency
             .unwrap_or(PRIORITY_RECONCILER_CONCURRENCY_DEFAULT),
+        safekeeper_reconciler_concurrency: args
+            .safekeeper_reconciler_concurrency
+            .unwrap_or(SAFEKEEPER_RECONCILER_CONCURRENCY_DEFAULT),
         tenant_rate_limit: args.tenant_rate_limit,
         split_threshold: args.split_threshold,
         max_split_shards: args.max_split_shards,

storage_controller/src/service.rs

@@ -194,6 +194,7 @@ pub(crate) enum LeadershipStatus {
 
 pub const RECONCILER_CONCURRENCY_DEFAULT: usize = 128;
 pub const PRIORITY_RECONCILER_CONCURRENCY_DEFAULT: usize = 256;
+pub const SAFEKEEPER_RECONCILER_CONCURRENCY_DEFAULT: usize = 32;
 
 // Depth of the channel used to enqueue shards for reconciliation when they can't do it immediately.
 // This channel is finite-size to avoid using excessive memory if we get into a state where reconciles are finishing more slowly
@@ -382,6 +383,9 @@ pub struct Config {
     /// How many high-priority Reconcilers may be spawned concurrently
     pub priority_reconciler_concurrency: usize,
 
+    /// How many safekeeper reconciles may happen concurrently (per safekeeper)
+    pub safekeeper_reconciler_concurrency: usize,
+
     /// How many API requests per second to allow per tenant, across all
     /// tenant-scoped API endpoints. Further API requests queue until ready.
     pub tenant_rate_limit: NonZeroU32,
@@ -3659,7 +3663,7 @@ impl Service {
                 locations: ShardMutationLocations,
                 http_client: reqwest::Client,
                 jwt: Option<String>,
-                create_req: TimelineCreateRequest,
+                mut create_req: TimelineCreateRequest,
             ) -> Result<TimelineInfo, ApiError> {
                 let latest = locations.latest.node;
 
@@ -3678,6 +3682,15 @@ impl Service {
                     .await
                     .map_err(|e| passthrough_api_error(&latest, e))?;
 
+                // If we are going to create the timeline on some stale locations for shard 0, then ask them to re-use
+                // the initdb generated by the latest location, rather than generating their own.  This avoids racing uploads
+                // of initdb to S3 which might not be binary-identical if different pageservers have different postgres binaries.
+                if tenant_shard_id.is_shard_zero() {
+                    if let models::TimelineCreateRequestMode::Bootstrap { existing_initdb_timeline_id, .. } = &mut create_req.mode {
+                        *existing_initdb_timeline_id = Some(create_req.new_timeline_id);
+                    }
+                }
+
                 // We propagate timeline creations to all attached locations such that a compute
                 // for the new timeline is able to start regardless of the current state of the
                 // tenant shard reconciliation.

storage_controller/src/service/safekeeper_reconciler.rs

@@ -3,7 +3,10 @@ use std::{collections::HashMap, str::FromStr, sync::Arc, time::Duration};
 use clashmap::{ClashMap, Entry};
 use safekeeper_api::models::PullTimelineRequest;
 use safekeeper_client::mgmt_api;
-use tokio::sync::mpsc::{self, UnboundedReceiver, UnboundedSender};
+use tokio::sync::{
+    Semaphore,
+    mpsc::{self, UnboundedReceiver, UnboundedSender},
+};
 use tokio_util::sync::CancellationToken;
 use tracing::Instrument;
 use utils::{
@@ -206,18 +209,27 @@ impl ReconcilerHandle {
 }
 
 pub(crate) struct SafekeeperReconciler {
-    service: Arc<Service>,
+    inner: SafekeeperReconcilerInner,
+    concurrency_limiter: Arc<Semaphore>,
     rx: UnboundedReceiver<(ScheduleRequest, CancellationToken)>,
     cancel: CancellationToken,
 }
 
+/// Thin wrapper over `Service` to not clutter its inherent functions
+#[derive(Clone)]
+struct SafekeeperReconcilerInner {
+    service: Arc<Service>,
+}
+
 impl SafekeeperReconciler {
     fn spawn(cancel: CancellationToken, service: Arc<Service>) -> ReconcilerHandle {
         // We hold the ServiceInner lock so we don't want to make sending to the reconciler channel to be blocking.
         let (tx, rx) = mpsc::unbounded_channel();
+        let concurrency = service.config.safekeeper_reconciler_concurrency;
         let mut reconciler = SafekeeperReconciler {
-            service,
+            inner: SafekeeperReconcilerInner { service },
             rx,
+            concurrency_limiter: Arc::new(Semaphore::new(concurrency)),
             cancel: cancel.clone(),
         };
         let handle = ReconcilerHandle {
@@ -230,31 +242,44 @@ impl SafekeeperReconciler {
     }
     async fn run(&mut self) {
         loop {
-            // TODO add parallelism with semaphore here
             let req = tokio::select! {
                 req = self.rx.recv() => req,
                 _ = self.cancel.cancelled() => break,
             };
             let Some((req, req_cancel)) = req else { break };
+
+            let permit_res = tokio::select! {
+                req = self.concurrency_limiter.clone().acquire_owned() => req,
+                _ = self.cancel.cancelled() => break,
+            };
+            let Ok(_permit) = permit_res else { return };
+
+            let inner = self.inner.clone();
             if req_cancel.is_cancelled() {
                 continue;
             }
 
-            let kind = req.kind;
-            let tenant_id = req.tenant_id;
-            let timeline_id = req.timeline_id;
-            let node_id = req.safekeeper.skp.id;
-            self.reconcile_one(req, req_cancel)
-                .instrument(tracing::info_span!(
-                    "reconcile_one",
-                    ?kind,
-                    %tenant_id,
-                    ?timeline_id,
-                    %node_id,
-                ))
-                .await;
+            tokio::task::spawn(async move {
+                let kind = req.kind;
+                let tenant_id = req.tenant_id;
+                let timeline_id = req.timeline_id;
+                let node_id = req.safekeeper.skp.id;
+                inner
+                    .reconcile_one(req, req_cancel)
+                    .instrument(tracing::info_span!(
+                        "reconcile_one",
+                        ?kind,
+                        %tenant_id,
+                        ?timeline_id,
+                        %node_id,
+                    ))
+                    .await;
+            });
         }
     }
+}
+
+impl SafekeeperReconcilerInner {
     async fn reconcile_one(&self, req: ScheduleRequest, req_cancel: CancellationToken) {
         let req_host = req.safekeeper.skp.host.clone();
         match req.kind {
@@ -281,10 +306,11 @@ impl SafekeeperReconciler {
                     req,
                     async |client| client.pull_timeline(&pull_req).await,
                     |resp| {
-                        tracing::info!(
-                            "pulled timeline from {} onto {req_host}",
-                            resp.safekeeper_host,
-                        );
+                        if let Some(host) = resp.safekeeper_host {
+                            tracing::info!("pulled timeline from {host} onto {req_host}");
+                        } else {
+                            tracing::info!("timeline already present on safekeeper on {req_host}");
+                        }
                     },
                     req_cancel,
                 )

test_runner/fixtures/neon_fixtures.py

@@ -1279,7 +1279,8 @@ class NeonEnv:
                 )
 
             tenant_config = ps_cfg.setdefault("tenant_config", {})
-            tenant_config["rel_size_v2_enabled"] = True  # Enable relsize_v2 by default in tests
+            # This feature is pending rollout.
+            # tenant_config["rel_size_v2_enabled"] = True
 
             if self.pageserver_remote_storage is not None:
                 ps_cfg["remote_storage"] = remote_storage_to_toml_dict(

test_runner/regress/test_attach_tenant_config.py

@@ -186,7 +186,7 @@ def test_fully_custom_config(positive_env: NeonEnv):
             "type": "interpreted",
             "args": {"format": "bincode", "compression": {"zstd": {"level": 1}}},
         },
-        "rel_size_v2_enabled": False,  # test suite enables it by default as of https://github.com/neondatabase/neon/issues/11081, so, custom config means disabling it
+        "rel_size_v2_enabled": True,
         "gc_compaction_enabled": True,
         "gc_compaction_verification": False,
         "gc_compaction_initial_threshold_kb": 1024000,

test_runner/regress/test_compaction.py

@@ -202,6 +202,8 @@ def test_pageserver_gc_compaction_preempt(
     env = neon_env_builder.init_start(initial_tenant_conf=conf)
 
     env.pageserver.allowed_errors.append(".*The timeline or pageserver is shutting down.*")
+    env.pageserver.allowed_errors.append(".*flush task cancelled.*")
+    env.pageserver.allowed_errors.append(".*failed to pipe.*")
 
     tenant_id = env.initial_tenant
     timeline_id = env.initial_timeline
@@ -229,7 +231,7 @@ def test_pageserver_gc_compaction_preempt(
 
 
 @skip_in_debug_build("only run with release build")
-@pytest.mark.timeout(600)  # This test is slow with sanitizers enabled, especially on ARM
+@pytest.mark.timeout(900)  # This test is slow with sanitizers enabled, especially on ARM
 @pytest.mark.parametrize(
     "with_branches",
     ["with_branches", "no_branches"],

test_runner/regress/test_pageserver_metric_collection.py

@@ -506,7 +506,6 @@ class SyntheticSizeVerifier:
 
 PER_METRIC_VERIFIERS = {
     "remote_storage_size": CannotVerifyAnything,
-    "resident_size": CannotVerifyAnything,
     "written_size": WrittenDataVerifier,
     "written_data_bytes_delta": WrittenDataDeltaVerifier,
     "timeline_logical_size": CannotVerifyAnything,

test_runner/regress/test_pg_regress.py

@@ -471,7 +471,7 @@ def test_tx_abort_with_many_relations(
         try:
             # Rollback phase should be fast: this is one WAL record that we should process efficiently
             fut = exec.submit(rollback_and_wait)
-            fut.result(timeout=15)
+            fut.result(timeout=15 if reldir_type == "v1" else 30)
         except:
             exec.shutdown(wait=False, cancel_futures=True)
             raise