proxy: introduce Acceptor and Connector traits

storcon: verbose logs in rare case of shards not attached yet (#10262 )
## Problem When we do a timeline CRUD operation, we check that the shards we need to mutate are currently attached to a pageserver, by reading `generation` and `generation_pageserver` from the database. If any don't appear to be attached, we respond with a a 503 and "One or more shards in tenant is not yet attached". This is happening more often than expected, and it's not obvious with current logging what's going on: specifically which shard has a problem, and exactly what we're seeing in these persistent generation columns. (Aside: it's possible that we broke something with the change in #10011 which clears generation_pageserver when we detach a shard, although if so the mechanism isn't trivial: what should happen is that if we stamp on generation_pageserver if a reconciler is running, then it shouldn't matter because we're about to ## Summary of changes - When we are in Attached mode but find that generation_pageserver/generation are unset, output details while looping over shards.
2026-07-07 22:20:36 +00:00 · 2025-01-03 15:53:39 +01:00 · 2025-01-03 10:55:15 +00:00 · 2025-01-03 10:55:07 +00:00 · 2025-01-03 10:20:18 +00:00 · 2025-01-03 10:04:58 +00:00
127 changed files with 2734 additions and 1202 deletions
--- a/.github/workflows/benchmarking.yml
+++ b/.github/workflows/benchmarking.yml
@@ -308,6 +308,7 @@ jobs:
          "image": [ "'"$image_default"'" ],
          "include": [{ "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-freetier",       "db_size": "3gb" ,"runner": '"$runner_default"', "image": "'"$image_default"'" },
                      { "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-new",            "db_size": "10gb","runner": '"$runner_default"', "image": "'"$image_default"'" },
+                      { "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-new-many-tables","db_size": "10gb","runner": '"$runner_default"', "image": "'"$image_default"'" },
                      { "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-new",            "db_size": "50gb","runner": '"$runner_default"', "image": "'"$image_default"'" },
                      { "pg_version": 16, "region_id": "azure-eastus2",          "platform": "neonvm-azure-captest-freetier", "db_size": "3gb" ,"runner": '"$runner_azure"',   "image": "neondatabase/build-tools:pinned-bookworm" },
                      { "pg_version": 16, "region_id": "azure-eastus2",          "platform": "neonvm-azure-captest-new",      "db_size": "10gb","runner": '"$runner_azure"',   "image": "neondatabase/build-tools:pinned-bookworm" },
@@ -410,7 +411,7 @@ jobs:
        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

    - name: Create Neon Project
-      if: contains(fromJson('["neonvm-captest-new", "neonvm-captest-freetier", "neonvm-azure-captest-freetier", "neonvm-azure-captest-new"]'), matrix.platform)
+      if: contains(fromJson('["neonvm-captest-new", "neonvm-captest-new-many-tables", "neonvm-captest-freetier", "neonvm-azure-captest-freetier", "neonvm-azure-captest-new"]'), matrix.platform)
      id: create-neon-project
      uses: ./.github/actions/neon-project-create
      with:
@@ -429,7 +430,7 @@ jobs:
          neonvm-captest-sharding-reuse)
            CONNSTR=${{ secrets.BENCHMARK_CAPTEST_SHARDING_CONNSTR }}
            ;;
-          neonvm-captest-new | neonvm-captest-freetier | neonvm-azure-captest-new | neonvm-azure-captest-freetier)
+          neonvm-captest-new | neonvm-captest-new-many-tables | neonvm-captest-freetier | neonvm-azure-captest-new | neonvm-azure-captest-freetier)
            CONNSTR=${{ steps.create-neon-project.outputs.dsn }}
            ;;
          rds-aurora)
@@ -446,6 +447,26 @@ jobs:

        echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT

+    # we want to compare Neon project OLTP throughput and latency at scale factor 10 GB 
+    # without (neonvm-captest-new)
+    # and with (neonvm-captest-new-many-tables) many relations in the database
+    - name: Create many relations before the run
+      if: contains(fromJson('["neonvm-captest-new-many-tables"]'), matrix.platform)
+      uses: ./.github/actions/run-python-test-set
+      with:
+        build_type: ${{ env.BUILD_TYPE }}
+        test_selection: performance
+        run_in_parallel: false
+        save_perf_report: ${{ env.SAVE_PERF_REPORT }}
+        extra_params: -m remote_cluster --timeout 21600 -k test_perf_many_relations
+        pg_version: ${{ env.DEFAULT_PG_VERSION }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+      env:
+        BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
+        VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
+        PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
+        TEST_NUM_RELATIONS: 10000
+
    - name: Benchmark init
      uses: ./.github/actions/run-python-test-set
      with:
--- a/.github/workflows/build_and_test.yml
+++ b/.github/workflows/build_and_test.yml
@@ -212,7 +212,7 @@ jobs:
          fi
          echo "CLIPPY_COMMON_ARGS=${CLIPPY_COMMON_ARGS}" >> $GITHUB_ENV
      - name: Run cargo clippy (debug)
-        run: cargo hack --feature-powerset clippy $CLIPPY_COMMON_ARGS
+        run: cargo hack --features default --ignore-unknown-features --feature-powerset clippy $CLIPPY_COMMON_ARGS

      - name: Check documentation generation
        run: cargo doc --workspace --no-deps --document-private-items
--- a/.github/workflows/cloud-regress.yml
+++ b/.github/workflows/cloud-regress.yml
@@ -21,6 +21,8 @@ concurrency:

 permissions:
  id-token: write # aws-actions/configure-aws-credentials
+  statuses: write
+  contents: write

 jobs:
  regress:
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -10,9 +10,9 @@ checksum = "8b5ace29ee3216de37c0546865ad08edef58b0f9e76838ed8959a84a990e58c5"

 [[package]]
 name = "addr2line"
-version = "0.21.0"
+version = "0.24.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "8a30b2e23b9e17a9f90641c7ab1549cd9b44f296d3ccbf309d2863cfe398a0cb"
+checksum = "dfbe277e56a376000877090da837660b4427aad530e3028d44e0bffe4f89a1c1"
 dependencies = [
 "gimli",
 ]
@@ -23,6 +23,12 @@ version = "1.0.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f26201604c87b1e01bd3d98f8d5d9a8fcbb815e8cedb41ffccbeb4bf593a35fe"

+[[package]]
+name = "adler2"
+version = "2.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "512761e0bb2578dd7380c6baaa0f4ce03e84f95e960231d1dec8bf4d7d6e2627"
+
 [[package]]
 name = "ahash"
 version = "0.8.11"
@@ -871,17 +877,17 @@ dependencies = [

 [[package]]
 name = "backtrace"
-version = "0.3.69"
+version = "0.3.74"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2089b7e3f35b9dd2d0ed921ead4f6d318c27680d4a5bd167b3ee120edb105837"
+checksum = "8d82cb332cdfaed17ae235a638438ac4d4839913cc2af585c3c6746e8f8bee1a"
 dependencies = [
 "addr2line",
- "cc",
 "cfg-if",
 "libc",
- "miniz_oxide",
+ "miniz_oxide 0.8.0",
 "object",
 "rustc-demangle",
+ "windows-targets 0.52.6",
 ]

 [[package]]
@@ -1127,7 +1133,7 @@ dependencies = [
 "num-traits",
 "serde",
 "wasm-bindgen",
- "windows-targets 0.52.4",
+ "windows-targets 0.52.6",
 ]

 [[package]]
@@ -1268,6 +1274,7 @@ dependencies = [
 "chrono",
 "clap",
 "compute_api",
+ "fail",
 "flate2",
 "futures",
 "hyper 0.14.30",
@@ -1726,9 +1733,9 @@ checksum = "ab03c107fafeb3ee9f5925686dbb7a73bc76e3932abb0d2b365cb64b169cf04c"

 [[package]]
 name = "diesel"
-version = "2.2.3"
+version = "2.2.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "65e13bab2796f412722112327f3e575601a3e9cdcbe426f0d30dbf43f3f5dc71"
+checksum = "ccf1bedf64cdb9643204a36dd15b19a6ce8e7aa7f7b105868e9f1fad5ffa7d12"
 dependencies = [
 "bitflags 2.4.1",
 "byteorder",
@@ -2107,7 +2114,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "3b9429470923de8e8cbd4d2dc513535400b4b3fef0319fb5c4e1f520a7bef743"
 dependencies = [
 "crc32fast",
- "miniz_oxide",
+ "miniz_oxide 0.7.1",
 ]

 [[package]]
@@ -2308,9 +2315,9 @@ dependencies = [

 [[package]]
 name = "gimli"
-version = "0.28.1"
+version = "0.31.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4271d37baee1b8c7e4b708028c57d816cf9d2434acb33a549475f78c181f6253"
+checksum = "07e28edb80900c19c28f1072f2e8aeca7fa06b23cd4169cefe1af5aa3260783f"

 [[package]]
 name = "git-version"
@@ -3404,6 +3411,15 @@ dependencies = [
 "adler",
 ]

+[[package]]
+name = "miniz_oxide"
+version = "0.8.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e2d80299ef12ff69b16a84bb182e3b9df68b5a91574d3d4fa6e41b65deec4df1"
+dependencies = [
+ "adler2",
+]
+
 [[package]]
 name = "mio"
 version = "0.8.11"
@@ -3638,9 +3654,9 @@ dependencies = [

 [[package]]
 name = "object"
-version = "0.32.2"
+version = "0.36.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "a6a622008b6e321afc04970976f62ee297fdbaa6f95318ca343e3eebb9648441"
+checksum = "aedf0a2d09c573ed1d8d85b30c119153926a2b36dce0ab28322c09a117a4683e"
 dependencies = [
 "memchr",
 ]
@@ -4478,9 +4494,9 @@ checksum = "5b40af805b3121feab8a3c29f04d8ad262fa8e0561883e7653e024ae4479e6de"

 [[package]]
 name = "pq-sys"
-version = "0.4.8"
+version = "0.6.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "31c0052426df997c0cbd30789eb44ca097e3541717a7b8fa36b1c464ee7edebd"
+checksum = "f6cc05d7ea95200187117196eee9edd0644424911821aeb28a18ce60ea0b8793"
 dependencies = [
 "vcpkg",
 ]
@@ -5323,9 +5339,9 @@ dependencies = [

 [[package]]
 name = "rustc-demangle"
-version = "0.1.23"
+version = "0.1.24"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "d626bb9dae77e28219937af045c257c28bfd3f69333c512553507f5f9798cb76"
+checksum = "719b953e2095829ee67db738b3bfa9fa368c94900df327b3f07fe6e794d2fe1f"

 [[package]]
 name = "rustc-hash"
@@ -7219,6 +7235,7 @@ dependencies = [
 "anyhow",
 "arc-swap",
 "async-compression",
+ "backtrace",
 "bincode",
 "byteorder",
 "bytes",
@@ -7229,12 +7246,14 @@ dependencies = [
 "criterion",
 "diatomic-waker",
 "fail",
+ "flate2",
 "futures",
 "git-version",
 "hex",
 "hex-literal",
 "humantime",
 "hyper 0.14.30",
+ "itertools 0.10.5",
 "jemalloc_pprof",
 "jsonwebtoken",
 "metrics",
@@ -7591,7 +7610,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e48a53791691ab099e5e2ad123536d0fff50652600abaf43bbf952894110d0be"
 dependencies = [
 "windows-core",
- "windows-targets 0.52.4",
+ "windows-targets 0.52.6",
 ]

 [[package]]
@@ -7600,7 +7619,7 @@ version = "0.52.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "33ab640c8d7e35bf8ba19b884ba838ceb4fba93a4e8c65a9059d08afcfc683d9"
 dependencies = [
- "windows-targets 0.52.4",
+ "windows-targets 0.52.6",
 ]

 [[package]]
@@ -7618,7 +7637,7 @@ version = "0.52.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "282be5f36a8ce781fad8c8ae18fa3f9beff57ec1b52cb3de0789201425d9a33d"
 dependencies = [
- "windows-targets 0.52.4",
+ "windows-targets 0.52.6",
 ]

 [[package]]
@@ -7638,17 +7657,18 @@ dependencies = [

 [[package]]
 name = "windows-targets"
-version = "0.52.4"
+version = "0.52.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7dd37b7e5ab9018759f893a1952c9420d060016fc19a472b4bb20d1bdd694d1b"
+checksum = "9b724f72796e036ab90c1021d4780d4d3d648aca59e491e6b98e725b84e99973"
 dependencies = [
- "windows_aarch64_gnullvm 0.52.4",
- "windows_aarch64_msvc 0.52.4",
- "windows_i686_gnu 0.52.4",
- "windows_i686_msvc 0.52.4",
- "windows_x86_64_gnu 0.52.4",
- "windows_x86_64_gnullvm 0.52.4",
- "windows_x86_64_msvc 0.52.4",
+ "windows_aarch64_gnullvm 0.52.6",
+ "windows_aarch64_msvc 0.52.6",
+ "windows_i686_gnu 0.52.6",
+ "windows_i686_gnullvm",
+ "windows_i686_msvc 0.52.6",
+ "windows_x86_64_gnu 0.52.6",
+ "windows_x86_64_gnullvm 0.52.6",
+ "windows_x86_64_msvc 0.52.6",
 ]

 [[package]]
@@ -7659,9 +7679,9 @@ checksum = "91ae572e1b79dba883e0d315474df7305d12f569b400fcf90581b06062f7e1bc"

 [[package]]
 name = "windows_aarch64_gnullvm"
-version = "0.52.4"
+version = "0.52.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "bcf46cf4c365c6f2d1cc93ce535f2c8b244591df96ceee75d8e83deb70a9cac9"
+checksum = "32a4622180e7a0ec044bb555404c800bc9fd9ec262ec147edd5989ccd0c02cd3"

 [[package]]
 name = "windows_aarch64_msvc"
@@ -7671,9 +7691,9 @@ checksum = "b2ef27e0d7bdfcfc7b868b317c1d32c641a6fe4629c171b8928c7b08d98d7cf3"

 [[package]]
 name = "windows_aarch64_msvc"
-version = "0.52.4"
+version = "0.52.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "da9f259dd3bcf6990b55bffd094c4f7235817ba4ceebde8e6d11cd0c5633b675"
+checksum = "09ec2a7bb152e2252b53fa7803150007879548bc709c039df7627cabbd05d469"

 [[package]]
 name = "windows_i686_gnu"
@@ -7683,9 +7703,15 @@ checksum = "622a1962a7db830d6fd0a69683c80a18fda201879f0f447f065a3b7467daa241"

 [[package]]
 name = "windows_i686_gnu"
-version = "0.52.4"
+version = "0.52.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b474d8268f99e0995f25b9f095bc7434632601028cf86590aea5c8a5cb7801d3"
+checksum = "8e9b5ad5ab802e97eb8e295ac6720e509ee4c243f69d781394014ebfe8bbfa0b"
+
+[[package]]
+name = "windows_i686_gnullvm"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0eee52d38c090b3caa76c563b86c3a4bd71ef1a819287c19d586d7334ae8ed66"

 [[package]]
 name = "windows_i686_msvc"
@@ -7695,9 +7721,9 @@ checksum = "4542c6e364ce21bf45d69fdd2a8e455fa38d316158cfd43b3ac1c5b1b19f8e00"

 [[package]]
 name = "windows_i686_msvc"
-version = "0.52.4"
+version = "0.52.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "1515e9a29e5bed743cb4415a9ecf5dfca648ce85ee42e15873c3cd8610ff8e02"
+checksum = "240948bc05c5e7c6dabba28bf89d89ffce3e303022809e73deaefe4f6ec56c66"

 [[package]]
 name = "windows_x86_64_gnu"
@@ -7707,9 +7733,9 @@ checksum = "ca2b8a661f7628cbd23440e50b05d705db3686f894fc9580820623656af974b1"

 [[package]]
 name = "windows_x86_64_gnu"
-version = "0.52.4"
+version = "0.52.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "5eee091590e89cc02ad514ffe3ead9eb6b660aedca2183455434b93546371a03"
+checksum = "147a5c80aabfbf0c7d901cb5895d1de30ef2907eb21fbbab29ca94c5b08b1a78"

 [[package]]
 name = "windows_x86_64_gnullvm"
@@ -7719,9 +7745,9 @@ checksum = "7896dbc1f41e08872e9d5e8f8baa8fdd2677f29468c4e156210174edc7f7b953"

 [[package]]
 name = "windows_x86_64_gnullvm"
-version = "0.52.4"
+version = "0.52.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "77ca79f2451b49fa9e2af39f0747fe999fcda4f5e241b2898624dca97a1f2177"
+checksum = "24d5b23dc417412679681396f2b49f3de8c1473deb516bd34410872eff51ed0d"

 [[package]]
 name = "windows_x86_64_msvc"
@@ -7731,9 +7757,9 @@ checksum = "1a515f5799fe4961cb532f983ce2b23082366b898e52ffbce459c86f67c8378a"

 [[package]]
 name = "windows_x86_64_msvc"
-version = "0.52.4"
+version = "0.52.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "32b752e52a2da0ddfbdbcc6fceadfeede4c939ed16d13e648833a61dfb611ed8"
+checksum = "589f6da84c646204747d1270a2a5661ea66ed1cced2631d546fdfb155959f9ec"

 [[package]]
 name = "winnow"
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -52,6 +52,7 @@ anyhow = { version = "1.0", features = ["backtrace"] }
 arc-swap = "1.6"
 async-compression = { version = "0.4.0", features = ["tokio", "gzip", "zstd"] }
 atomic-take = "1.1.0"
+backtrace = "0.3.74"
 flate2 = "1.0.26"
 async-stream = "0.3"
 async-trait = "0.1"
--- a/compute/compute-node.Dockerfile
+++ b/compute/compute-node.Dockerfile
@@ -35,10 +35,12 @@ RUN case $DEBIAN_VERSION in \
      ;; \
    esac && \
    apt update &&  \
-    apt install --no-install-recommends -y git autoconf automake libtool build-essential bison flex libreadline-dev \
+    apt install --no-install-recommends --no-install-suggests -y \
+    ninja-build git autoconf automake libtool build-essential bison flex libreadline-dev \
    zlib1g-dev libxml2-dev libcurl4-openssl-dev libossp-uuid-dev wget ca-certificates pkg-config libssl-dev \
    libicu-dev libxslt1-dev liblz4-dev libzstd-dev zstd \
-    $VERSION_INSTALLS
+    $VERSION_INSTALLS \
+    && apt clean && rm -rf /var/lib/apt/lists/*

 #########################################################################################
 #
@@ -113,10 +115,12 @@ ARG DEBIAN_VERSION
 ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 RUN apt update && \
-    apt install --no-install-recommends -y gdal-bin libboost-dev libboost-thread-dev libboost-filesystem-dev \
+    apt install --no-install-recommends --no-install-suggests -y \
+    gdal-bin libboost-dev libboost-thread-dev libboost-filesystem-dev \
    libboost-system-dev libboost-iostreams-dev libboost-program-options-dev libboost-timer-dev \
    libcgal-dev libgdal-dev libgmp-dev libmpfr-dev libopenscenegraph-dev libprotobuf-c-dev \
-    protobuf-c-compiler xsltproc
+    protobuf-c-compiler xsltproc \
+    && apt clean && rm -rf /var/lib/apt/lists/*


 # Postgis 3.5.0 requires SFCGAL 1.4+
@@ -143,9 +147,9 @@ RUN case "${DEBIAN_VERSION}" in \
    wget https://gitlab.com/sfcgal/SFCGAL/-/archive/v${SFCGAL_VERSION}/SFCGAL-v${SFCGAL_VERSION}.tar.gz -O SFCGAL.tar.gz && \
    echo "${SFCGAL_CHECKSUM} SFCGAL.tar.gz" | sha256sum --check && \
    mkdir sfcgal-src && cd sfcgal-src && tar xzf ../SFCGAL.tar.gz --strip-components=1 -C . && \
-    cmake -DCMAKE_BUILD_TYPE=Release . && make -j $(getconf _NPROCESSORS_ONLN) && \
-    DESTDIR=/sfcgal make install -j $(getconf _NPROCESSORS_ONLN) && \
-    make clean && cp -R /sfcgal/* /
+    cmake -DCMAKE_BUILD_TYPE=Release -GNinja . && ninja -j $(getconf _NPROCESSORS_ONLN) && \
+    DESTDIR=/sfcgal ninja install -j $(getconf _NPROCESSORS_ONLN) && \
+    ninja clean && cp -R /sfcgal/* /

 ENV PATH="/usr/local/pgsql/bin:$PATH"

@@ -213,9 +217,9 @@ RUN case "${PG_VERSION}" in \
    echo "${PGROUTING_CHECKSUM} pgrouting.tar.gz" | sha256sum --check && \
    mkdir pgrouting-src && cd pgrouting-src && tar xzf ../pgrouting.tar.gz --strip-components=1 -C . && \
    mkdir build && cd build && \
-    cmake -DCMAKE_BUILD_TYPE=Release .. && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    cmake -GNinja -DCMAKE_BUILD_TYPE=Release .. && \
+    ninja -j $(getconf _NPROCESSORS_ONLN) && \
+    ninja -j $(getconf _NPROCESSORS_ONLN) install && \
    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgrouting.control && \
    find /usr/local/pgsql -type f | sed 's|^/usr/local/pgsql/||' > /after.txt &&\
    cp /usr/local/pgsql/share/extension/pgrouting.control /extensions/postgis && \
@@ -235,7 +239,9 @@ COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY compute/patches/plv8-3.1.10.patch /plv8-3.1.10.patch

 RUN apt update && \
-    apt install --no-install-recommends -y ninja-build python3-dev libncurses5 binutils clang
+    apt install --no-install-recommends --no-install-suggests -y \
+    ninja-build python3-dev libncurses5 binutils clang \
+    && apt clean && rm -rf /var/lib/apt/lists/*

 # plv8 3.2.3 supports v17
 # last release v3.2.3 - Sep 7, 2024
@@ -301,9 +307,10 @@ RUN mkdir -p /h3/usr/ && \
    echo "ec99f1f5974846bde64f4513cf8d2ea1b8d172d2218ab41803bf6a63532272bc h3.tar.gz" | sha256sum --check && \
    mkdir h3-src && cd h3-src && tar xzf ../h3.tar.gz --strip-components=1 -C . && \
    mkdir build && cd build && \
-    cmake .. -DCMAKE_BUILD_TYPE=Release && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    DESTDIR=/h3 make install && \
+    cmake .. -GNinja -DBUILD_BENCHMARKS=0 -DCMAKE_BUILD_TYPE=Release \
+        -DBUILD_FUZZERS=0 -DBUILD_FILTERS=0 -DBUILD_GENERATORS=0 -DBUILD_TESTING=0 \
+    && ninja -j $(getconf _NPROCESSORS_ONLN) && \
+    DESTDIR=/h3 ninja install && \
    cp -R /h3/usr / && \
    rm -rf build

@@ -650,14 +657,15 @@ FROM build-deps AS rdkit-pg-build
 ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-RUN apt-get update && \
-    apt-get install --no-install-recommends -y \
+RUN apt update && \
+    apt install --no-install-recommends --no-install-suggests -y \
        libboost-iostreams1.74-dev \
        libboost-regex1.74-dev \
        libboost-serialization1.74-dev \
        libboost-system1.74-dev \
        libeigen3-dev \
-        libboost-all-dev
+        libboost-all-dev \
+    && apt clean && rm -rf /var/lib/apt/lists/*

 # rdkit Release_2024_09_1 supports v17
 # last release Release_2024_09_1 - Sep 27, 2024
@@ -693,6 +701,8 @@ RUN case "${PG_VERSION}" in \
        -D RDK_BUILD_MOLINTERCHANGE_SUPPORT=OFF \
        -D RDK_BUILD_YAEHMOP_SUPPORT=OFF \
        -D RDK_BUILD_STRUCTCHECKER_SUPPORT=OFF \
+        -D RDK_TEST_MULTITHREADED=OFF \
+        -D RDK_BUILD_CPP_TESTS=OFF \
        -D RDK_USE_URF=OFF \
        -D RDK_BUILD_PGSQL=ON \
        -D RDK_PGSQL_STATIC=ON \
@@ -704,9 +714,10 @@ RUN case "${PG_VERSION}" in \
        -D RDK_INSTALL_COMIC_FONTS=OFF \
        -D RDK_BUILD_FREETYPE_SUPPORT=OFF \
        -D CMAKE_BUILD_TYPE=Release \
+        -GNinja \
        . && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    ninja -j $(getconf _NPROCESSORS_ONLN) && \
+    ninja -j $(getconf _NPROCESSORS_ONLN) install && \
    echo 'trusted = true' >> /usr/local/pgsql/share/extension/rdkit.control

 #########################################################################################
@@ -849,8 +860,9 @@ FROM build-deps AS rust-extensions-build
 ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-RUN apt-get update && \
-    apt-get install --no-install-recommends -y curl libclang-dev && \
+RUN apt update && \
+    apt install --no-install-recommends --no-install-suggests -y curl libclang-dev && \
+    apt clean && rm -rf /var/lib/apt/lists/* && \
    useradd -ms /bin/bash nonroot -b /home

 ENV HOME=/home/nonroot
@@ -885,8 +897,9 @@ FROM build-deps AS rust-extensions-build-pgrx12
 ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-RUN apt-get update && \
-    apt-get install --no-install-recommends -y curl libclang-dev && \
+RUN apt update && \
+    apt install --no-install-recommends --no-install-suggests -y curl libclang-dev && \
+    apt clean && rm -rf /var/lib/apt/lists/* && \
    useradd -ms /bin/bash nonroot -b /home

 ENV HOME=/home/nonroot
@@ -914,18 +927,22 @@ FROM rust-extensions-build-pgrx12 AS pg-onnx-build

 # cmake 3.26 or higher is required, so installing it using pip (bullseye-backports has cmake 3.25).
 # Install it using virtual environment, because Python 3.11 (the default version on Debian 12 (Bookworm)) complains otherwise
-RUN apt-get update && apt-get install -y python3 python3-pip python3-venv && \
+RUN apt update && apt install --no-install-recommends --no-install-suggests -y \
+    python3 python3-pip python3-venv && \
+    apt clean && rm -rf /var/lib/apt/lists/* && \
    python3 -m venv venv && \
    . venv/bin/activate && \
    python3 -m pip install cmake==3.30.5 && \
    wget https://github.com/microsoft/onnxruntime/archive/refs/tags/v1.18.1.tar.gz -O onnxruntime.tar.gz && \
    mkdir onnxruntime-src && cd onnxruntime-src && tar xzf ../onnxruntime.tar.gz --strip-components=1 -C . && \
-    ./build.sh --config Release --parallel --skip_submodule_sync --skip_tests --allow_running_as_root
+    ./build.sh --config Release --parallel --cmake_generator Ninja \
+    --skip_submodule_sync --skip_tests --allow_running_as_root


 FROM pg-onnx-build AS pgrag-pg-build

-RUN apt-get install -y protobuf-compiler && \
+RUN apt update && apt install --no-install-recommends --no-install-suggests -y protobuf-compiler \
+    && apt clean && rm -rf /var/lib/apt/lists/* && \
    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 && \
    mkdir pgrag-src && cd pgrag-src && tar xzf ../pgrag.tar.gz --strip-components=1 -C . && \
@@ -1168,6 +1185,25 @@ RUN case "${PG_VERSION}" in \
    make BUILD_TYPE=release -j $(getconf _NPROCESSORS_ONLN) install && \
    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pg_mooncake.control

+#########################################################################################
+#
+# Layer "pg_repack"
+# compile pg_repack extension
+#
+#########################################################################################
+
+FROM build-deps AS pg-repack-build
+ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
+
+RUN wget https://github.com/reorg/pg_repack/archive/refs/tags/ver_1.5.2.tar.gz -O pg_repack.tar.gz && \
+    echo '4516cad42251ed3ad53ff619733004db47d5755acac83f75924cd94d1c4fb681 pg_repack.tar.gz' | sha256sum --check && \
+    mkdir pg_repack-src && cd pg_repack-src && tar xzf ../pg_repack.tar.gz --strip-components=1 -C . && \
+    make -j $(getconf _NPROCESSORS_ONLN) && \
+    make -j $(getconf _NPROCESSORS_ONLN) install
+
 #########################################################################################
 #
 # Layer "neon-pg-ext-build"
@@ -1213,6 +1249,7 @@ 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/
+COPY --from=pg-repack-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY pgxn/ pgxn/

 RUN make -j $(getconf _NPROCESSORS_ONLN) \
@@ -1248,7 +1285,7 @@ RUN make -j $(getconf _NPROCESSORS_ONLN) \

 #########################################################################################
 #
-# Compile and run the Neon-specific `compute_ctl` and `fast_import` binaries
+# Compile the Neon-specific `compute_ctl`, `fast_import`, and `local_proxy` binaries
 #
 #########################################################################################
 FROM $REPOSITORY/$IMAGE:$TAG AS compute-tools
@@ -1258,7 +1295,7 @@ ENV BUILD_TAG=$BUILD_TAG
 USER nonroot
 # Copy entire project to get Cargo.* files with proper dependencies for the whole project
 COPY --chown=nonroot . .
-RUN cd compute_tools && mold -run cargo build --locked --profile release-line-debug-size-lto
+RUN mold -run cargo build --locked --profile release-line-debug-size-lto --bin compute_ctl --bin fast_import --bin local_proxy

 #########################################################################################
 #
@@ -1279,8 +1316,8 @@ COPY --from=compute-tools /home/nonroot/target/release-line-debug-size-lto/fast_

 FROM debian:$DEBIAN_FLAVOR AS pgbouncer
 RUN set -e \
-    && apt-get update \
-    && apt-get install --no-install-recommends -y \
+    && apt update \
+    && apt install --no-install-suggests --no-install-recommends -y \
        build-essential \
        git \
        ca-certificates \
@@ -1288,7 +1325,8 @@ RUN set -e \
        automake \
        libevent-dev \
        libtool \
-        pkg-config
+        pkg-config \
+    && apt clean && rm -rf /var/lib/apt/lists/*

 # Use `dist_man_MANS=` to skip manpage generation (which requires python3/pandoc)
 ENV PGBOUNCER_TAG=pgbouncer_1_22_1
@@ -1300,20 +1338,6 @@ RUN set -e \
    && make -j $(nproc) dist_man_MANS= \
    && make install dist_man_MANS=

-#########################################################################################
-#
-# Compile the Neon-specific `local_proxy` binary
-#
-#########################################################################################
-FROM $REPOSITORY/$IMAGE:$TAG AS local_proxy
-ARG BUILD_TAG
-ENV BUILD_TAG=$BUILD_TAG
-
-USER nonroot
-# Copy entire project to get Cargo.* files with proper dependencies for the whole project
-COPY --chown=nonroot . .
-RUN mold -run cargo build --locked --profile release-line-debug-size-lto --bin local_proxy
-
 #########################################################################################
 #
 # Layers "postgres-exporter" and "sql-exporter"
@@ -1453,7 +1477,7 @@ COPY --from=pgbouncer         /usr/local/pgbouncer/bin/pgbouncer /usr/local/bin/
 COPY --chmod=0666 --chown=postgres compute/etc/pgbouncer.ini /etc/pgbouncer.ini

 # local_proxy and its config
-COPY --from=local_proxy --chown=postgres /home/nonroot/target/release-line-debug-size-lto/local_proxy /usr/local/bin/local_proxy
+COPY --from=compute-tools --chown=postgres /home/nonroot/target/release-line-debug-size-lto/local_proxy /usr/local/bin/local_proxy
 RUN mkdir -p /etc/local_proxy && chown postgres:postgres /etc/local_proxy

 # Metrics exporter binaries and  configuration files
@@ -1518,28 +1542,30 @@ RUN apt update && \
        locales \
        procps \
        ca-certificates \
+        curl \
+        unzip \
        $VERSION_INSTALLS && \
-    rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/* && \
+    apt clean && 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

-# s5cmd 2.2.2 from https://github.com/peak/s5cmd/releases/tag/v2.2.2
-# used by fast_import
+# aws cli is used by fast_import (curl and unzip above are at this time only used for this installation step)
 ARG TARGETARCH
-ADD https://github.com/peak/s5cmd/releases/download/v2.2.2/s5cmd_2.2.2_linux_$TARGETARCH.deb /tmp/s5cmd.deb
 RUN set -ex; \
-    \
-    # Determine the expected checksum based on TARGETARCH
    if [ "${TARGETARCH}" = "amd64" ]; then \
-        CHECKSUM="392c385320cd5ffa435759a95af77c215553d967e4b1c0fffe52e4f14c29cf85"; \
+        TARGETARCH_ALT="x86_64"; \
+        CHECKSUM="c9a9df3770a3ff9259cb469b6179e02829687a464e0824d5c32d378820b53a00"; \
    elif [ "${TARGETARCH}" = "arm64" ]; then \
-        CHECKSUM="939bee3cf4b5604ddb00e67f8c157b91d7c7a5b553d1fbb6890fad32894b7b46"; \
+        TARGETARCH_ALT="aarch64"; \
+        CHECKSUM="8181730be7891582b38b028112e81b4899ca817e8c616aad807c9e9d1289223a"; \
    else \
        echo "Unsupported architecture: ${TARGETARCH}"; exit 1; \
    fi; \
-    \
-    # Compute and validate the checksum
-    echo "${CHECKSUM}  /tmp/s5cmd.deb" | sha256sum -c -
-RUN dpkg -i /tmp/s5cmd.deb && rm /tmp/s5cmd.deb
+    curl -L "https://awscli.amazonaws.com/awscli-exe-linux-${TARGETARCH_ALT}-2.17.5.zip" -o /tmp/awscliv2.zip; \
+    echo "${CHECKSUM}  /tmp/awscliv2.zip" | sha256sum -c -; \
+    unzip /tmp/awscliv2.zip -d /tmp/awscliv2; \
+    /tmp/awscliv2/aws/install; \
+    rm -rf /tmp/awscliv2.zip /tmp/awscliv2; \
+    true

 ENV LANG=en_US.utf8
 USER postgres
--- a/compute/etc/neon_collector.jsonnet
+++ b/compute/etc/neon_collector.jsonnet
@@ -3,7 +3,7 @@
  metrics: [
    import 'sql_exporter/checkpoints_req.libsonnet',
    import 'sql_exporter/checkpoints_timed.libsonnet',
-    import 'sql_exporter/compute_backpressure_throttling_seconds.libsonnet',
+    import 'sql_exporter/compute_backpressure_throttling_seconds_total.libsonnet',
    import 'sql_exporter/compute_current_lsn.libsonnet',
    import 'sql_exporter/compute_logical_snapshot_files.libsonnet',
    import 'sql_exporter/compute_logical_snapshots_bytes.libsonnet',
--- a/compute/etc/sql_exporter/compute_backpressure_throttling_seconds_total.libsonnet
+++ b/compute/etc/sql_exporter/compute_backpressure_throttling_seconds_total.libsonnet
@@ -1,10 +1,10 @@
 {
-  metric_name: 'compute_backpressure_throttling_seconds',
-  type: 'gauge',
+  metric_name: 'compute_backpressure_throttling_seconds_total',
+  type: 'counter',
  help: 'Time compute has spent throttled',
  key_labels: null,
  values: [
    'throttled',
  ],
-  query: importstr 'sql_exporter/compute_backpressure_throttling_seconds.sql',
+  query: importstr 'sql_exporter/compute_backpressure_throttling_seconds_total.sql',
 }
--- a/compute/etc/sql_exporter/compute_backpressure_throttling_seconds_total.sql
+++ b/compute/etc/sql_exporter/compute_backpressure_throttling_seconds_total.sql
--- a/compute_tools/Cargo.toml
+++ b/compute_tools/Cargo.toml
@@ -7,7 +7,7 @@ license.workspace = true
 [features]
 default = []
 # Enables test specific features.
-testing = []
+testing = ["fail/failpoints"]

 [dependencies]
 base64.workspace = true
@@ -19,6 +19,7 @@ camino.workspace = true
 chrono.workspace = true
 cfg-if.workspace = true
 clap.workspace = true
+fail.workspace = true
 flate2.workspace = true
 futures.workspace = true
 hyper0 = { workspace = true, features = ["full"] }
--- a/compute_tools/src/bin/compute_ctl.rs
+++ b/compute_tools/src/bin/compute_ctl.rs
@@ -67,12 +67,15 @@ use compute_tools::params::*;
 use compute_tools::spec::*;
 use compute_tools::swap::resize_swap;
 use rlimit::{setrlimit, Resource};
+use utils::failpoint_support;

 // this is an arbitrary build tag. Fine as a default / for testing purposes
 // in-case of not-set environment var
 const BUILD_TAG_DEFAULT: &str = "latest";

 fn main() -> Result<()> {
+    let scenario = failpoint_support::init();
+
    let (build_tag, clap_args) = init()?;

    // enable core dumping for all child processes
@@ -100,6 +103,8 @@ fn main() -> Result<()> {

    maybe_delay_exit(delay_exit);

+    scenario.teardown();
+
    deinit_and_exit(wait_pg_result);
 }

@@ -419,9 +424,13 @@ fn start_postgres(
        "running compute with features: {:?}",
        state.pspec.as_ref().unwrap().spec.features
    );
-    // before we release the mutex, fetch the swap size (if any) for later.
-    let swap_size_bytes = state.pspec.as_ref().unwrap().spec.swap_size_bytes;
-    let disk_quota_bytes = state.pspec.as_ref().unwrap().spec.disk_quota_bytes;
+    // before we release the mutex, fetch some parameters for later.
+    let &ComputeSpec {
+        swap_size_bytes,
+        disk_quota_bytes,
+        disable_lfc_resizing,
+        ..
+    } = &state.pspec.as_ref().unwrap().spec;
    drop(state);

    // Launch remaining service threads
@@ -526,11 +535,18 @@ fn start_postgres(
            // This token is used internally by the monitor to clean up all threads
            let token = CancellationToken::new();

+            // don't pass postgres connection string to vm-monitor if we don't want it to resize LFC
+            let pgconnstr = if disable_lfc_resizing.unwrap_or(false) {
+                None
+            } else {
+                file_cache_connstr.cloned()
+            };
+
            let vm_monitor = rt.as_ref().map(|rt| {
                rt.spawn(vm_monitor::start(
                    Box::leak(Box::new(vm_monitor::Args {
                        cgroup: cgroup.cloned(),
-                        pgconnstr: file_cache_connstr.cloned(),
+                        pgconnstr,
                        addr: vm_monitor_addr.clone(),
                    })),
                    token.clone(),
--- a/compute_tools/src/bin/fast_import.rs
+++ b/compute_tools/src/bin/fast_import.rs
@@ -34,12 +34,12 @@ use nix::unistd::Pid;
 use tracing::{info, info_span, warn, Instrument};
 use utils::fs_ext::is_directory_empty;

+#[path = "fast_import/aws_s3_sync.rs"]
+mod aws_s3_sync;
 #[path = "fast_import/child_stdio_to_log.rs"]
 mod child_stdio_to_log;
 #[path = "fast_import/s3_uri.rs"]
 mod s3_uri;
-#[path = "fast_import/s5cmd.rs"]
-mod s5cmd;

 #[derive(clap::Parser)]
 struct Args {
@@ -326,7 +326,7 @@ pub(crate) async fn main() -> anyhow::Result<()> {
    }

    info!("upload pgdata");
-    s5cmd::sync(Utf8Path::new(&pgdata_dir), &s3_prefix.append("/"))
+    aws_s3_sync::sync(Utf8Path::new(&pgdata_dir), &s3_prefix.append("/pgdata/"))
        .await
        .context("sync dump directory to destination")?;

@@ -334,10 +334,10 @@ pub(crate) async fn main() -> anyhow::Result<()> {
    {
        let status_dir = working_directory.join("status");
        std::fs::create_dir(&status_dir).context("create status directory")?;
-        let status_file = status_dir.join("status");
+        let status_file = status_dir.join("pgdata");
        std::fs::write(&status_file, serde_json::json!({"done": true}).to_string())
            .context("write status file")?;
-        s5cmd::sync(&status_file, &s3_prefix.append("/status/pgdata"))
+        aws_s3_sync::sync(&status_dir, &s3_prefix.append("/status/"))
            .await
            .context("sync status directory to destination")?;
    }
--- a/compute_tools/src/bin/fast_import/aws_s3_sync.rs
+++ b/compute_tools/src/bin/fast_import/aws_s3_sync.rs
@@ -4,24 +4,21 @@ use camino::Utf8Path;
 use super::s3_uri::S3Uri;

 pub(crate) async fn sync(local: &Utf8Path, remote: &S3Uri) -> anyhow::Result<()> {
-    let mut builder = tokio::process::Command::new("s5cmd");
-    // s5cmd uses aws-sdk-go v1, hence doesn't support AWS_ENDPOINT_URL
-    if let Some(val) = std::env::var_os("AWS_ENDPOINT_URL") {
-        builder.arg("--endpoint-url").arg(val);
-    }
+    let mut builder = tokio::process::Command::new("aws");
    builder
+        .arg("s3")
        .arg("sync")
        .arg(local.as_str())
        .arg(remote.to_string());
    let st = builder
        .spawn()
-        .context("spawn s5cmd")?
+        .context("spawn aws s3 sync")?
        .wait()
        .await
-        .context("wait for s5cmd")?;
+        .context("wait for aws s3 sync")?;
    if st.success() {
        Ok(())
    } else {
-        Err(anyhow::anyhow!("s5cmd failed"))
+        Err(anyhow::anyhow!("aws s3 sync failed"))
    }
 }
--- a/compute_tools/src/compute.rs
+++ b/compute_tools/src/compute.rs
@@ -1181,8 +1181,19 @@ impl ComputeNode {
            let mut conf = postgres::config::Config::from(conf);
            conf.application_name("compute_ctl:migrations");

-            let mut client = conf.connect(NoTls)?;
-            handle_migrations(&mut client).context("apply_config handle_migrations")
+            match conf.connect(NoTls) {
+                Ok(mut client) => {
+                    if let Err(e) = handle_migrations(&mut client) {
+                        error!("Failed to run migrations: {}", e);
+                    }
+                }
+                Err(e) => {
+                    error!(
+                        "Failed to connect to the compute for running migrations: {}",
+                        e
+                    );
+                }
+            };
        });

        Ok::<(), anyhow::Error>(())
--- a/compute_tools/src/http/api.rs
+++ b/compute_tools/src/http/api.rs
@@ -24,8 +24,11 @@ use metrics::proto::MetricFamily;
 use metrics::Encoder;
 use metrics::TextEncoder;
 use tokio::task;
+use tokio_util::sync::CancellationToken;
 use tracing::{debug, error, info, warn};
 use tracing_utils::http::OtelName;
+use utils::failpoint_support::failpoints_handler;
+use utils::http::error::ApiError;
 use utils::http::request::must_get_query_param;

 fn status_response_from_state(state: &ComputeState) -> ComputeStatusResponse {
@@ -310,6 +313,18 @@ async fn routes(req: Request<Body>, compute: &Arc<ComputeNode>) -> Response<Body
            }
        }

+        (&Method::POST, "/failpoints") if cfg!(feature = "testing") => {
+            match failpoints_handler(req, CancellationToken::new()).await {
+                Ok(r) => r,
+                Err(ApiError::BadRequest(e)) => {
+                    render_json_error(&e.to_string(), StatusCode::BAD_REQUEST)
+                }
+                Err(_) => {
+                    render_json_error("Internal server error", StatusCode::INTERNAL_SERVER_ERROR)
+                }
+            }
+        }
+
        // download extension files from remote extension storage on demand
        (&Method::POST, route) if route.starts_with("/extension_server/") => {
            info!("serving {:?} POST request", route);
--- a/compute_tools/src/migration.rs
+++ b/compute_tools/src/migration.rs
@@ -1,13 +1,16 @@
 use anyhow::{Context, Result};
+use fail::fail_point;
 use postgres::Client;
 use tracing::info;

+/// Runs a series of migrations on a target database
 pub(crate) struct MigrationRunner<'m> {
    client: &'m mut Client,
    migrations: &'m [&'m str],
 }

 impl<'m> MigrationRunner<'m> {
+    /// Create a new migration runner
    pub fn new(client: &'m mut Client, migrations: &'m [&'m str]) -> Self {
        // The neon_migration.migration_id::id column is a bigint, which is equivalent to an i64
        assert!(migrations.len() + 1 < i64::MAX as usize);
@@ -15,6 +18,7 @@ impl<'m> MigrationRunner<'m> {
        Self { client, migrations }
    }

+    /// Get the current value neon_migration.migration_id
    fn get_migration_id(&mut self) -> Result<i64> {
        let query = "SELECT id FROM neon_migration.migration_id";
        let row = self
@@ -25,37 +29,61 @@ impl<'m> MigrationRunner<'m> {
        Ok(row.get::<&str, i64>("id"))
    }

+    /// Update the neon_migration.migration_id value
+    ///
+    /// This function has a fail point called compute-migration, which can be
+    /// used if you would like to fail the application of a series of migrations
+    /// at some point.
    fn update_migration_id(&mut self, migration_id: i64) -> Result<()> {
-        let setval = format!("UPDATE neon_migration.migration_id SET id={}", migration_id);
+        // We use this fail point in order to check that failing in the
+        // middle of applying a series of migrations fails in an expected
+        // manner
+        if cfg!(feature = "testing") {
+            let fail = (|| {
+                fail_point!("compute-migration", |fail_migration_id| {
+                    migration_id == fail_migration_id.unwrap().parse::<i64>().unwrap()
+                });
+
+                false
+            })();
+
+            if fail {
+                return Err(anyhow::anyhow!(format!(
+                    "migration {} was configured to fail because of a failpoint",
+                    migration_id
+                )));
+            }
+        }

        self.client
-            .simple_query(&setval)
+            .query(
+                "UPDATE neon_migration.migration_id SET id = $1",
+                &[&migration_id],
+            )
            .context("run_migrations update id")?;

        Ok(())
    }

-    fn prepare_migrations(&mut self) -> Result<()> {
-        let query = "CREATE SCHEMA IF NOT EXISTS neon_migration";
-        self.client.simple_query(query)?;
-
-        let query = "CREATE TABLE IF NOT EXISTS neon_migration.migration_id (key INT NOT NULL PRIMARY KEY, id bigint NOT NULL DEFAULT 0)";
-        self.client.simple_query(query)?;
-
-        let query = "INSERT INTO neon_migration.migration_id VALUES (0, 0) ON CONFLICT DO NOTHING";
-        self.client.simple_query(query)?;
-
-        let query = "ALTER SCHEMA neon_migration OWNER TO cloud_admin";
-        self.client.simple_query(query)?;
-
-        let query = "REVOKE ALL ON SCHEMA neon_migration FROM PUBLIC";
-        self.client.simple_query(query)?;
+    /// Prepare the migrations the target database for handling migrations
+    fn prepare_database(&mut self) -> Result<()> {
+        self.client
+            .simple_query("CREATE SCHEMA IF NOT EXISTS neon_migration")?;
+        self.client.simple_query("CREATE TABLE IF NOT EXISTS neon_migration.migration_id (key INT NOT NULL PRIMARY KEY, id bigint NOT NULL DEFAULT 0)")?;
+        self.client.simple_query(
+            "INSERT INTO neon_migration.migration_id VALUES (0, 0) ON CONFLICT DO NOTHING",
+        )?;
+        self.client
+            .simple_query("ALTER SCHEMA neon_migration OWNER TO cloud_admin")?;
+        self.client
+            .simple_query("REVOKE ALL ON SCHEMA neon_migration FROM PUBLIC")?;

        Ok(())
    }

+    /// Run the configrured set of migrations
    pub fn run_migrations(mut self) -> Result<()> {
-        self.prepare_migrations()?;
+        self.prepare_database()?;

        let mut current_migration = self.get_migration_id()? as usize;
        while current_migration < self.migrations.len() {
@@ -69,6 +97,11 @@ impl<'m> MigrationRunner<'m> {

            if migration.starts_with("-- SKIP") {
                info!("Skipping migration id={}", migration_id!(current_migration));
+
+                // Even though we are skipping the migration, updating the
+                // migration ID should help keep logic easy to understand when
+                // trying to understand the state of a cluster.
+                self.update_migration_id(migration_id!(current_migration))?;
            } else {
                info!(
                    "Running migration id={}:\n{}\n",
@@ -87,7 +120,6 @@ impl<'m> MigrationRunner<'m> {
                    )
                })?;

-                // Migration IDs start at 1
                self.update_migration_id(migration_id!(current_migration))?;

                self.client
--- a/compute_tools/src/migrations/tests/0001-neon_superuser_bypass_rls.sql
+++ b/compute_tools/src/migrations/tests/0001-neon_superuser_bypass_rls.sql
@@ -0,0 +1,9 @@
+DO $$
+DECLARE
+    bypassrls boolean;
+BEGIN
+    SELECT rolbypassrls INTO bypassrls FROM pg_roles WHERE rolname = 'neon_superuser';
+    IF NOT bypassrls THEN
+        RAISE EXCEPTION 'neon_superuser cannot bypass RLS';
+    END IF;
+END $$;
--- a/compute_tools/src/migrations/tests/0002-alter_roles.sql
+++ b/compute_tools/src/migrations/tests/0002-alter_roles.sql
@@ -0,0 +1,25 @@
+DO $$
+DECLARE
+    role record;
+BEGIN
+    FOR role IN
+        SELECT rolname AS name, rolinherit AS inherit
+        FROM pg_roles
+        WHERE pg_has_role(rolname, 'neon_superuser', 'member')
+    LOOP
+        IF NOT role.inherit THEN
+            RAISE EXCEPTION '% cannot inherit', quote_ident(role.name);
+        END IF;
+    END LOOP;
+
+    FOR role IN
+        SELECT rolname AS name, rolbypassrls AS bypassrls
+        FROM pg_roles
+        WHERE NOT pg_has_role(rolname, 'neon_superuser', 'member')
+            AND NOT starts_with(rolname, 'pg_')
+    LOOP
+        IF role.bypassrls THEN
+            RAISE EXCEPTION  '% can bypass RLS', quote_ident(role.name);
+        END IF;
+    END LOOP;
+END $$;
--- a/compute_tools/src/migrations/tests/0003-grant_pg_create_subscription_to_neon_superuser.sql
+++ b/compute_tools/src/migrations/tests/0003-grant_pg_create_subscription_to_neon_superuser.sql
@@ -0,0 +1,10 @@
+DO $$
+BEGIN
+    IF (SELECT current_setting('server_version_num')::numeric < 160000) THEN
+        RETURN;
+    END IF;
+
+    IF NOT (SELECT pg_has_role('neon_superuser', 'pg_create_subscription', 'member')) THEN
+        RAISE EXCEPTION 'neon_superuser cannot execute pg_create_subscription';
+    END IF;
+END $$;
--- a/compute_tools/src/migrations/tests/0004-grant_pg_monitor_to_neon_superuser.sql
+++ b/compute_tools/src/migrations/tests/0004-grant_pg_monitor_to_neon_superuser.sql
@@ -0,0 +1,19 @@
+DO $$
+DECLARE
+    monitor record;
+BEGIN
+    SELECT pg_has_role('neon_superuser', 'pg_monitor', 'member') AS member,
+            admin_option AS admin
+        INTO monitor
+        FROM pg_auth_members
+        WHERE roleid = 'pg_monitor'::regrole
+            AND member = 'pg_monitor'::regrole;
+
+    IF NOT monitor.member THEN
+        RAISE EXCEPTION 'neon_superuser is not a member of pg_monitor';
+    END IF;
+
+    IF NOT monitor.admin THEN
+        RAISE EXCEPTION 'neon_superuser cannot grant pg_monitor';
+    END IF;
+END $$;
--- a/compute_tools/src/migrations/tests/0005-grant_all_on_tables_to_neon_superuser.sql
+++ b/compute_tools/src/migrations/tests/0005-grant_all_on_tables_to_neon_superuser.sql
@@ -0,0 +1,2 @@
+-- This test was never written becuase at the time migration tests were added
+-- the accompanying migration was already skipped.
--- a/compute_tools/src/migrations/tests/0006-grant_all_on_sequences_to_neon_superuser.sql
+++ b/compute_tools/src/migrations/tests/0006-grant_all_on_sequences_to_neon_superuser.sql
@@ -0,0 +1,2 @@
+-- This test was never written becuase at the time migration tests were added
+-- the accompanying migration was already skipped.
--- a/compute_tools/src/migrations/tests/0007-grant_all_on_tables_to_neon_superuser_with_grant_option.sql
+++ b/compute_tools/src/migrations/tests/0007-grant_all_on_tables_to_neon_superuser_with_grant_option.sql
@@ -0,0 +1,2 @@
+-- This test was never written becuase at the time migration tests were added
+-- the accompanying migration was already skipped.
--- a/compute_tools/src/migrations/tests/0008-grant_all_on_sequences_to_neon_superuser_with_grant_option.sql
+++ b/compute_tools/src/migrations/tests/0008-grant_all_on_sequences_to_neon_superuser_with_grant_option.sql
@@ -0,0 +1,2 @@
+-- This test was never written becuase at the time migration tests were added
+-- the accompanying migration was already skipped.
--- a/compute_tools/src/migrations/tests/0009-revoke_replication_for_previously_allowed_roles.sql
+++ b/compute_tools/src/migrations/tests/0009-revoke_replication_for_previously_allowed_roles.sql
@@ -0,0 +1,2 @@
+-- This test was never written becuase at the time migration tests were added
+-- the accompanying migration was already skipped.
--- a/compute_tools/src/migrations/tests/0010-grant_snapshot_synchronization_funcs_to_neon_superuser.sql
+++ b/compute_tools/src/migrations/tests/0010-grant_snapshot_synchronization_funcs_to_neon_superuser.sql
@@ -0,0 +1,13 @@
+DO $$
+DECLARE
+    can_execute boolean;
+BEGIN
+    SELECT bool_and(has_function_privilege('neon_superuser', oid, 'execute'))
+       INTO can_execute
+       FROM pg_proc
+       WHERE proname IN ('pg_export_snapshot', 'pg_log_standby_snapshot')
+           AND pronamespace = 'pg_catalog'::regnamespace;
+    IF NOT can_execute THEN
+        RAISE EXCEPTION 'neon_superuser cannot execute both pg_export_snapshot and pg_log_standby_snapshot';
+    END IF;
+END $$;
--- a/compute_tools/src/migrations/tests/0011-grant_pg_show_replication_origin_status_to_neon_superuser.sql
+++ b/compute_tools/src/migrations/tests/0011-grant_pg_show_replication_origin_status_to_neon_superuser.sql
@@ -0,0 +1,13 @@
+DO $$
+DECLARE
+    can_execute boolean;
+BEGIN
+    SELECT has_function_privilege('neon_superuser', oid, 'execute')
+       INTO can_execute
+       FROM pg_proc
+       WHERE proname = 'pg_show_replication_origin_status'
+           AND pronamespace = 'pg_catalog'::regnamespace;
+    IF NOT can_execute THEN
+        RAISE EXCEPTION 'neon_superuser cannot execute pg_show_replication_origin_status';
+    END IF;
+END $$;
--- a/control_plane/src/bin/neon_local.rs
+++ b/control_plane/src/bin/neon_local.rs
@@ -19,6 +19,7 @@ use control_plane::storage_controller::{
    NeonStorageControllerStartArgs, NeonStorageControllerStopArgs, StorageController,
 };
 use control_plane::{broker, local_env};
+use nix::fcntl::{flock, FlockArg};
 use pageserver_api::config::{
    DEFAULT_HTTP_LISTEN_PORT as DEFAULT_PAGESERVER_HTTP_PORT,
    DEFAULT_PG_LISTEN_PORT as DEFAULT_PAGESERVER_PG_PORT,
@@ -36,6 +37,8 @@ use safekeeper_api::{
 };
 use std::borrow::Cow;
 use std::collections::{BTreeSet, HashMap};
+use std::fs::File;
+use std::os::fd::AsRawFd;
 use std::path::PathBuf;
 use std::process::exit;
 use std::str::FromStr;
@@ -689,6 +692,21 @@ struct TimelineTreeEl {
    pub children: BTreeSet<TimelineId>,
 }

+/// A flock-based guard over the neon_local repository directory
+struct RepoLock {
+    _file: File,
+}
+
+impl RepoLock {
+    fn new() -> Result<Self> {
+        let repo_dir = File::open(local_env::base_path())?;
+        let repo_dir_fd = repo_dir.as_raw_fd();
+        flock(repo_dir_fd, FlockArg::LockExclusive)?;
+
+        Ok(Self { _file: repo_dir })
+    }
+}
+
 // Main entry point for the 'neon_local' CLI utility
 //
 // This utility helps to manage neon installation. That includes following:
@@ -700,9 +718,14 @@ fn main() -> Result<()> {
    let cli = Cli::parse();

    // Check for 'neon init' command first.
-    let subcommand_result = if let NeonLocalCmd::Init(args) = cli.command {
-        handle_init(&args).map(|env| Some(Cow::Owned(env)))
+    let (subcommand_result, _lock) = if let NeonLocalCmd::Init(args) = cli.command {
+        (handle_init(&args).map(|env| Some(Cow::Owned(env))), None)
    } else {
+        // This tool uses a collection of simple files to store its state, and consequently
+        // it is not generally safe to run multiple commands concurrently.  Rather than expect
+        // all callers to know this, use a lock file to protect against concurrent execution.
+        let _repo_lock = RepoLock::new().unwrap();
+
        // all other commands need an existing config
        let env = LocalEnv::load_config(&local_env::base_path()).context("Error loading config")?;
        let original_env = env.clone();
@@ -728,11 +751,12 @@ fn main() -> Result<()> {
            NeonLocalCmd::Mappings(subcmd) => handle_mappings(&subcmd, env),
        };

-        if &original_env != env {
+        let subcommand_result = if &original_env != env {
            subcommand_result.map(|()| Some(Cow::Borrowed(env)))
        } else {
            subcommand_result.map(|()| None)
-        }
+        };
+        (subcommand_result, Some(_repo_lock))
    };

    match subcommand_result {
@@ -922,7 +946,7 @@ fn handle_init(args: &InitCmdArgs) -> anyhow::Result<LocalEnv> {
    } else {
        // User (likely interactive) did not provide a description of the environment, give them the default
        NeonLocalInitConf {
-            control_plane_api: Some(Some(DEFAULT_PAGESERVER_CONTROL_PLANE_API.parse().unwrap())),
+            control_plane_api: Some(DEFAULT_PAGESERVER_CONTROL_PLANE_API.parse().unwrap()),
            broker: NeonBroker {
                listen_addr: DEFAULT_BROKER_ADDR.parse().unwrap(),
            },
@@ -1718,18 +1742,15 @@ async fn handle_start_all_impl(
            broker::start_broker_process(env, &retry_timeout).await
        });

-        // Only start the storage controller if the pageserver is configured to need it
-        if env.control_plane_api.is_some() {
-            js.spawn(async move {
-                let storage_controller = StorageController::from_env(env);
-                storage_controller
-                    .start(NeonStorageControllerStartArgs::with_default_instance_id(
-                        retry_timeout,
-                    ))
-                    .await
-                    .map_err(|e| e.context("start storage_controller"))
-            });
-        }
+        js.spawn(async move {
+            let storage_controller = StorageController::from_env(env);
+            storage_controller
+                .start(NeonStorageControllerStartArgs::with_default_instance_id(
+                    retry_timeout,
+                ))
+                .await
+                .map_err(|e| e.context("start storage_controller"))
+        });

        for ps_conf in &env.pageservers {
            js.spawn(async move {
@@ -1774,10 +1795,6 @@ async fn neon_start_status_check(
    const RETRY_INTERVAL: Duration = Duration::from_millis(100);
    const NOTICE_AFTER_RETRIES: Duration = Duration::from_secs(5);

-    if env.control_plane_api.is_none() {
-        return Ok(());
-    }
-
    let storcon = StorageController::from_env(env);

    let retries = retry_timeout.as_millis() / RETRY_INTERVAL.as_millis();
--- a/control_plane/src/endpoint.rs
+++ b/control_plane/src/endpoint.rs
@@ -316,6 +316,10 @@ impl Endpoint {
        // and can cause errors like 'no unpinned buffers available', see
        // <https://github.com/neondatabase/neon/issues/9956>
        conf.append("shared_buffers", "1MB");
+        // Postgres defaults to effective_io_concurrency=1, which does not exercise the pageserver's
+        // batching logic.  Set this to 2 so that we exercise the code a bit without letting
+        // individual tests do a lot of concurrent work on underpowered test machines
+        conf.append("effective_io_concurrency", "2");
        conf.append("fsync", "off");
        conf.append("max_connections", "100");
        conf.append("wal_level", "logical");
@@ -581,6 +585,7 @@ impl Endpoint {
            features: self.features.clone(),
            swap_size_bytes: None,
            disk_quota_bytes: None,
+            disable_lfc_resizing: None,
            cluster: Cluster {
                cluster_id: None, // project ID: not used
                name: None,       // project name: not used
--- a/control_plane/src/local_env.rs
+++ b/control_plane/src/local_env.rs
@@ -76,7 +76,7 @@ pub struct LocalEnv {

    // Control plane upcall API for pageserver: if None, we will not run storage_controller  If set, this will
    // be propagated into each pageserver's configuration.
-    pub control_plane_api: Option<Url>,
+    pub control_plane_api: Url,

    // Control plane upcall API for storage controller.  If set, this will be propagated into the
    // storage controller's configuration.
@@ -133,7 +133,7 @@ pub struct NeonLocalInitConf {
    pub storage_controller: Option<NeonStorageControllerConf>,
    pub pageservers: Vec<NeonLocalInitPageserverConf>,
    pub safekeepers: Vec<SafekeeperConf>,
-    pub control_plane_api: Option<Option<Url>>,
+    pub control_plane_api: Option<Url>,
    pub control_plane_compute_hook_api: Option<Option<Url>>,
 }

@@ -180,7 +180,7 @@ impl NeonStorageControllerConf {
    const DEFAULT_MAX_WARMING_UP_INTERVAL: std::time::Duration = std::time::Duration::from_secs(30);

    // Very tight heartbeat interval to speed up tests
-    const DEFAULT_HEARTBEAT_INTERVAL: std::time::Duration = std::time::Duration::from_millis(100);
+    const DEFAULT_HEARTBEAT_INTERVAL: std::time::Duration = std::time::Duration::from_millis(1000);
 }

 impl Default for NeonStorageControllerConf {
@@ -535,7 +535,7 @@ impl LocalEnv {
                storage_controller,
                pageservers,
                safekeepers,
-                control_plane_api,
+                control_plane_api: control_plane_api.unwrap(),
                control_plane_compute_hook_api,
                branch_name_mappings,
            }
@@ -638,7 +638,7 @@ impl LocalEnv {
                storage_controller: self.storage_controller.clone(),
                pageservers: vec![], // it's skip_serializing anyway
                safekeepers: self.safekeepers.clone(),
-                control_plane_api: self.control_plane_api.clone(),
+                control_plane_api: Some(self.control_plane_api.clone()),
                control_plane_compute_hook_api: self.control_plane_compute_hook_api.clone(),
                branch_name_mappings: self.branch_name_mappings.clone(),
            },
@@ -768,7 +768,7 @@ impl LocalEnv {
            storage_controller: storage_controller.unwrap_or_default(),
            pageservers: pageservers.iter().map(Into::into).collect(),
            safekeepers,
-            control_plane_api: control_plane_api.unwrap_or_default(),
+            control_plane_api: control_plane_api.unwrap(),
            control_plane_compute_hook_api: control_plane_compute_hook_api.unwrap_or_default(),
            branch_name_mappings: Default::default(),
        };
--- a/control_plane/src/pageserver.rs
+++ b/control_plane/src/pageserver.rs
@@ -95,21 +95,19 @@ impl PageServerNode {

        let mut overrides = vec![pg_distrib_dir_param, broker_endpoint_param];

-        if let Some(control_plane_api) = &self.env.control_plane_api {
-            overrides.push(format!(
-                "control_plane_api='{}'",
-                control_plane_api.as_str()
-            ));
+        overrides.push(format!(
+            "control_plane_api='{}'",
+            self.env.control_plane_api.as_str()
+        ));

-            // Storage controller uses the same auth as pageserver: if JWT is enabled
-            // for us, we will also need it to talk to them.
-            if matches!(conf.http_auth_type, AuthType::NeonJWT) {
-                let jwt_token = self
-                    .env
-                    .generate_auth_token(&Claims::new(None, Scope::GenerationsApi))
-                    .unwrap();
-                overrides.push(format!("control_plane_api_token='{}'", jwt_token));
-            }
+        // Storage controller uses the same auth as pageserver: if JWT is enabled
+        // for us, we will also need it to talk to them.
+        if matches!(conf.http_auth_type, AuthType::NeonJWT) {
+            let jwt_token = self
+                .env
+                .generate_auth_token(&Claims::new(None, Scope::GenerationsApi))
+                .unwrap();
+            overrides.push(format!("control_plane_api_token='{}'", jwt_token));
        }

        if !conf.other.contains_key("remote_storage") {
--- a/control_plane/src/storage_controller.rs
+++ b/control_plane/src/storage_controller.rs
@@ -338,7 +338,7 @@ impl StorageController {
                        .port(),
                )
            } else {
-                let listen_url = self.env.control_plane_api.clone().unwrap();
+                let listen_url = self.env.control_plane_api.clone();

                let listen = format!(
                    "{}:{}",
@@ -708,7 +708,7 @@ impl StorageController {
        } else {
            // The configured URL has the /upcall path prefix for pageservers to use: we will strip that out
            // for general purpose API access.
-            let listen_url = self.env.control_plane_api.clone().unwrap();
+            let listen_url = self.env.control_plane_api.clone();
            Url::from_str(&format!(
                "http://{}:{}/{path}",
                listen_url.host_str().unwrap(),
--- a/control_plane/storcon_cli/src/main.rs
+++ b/control_plane/storcon_cli/src/main.rs
@@ -5,7 +5,8 @@ use clap::{Parser, Subcommand};
 use pageserver_api::{
    controller_api::{
        AvailabilityZone, NodeAvailabilityWrapper, NodeDescribeResponse, NodeShardResponse,
-        ShardSchedulingPolicy, TenantCreateRequest, TenantDescribeResponse, TenantPolicyRequest,
+        SafekeeperDescribeResponse, ShardSchedulingPolicy, TenantCreateRequest,
+        TenantDescribeResponse, TenantPolicyRequest,
    },
    models::{
        EvictionPolicy, EvictionPolicyLayerAccessThreshold, LocationConfigSecondary,
@@ -211,6 +212,8 @@ enum Command {
        #[arg(long)]
        timeout: humantime::Duration,
    },
+    /// List safekeepers known to the storage controller
+    Safekeepers {},
 }

 #[derive(Parser)]
@@ -1020,6 +1023,31 @@ async fn main() -> anyhow::Result<()> {
                "Fill was cancelled for node {node_id}. Schedulling policy is now {final_policy:?}"
            );
        }
+        Command::Safekeepers {} => {
+            let mut resp = storcon_client
+                .dispatch::<(), Vec<SafekeeperDescribeResponse>>(
+                    Method::GET,
+                    "control/v1/safekeeper".to_string(),
+                    None,
+                )
+                .await?;
+
+            resp.sort_by(|a, b| a.id.cmp(&b.id));
+
+            let mut table = comfy_table::Table::new();
+            table.set_header(["Id", "Version", "Host", "Port", "Http Port", "AZ Id"]);
+            for sk in resp {
+                table.add_row([
+                    format!("{}", sk.id),
+                    format!("{}", sk.version),
+                    sk.host,
+                    format!("{}", sk.port),
+                    format!("{}", sk.http_port),
+                    sk.availability_zone_id.to_string(),
+                ]);
+            }
+            println!("{table}");
+        }
    }

    Ok(())
--- a/libs/compute_api/src/spec.rs
+++ b/libs/compute_api/src/spec.rs
@@ -67,6 +67,15 @@ pub struct ComputeSpec {
    #[serde(default)]
    pub disk_quota_bytes: Option<u64>,

+    /// Disables the vm-monitor behavior that resizes LFC on upscale/downscale, instead relying on
+    /// the initial size of LFC.
+    ///
+    /// This is intended for use when the LFC size is being overridden from the default but
+    /// autoscaling is still enabled, and we don't want the vm-monitor to interfere with the custom
+    /// LFC sizing.
+    #[serde(default)]
+    pub disable_lfc_resizing: Option<bool>,
+
    /// Expected cluster state at the end of transition process.
    pub cluster: Cluster,
    pub delta_operations: Option<Vec<DeltaOp>>,
--- a/libs/pageserver_api/src/controller_api.rs
+++ b/libs/pageserver_api/src/controller_api.rs
@@ -372,6 +372,23 @@ pub struct MetadataHealthListOutdatedResponse {
    pub health_records: Vec<MetadataHealthRecord>,
 }

+/// Publicly exposed safekeeper description
+///
+/// The `active` flag which we have in the DB is not included on purpose: it is deprecated.
+#[derive(Serialize, Deserialize, Clone)]
+pub struct SafekeeperDescribeResponse {
+    pub id: NodeId,
+    pub region_id: String,
+    /// 1 is special, it means just created (not currently posted to storcon).
+    /// Zero or negative is not really expected.
+    /// Otherwise the number from `release-$(number_of_commits_on_branch)` tag.
+    pub version: i64,
+    pub host: String,
+    pub port: i32,
+    pub http_port: i32,
+    pub availability_zone_id: String,
+}
+
 #[cfg(test)]
 mod test {
    use super::*;
--- a/libs/pageserver_api/src/models.rs
+++ b/libs/pageserver_api/src/models.rs
@@ -6,6 +6,7 @@ pub mod utilization;
 use camino::Utf8PathBuf;
 pub use utilization::PageserverUtilization;

+use core::ops::Range;
 use std::{
    collections::HashMap,
    fmt::Display,
@@ -28,6 +29,7 @@ use utils::{
 };

 use crate::{
+    key::Key,
    reltag::RelTag,
    shard::{ShardCount, ShardStripeSize, TenantShardId},
 };
@@ -210,6 +212,68 @@ pub enum TimelineState {
    Broken { reason: String, backtrace: String },
 }

+#[serde_with::serde_as]
+#[derive(Debug, Clone, serde::Deserialize, serde::Serialize)]
+pub struct CompactLsnRange {
+    pub start: Lsn,
+    pub end: Lsn,
+}
+
+#[serde_with::serde_as]
+#[derive(Debug, Clone, serde::Deserialize, serde::Serialize)]
+pub struct CompactKeyRange {
+    #[serde_as(as = "serde_with::DisplayFromStr")]
+    pub start: Key,
+    #[serde_as(as = "serde_with::DisplayFromStr")]
+    pub end: Key,
+}
+
+impl From<Range<Lsn>> for CompactLsnRange {
+    fn from(range: Range<Lsn>) -> Self {
+        Self {
+            start: range.start,
+            end: range.end,
+        }
+    }
+}
+
+impl From<Range<Key>> for CompactKeyRange {
+    fn from(range: Range<Key>) -> Self {
+        Self {
+            start: range.start,
+            end: range.end,
+        }
+    }
+}
+
+impl From<CompactLsnRange> for Range<Lsn> {
+    fn from(range: CompactLsnRange) -> Self {
+        range.start..range.end
+    }
+}
+
+impl From<CompactKeyRange> for Range<Key> {
+    fn from(range: CompactKeyRange) -> Self {
+        range.start..range.end
+    }
+}
+
+impl CompactLsnRange {
+    pub fn above(lsn: Lsn) -> Self {
+        Self {
+            start: lsn,
+            end: Lsn::MAX,
+        }
+    }
+}
+
+#[derive(Debug, Clone, Serialize)]
+pub struct CompactInfoResponse {
+    pub compact_key_range: Option<CompactKeyRange>,
+    pub compact_lsn_range: Option<CompactLsnRange>,
+    pub sub_compaction: bool,
+}
+
 #[derive(Serialize, Deserialize, Clone)]
 pub struct TimelineCreateRequest {
    pub new_timeline_id: TimelineId,
--- a/libs/postgres_ffi/src/wal_generator.rs
+++ b/libs/postgres_ffi/src/wal_generator.rs
@@ -106,11 +106,11 @@ impl<R: RecordGenerator> WalGenerator<R> {
    const TIMELINE_ID: u32 = 1;

    /// Creates a new WAL generator with the given record generator.
-    pub fn new(record_generator: R) -> WalGenerator<R> {
+    pub fn new(record_generator: R, start_lsn: Lsn) -> WalGenerator<R> {
        Self {
            record_generator,
-            lsn: Lsn(0),
-            prev_lsn: Lsn(0),
+            lsn: start_lsn,
+            prev_lsn: start_lsn,
        }
    }

--- a/libs/proxy/postgres-protocol2/Cargo.toml
+++ b/libs/proxy/postgres-protocol2/Cargo.toml
@@ -1,7 +1,7 @@
 [package]
 name = "postgres-protocol2"
 version = "0.1.0"
-edition = "2018"
+edition = "2021"
 license = "MIT/Apache-2.0"

 [dependencies]
--- a/libs/proxy/postgres-protocol2/src/lib.rs
+++ b/libs/proxy/postgres-protocol2/src/lib.rs
@@ -9,8 +9,7 @@
 //!
 //! This library assumes that the `client_encoding` backend parameter has been
 //! set to `UTF8`. It will most likely not behave properly if that is not the case.
-#![doc(html_root_url = "https://docs.rs/postgres-protocol/0.6")]
-#![warn(missing_docs, rust_2018_idioms, clippy::all)]
+#![warn(missing_docs, clippy::all)]

 use byteorder::{BigEndian, ByteOrder};
 use bytes::{BufMut, BytesMut};
--- a/libs/proxy/postgres-protocol2/src/message/frontend.rs
+++ b/libs/proxy/postgres-protocol2/src/message/frontend.rs
@@ -3,7 +3,6 @@

 use byteorder::{BigEndian, ByteOrder};
 use bytes::{Buf, BufMut, BytesMut};
-use std::convert::TryFrom;
 use std::error::Error;
 use std::io;
 use std::marker;
--- a/libs/proxy/postgres-types2/Cargo.toml
+++ b/libs/proxy/postgres-types2/Cargo.toml
@@ -1,7 +1,7 @@
 [package]
 name = "postgres-types2"
 version = "0.1.0"
-edition = "2018"
+edition = "2021"
 license = "MIT/Apache-2.0"

 [dependencies]
--- a/libs/proxy/postgres-types2/src/lib.rs
+++ b/libs/proxy/postgres-types2/src/lib.rs
@@ -2,8 +2,7 @@
 //!
 //! This crate is used by the `tokio-postgres` and `postgres` crates. You normally don't need to depend directly on it
 //! unless you want to define your own `ToSql` or `FromSql` definitions.
-#![doc(html_root_url = "https://docs.rs/postgres-types/0.2")]
-#![warn(clippy::all, rust_2018_idioms, missing_docs)]
+#![warn(clippy::all, missing_docs)]

 use fallible_iterator::FallibleIterator;
 use postgres_protocol2::types;
--- a/libs/proxy/tokio-postgres2/Cargo.toml
+++ b/libs/proxy/tokio-postgres2/Cargo.toml
@@ -1,7 +1,7 @@
 [package]
 name = "tokio-postgres2"
 version = "0.1.0"
-edition = "2018"
+edition = "2021"
 license = "MIT/Apache-2.0"

 [dependencies]
--- a/libs/proxy/tokio-postgres2/src/connection.rs
+++ b/libs/proxy/tokio-postgres2/src/connection.rs
@@ -33,10 +33,14 @@ pub struct Response {
 #[derive(PartialEq, Debug)]
 enum State {
    Active,
-    Terminating,
    Closing,
 }

+enum WriteReady {
+    Terminating,
+    WaitingOnRead,
+}
+
 /// A connection to a PostgreSQL database.
 ///
 /// This is one half of what is returned when a new connection is established. It performs the actual IO with the
@@ -51,7 +55,6 @@ pub struct Connection<S, T> {
    /// HACK: we need this in the Neon Proxy to forward params.
    pub parameters: HashMap<String, String>,
    receiver: mpsc::UnboundedReceiver<Request>,
-    pending_request: Option<RequestMessages>,
    pending_responses: VecDeque<BackendMessage>,
    responses: VecDeque<Response>,
    state: State,
@@ -72,7 +75,6 @@ where
            stream,
            parameters,
            receiver,
-            pending_request: None,
            pending_responses,
            responses: VecDeque::new(),
            state: State::Active,
@@ -93,26 +95,23 @@ where
            .map(|o| o.map(|r| r.map_err(Error::io)))
    }

-    fn poll_read(&mut self, cx: &mut Context<'_>) -> Result<Option<AsyncMessage>, Error> {
-        if self.state != State::Active {
-            trace!("poll_read: done");
-            return Ok(None);
-        }
-
+    /// Read and process messages from the connection to postgres.
+    /// client <- postgres
+    fn poll_read(&mut self, cx: &mut Context<'_>) -> Poll<Result<AsyncMessage, Error>> {
        loop {
            let message = match self.poll_response(cx)? {
                Poll::Ready(Some(message)) => message,
-                Poll::Ready(None) => return Err(Error::closed()),
+                Poll::Ready(None) => return Poll::Ready(Err(Error::closed())),
                Poll::Pending => {
                    trace!("poll_read: waiting on response");
-                    return Ok(None);
+                    return Poll::Pending;
                }
            };

            let (mut messages, request_complete) = match message {
                BackendMessage::Async(Message::NoticeResponse(body)) => {
                    let error = DbError::parse(&mut body.fields()).map_err(Error::parse)?;
-                    return Ok(Some(AsyncMessage::Notice(error)));
+                    return Poll::Ready(Ok(AsyncMessage::Notice(error)));
                }
                BackendMessage::Async(Message::NotificationResponse(body)) => {
                    let notification = Notification {
@@ -120,7 +119,7 @@ where
                        channel: body.channel().map_err(Error::parse)?.to_string(),
                        payload: body.message().map_err(Error::parse)?.to_string(),
                    };
-                    return Ok(Some(AsyncMessage::Notification(notification)));
+                    return Poll::Ready(Ok(AsyncMessage::Notification(notification)));
                }
                BackendMessage::Async(Message::ParameterStatus(body)) => {
                    self.parameters.insert(
@@ -139,8 +138,10 @@ where
            let mut response = match self.responses.pop_front() {
                Some(response) => response,
                None => match messages.next().map_err(Error::parse)? {
-                    Some(Message::ErrorResponse(error)) => return Err(Error::db(error)),
-                    _ => return Err(Error::unexpected_message()),
+                    Some(Message::ErrorResponse(error)) => {
+                        return Poll::Ready(Err(Error::db(error)))
+                    }
+                    _ => return Poll::Ready(Err(Error::unexpected_message())),
                },
            };

@@ -164,18 +165,14 @@ where
                        request_complete,
                    });
                    trace!("poll_read: waiting on sender");
-                    return Ok(None);
+                    return Poll::Pending;
                }
            }
        }
    }

+    /// Fetch the next client request and enqueue the response sender.
    fn poll_request(&mut self, cx: &mut Context<'_>) -> Poll<Option<RequestMessages>> {
-        if let Some(messages) = self.pending_request.take() {
-            trace!("retrying pending request");
-            return Poll::Ready(Some(messages));
-        }
-
        if self.receiver.is_closed() {
            return Poll::Ready(None);
        }
@@ -193,74 +190,80 @@ where
        }
    }

-    fn poll_write(&mut self, cx: &mut Context<'_>) -> Result<bool, Error> {
+    /// Process client requests and write them to the postgres connection, flushing if necessary.
+    /// client -> postgres
+    fn poll_write(&mut self, cx: &mut Context<'_>) -> Poll<Result<WriteReady, Error>> {
        loop {
-            if self.state == State::Closing {
-                trace!("poll_write: done");
-                return Ok(false);
-            }
-
            if Pin::new(&mut self.stream)
                .poll_ready(cx)
                .map_err(Error::io)?
                .is_pending()
            {
                trace!("poll_write: waiting on socket");
-                return Ok(false);
+
+                // poll_ready is self-flushing.
+                return Poll::Pending;
            }

-            let request = match self.poll_request(cx) {
-                Poll::Ready(Some(request)) => request,
-                Poll::Ready(None) if self.responses.is_empty() && self.state == State::Active => {
+            match self.poll_request(cx) {
+                // send the message to postgres
+                Poll::Ready(Some(RequestMessages::Single(request))) => {
+                    Pin::new(&mut self.stream)
+                        .start_send(request)
+                        .map_err(Error::io)?;
+                }
+                // No more messages from the client, and no more responses to wait for.
+                // Send a terminate message to postgres
+                Poll::Ready(None) if self.responses.is_empty() => {
                    trace!("poll_write: at eof, terminating");
-                    self.state = State::Terminating;
                    let mut request = BytesMut::new();
                    frontend::terminate(&mut request);
-                    RequestMessages::Single(FrontendMessage::Raw(request.freeze()))
+                    let request = FrontendMessage::Raw(request.freeze());
+
+                    Pin::new(&mut self.stream)
+                        .start_send(request)
+                        .map_err(Error::io)?;
+
+                    trace!("poll_write: sent eof, closing");
+                    trace!("poll_write: done");
+                    return Poll::Ready(Ok(WriteReady::Terminating));
                }
+                // No more messages from the client, but there are still some responses to wait for.
                Poll::Ready(None) => {
                    trace!(
                        "poll_write: at eof, pending responses {}",
                        self.responses.len()
                    );
-                    return Ok(true);
+                    ready!(self.poll_flush(cx))?;
+                    return Poll::Ready(Ok(WriteReady::WaitingOnRead));
                }
+                // Still waiting for a message from the client.
                Poll::Pending => {
                    trace!("poll_write: waiting on request");
-                    return Ok(true);
-                }
-            };
-
-            match request {
-                RequestMessages::Single(request) => {
-                    Pin::new(&mut self.stream)
-                        .start_send(request)
-                        .map_err(Error::io)?;
-                    if self.state == State::Terminating {
-                        trace!("poll_write: sent eof, closing");
-                        self.state = State::Closing;
-                    }
+                    ready!(self.poll_flush(cx))?;
+                    return Poll::Pending;
                }
            }
        }
    }

-    fn poll_flush(&mut self, cx: &mut Context<'_>) -> Result<(), Error> {
+    fn poll_flush(&mut self, cx: &mut Context<'_>) -> Poll<Result<(), Error>> {
        match Pin::new(&mut self.stream)
            .poll_flush(cx)
            .map_err(Error::io)?
        {
-            Poll::Ready(()) => trace!("poll_flush: flushed"),
-            Poll::Pending => trace!("poll_flush: waiting on socket"),
+            Poll::Ready(()) => {
+                trace!("poll_flush: flushed");
+                Poll::Ready(Ok(()))
+            }
+            Poll::Pending => {
+                trace!("poll_flush: waiting on socket");
+                Poll::Pending
+            }
        }
-        Ok(())
    }

    fn poll_shutdown(&mut self, cx: &mut Context<'_>) -> Poll<Result<(), Error>> {
-        if self.state != State::Closing {
-            return Poll::Pending;
-        }
-
        match Pin::new(&mut self.stream)
            .poll_close(cx)
            .map_err(Error::io)?
@@ -289,18 +292,30 @@ where
        &mut self,
        cx: &mut Context<'_>,
    ) -> Poll<Option<Result<AsyncMessage, Error>>> {
-        let message = self.poll_read(cx)?;
-        let want_flush = self.poll_write(cx)?;
-        if want_flush {
-            self.poll_flush(cx)?;
+        if self.state != State::Closing {
+            // if the state is still active, try read from and write to postgres.
+            let message = self.poll_read(cx)?;
+            let closing = self.poll_write(cx)?;
+            if let Poll::Ready(WriteReady::Terminating) = closing {
+                self.state = State::Closing;
+            }
+
+            if let Poll::Ready(message) = message {
+                return Poll::Ready(Some(Ok(message)));
+            }
+
+            // poll_read returned Pending.
+            // poll_write returned Pending or Ready(WriteReady::WaitingOnRead).
+            // if poll_write returned Ready(WriteReady::WaitingOnRead), then we are waiting to read more data from postgres.
+            if self.state != State::Closing {
+                return Poll::Pending;
+            }
        }
-        match message {
-            Some(message) => Poll::Ready(Some(Ok(message))),
-            None => match self.poll_shutdown(cx) {
-                Poll::Ready(Ok(())) => Poll::Ready(None),
-                Poll::Ready(Err(e)) => Poll::Ready(Some(Err(e))),
-                Poll::Pending => Poll::Pending,
-            },
+
+        match self.poll_shutdown(cx) {
+            Poll::Ready(Ok(())) => Poll::Ready(None),
+            Poll::Ready(Err(e)) => Poll::Ready(Some(Err(e))),
+            Poll::Pending => Poll::Pending,
        }
    }
 }
--- a/libs/proxy/tokio-postgres2/src/lib.rs
+++ b/libs/proxy/tokio-postgres2/src/lib.rs
@@ -1,5 +1,5 @@
 //! An asynchronous, pipelined, PostgreSQL client.
-#![warn(rust_2018_idioms, clippy::all)]
+#![warn(clippy::all)]

 pub use crate::cancel_token::CancelToken;
 pub use crate::client::{Client, SocketConfig};
--- a/libs/utils/Cargo.toml
+++ b/libs/utils/Cargo.toml
@@ -15,17 +15,20 @@ arc-swap.workspace = true
 sentry.workspace = true
 async-compression.workspace = true
 anyhow.workspace = true
+backtrace.workspace = true
 bincode.workspace = true
 bytes.workspace = true
 camino.workspace = true
 chrono.workspace = true
 diatomic-waker.workspace = true
+flate2.workspace = true
 git-version.workspace = true
 hex = { workspace = true, features = ["serde"] }
 humantime.workspace = true
 hyper0 = { workspace = true, features = ["full"] }
+itertools.workspace = true
 fail.workspace = true
-futures = { workspace = true}
+futures = { workspace = true }
 jemalloc_pprof.workspace = true
 jsonwebtoken.workspace = true
 nix.workspace = true
--- a/libs/utils/src/failpoint_support.rs
+++ b/libs/utils/src/failpoint_support.rs
@@ -9,7 +9,7 @@ use serde::{Deserialize, Serialize};
 use tokio_util::sync::CancellationToken;
 use tracing::*;

-/// Declare a failpoint that can use the `pause` failpoint action.
+/// Declare a failpoint that can use to `pause` failpoint action.
 /// We don't want to block the executor thread, hence, spawn_blocking + await.
 #[macro_export]
 macro_rules! pausable_failpoint {
@@ -181,7 +181,7 @@ pub async fn failpoints_handler(
 ) -> Result<Response<Body>, ApiError> {
    if !fail::has_failpoints() {
        return Err(ApiError::BadRequest(anyhow::anyhow!(
-            "Cannot manage failpoints because storage was compiled without failpoints support"
+            "Cannot manage failpoints because neon was compiled without failpoints support"
        )));
    }

--- a/libs/utils/src/http/endpoint.rs
+++ b/libs/utils/src/http/endpoint.rs
@@ -1,15 +1,22 @@
 use crate::auth::{AuthError, Claims, SwappableJwtAuth};
 use crate::http::error::{api_error_handler, route_error_handler, ApiError};
 use crate::http::request::{get_query_param, parse_query_param};
+use crate::pprof;
+use ::pprof::protos::Message as _;
+use ::pprof::ProfilerGuardBuilder;
 use anyhow::{anyhow, Context};
+use bytes::{Bytes, BytesMut};
 use hyper::header::{HeaderName, AUTHORIZATION, CONTENT_DISPOSITION};
 use hyper::http::HeaderValue;
 use hyper::Method;
 use hyper::{header::CONTENT_TYPE, Body, Request, Response};
 use metrics::{register_int_counter, Encoder, IntCounter, TextEncoder};
 use once_cell::sync::Lazy;
+use regex::Regex;
 use routerify::ext::RequestExt;
 use routerify::{Middleware, RequestInfo, Router, RouterBuilder};
+use tokio::sync::{mpsc, Mutex};
+use tokio_stream::wrappers::ReceiverStream;
 use tokio_util::io::ReaderStream;
 use tracing::{debug, info, info_span, warn, Instrument};

@@ -18,11 +25,6 @@ use std::io::Write as _;
 use std::str::FromStr;
 use std::time::Duration;

-use bytes::{Bytes, BytesMut};
-use pprof::protos::Message as _;
-use tokio::sync::{mpsc, Mutex};
-use tokio_stream::wrappers::ReceiverStream;
-
 static SERVE_METRICS_COUNT: Lazy<IntCounter> = Lazy::new(|| {
    register_int_counter!(
        "libmetrics_metric_handler_requests_total",
@@ -365,7 +367,7 @@ pub async fn profile_cpu_handler(req: Request<Body>) -> Result<Response<Body>, A

    // Take the profile.
    let report = tokio::task::spawn_blocking(move || {
-        let guard = pprof::ProfilerGuardBuilder::default()
+        let guard = ProfilerGuardBuilder::default()
            .frequency(frequency_hz)
            .blocklist(&["libc", "libgcc", "pthread", "vdso"])
            .build()?;
@@ -457,10 +459,34 @@ pub async fn profile_heap_handler(req: Request<Body>) -> Result<Response<Body>,
        }

        Format::Pprof => {
-            let data = tokio::task::spawn_blocking(move || prof_ctl.dump_pprof())
-                .await
-                .map_err(|join_err| ApiError::InternalServerError(join_err.into()))?
-                .map_err(ApiError::InternalServerError)?;
+            let data = tokio::task::spawn_blocking(move || {
+                let bytes = prof_ctl.dump_pprof()?;
+                // Symbolize the profile.
+                // TODO: consider moving this upstream to jemalloc_pprof and avoiding the
+                // serialization roundtrip.
+                static STRIP_FUNCTIONS: Lazy<Vec<(Regex, bool)>> = Lazy::new(|| {
+                    // Functions to strip from profiles. If true, also remove child frames.
+                    vec![
+                        (Regex::new("^__rust").unwrap(), false),
+                        (Regex::new("^_start$").unwrap(), false),
+                        (Regex::new("^irallocx_prof").unwrap(), true),
+                        (Regex::new("^prof_alloc_prep").unwrap(), true),
+                        (Regex::new("^std::rt::lang_start").unwrap(), false),
+                        (Regex::new("^std::sys::backtrace::__rust").unwrap(), false),
+                    ]
+                });
+                let profile = pprof::decode(&bytes)?;
+                let profile = pprof::symbolize(profile)?;
+                let profile = pprof::strip_locations(
+                    profile,
+                    &["libc", "libgcc", "pthread", "vdso"],
+                    &STRIP_FUNCTIONS,
+                );
+                pprof::encode(&profile)
+            })
+            .await
+            .map_err(|join_err| ApiError::InternalServerError(join_err.into()))?
+            .map_err(ApiError::InternalServerError)?;
            Response::builder()
                .status(200)
                .header(CONTENT_TYPE, "application/octet-stream")
--- a/libs/utils/src/lib.rs
+++ b/libs/utils/src/lib.rs
@@ -96,6 +96,8 @@ pub mod circuit_breaker;

 pub mod try_rcu;

+pub mod pprof;
+
 // Re-export used in macro. Avoids adding git-version as dep in target crates.
 #[doc(hidden)]
 pub use git_version;
--- a/libs/utils/src/pprof.rs
+++ b/libs/utils/src/pprof.rs
@@ -0,0 +1,190 @@
+use flate2::write::{GzDecoder, GzEncoder};
+use flate2::Compression;
+use itertools::Itertools as _;
+use once_cell::sync::Lazy;
+use pprof::protos::{Function, Line, Message as _, Profile};
+use regex::Regex;
+
+use std::borrow::Cow;
+use std::collections::{HashMap, HashSet};
+use std::ffi::c_void;
+use std::io::Write as _;
+
+/// Decodes a gzip-compressed Protobuf-encoded pprof profile.
+pub fn decode(bytes: &[u8]) -> anyhow::Result<Profile> {
+    let mut gz = GzDecoder::new(Vec::new());
+    gz.write_all(bytes)?;
+    Ok(Profile::parse_from_bytes(&gz.finish()?)?)
+}
+
+/// Encodes a pprof profile as gzip-compressed Protobuf.
+pub fn encode(profile: &Profile) -> anyhow::Result<Vec<u8>> {
+    let mut gz = GzEncoder::new(Vec::new(), Compression::default());
+    profile.write_to_writer(&mut gz)?;
+    Ok(gz.finish()?)
+}
+
+/// Symbolizes a pprof profile using the current binary.
+pub fn symbolize(mut profile: Profile) -> anyhow::Result<Profile> {
+    if !profile.function.is_empty() {
+        return Ok(profile); // already symbolized
+    }
+
+    // Collect function names.
+    let mut functions: HashMap<String, Function> = HashMap::new();
+    let mut strings: HashMap<String, i64> = profile
+        .string_table
+        .into_iter()
+        .enumerate()
+        .map(|(i, s)| (s, i as i64))
+        .collect();
+
+    // Helper to look up or register a string.
+    let mut string_id = |s: &str| -> i64 {
+        // Don't use .entry() to avoid unnecessary allocations.
+        if let Some(id) = strings.get(s) {
+            return *id;
+        }
+        let id = strings.len() as i64;
+        strings.insert(s.to_string(), id);
+        id
+    };
+
+    for loc in &mut profile.location {
+        if !loc.line.is_empty() {
+            continue;
+        }
+
+        // Resolve the line and function for each location.
+        backtrace::resolve(loc.address as *mut c_void, |symbol| {
+            let Some(symname) = symbol.name() else {
+                return;
+            };
+            let mut name = symname.to_string();
+
+            // Strip the Rust monomorphization suffix from the symbol name.
+            static SUFFIX_REGEX: Lazy<Regex> =
+                Lazy::new(|| Regex::new("::h[0-9a-f]{16}$").expect("invalid regex"));
+            if let Some(m) = SUFFIX_REGEX.find(&name) {
+                name.truncate(m.start());
+            }
+
+            let function_id = match functions.get(&name) {
+                Some(function) => function.id,
+                None => {
+                    let id = functions.len() as u64 + 1;
+                    let system_name = String::from_utf8_lossy(symname.as_bytes());
+                    let filename = symbol
+                        .filename()
+                        .map(|path| path.to_string_lossy())
+                        .unwrap_or(Cow::Borrowed(""));
+                    let function = Function {
+                        id,
+                        name: string_id(&name),
+                        system_name: string_id(&system_name),
+                        filename: string_id(&filename),
+                        ..Default::default()
+                    };
+                    functions.insert(name, function);
+                    id
+                }
+            };
+            loc.line.push(Line {
+                function_id,
+                line: symbol.lineno().unwrap_or(0) as i64,
+                ..Default::default()
+            });
+        });
+    }
+
+    // Store the resolved functions, and mark the mapping as resolved.
+    profile.function = functions.into_values().sorted_by_key(|f| f.id).collect();
+    profile.string_table = strings
+        .into_iter()
+        .sorted_by_key(|(_, i)| *i)
+        .map(|(s, _)| s)
+        .collect();
+
+    for mapping in &mut profile.mapping {
+        mapping.has_functions = true;
+        mapping.has_filenames = true;
+    }
+
+    Ok(profile)
+}
+
+/// Strips locations (stack frames) matching the given mappings (substring) or function names
+/// (regex). The function bool specifies whether child frames should be stripped as well.
+///
+/// The string definitions are left behind in the profile for simplicity, to avoid rewriting all
+/// string references.
+pub fn strip_locations(
+    mut profile: Profile,
+    mappings: &[&str],
+    functions: &[(Regex, bool)],
+) -> Profile {
+    // Strip mappings.
+    let mut strip_mappings: HashSet<u64> = HashSet::new();
+
+    profile.mapping.retain(|mapping| {
+        let Some(name) = profile.string_table.get(mapping.filename as usize) else {
+            return true;
+        };
+        if mappings.iter().any(|substr| name.contains(substr)) {
+            strip_mappings.insert(mapping.id);
+            return false;
+        }
+        true
+    });
+
+    // Strip functions.
+    let mut strip_functions: HashMap<u64, bool> = HashMap::new();
+
+    profile.function.retain(|function| {
+        let Some(name) = profile.string_table.get(function.name as usize) else {
+            return true;
+        };
+        for (regex, strip_children) in functions {
+            if regex.is_match(name) {
+                strip_functions.insert(function.id, *strip_children);
+                return false;
+            }
+        }
+        true
+    });
+
+    // Strip locations. The bool specifies whether child frames should be stripped too.
+    let mut strip_locations: HashMap<u64, bool> = HashMap::new();
+
+    profile.location.retain(|location| {
+        for line in &location.line {
+            if let Some(strip_children) = strip_functions.get(&line.function_id) {
+                strip_locations.insert(location.id, *strip_children);
+                return false;
+            }
+        }
+        if strip_mappings.contains(&location.mapping_id) {
+            strip_locations.insert(location.id, false);
+            return false;
+        }
+        true
+    });
+
+    // Strip sample locations.
+    for sample in &mut profile.sample {
+        // First, find the uppermost function with child removal and truncate the stack.
+        if let Some(truncate) = sample
+            .location_id
+            .iter()
+            .rposition(|id| strip_locations.get(id) == Some(&true))
+        {
+            sample.location_id.drain(..=truncate);
+        }
+        // Next, strip any individual frames without child removal.
+        sample
+            .location_id
+            .retain(|id| !strip_locations.contains_key(id));
+    }
+
+    profile
+}
--- a/pageserver/src/http/routes.rs
+++ b/pageserver/src/http/routes.rs
@@ -97,8 +97,8 @@ use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError};
 use crate::DEFAULT_PG_VERSION;
 use crate::{disk_usage_eviction_task, tenant};
 use pageserver_api::models::{
-    StatusResponse, TenantConfigRequest, TenantInfo, TimelineCreateRequest, TimelineGcRequest,
-    TimelineInfo,
+    CompactInfoResponse, StatusResponse, TenantConfigRequest, TenantInfo, TimelineCreateRequest,
+    TimelineGcRequest, TimelineInfo,
 };
 use utils::{
    auth::SwappableJwtAuth,
@@ -2039,6 +2039,34 @@ async fn timeline_cancel_compact_handler(
    .await
 }

+// Get compact info of a timeline
+async fn timeline_compact_info_handler(
+    request: Request<Body>,
+    _cancel: CancellationToken,
+) -> Result<Response<Body>, ApiError> {
+    let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
+    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
+    check_permission(&request, Some(tenant_shard_id.tenant_id))?;
+    let state = get_state(&request);
+    async {
+        let tenant = state
+            .tenant_manager
+            .get_attached_tenant_shard(tenant_shard_id)?;
+        let res = tenant.get_scheduled_compaction_tasks(timeline_id);
+        let mut resp = Vec::new();
+        for item in res {
+            resp.push(CompactInfoResponse {
+                compact_key_range: item.compact_key_range,
+                compact_lsn_range: item.compact_lsn_range,
+                sub_compaction: item.sub_compaction,
+            });
+        }
+        json_response(StatusCode::OK, resp)
+    }
+    .instrument(info_span!("timeline_compact_info", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
+    .await
+}
+
 // Run compaction immediately on given timeline.
 async fn timeline_compact_handler(
    mut request: Request<Body>,
@@ -3400,6 +3428,10 @@ pub fn make_router(
            "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/do_gc",
            |r| api_handler(r, timeline_gc_handler),
        )
+        .get(
+            "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
+            |r| api_handler(r, timeline_compact_info_handler),
+        )
        .put(
            "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
            |r| api_handler(r, timeline_compact_handler),
--- a/pageserver/src/tenant.rs
+++ b/pageserver/src/tenant.rs
@@ -3122,6 +3122,23 @@ impl Tenant {
        }
    }

+    pub(crate) fn get_scheduled_compaction_tasks(
+        &self,
+        timeline_id: TimelineId,
+    ) -> Vec<CompactOptions> {
+        use itertools::Itertools;
+        let guard = self.scheduled_compaction_tasks.lock().unwrap();
+        guard
+            .get(&timeline_id)
+            .map(|tline_pending_tasks| {
+                tline_pending_tasks
+                    .iter()
+                    .map(|x| x.options.clone())
+                    .collect_vec()
+            })
+            .unwrap_or_default()
+    }
+
    /// Schedule a compaction task for a timeline.
    pub(crate) async fn schedule_compaction(
        &self,
@@ -5759,13 +5776,13 @@ mod tests {
    use timeline::{CompactOptions, DeltaLayerTestDesc};
    use utils::id::TenantId;

+    #[cfg(feature = "testing")]
+    use models::CompactLsnRange;
    #[cfg(feature = "testing")]
    use pageserver_api::record::NeonWalRecord;
    #[cfg(feature = "testing")]
    use timeline::compaction::{KeyHistoryRetention, KeyLogAtLsn};
    #[cfg(feature = "testing")]
-    use timeline::CompactLsnRange;
-    #[cfg(feature = "testing")]
    use timeline::GcInfo;

    static TEST_KEY: Lazy<Key> =
@@ -9634,7 +9651,7 @@ mod tests {
    #[cfg(feature = "testing")]
    #[tokio::test]
    async fn test_simple_bottom_most_compaction_on_branch() -> anyhow::Result<()> {
-        use timeline::CompactLsnRange;
+        use models::CompactLsnRange;

        let harness = TenantHarness::create("test_simple_bottom_most_compaction_on_branch").await?;
        let (tenant, ctx) = harness.load().await;
--- a/pageserver/src/tenant/checks.rs
+++ b/pageserver/src/tenant/checks.rs
@@ -1,12 +1,15 @@
 use std::collections::BTreeSet;

 use itertools::Itertools;
+use pageserver_compaction::helpers::overlaps_with;

 use super::storage_layer::LayerName;

 /// Checks whether a layer map is valid (i.e., is a valid result of the current compaction algorithm if nothing goes wrong).
 ///
-/// The function checks if we can split the LSN range of a delta layer only at the LSNs of the delta layers. For example,
+/// The function implements a fast path check and a slow path check.
+///
+/// The fast path checks if we can split the LSN range of a delta layer only at the LSNs of the delta layers. For example,
 ///
 /// ```plain
 /// |       |                 |       |
@@ -25,31 +28,47 @@ use super::storage_layer::LayerName;
 /// |       |    |   4   |    |       |
 ///
 /// If layer 2 and 4 contain the same single key, this is also a valid layer map.
+///
+/// However, if a partial compaction is still going on, it is possible that we get a layer map not satisfying the above condition.
+/// Therefore, we fallback to simply check if any of the two delta layers overlap. (See "A slow path...")
 pub fn check_valid_layermap(metadata: &[LayerName]) -> Option<String> {
    let mut lsn_split_point = BTreeSet::new(); // TODO: use a better data structure (range tree / range set?)
    let mut all_delta_layers = Vec::new();
    for name in metadata {
        if let LayerName::Delta(layer) = name {
-            if layer.key_range.start.next() != layer.key_range.end {
-                all_delta_layers.push(layer.clone());
-            }
+            all_delta_layers.push(layer.clone());
        }
    }
    for layer in &all_delta_layers {
-        let lsn_range = &layer.lsn_range;
-        lsn_split_point.insert(lsn_range.start);
-        lsn_split_point.insert(lsn_range.end);
+        if layer.key_range.start.next() != layer.key_range.end {
+            let lsn_range = &layer.lsn_range;
+            lsn_split_point.insert(lsn_range.start);
+            lsn_split_point.insert(lsn_range.end);
+        }
    }
-    for layer in &all_delta_layers {
+    for (idx, layer) in all_delta_layers.iter().enumerate() {
+        if layer.key_range.start.next() == layer.key_range.end {
+            continue;
+        }
        let lsn_range = layer.lsn_range.clone();
        let intersects = lsn_split_point.range(lsn_range).collect_vec();
        if intersects.len() > 1 {
-            let err = format!(
-                "layer violates the layer map LSN split assumption: layer {} intersects with LSN [{}]",
-                layer,
-                intersects.into_iter().map(|lsn| lsn.to_string()).join(", ")
-            );
-            return Some(err);
+            // A slow path to check if the layer intersects with any other delta layer.
+            for (other_idx, other_layer) in all_delta_layers.iter().enumerate() {
+                if other_idx == idx {
+                    // do not check self intersects with self
+                    continue;
+                }
+                if overlaps_with(&layer.lsn_range, &other_layer.lsn_range)
+                    && overlaps_with(&layer.key_range, &other_layer.key_range)
+                {
+                    let err = format!(
+                            "layer violates the layer map LSN split assumption: layer {} intersects with layer {}",
+                            layer, other_layer
+                        );
+                    return Some(err);
+                }
+            }
        }
    }
    None
--- a/pageserver/src/tenant/timeline.rs
+++ b/pageserver/src/tenant/timeline.rs
@@ -31,9 +31,9 @@ use pageserver_api::{
    },
    keyspace::{KeySpaceAccum, KeySpaceRandomAccum, SparseKeyPartitioning},
    models::{
-        CompactionAlgorithm, CompactionAlgorithmSettings, DownloadRemoteLayersTaskInfo,
-        DownloadRemoteLayersTaskSpawnRequest, EvictionPolicy, InMemoryLayerInfo, LayerMapInfo,
-        LsnLease, TimelineState,
+        CompactKeyRange, CompactLsnRange, CompactionAlgorithm, CompactionAlgorithmSettings,
+        DownloadRemoteLayersTaskInfo, DownloadRemoteLayersTaskSpawnRequest, EvictionPolicy,
+        InMemoryLayerInfo, LayerMapInfo, LsnLease, TimelineState,
    },
    reltag::BlockNumber,
    shard::{ShardIdentity, ShardNumber, TenantShardId},
@@ -788,63 +788,6 @@ pub(crate) struct CompactRequest {
    pub sub_compaction_max_job_size_mb: Option<u64>,
 }

-#[serde_with::serde_as]
-#[derive(Debug, Clone, serde::Deserialize)]
-pub(crate) struct CompactLsnRange {
-    pub start: Lsn,
-    pub end: Lsn,
-}
-
-#[serde_with::serde_as]
-#[derive(Debug, Clone, serde::Deserialize)]
-pub(crate) struct CompactKeyRange {
-    #[serde_as(as = "serde_with::DisplayFromStr")]
-    pub start: Key,
-    #[serde_as(as = "serde_with::DisplayFromStr")]
-    pub end: Key,
-}
-
-impl From<Range<Lsn>> for CompactLsnRange {
-    fn from(range: Range<Lsn>) -> Self {
-        Self {
-            start: range.start,
-            end: range.end,
-        }
-    }
-}
-
-impl From<Range<Key>> for CompactKeyRange {
-    fn from(range: Range<Key>) -> Self {
-        Self {
-            start: range.start,
-            end: range.end,
-        }
-    }
-}
-
-impl From<CompactLsnRange> for Range<Lsn> {
-    fn from(range: CompactLsnRange) -> Self {
-        range.start..range.end
-    }
-}
-
-impl From<CompactKeyRange> for Range<Key> {
-    fn from(range: CompactKeyRange) -> Self {
-        range.start..range.end
-    }
-}
-
-impl CompactLsnRange {
-    #[cfg(test)]
-    #[cfg(feature = "testing")]
-    pub fn above(lsn: Lsn) -> Self {
-        Self {
-            start: lsn,
-            end: Lsn::MAX,
-        }
-    }
-}
-
 #[derive(Debug, Clone, Default)]
 pub(crate) struct CompactOptions {
    pub flags: EnumSet<CompactFlags>,
--- a/pageserver/src/tenant/timeline/compaction.rs
+++ b/pageserver/src/tenant/timeline/compaction.rs
@@ -29,6 +29,7 @@ use utils::id::TimelineId;
 use crate::context::{AccessStatsBehavior, RequestContext, RequestContextBuilder};
 use crate::page_cache;
 use crate::statvfs::Statvfs;
+use crate::tenant::checks::check_valid_layermap;
 use crate::tenant::remote_timeline_client::WaitCompletionError;
 use crate::tenant::storage_layer::batch_split_writer::{
    BatchWriterResult, SplitDeltaLayerWriter, SplitImageLayerWriter,
@@ -1823,7 +1824,7 @@ impl Timeline {
        // by estimating the amount of files read for a compaction job. We should also partition on LSN.
        let ((dense_ks, sparse_ks), _) = {
            let Ok(partition) = self.partitioning.try_lock() else {
-                bail!("failed to acquire partition lock");
+                bail!("failed to acquire partition lock during gc-compaction");
            };
            partition.clone()
        };
@@ -2156,15 +2157,14 @@ impl Timeline {

        // Step 1: construct a k-merge iterator over all layers.
        // Also, verify if the layer map can be split by drawing a horizontal line at every LSN start/end split point.
-        // disable the check for now because we need to adjust the check for partial compactions, will enable later.
-        // let layer_names = job_desc
-        //     .selected_layers
-        //     .iter()
-        //     .map(|layer| layer.layer_desc().layer_name())
-        //     .collect_vec();
-        // if let Some(err) = check_valid_layermap(&layer_names) {
-        //     warn!("gc-compaction layer map check failed because {}, this is normal if partial compaction is not finished yet", err);
-        // }
+        let layer_names = job_desc
+            .selected_layers
+            .iter()
+            .map(|layer| layer.layer_desc().layer_name())
+            .collect_vec();
+        if let Some(err) = check_valid_layermap(&layer_names) {
+            bail!("gc-compaction layer map check failed because {}, cannot proceed with compaction due to potential data loss", err);
+        }
        // The maximum LSN we are processing in this compaction loop
        let end_lsn = job_desc
            .selected_layers
@@ -2546,13 +2546,48 @@ impl Timeline {
        );

        // Step 3: Place back to the layer map.
+
+        // First, do a sanity check to ensure the newly-created layer map does not contain overlaps.
+        let all_layers = {
+            let guard = self.layers.read().await;
+            let layer_map = guard.layer_map()?;
+            layer_map.iter_historic_layers().collect_vec()
+        };
+
+        let mut final_layers = all_layers
+            .iter()
+            .map(|layer| layer.layer_name())
+            .collect::<HashSet<_>>();
+        for layer in &layer_selection {
+            final_layers.remove(&layer.layer_desc().layer_name());
+        }
+        for layer in &compact_to {
+            final_layers.insert(layer.layer_desc().layer_name());
+        }
+        let final_layers = final_layers.into_iter().collect_vec();
+
+        // TODO: move this check before we call `finish` on image layer writers. However, this will require us to get the layer name before we finish
+        // the writer, so potentially, we will need a function like `ImageLayerBatchWriter::get_all_pending_layer_keys` to get all the keys that are
+        // in the writer before finalizing the persistent layers. Now we would leave some dangling layers on the disk if the check fails.
+        if let Some(err) = check_valid_layermap(&final_layers) {
+            bail!("gc-compaction layer map check failed after compaction because {}, compaction result not applied to the layer map due to potential data loss", err);
+        }
+
+        // Between the sanity check and this compaction update, there could be new layers being flushed, but it should be fine because we only
+        // operate on L1 layers.
        {
-            // TODO: sanity check if the layer map is valid (i.e., should not have overlaps)
            let mut guard = self.layers.write().await;
            guard
                .open_mut()?
                .finish_gc_compaction(&layer_selection, &compact_to, &self.metrics)
        };
+
+        // Schedule an index-only upload to update the `latest_gc_cutoff` in the index_part.json.
+        // Otherwise, after restart, the index_part only contains the old `latest_gc_cutoff` and
+        // find_gc_cutoffs will try accessing things below the cutoff. TODO: ideally, this should
+        // be batched into `schedule_compaction_update`.
+        let disk_consistent_lsn = self.disk_consistent_lsn.load();
+        self.schedule_uploads(disk_consistent_lsn, None)?;
        self.remote_client
            .schedule_compaction_update(&layer_selection, &compact_to)?;

--- a/pgxn/neon/file_cache.c
+++ b/pgxn/neon/file_cache.c
@@ -541,6 +541,7 @@ lfc_cache_containsv(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
 		}
 		else
 		{
+			LWLockRelease(lfc_lock);
 			return found;
 		}

--- a/pgxn/neon/libpagestore.c
+++ b/pgxn/neon/libpagestore.c
@@ -827,7 +827,6 @@ pageserver_send(shardno_t shard_no, NeonRequest *request)
 	{
 		while (!pageserver_connect(shard_no, shard->n_reconnect_attempts < max_reconnect_attempts ? LOG : ERROR))
 		{
-			HandleMainLoopInterrupts();
 			shard->n_reconnect_attempts += 1;
 		}
 		shard->n_reconnect_attempts = 0;
--- a/proxy/README.md
+++ b/proxy/README.md
@@ -102,23 +102,39 @@ User can pass several optional headers that will affect resulting json.
 2. `Neon-Array-Mode: true`. Return postgres rows as arrays instead of objects. That is more compact representation and also helps in some edge
 cases where it is hard to use rows represented as objects (e.g. when several fields have the same name).

+## Test proxy locally

-## Using SNI-based routing on localhost
-
-Now proxy determines project name from the subdomain, request to the `round-rice-566201.somedomain.tld` will be routed to the project named `round-rice-566201`. Unfortunately, `/etc/hosts` does not support domain wildcards, so I usually use `*.localtest.me` which resolves to `127.0.0.1`. Now we can create self-signed certificate and play with proxy:
+Proxy determines project name from the subdomain, request to the `round-rice-566201.somedomain.tld` will be routed to the project named `round-rice-566201`. Unfortunately, `/etc/hosts` does not support domain wildcards, so we can use *.localtest.me` which resolves to `127.0.0.1`.

+Let's create self-signed certificate by running:
 ```sh
 openssl req -new -x509 -days 365 -nodes -text -out server.crt -keyout server.key -subj "/CN=*.localtest.me"
 ```

-start proxy
-
+Then we need to build proxy with 'testing' feature and run, e.g.:
 ```sh
-./target/debug/proxy -c server.crt -k server.key
+RUST_LOG=proxy cargo run -p proxy --bin proxy --features testing -- --auth-backend postgres --auth-endpoint 'postgresql://proxy:password@endpoint.localtest.me:5432/postgres' --is-private-access-proxy true -c server.crt -k server.key
 ```

-and connect to it
+We will also need to have a postgres instance. Assuming that we have setted up docker we can set it up as follows:
+```sh
+docker run \
+  --detach \
+  --name proxy-postgres \
+  --env POSTGRES_PASSWORD=proxy-postgres \
+  --publish 5432:5432 \
+  postgres:17-bookworm
+```
+
+Next step is setting up auth table and schema as well as creating role (without the JWT table):
+```sh
+docker exec -it proxy-postgres psql -U postgres -c "CREATE SCHEMA IF NOT EXISTS neon_control_plane"
+docker exec -it proxy-postgres psql -U postgres -c "CREATE TABLE neon_control_plane.endpoints (endpoint_id VARCHAR(255) PRIMARY KEY, allowed_ips VARCHAR(255))"
+docker exec -it proxy-postgres psql -U postgres -c "CREATE ROLE proxy WITH SUPERUSER LOGIN PASSWORD 'password';"
+```
+
+Now from client you can start a new session:

 ```sh
-PGSSLROOTCERT=./server.crt psql 'postgres://my-cluster-42.localtest.me:1234?sslmode=verify-full'
-```
+PGSSLROOTCERT=./server.crt psql  "postgresql://proxy:password@endpoint.localtest.me:4432/postgres?sslmode=verify-full"
+```
--- a/proxy/src/auth/backend/mod.rs
+++ b/proxy/src/auth/backend/mod.rs
@@ -678,6 +678,9 @@ mod tests {
        .await
        .unwrap();

+        // flush the final server message
+        stream.flush().await.unwrap();
+
        handle.await.unwrap();
    }

--- a/proxy/src/auth/flow.rs
+++ b/proxy/src/auth/flow.rs
@@ -10,7 +10,6 @@ use tracing::info;

 use super::backend::ComputeCredentialKeys;
 use super::{AuthError, PasswordHackPayload};
-use crate::config::TlsServerEndPoint;
 use crate::context::RequestContext;
 use crate::control_plane::AuthSecret;
 use crate::intern::EndpointIdInt;
@@ -18,6 +17,7 @@ use crate::sasl;
 use crate::scram::threadpool::ThreadPool;
 use crate::scram::{self};
 use crate::stream::{PqStream, Stream};
+use crate::tls::TlsServerEndPoint;

 /// Every authentication selector is supposed to implement this trait.
 pub(crate) trait AuthMethod {
--- a/proxy/src/bin/local_proxy.rs
+++ b/proxy/src/bin/local_proxy.rs
@@ -13,7 +13,10 @@ use proxy::auth::backend::jwt::JwkCache;
 use proxy::auth::backend::local::{LocalBackend, JWKS_ROLE_MAP};
 use proxy::auth::{self};
 use proxy::cancellation::CancellationHandlerMain;
-use proxy::config::{self, AuthenticationConfig, HttpConfig, ProxyConfig, RetryConfig};
+use proxy::config::{
+    self, AuthenticationConfig, ComputeConfig, HttpConfig, ProxyConfig, RetryConfig,
+};
+use proxy::conn::TokioTcpAcceptor;
 use proxy::control_plane::locks::ApiLocks;
 use proxy::control_plane::messages::{EndpointJwksResponse, JwksSettings};
 use proxy::http::health_server::AppMetrics;
@@ -25,6 +28,7 @@ use proxy::rate_limiter::{
 use proxy::scram::threadpool::ThreadPool;
 use proxy::serverless::cancel_set::CancelSet;
 use proxy::serverless::{self, GlobalConnPoolOptions};
+use proxy::tls::client_config::compute_client_config_with_root_certs;
 use proxy::types::RoleName;
 use proxy::url::ApiUrl;

@@ -33,7 +37,6 @@ project_build_tag!(BUILD_TAG);

 use clap::Parser;
 use thiserror::Error;
-use tokio::net::TcpListener;
 use tokio::sync::Notify;
 use tokio::task::JoinSet;
 use tokio_util::sync::CancellationToken;
@@ -163,8 +166,8 @@ async fn main() -> anyhow::Result<()> {
        }
    };

-    let metrics_listener = TcpListener::bind(args.metrics).await?.into_std()?;
-    let http_listener = TcpListener::bind(args.http).await?;
+    let metrics_listener = TokioTcpAcceptor::bind(args.metrics).await?;
+    let http_listener = TokioTcpAcceptor::bind(args.http).await?;
    let shutdown = CancellationToken::new();

    // todo: should scale with CU
@@ -209,6 +212,7 @@ async fn main() -> anyhow::Result<()> {
        http_listener,
        shutdown.clone(),
        Arc::new(CancellationHandlerMain::new(
+            &config.connect_to_compute,
            Arc::new(DashMap::new()),
            None,
            proxy::metrics::CancellationSource::Local,
@@ -268,10 +272,15 @@ fn build_config(args: &LocalProxyCliArgs) -> anyhow::Result<&'static ProxyConfig
        max_response_size_bytes: args.sql_over_http.sql_over_http_max_response_size_bytes,
    };

+    let compute_config = ComputeConfig {
+        retry: RetryConfig::parse(RetryConfig::CONNECT_TO_COMPUTE_DEFAULT_VALUES)?,
+        tls: Arc::new(compute_client_config_with_root_certs()?),
+        timeout: Duration::from_secs(2),
+    };
+
    Ok(Box::leak(Box::new(ProxyConfig {
        tls_config: None,
        metric_collection: None,
-        allow_self_signed_compute: false,
        http_config,
        authentication_config: AuthenticationConfig {
            jwks_cache: JwkCache::default(),
@@ -290,9 +299,7 @@ fn build_config(args: &LocalProxyCliArgs) -> anyhow::Result<&'static ProxyConfig
        region: "local".into(),
        wake_compute_retry_config: RetryConfig::parse(RetryConfig::WAKE_COMPUTE_DEFAULT_VALUES)?,
        connect_compute_locks,
-        connect_to_compute_retry_config: RetryConfig::parse(
-            RetryConfig::CONNECT_TO_COMPUTE_DEFAULT_VALUES,
-        )?,
+        connect_to_compute: compute_config,
    })))
 }

--- a/proxy/src/bin/pg_sni_router.rs
+++ b/proxy/src/bin/pg_sni_router.rs
@@ -10,12 +10,13 @@ use clap::Arg;
 use futures::future::Either;
 use futures::TryFutureExt;
 use itertools::Itertools;
-use proxy::config::TlsServerEndPoint;
+use proxy::conn::{Acceptor, TokioTcpAcceptor};
 use proxy::context::RequestContext;
 use proxy::metrics::{Metrics, ThreadPoolMetrics};
 use proxy::protocol2::ConnectionInfo;
 use proxy::proxy::{copy_bidirectional_client_compute, run_until_cancelled, ErrorSource};
 use proxy::stream::{PqStream, Stream};
+use proxy::tls::TlsServerEndPoint;
 use rustls::crypto::ring;
 use rustls::pki_types::PrivateKeyDer;
 use tokio::io::{AsyncRead, AsyncWrite};
@@ -122,7 +123,7 @@ async fn main() -> anyhow::Result<()> {
    // Start listening for incoming client connections
    let proxy_address: SocketAddr = args.get_one::<String>("listen").unwrap().parse()?;
    info!("Starting sni router on {proxy_address}");
-    let proxy_listener = TcpListener::bind(proxy_address).await?;
+    let proxy_listener = TokioTcpAcceptor::bind(proxy_address).await?;

    let cancellation_token = CancellationToken::new();

@@ -152,17 +153,13 @@ async fn task_main(
    dest_suffix: Arc<String>,
    tls_config: Arc<rustls::ServerConfig>,
    tls_server_end_point: TlsServerEndPoint,
-    listener: tokio::net::TcpListener,
+    acceptor: TokioTcpAcceptor,
    cancellation_token: CancellationToken,
 ) -> anyhow::Result<()> {
-    // When set for the server socket, the keepalive setting
-    // will be inherited by all accepted client sockets.
-    socket2::SockRef::from(&listener).set_keepalive(true)?;
-
    let connections = tokio_util::task::task_tracker::TaskTracker::new();

    while let Some(accept_result) =
-        run_until_cancelled(listener.accept(), &cancellation_token).await
+        run_until_cancelled(acceptor.accept(), &cancellation_token).await
    {
        let (socket, peer_addr) = accept_result?;

@@ -172,10 +169,6 @@ async fn task_main(

        connections.spawn(
            async move {
-                socket
-                    .set_nodelay(true)
-                    .context("failed to set socket option")?;
-
                info!(%peer_addr, "serving");
                let ctx = RequestContext::new(
                    session_id,
@@ -197,7 +190,7 @@ async fn task_main(
    }

    connections.close();
-    drop(listener);
+    drop(acceptor);

    connections.wait().await;

@@ -229,7 +222,7 @@ async fn ssl_handshake<S: AsyncRead + AsyncWrite + Unpin>(

            let (raw, read_buf) = stream.into_inner();
            // TODO: Normally, client doesn't send any data before
-            // server says TLS handshake is ok and read_buf is empy.
+            // server says TLS handshake is ok and read_buf is empty.
            // However, you could imagine pipelining of postgres
            // SSLRequest + TLS ClientHello in one hunk similar to
            // pipelining in our node js driver. We should probably
--- a/proxy/src/bin/proxy.rs
+++ b/proxy/src/bin/proxy.rs
@@ -1,6 +1,7 @@
 use std::net::SocketAddr;
 use std::pin::pin;
 use std::sync::Arc;
+use std::time::Duration;

 use anyhow::bail;
 use futures::future::Either;
@@ -8,9 +9,10 @@ use proxy::auth::backend::jwt::JwkCache;
 use proxy::auth::backend::{AuthRateLimiter, ConsoleRedirectBackend, MaybeOwned};
 use proxy::cancellation::{CancelMap, CancellationHandler};
 use proxy::config::{
-    self, remote_storage_from_toml, AuthenticationConfig, CacheOptions, HttpConfig,
+    self, remote_storage_from_toml, AuthenticationConfig, CacheOptions, ComputeConfig, HttpConfig,
    ProjectInfoCacheOptions, ProxyConfig, ProxyProtocolV2,
 };
+use proxy::conn::TokioTcpAcceptor;
 use proxy::context::parquet::ParquetUploadArgs;
 use proxy::http::health_server::AppMetrics;
 use proxy::metrics::Metrics;
@@ -23,9 +25,9 @@ use proxy::redis::{elasticache, notifications};
 use proxy::scram::threadpool::ThreadPool;
 use proxy::serverless::cancel_set::CancelSet;
 use proxy::serverless::GlobalConnPoolOptions;
+use proxy::tls::client_config::compute_client_config_with_root_certs;
 use proxy::{auth, control_plane, http, serverless, usage_metrics};
 use remote_storage::RemoteStorageConfig;
-use tokio::net::TcpListener;
 use tokio::sync::Mutex;
 use tokio::task::JoinSet;
 use tokio_util::sync::CancellationToken;
@@ -129,9 +131,6 @@ struct ProxyCliArgs {
    /// lock for `connect_compute` api method. example: "shards=32,permits=4,epoch=10m,timeout=1s". (use `permits=0` to disable).
    #[clap(long, default_value = config::ConcurrencyLockOptions::DEFAULT_OPTIONS_CONNECT_COMPUTE_LOCK)]
    connect_compute_lock: String,
-    /// Allow self-signed certificates for compute nodes (for testing)
-    #[clap(long, default_value_t = false, value_parser = clap::builder::BoolishValueParser::new(), action = clap::ArgAction::Set)]
-    allow_self_signed_compute: bool,
    #[clap(flatten)]
    sql_over_http: SqlOverHttpArgs,
    /// timeout for scram authentication protocol
@@ -354,17 +353,17 @@ async fn main() -> anyhow::Result<()> {
    // Check that we can bind to address before further initialization
    let http_address: SocketAddr = args.http.parse()?;
    info!("Starting http on {http_address}");
-    let http_listener = TcpListener::bind(http_address).await?.into_std()?;
+    let http_listener = TokioTcpAcceptor::bind(http_address).await?;

    let mgmt_address: SocketAddr = args.mgmt.parse()?;
    info!("Starting mgmt on {mgmt_address}");
-    let mgmt_listener = TcpListener::bind(mgmt_address).await?;
+    let mgmt_listener = TokioTcpAcceptor::bind(mgmt_address).await?;

    let proxy_listener = if !args.is_auth_broker {
        let proxy_address: SocketAddr = args.proxy.parse()?;
        info!("Starting proxy on {proxy_address}");

-        Some(TcpListener::bind(proxy_address).await?)
+        Some(TokioTcpAcceptor::bind(proxy_address).await?)
    } else {
        None
    };
@@ -374,7 +373,7 @@ async fn main() -> anyhow::Result<()> {
    let serverless_listener = if let Some(serverless_address) = args.wss {
        let serverless_address: SocketAddr = serverless_address.parse()?;
        info!("Starting wss on {serverless_address}");
-        Some(TcpListener::bind(serverless_address).await?)
+        Some(TokioTcpAcceptor::bind(serverless_address).await?)
    } else if args.is_auth_broker {
        bail!("wss arg must be present for auth-broker")
    } else {
@@ -400,6 +399,7 @@ async fn main() -> anyhow::Result<()> {
    let cancellation_handler = Arc::new(CancellationHandler::<
        Option<Arc<Mutex<RedisPublisherClient>>>,
    >::new(
+        &config.connect_to_compute,
        cancel_map.clone(),
        redis_publisher,
        proxy::metrics::CancellationSource::FromClient,
@@ -495,6 +495,7 @@ async fn main() -> anyhow::Result<()> {
                    let cache = api.caches.project_info.clone();
                    if let Some(client) = client1 {
                        maintenance_tasks.spawn(notifications::task_main(
+                            config,
                            client,
                            cache.clone(),
                            cancel_map.clone(),
@@ -503,6 +504,7 @@ async fn main() -> anyhow::Result<()> {
                    }
                    if let Some(client) = client2 {
                        maintenance_tasks.spawn(notifications::task_main(
+                            config,
                            client,
                            cache.clone(),
                            cancel_map.clone(),
@@ -564,9 +566,6 @@ fn build_config(args: &ProxyCliArgs) -> anyhow::Result<&'static ProxyConfig> {
        _ => bail!("either both or neither tls-key and tls-cert must be specified"),
    };

-    if args.allow_self_signed_compute {
-        warn!("allowing self-signed compute certificates");
-    }
    let backup_metric_collection_config = config::MetricBackupCollectionConfig {
        interval: args.metric_backup_collection_interval,
        remote_storage_config: args.metric_backup_collection_remote_storage.clone(),
@@ -638,10 +637,15 @@ fn build_config(args: &ProxyCliArgs) -> anyhow::Result<&'static ProxyConfig> {
        console_redirect_confirmation_timeout: args.webauth_confirmation_timeout,
    };

+    let compute_config = ComputeConfig {
+        retry: config::RetryConfig::parse(&args.connect_to_compute_retry)?,
+        tls: Arc::new(compute_client_config_with_root_certs()?),
+        timeout: Duration::from_secs(2),
+    };
+
    let config = ProxyConfig {
        tls_config,
        metric_collection,
-        allow_self_signed_compute: args.allow_self_signed_compute,
        http_config,
        authentication_config,
        proxy_protocol_v2: args.proxy_protocol_v2,
@@ -649,9 +653,7 @@ fn build_config(args: &ProxyCliArgs) -> anyhow::Result<&'static ProxyConfig> {
        region: args.region.clone(),
        wake_compute_retry_config: config::RetryConfig::parse(&args.wake_compute_retry)?,
        connect_compute_locks,
-        connect_to_compute_retry_config: config::RetryConfig::parse(
-            &args.connect_to_compute_retry,
-        )?,
+        connect_to_compute: compute_config,
    };

    let config = Box::leak(Box::new(config));
--- a/proxy/src/cancellation.rs
+++ b/proxy/src/cancellation.rs
@@ -3,7 +3,8 @@ use std::sync::Arc;

 use dashmap::DashMap;
 use ipnet::{IpNet, Ipv4Net, Ipv6Net};
-use postgres_client::{CancelToken, NoTls};
+use postgres_client::tls::MakeTlsConnect;
+use postgres_client::CancelToken;
 use pq_proto::CancelKeyData;
 use thiserror::Error;
 use tokio::net::TcpStream;
@@ -12,6 +13,7 @@ use tracing::{debug, info};
 use uuid::Uuid;

 use crate::auth::{check_peer_addr_is_in_list, IpPattern};
+use crate::config::ComputeConfig;
 use crate::error::ReportableError;
 use crate::ext::LockExt;
 use crate::metrics::{CancellationRequest, CancellationSource, Metrics};
@@ -19,6 +21,7 @@ use crate::rate_limiter::LeakyBucketRateLimiter;
 use crate::redis::cancellation_publisher::{
    CancellationPublisher, CancellationPublisherMut, RedisPublisherClient,
 };
+use crate::tls::postgres_rustls::MakeRustlsConnect;

 pub type CancelMap = Arc<DashMap<CancelKeyData, Option<CancelClosure>>>;
 pub type CancellationHandlerMain = CancellationHandler<Option<Arc<Mutex<RedisPublisherClient>>>>;
@@ -30,6 +33,7 @@ type IpSubnetKey = IpNet;
 ///
 /// If `CancellationPublisher` is available, cancel request will be used to publish the cancellation key to other proxy instances.
 pub struct CancellationHandler<P> {
+    compute_config: &'static ComputeConfig,
    map: CancelMap,
    client: P,
    /// This field used for the monitoring purposes.
@@ -174,8 +178,11 @@ impl<P: CancellationPublisher> CancellationHandler<P> {
                source: self.from,
                kind: crate::metrics::CancellationOutcome::Found,
            });
-        info!("cancelling query per user's request using key {key}");
-        cancel_closure.try_cancel_query().await
+        info!(
+            "cancelling query per user's request using key {key}, hostname {}, address: {}",
+            cancel_closure.hostname, cancel_closure.socket_addr
+        );
+        cancel_closure.try_cancel_query(self.compute_config).await
    }

    #[cfg(test)]
@@ -190,8 +197,13 @@ impl<P: CancellationPublisher> CancellationHandler<P> {
 }

 impl CancellationHandler<()> {
-    pub fn new(map: CancelMap, from: CancellationSource) -> Self {
+    pub fn new(
+        compute_config: &'static ComputeConfig,
+        map: CancelMap,
+        from: CancellationSource,
+    ) -> Self {
        Self {
+            compute_config,
            map,
            client: (),
            from,
@@ -206,8 +218,14 @@ impl CancellationHandler<()> {
 }

 impl<P: CancellationPublisherMut> CancellationHandler<Option<Arc<Mutex<P>>>> {
-    pub fn new(map: CancelMap, client: Option<Arc<Mutex<P>>>, from: CancellationSource) -> Self {
+    pub fn new(
+        compute_config: &'static ComputeConfig,
+        map: CancelMap,
+        client: Option<Arc<Mutex<P>>>,
+        from: CancellationSource,
+    ) -> Self {
        Self {
+            compute_config,
            map,
            client,
            from,
@@ -229,6 +247,7 @@ pub struct CancelClosure {
    socket_addr: SocketAddr,
    cancel_token: CancelToken,
    ip_allowlist: Vec<IpPattern>,
+    hostname: String, // for pg_sni router
 }

 impl CancelClosure {
@@ -236,17 +255,36 @@ impl CancelClosure {
        socket_addr: SocketAddr,
        cancel_token: CancelToken,
        ip_allowlist: Vec<IpPattern>,
+        hostname: String,
    ) -> Self {
        Self {
            socket_addr,
            cancel_token,
            ip_allowlist,
+            hostname,
        }
    }
    /// Cancels the query running on user's compute node.
-    pub(crate) async fn try_cancel_query(self) -> Result<(), CancelError> {
+    pub(crate) async fn try_cancel_query(
+        self,
+        compute_config: &ComputeConfig,
+    ) -> Result<(), CancelError> {
        let socket = TcpStream::connect(self.socket_addr).await?;
-        self.cancel_token.cancel_query_raw(socket, NoTls).await?;
+
+        let mut mk_tls =
+            crate::tls::postgres_rustls::MakeRustlsConnect::new(compute_config.tls.clone());
+        let tls = <MakeRustlsConnect as MakeTlsConnect<tokio::net::TcpStream>>::make_tls_connect(
+            &mut mk_tls,
+            &self.hostname,
+        )
+        .map_err(|e| {
+            CancelError::IO(std::io::Error::new(
+                std::io::ErrorKind::Other,
+                e.to_string(),
+            ))
+        })?;
+
+        self.cancel_token.cancel_query_raw(socket, tls).await?;
        debug!("query was cancelled");
        Ok(())
    }
@@ -286,11 +324,30 @@ impl<P> Drop for Session<P> {
 #[cfg(test)]
 #[expect(clippy::unwrap_used)]
 mod tests {
+    use std::time::Duration;
+
    use super::*;
+    use crate::config::RetryConfig;
+    use crate::tls::client_config::compute_client_config_with_certs;
+
+    fn config() -> ComputeConfig {
+        let retry = RetryConfig {
+            base_delay: Duration::from_secs(1),
+            max_retries: 5,
+            backoff_factor: 2.0,
+        };
+
+        ComputeConfig {
+            retry,
+            tls: Arc::new(compute_client_config_with_certs(std::iter::empty())),
+            timeout: Duration::from_secs(2),
+        }
+    }

    #[tokio::test]
    async fn check_session_drop() -> anyhow::Result<()> {
        let cancellation_handler = Arc::new(CancellationHandler::<()>::new(
+            Box::leak(Box::new(config())),
            CancelMap::default(),
            CancellationSource::FromRedis,
        ));
@@ -306,8 +363,11 @@ mod tests {

    #[tokio::test]
    async fn cancel_session_noop_regression() {
-        let handler =
-            CancellationHandler::<()>::new(CancelMap::default(), CancellationSource::Local);
+        let handler = CancellationHandler::<()>::new(
+            Box::leak(Box::new(config())),
+            CancelMap::default(),
+            CancellationSource::Local,
+        );
        handler
            .cancel_session(
                CancelKeyData {
--- a/proxy/src/compute.rs
+++ b/proxy/src/compute.rs
@@ -1,17 +1,13 @@
 use std::io;
 use std::net::SocketAddr;
-use std::sync::Arc;
 use std::time::Duration;

 use futures::{FutureExt, TryFutureExt};
 use itertools::Itertools;
-use once_cell::sync::OnceCell;
 use postgres_client::tls::MakeTlsConnect;
 use postgres_client::{CancelToken, RawConnection};
 use postgres_protocol::message::backend::NoticeResponseBody;
 use pq_proto::StartupMessageParams;
-use rustls::client::danger::ServerCertVerifier;
-use rustls::crypto::ring;
 use rustls::pki_types::InvalidDnsNameError;
 use thiserror::Error;
 use tokio::net::TcpStream;
@@ -19,14 +15,15 @@ use tracing::{debug, error, info, warn};

 use crate::auth::parse_endpoint_param;
 use crate::cancellation::CancelClosure;
+use crate::config::ComputeConfig;
 use crate::context::RequestContext;
 use crate::control_plane::client::ApiLockError;
 use crate::control_plane::errors::WakeComputeError;
 use crate::control_plane::messages::MetricsAuxInfo;
 use crate::error::{ReportableError, UserFacingError};
 use crate::metrics::{Metrics, NumDbConnectionsGuard};
-use crate::postgres_rustls::MakeRustlsConnect;
 use crate::proxy::neon_option;
+use crate::tls::postgres_rustls::MakeRustlsConnect;
 use crate::types::Host;

 pub const COULD_NOT_CONNECT: &str = "Couldn't connect to compute node";
@@ -41,9 +38,6 @@ pub(crate) enum ConnectionError {
    #[error("{COULD_NOT_CONNECT}: {0}")]
    CouldNotConnect(#[from] io::Error),

-    #[error("Couldn't load native TLS certificates: {0:?}")]
-    TlsCertificateError(Vec<rustls_native_certs::Error>),
-
    #[error("{COULD_NOT_CONNECT}: {0}")]
    TlsError(#[from] InvalidDnsNameError),

@@ -90,7 +84,6 @@ impl ReportableError for ConnectionError {
            }
            ConnectionError::Postgres(_) => crate::error::ErrorKind::Compute,
            ConnectionError::CouldNotConnect(_) => crate::error::ErrorKind::Compute,
-            ConnectionError::TlsCertificateError(_) => crate::error::ErrorKind::Service,
            ConnectionError::TlsError(_) => crate::error::ErrorKind::Compute,
            ConnectionError::WakeComputeError(e) => e.get_error_kind(),
            ConnectionError::TooManyConnectionAttempts(e) => e.get_error_kind(),
@@ -200,11 +193,15 @@ impl ConnCfg {

        let connect_once = |host, port| {
            debug!("trying to connect to compute node at {host}:{port}");
-            connect_with_timeout(host, port).and_then(|socket| async {
-                let socket_addr = socket.peer_addr()?;
+            connect_with_timeout(host, port).and_then(|stream| async {
+                let socket_addr = stream.peer_addr()?;
+                let socket = socket2::SockRef::from(&stream);
+                // Disable Nagle's algorithm to not introduce latency between
+                // client and compute.
+                socket.set_nodelay(true)?;
                // This prevents load balancer from severing the connection.
-                socket2::SockRef::from(&socket).set_keepalive(true)?;
-                Ok((socket_addr, socket))
+                socket.set_keepalive(true)?;
+                Ok((socket_addr, stream))
            })
        };

@@ -251,35 +248,14 @@ impl ConnCfg {
    pub(crate) async fn connect(
        &self,
        ctx: &RequestContext,
-        allow_self_signed_compute: bool,
        aux: MetricsAuxInfo,
-        timeout: Duration,
+        config: &ComputeConfig,
    ) -> Result<PostgresConnection, ConnectionError> {
        let pause = ctx.latency_timer_pause(crate::metrics::Waiting::Compute);
-        let (socket_addr, stream, host) = self.connect_raw(timeout).await?;
+        let (socket_addr, stream, host) = self.connect_raw(config.timeout).await?;
        drop(pause);

-        let client_config = if allow_self_signed_compute {
-            // Allow all certificates for creating the connection
-            let verifier = Arc::new(AcceptEverythingVerifier);
-            rustls::ClientConfig::builder_with_provider(Arc::new(ring::default_provider()))
-                .with_safe_default_protocol_versions()
-                .expect("ring should support the default protocol versions")
-                .dangerous()
-                .with_custom_certificate_verifier(verifier)
-        } else {
-            let root_store = TLS_ROOTS
-                .get_or_try_init(load_certs)
-                .map_err(ConnectionError::TlsCertificateError)?
-                .clone();
-            rustls::ClientConfig::builder_with_provider(Arc::new(ring::default_provider()))
-                .with_safe_default_protocol_versions()
-                .expect("ring should support the default protocol versions")
-                .with_root_certificates(root_store)
-        };
-        let client_config = client_config.with_no_client_auth();
-
-        let mut mk_tls = crate::postgres_rustls::MakeRustlsConnect::new(client_config);
+        let mut mk_tls = crate::tls::postgres_rustls::MakeRustlsConnect::new(config.tls.clone());
        let tls = <MakeRustlsConnect as MakeTlsConnect<tokio::net::TcpStream>>::make_tls_connect(
            &mut mk_tls,
            host,
@@ -319,6 +295,7 @@ impl ConnCfg {
                secret_key,
            },
            vec![],
+            host.to_string(),
        );

        let connection = PostgresConnection {
@@ -350,63 +327,6 @@ fn filtered_options(options: &str) -> Option<String> {
    Some(options)
 }

-fn load_certs() -> Result<Arc<rustls::RootCertStore>, Vec<rustls_native_certs::Error>> {
-    let der_certs = rustls_native_certs::load_native_certs();
-
-    if !der_certs.errors.is_empty() {
-        return Err(der_certs.errors);
-    }
-
-    let mut store = rustls::RootCertStore::empty();
-    store.add_parsable_certificates(der_certs.certs);
-    Ok(Arc::new(store))
-}
-static TLS_ROOTS: OnceCell<Arc<rustls::RootCertStore>> = OnceCell::new();
-
-#[derive(Debug)]
-struct AcceptEverythingVerifier;
-impl ServerCertVerifier for AcceptEverythingVerifier {
-    fn supported_verify_schemes(&self) -> Vec<rustls::SignatureScheme> {
-        use rustls::SignatureScheme;
-        // The schemes for which `SignatureScheme::supported_in_tls13` returns true.
-        vec![
-            SignatureScheme::ECDSA_NISTP521_SHA512,
-            SignatureScheme::ECDSA_NISTP384_SHA384,
-            SignatureScheme::ECDSA_NISTP256_SHA256,
-            SignatureScheme::RSA_PSS_SHA512,
-            SignatureScheme::RSA_PSS_SHA384,
-            SignatureScheme::RSA_PSS_SHA256,
-            SignatureScheme::ED25519,
-        ]
-    }
-    fn verify_server_cert(
-        &self,
-        _end_entity: &rustls::pki_types::CertificateDer<'_>,
-        _intermediates: &[rustls::pki_types::CertificateDer<'_>],
-        _server_name: &rustls::pki_types::ServerName<'_>,
-        _ocsp_response: &[u8],
-        _now: rustls::pki_types::UnixTime,
-    ) -> Result<rustls::client::danger::ServerCertVerified, rustls::Error> {
-        Ok(rustls::client::danger::ServerCertVerified::assertion())
-    }
-    fn verify_tls12_signature(
-        &self,
-        _message: &[u8],
-        _cert: &rustls::pki_types::CertificateDer<'_>,
-        _dss: &rustls::DigitallySignedStruct,
-    ) -> Result<rustls::client::danger::HandshakeSignatureValid, rustls::Error> {
-        Ok(rustls::client::danger::HandshakeSignatureValid::assertion())
-    }
-    fn verify_tls13_signature(
-        &self,
-        _message: &[u8],
-        _cert: &rustls::pki_types::CertificateDer<'_>,
-        _dss: &rustls::DigitallySignedStruct,
-    ) -> Result<rustls::client::danger::HandshakeSignatureValid, rustls::Error> {
-        Ok(rustls::client::danger::HandshakeSignatureValid::assertion())
-    }
-}
-
 #[cfg(test)]
 mod tests {
    use super::*;
--- a/proxy/src/config.rs
+++ b/proxy/src/config.rs
@@ -1,17 +1,10 @@
-use std::collections::{HashMap, HashSet};
 use std::str::FromStr;
 use std::sync::Arc;
 use std::time::Duration;

 use anyhow::{bail, ensure, Context, Ok};
 use clap::ValueEnum;
-use itertools::Itertools;
 use remote_storage::RemoteStorageConfig;
-use rustls::crypto::ring::{self, sign};
-use rustls::pki_types::{CertificateDer, PrivateKeyDer};
-use sha2::{Digest, Sha256};
-use tracing::{error, info};
-use x509_parser::oid_registry;

 use crate::auth::backend::jwt::JwkCache;
 use crate::auth::backend::AuthRateLimiter;
@@ -20,12 +13,12 @@ use crate::rate_limiter::{RateBucketInfo, RateLimitAlgorithm, RateLimiterConfig}
 use crate::scram::threadpool::ThreadPool;
 use crate::serverless::cancel_set::CancelSet;
 use crate::serverless::GlobalConnPoolOptions;
+pub use crate::tls::server_config::{configure_tls, TlsConfig};
 use crate::types::Host;

 pub struct ProxyConfig {
    pub tls_config: Option<TlsConfig>,
    pub metric_collection: Option<MetricCollectionConfig>,
-    pub allow_self_signed_compute: bool,
    pub http_config: HttpConfig,
    pub authentication_config: AuthenticationConfig,
    pub proxy_protocol_v2: ProxyProtocolV2,
@@ -33,7 +26,13 @@ pub struct ProxyConfig {
    pub handshake_timeout: Duration,
    pub wake_compute_retry_config: RetryConfig,
    pub connect_compute_locks: ApiLocks<Host>,
-    pub connect_to_compute_retry_config: RetryConfig,
+    pub connect_to_compute: ComputeConfig,
+}
+
+pub struct ComputeConfig {
+    pub retry: RetryConfig,
+    pub tls: Arc<rustls::ClientConfig>,
+    pub timeout: Duration,
 }

 #[derive(Copy, Clone, Debug, ValueEnum, PartialEq)]
@@ -53,12 +52,6 @@ pub struct MetricCollectionConfig {
    pub backup_metric_collection_config: MetricBackupCollectionConfig,
 }

-pub struct TlsConfig {
-    pub config: Arc<rustls::ServerConfig>,
-    pub common_names: HashSet<String>,
-    pub cert_resolver: Arc<CertResolver>,
-}
-
 pub struct HttpConfig {
    pub accept_websockets: bool,
    pub pool_options: GlobalConnPoolOptions,
@@ -81,272 +74,6 @@ pub struct AuthenticationConfig {
    pub console_redirect_confirmation_timeout: tokio::time::Duration,
 }

-impl TlsConfig {
-    pub fn to_server_config(&self) -> Arc<rustls::ServerConfig> {
-        self.config.clone()
-    }
-}
-
-/// <https://github.com/postgres/postgres/blob/ca481d3c9ab7bf69ff0c8d71ad3951d407f6a33c/src/include/libpq/pqcomm.h#L159>
-pub const PG_ALPN_PROTOCOL: &[u8] = b"postgresql";
-
-/// Configure TLS for the main endpoint.
-pub fn configure_tls(
-    key_path: &str,
-    cert_path: &str,
-    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)?;
-
-    // add extra certificates
-    if let Some(certs_dir) = certs_dir {
-        for entry in std::fs::read_dir(certs_dir)? {
-            let entry = entry?;
-            let path = entry.path();
-            if path.is_dir() {
-                // file names aligned with default cert-manager names
-                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,
-                    )?;
-                }
-            }
-        }
-    }
-
-    let common_names = cert_resolver.get_common_names();
-
-    let cert_resolver = Arc::new(cert_resolver);
-
-    // allow TLS 1.2 to be compatible with older client libraries
-    let mut config =
-        rustls::ServerConfig::builder_with_provider(Arc::new(ring::default_provider()))
-            .with_protocol_versions(&[&rustls::version::TLS13, &rustls::version::TLS12])
-            .context("ring should support TLS1.2 and TLS1.3")?
-            .with_no_client_auth()
-            .with_cert_resolver(cert_resolver.clone());
-
-    config.alpn_protocols = vec![PG_ALPN_PROTOCOL.to_vec()];
-
-    if allow_tls_keylogfile {
-        // KeyLogFile will check for the SSLKEYLOGFILE environment variable.
-        config.key_log = Arc::new(rustls::KeyLogFile::new());
-    }
-
-    Ok(TlsConfig {
-        config: Arc::new(config),
-        common_names,
-        cert_resolver,
-    })
-}
-
-/// Channel binding parameter
-///
-/// <https://www.rfc-editor.org/rfc/rfc5929#section-4>
-/// Description: The hash of the TLS server's certificate as it
-/// appears, octet for octet, in the server's Certificate message.  Note
-/// that the Certificate message contains a certificate_list, in which
-/// the first element is the server's certificate.
-///
-/// The hash function is to be selected as follows:
-///
-/// * if the certificate's signatureAlgorithm uses a single hash
-///   function, and that hash function is either MD5 or SHA-1, then use SHA-256;
-///
-/// * if the certificate's signatureAlgorithm uses a single hash
-///   function and that hash function neither MD5 nor SHA-1, then use
-///   the hash function associated with the certificate's
-///   signatureAlgorithm;
-///
-/// * if the certificate's signatureAlgorithm uses no hash functions or
-///   uses multiple hash functions, then this channel binding type's
-///   channel bindings are undefined at this time (updates to is channel
-///   binding type may occur to address this issue if it ever arises).
-#[derive(Debug, Clone, Copy)]
-pub enum TlsServerEndPoint {
-    Sha256([u8; 32]),
-    Undefined,
-}
-
-impl TlsServerEndPoint {
-    pub fn new(cert: &CertificateDer<'_>) -> anyhow::Result<Self> {
-        let sha256_oids = [
-            // I'm explicitly not adding MD5 or SHA1 here... They're bad.
-            oid_registry::OID_SIG_ECDSA_WITH_SHA256,
-            oid_registry::OID_PKCS1_SHA256WITHRSA,
-        ];
-
-        let pem = x509_parser::parse_x509_certificate(cert)
-            .context("Failed to parse PEM object from cerficiate")?
-            .1;
-
-        info!(subject = %pem.subject, "parsing TLS certificate");
-
-        let reg = oid_registry::OidRegistry::default().with_all_crypto();
-        let oid = pem.signature_algorithm.oid();
-        let alg = reg.get(oid);
-        if sha256_oids.contains(oid) {
-            let tls_server_end_point: [u8; 32] = Sha256::new().chain_update(cert).finalize().into();
-            info!(subject = %pem.subject, signature_algorithm = alg.map(|a| a.description()), tls_server_end_point = %base64::encode(tls_server_end_point), "determined channel binding");
-            Ok(Self::Sha256(tls_server_end_point))
-        } else {
-            error!(subject = %pem.subject, signature_algorithm = alg.map(|a| a.description()), "unknown channel binding");
-            Ok(Self::Undefined)
-        }
-    }
-
-    pub fn supported(&self) -> bool {
-        !matches!(self, TlsServerEndPoint::Undefined)
-    }
-}
-
-#[derive(Default, Debug)]
-pub struct CertResolver {
-    certs: HashMap<String, (Arc<rustls::sign::CertifiedKey>, TlsServerEndPoint)>,
-    default: Option<(Arc<rustls::sign::CertifiedKey>, TlsServerEndPoint)>,
-}
-
-impl CertResolver {
-    pub fn new() -> Self {
-        Self::default()
-    }
-
-    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}'"))?
-        };
-
-        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)
-    }
-
-    pub 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 pem = x509_parser::parse_x509_certificate(first_cert)
-            .context("Failed to parse PEM object from cerficiate")?
-            .1;
-
-        let common_name = pem.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));
-        }
-
-        self.certs.insert(common_name, (cert, tls_server_end_point));
-
-        Ok(())
-    }
-
-    pub fn get_common_names(&self) -> HashSet<String> {
-        self.certs.keys().map(|s| s.to_string()).collect()
-    }
-}
-
-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)
-    }
-}
-
-impl CertResolver {
-    pub fn resolve(
-        &self,
-        server_name: Option<&str>,
-    ) -> Option<(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.
-        //
-        // With the current coding foo.com will match *.foo.com and that
-        // repeats behavior of the old code.
-        if let Some(mut sni_name) = server_name {
-            loop {
-                if let Some(cert) = self.certs.get(sni_name) {
-                    return Some(cert.clone());
-                }
-                if let Some((_, rest)) = sni_name.split_once('.') {
-                    sni_name = rest;
-                } else {
-                    return None;
-                }
-            }
-        } else {
-            // No SNI, use the default certificate, otherwise we can't get to
-            // options parameter which can be used to set endpoint name too.
-            // That means that non-SNI flow will not work for CNAME domains in
-            // verify-full mode.
-            //
-            // If that will be a problem we can:
-            //
-            // a) Instead of multi-cert approach use single cert with extra
-            //    domains listed in Subject Alternative Name (SAN).
-            // b) Deploy separate proxy instances for extra domains.
-            self.default.clone()
-        }
-    }
-}
-
 #[derive(Debug)]
 pub struct EndpointCacheConfig {
    /// Batch size to receive all endpoints on the startup.
--- a/proxy/src/conn.rs
+++ b/proxy/src/conn.rs
@@ -0,0 +1,221 @@
+use std::future::{poll_fn, Future};
+use std::io;
+use std::net::SocketAddr;
+use std::pin::Pin;
+use std::task::{Context, Poll};
+
+use tokio::net::{TcpListener, TcpStream, ToSocketAddrs};
+
+pub trait Acceptor {
+    type Connection: AsyncRead + AsyncWrite + Send + Unpin + 'static;
+    type Error: std::error::Error + Send + Sync + 'static;
+
+    #[inline]
+    fn poll_ready(&self, cx: &mut Context<'_>) -> Poll<Result<(), Self::Error>> {
+        let _ = cx;
+        Poll::Ready(Ok(()))
+    }
+
+    fn accept(
+        &self,
+    ) -> impl Future<Output = Result<(Self::Connection, SocketAddr), Self::Error>> + Send;
+}
+
+pub trait Connector {
+    type Connection: AsyncRead + AsyncWrite + Send + Unpin + 'static;
+    type Error: std::error::Error + Send + Sync + 'static;
+
+    #[inline]
+    fn poll_ready(&self, cx: &mut Context<'_>) -> Poll<Result<(), Self::Error>> {
+        let _ = cx;
+        Poll::Ready(Ok(()))
+    }
+
+    fn connect(
+        &self,
+        addr: SocketAddr,
+    ) -> impl Future<Output = Result<Self::Connection, Self::Error>> + Send;
+}
+
+pub struct TokioTcpAcceptor {
+    listener: TcpListener,
+    tcp_nodelay: Option<bool>,
+    tcp_keepalive: Option<bool>,
+}
+
+impl TokioTcpAcceptor {
+    pub async fn bind<A: ToSocketAddrs>(addr: A) -> io::Result<Self> {
+        let listener = TcpListener::bind(addr).await?;
+        // When set for the server socket, the keepalive setting
+        // will be inherited by all accepted client sockets.
+        socket2::SockRef::from(&listener).set_keepalive(true)?;
+        Ok(Self {
+            listener,
+            tcp_nodelay: Some(true),
+            tcp_keepalive: None,
+        })
+    }
+
+    pub fn into_std(self) -> io::Result<std::net::TcpListener> {
+        self.listener.into_std()
+    }
+}
+
+impl Acceptor for TokioTcpAcceptor {
+    type Connection = TcpStream;
+    type Error = io::Error;
+
+    fn accept(&self) -> impl Future<Output = Result<(Self::Connection, SocketAddr), Self::Error>> {
+        async move {
+            let (stream, addr) = self.listener.accept().await?;
+
+            let socket = socket2::SockRef::from(&stream);
+            if let Some(nodelay) = self.tcp_nodelay {
+                socket.set_nodelay(nodelay)?;
+            }
+            if let Some(keepalive) = self.tcp_keepalive {
+                socket.set_keepalive(keepalive)?;
+            }
+
+            Ok((stream, addr))
+        }
+    }
+}
+
+pub struct TokioTcpConnector;
+
+impl Connector for TokioTcpConnector {
+    type Connection = TcpStream;
+    type Error = io::Error;
+
+    fn connect(
+        &self,
+        addr: SocketAddr,
+    ) -> impl Future<Output = Result<Self::Connection, Self::Error>> {
+        async move {
+            let socket = TcpStream::connect(addr).await?;
+            socket.set_nodelay(true)?;
+            Ok(socket)
+        }
+    }
+}
+
+pub trait Stream: AsyncRead + AsyncWrite + Send + Unpin + 'static {}
+
+impl<T: AsyncRead + AsyncWrite + Send + Unpin + 'static> Stream for T {}
+
+pub trait AsyncRead {
+    fn readable(&self) -> impl Future<Output = io::Result<()>> + Send
+    where
+        Self: Send + Sync,
+    {
+        poll_fn(move |cx| self.poll_read_ready(cx))
+    }
+
+    fn poll_read_ready(&self, cx: &mut Context<'_>) -> Poll<io::Result<()>>;
+
+    fn poll_read(
+        self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        buf: &mut [u8],
+    ) -> Poll<io::Result<usize>>;
+
+    fn poll_read_vectored(
+        self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        bufs: &mut [io::IoSliceMut<'_>],
+    ) -> Poll<io::Result<usize>>;
+}
+
+pub trait AsyncWrite {
+    fn writable(&self) -> impl Future<Output = io::Result<()>> + Send
+    where
+        Self: Send + Sync,
+    {
+        poll_fn(move |cx| self.poll_write_ready(cx))
+    }
+
+    fn poll_write_ready(&self, cx: &mut Context<'_>) -> Poll<io::Result<()>>;
+
+    fn poll_write(
+        self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        buf: &[u8],
+    ) -> Poll<io::Result<usize>>;
+
+    fn poll_write_vectored(
+        self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        bufs: &[io::IoSlice<'_>],
+    ) -> Poll<io::Result<usize>>;
+
+    fn poll_flush(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>>;
+
+    fn poll_close(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>>;
+}
+
+impl AsyncRead for tokio::net::TcpStream {
+    fn poll_read_ready(&self, cx: &mut Context<'_>) -> Poll<io::Result<()>> {
+        tokio::net::TcpStream::poll_read_ready(self, cx)
+    }
+
+    fn poll_read(
+        mut self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        buf: &mut [u8],
+    ) -> Poll<io::Result<usize>> {
+        match tokio::net::TcpStream::try_read(Pin::new(&mut *self).get_mut(), buf) {
+            Ok(n) => Poll::Ready(Ok(n)),
+            Err(e) if e.kind() == io::ErrorKind::WouldBlock => {
+                cx.waker().wake_by_ref();
+                Poll::Pending
+            }
+            Err(e) => Poll::Ready(Err(e)),
+        }
+    }
+
+    fn poll_read_vectored(
+        mut self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        bufs: &mut [io::IoSliceMut<'_>],
+    ) -> Poll<io::Result<usize>> {
+        match tokio::net::TcpStream::try_read_vectored(Pin::new(&mut *self).get_mut(), bufs) {
+            Ok(n) => Poll::Ready(Ok(n)),
+            Err(e) if e.kind() == io::ErrorKind::WouldBlock => {
+                cx.waker().wake_by_ref();
+                Poll::Pending
+            }
+            Err(e) => Poll::Ready(Err(e)),
+        }
+    }
+}
+
+impl AsyncWrite for tokio::net::TcpStream {
+    fn poll_write_ready(&self, cx: &mut Context<'_>) -> Poll<io::Result<()>> {
+        tokio::net::TcpStream::poll_write_ready(self, cx)
+    }
+
+    fn poll_write(
+        self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        buf: &[u8],
+    ) -> Poll<io::Result<usize>> {
+        <Self as tokio::io::AsyncWrite>::poll_write(self, cx, buf)
+    }
+
+    fn poll_write_vectored(
+        self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        bufs: &[io::IoSlice<'_>],
+    ) -> Poll<io::Result<usize>> {
+        <Self as tokio::io::AsyncWrite>::poll_write_vectored(self, cx, bufs)
+    }
+
+    fn poll_flush(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>> {
+        <Self as tokio::io::AsyncWrite>::poll_flush(self, cx)
+    }
+
+    fn poll_close(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>> {
+        <Self as tokio::io::AsyncWrite>::poll_shutdown(self, cx)
+    }
+}
--- a/proxy/src/console_redirect_proxy.rs
+++ b/proxy/src/console_redirect_proxy.rs
@@ -8,6 +8,7 @@ use tracing::{debug, error, info, Instrument};
 use crate::auth::backend::ConsoleRedirectBackend;
 use crate::cancellation::{CancellationHandlerMain, CancellationHandlerMainInternal};
 use crate::config::{ProxyConfig, ProxyProtocolV2};
+use crate::conn::{Acceptor, TokioTcpAcceptor};
 use crate::context::RequestContext;
 use crate::error::ReportableError;
 use crate::metrics::{Metrics, NumClientConnectionsGuard};
@@ -22,7 +23,7 @@ use crate::proxy::{
 pub async fn task_main(
    config: &'static ProxyConfig,
    backend: &'static ConsoleRedirectBackend,
-    listener: tokio::net::TcpListener,
+    acceptor: TokioTcpAcceptor,
    cancellation_token: CancellationToken,
    cancellation_handler: Arc<CancellationHandlerMain>,
 ) -> anyhow::Result<()> {
@@ -30,15 +31,11 @@ pub async fn task_main(
        info!("proxy has shut down");
    }

-    // When set for the server socket, the keepalive setting
-    // will be inherited by all accepted client sockets.
-    socket2::SockRef::from(&listener).set_keepalive(true)?;
-
    let connections = tokio_util::task::task_tracker::TaskTracker::new();
    let cancellations = tokio_util::task::task_tracker::TaskTracker::new();

    while let Some(accept_result) =
-        run_until_cancelled(listener.accept(), &cancellation_token).await
+        run_until_cancelled(acceptor.accept(), &cancellation_token).await
    {
        let (socket, peer_addr) = accept_result?;

@@ -115,7 +112,7 @@ pub async fn task_main(
                Ok(Some(p)) => {
                    ctx.set_success();
                    let _disconnect = ctx.log_connect();
-                    match p.proxy_pass().await {
+                    match p.proxy_pass(&config.connect_to_compute).await {
                        Ok(()) => {}
                        Err(ErrorSource::Client(e)) => {
                            error!(?session_id, "per-client task finished with an IO error from the client: {e:#}");
@@ -131,7 +128,7 @@ pub async fn task_main(

    connections.close();
    cancellations.close();
-    drop(listener);
+    drop(acceptor);

    // Drain connections
    connections.wait().await;
@@ -213,11 +210,10 @@ pub(crate) async fn handle_client<S: AsyncRead + AsyncWrite + Unpin>(
            params_compat: true,
            params: &params,
            locks: &config.connect_compute_locks,
-            allow_self_signed_compute: config.allow_self_signed_compute,
        },
        &user_info,
        config.wake_compute_retry_config,
-        config.connect_to_compute_retry_config,
+        &config.connect_to_compute,
    )
    .or_else(|e| stream.throw_error(e))
    .await?;
--- a/proxy/src/control_plane/client/mock.rs
+++ b/proxy/src/control_plane/client/mock.rs
@@ -74,39 +74,44 @@ impl MockControlPlane {

            tokio::spawn(connection);

-            let secret = {
-                let query = "select rolpassword from pg_catalog.pg_authid where rolname = $1";
-                if let Some(row) = client.query_opt(query, &[&&*user_info.user]).await? {
-                    let entry: String = row.get("rolpassword");
-
-                    info!("got a secret: {entry}"); // safe since it's not a prod scenario
-                    let secret = scram::ServerSecret::parse(&entry).map(AuthSecret::Scram);
-                    secret.or_else(|| parse_md5(&entry).map(AuthSecret::Md5))
-                } else {
-                    warn!("user '{}' does not exist", user_info.user);
-                    None
-                }
+            let secret = if let Some(entry) = get_execute_postgres_query(
+                &client,
+                "select rolpassword from pg_catalog.pg_authid where rolname = $1",
+                &[&&*user_info.user],
+                "rolpassword",
+            )
+            .await?
+            {
+                info!("got a secret: {entry}"); // safe since it's not a prod scenario
+                let secret = scram::ServerSecret::parse(&entry).map(AuthSecret::Scram);
+                secret.or_else(|| parse_md5(&entry).map(AuthSecret::Md5))
+            } else {
+                warn!("user '{}' does not exist", user_info.user);
+                None
            };

-            let mut allowed_ips = vec![];
-            if self.ip_allowlist_check_enabled {
-                let query =
-                    "select allowed_ips from neon_control_plane.endpoints where endpoint_id = $1";
-                let row = client
-                    .query_one(query, &[&user_info.endpoint.as_str()])
-                    .await?;
-
-                let s: Option<String> = row.get("allowed_ips");
-                if let Some(s) = s {
-                    info!("got allowed_ips: {s}");
-                    allowed_ips = s
-                        .split(',')
-                        .map(|s| {
-                            IpPattern::from_str(s).expect("mocked ip pattern should be correct")
-                        })
-                        .collect();
+            let allowed_ips = if self.ip_allowlist_check_enabled {
+                match get_execute_postgres_query(
+                    &client,
+                    "select allowed_ips from neon_control_plane.endpoints where endpoint_id = $1",
+                    &[&user_info.endpoint.as_str()],
+                    "allowed_ips",
+                )
+                .await?
+                {
+                    Some(s) => {
+                        info!("got allowed_ips: {s}");
+                        s.split(',')
+                            .map(|s| {
+                                IpPattern::from_str(s).expect("mocked ip pattern should be correct")
+                            })
+                            .collect()
+                    }
+                    None => vec![],
                }
-            }
+            } else {
+                vec![]
+            };

            Ok((secret, allowed_ips))
        }
@@ -129,8 +134,11 @@ impl MockControlPlane {

        let connection = tokio::spawn(connection);

-        let query = "select id, jwks_url, audience, role_names from neon_control_plane.endpoint_jwks where endpoint_id = $1";
-        let res = client.query(query, &[&endpoint.as_str()]).await?;
+        let res = client.query(
+                "select id, jwks_url, audience, role_names from neon_control_plane.endpoint_jwks where endpoint_id = $1",
+                &[&endpoint.as_str()],
+            )
+            .await?;

        let mut rows = vec![];
        for row in res {
@@ -153,23 +161,11 @@ impl MockControlPlane {
        Ok(rows)
    }

-    async fn do_wake_compute(
-        &self,
-        user_info: &ComputeUserInfo,
-    ) -> Result<NodeInfo, WakeComputeError> {
-        let (client, connection) =
-            tokio_postgres::connect(self.endpoint.as_str(), tokio_postgres::NoTls).await?;
-        tokio::spawn(connection);
-
-        let query = "select host, port from neon_control_plane.endpoints where endpoint_id = $1";
-        let row = client
-            .query_one(query, &[&user_info.endpoint.as_str()])
-            .await?;
-
-        let host: String = row.get("host");
-        let port: i32 = row.get("port");
-
-        let mut config = compute::ConnCfg::new(host, port as u16);
+    async fn do_wake_compute(&self) -> Result<NodeInfo, WakeComputeError> {
+        let mut config = compute::ConnCfg::new(
+            self.endpoint.host_str().unwrap_or("localhost").to_owned(),
+            self.endpoint.port().unwrap_or(5432),
+        );
        config.ssl_mode(postgres_client::config::SslMode::Disable);

        let node = NodeInfo {
@@ -186,6 +182,26 @@ impl MockControlPlane {
    }
 }

+async fn get_execute_postgres_query(
+    client: &Client,
+    query: &str,
+    params: &[&(dyn tokio_postgres::types::ToSql + Sync)],
+    idx: &str,
+) -> Result<Option<String>, GetAuthInfoError> {
+    let rows = client.query(query, params).await?;
+
+    // We can get at most one row, because `rolname` is unique.
+    let Some(row) = rows.first() else {
+        // This means that the user doesn't exist, so there can be no secret.
+        // However, this is still a *valid* outcome which is very similar
+        // to getting `404 Not found` from the Neon console.
+        return Ok(None);
+    };
+
+    let entry = row.try_get(idx).map_err(MockApiError::PasswordNotSet)?;
+    Ok(Some(entry))
+}
+
 impl super::ControlPlaneApi for MockControlPlane {
    #[tracing::instrument(skip_all)]
    async fn get_role_secret(
@@ -223,11 +239,9 @@ impl super::ControlPlaneApi for MockControlPlane {
    async fn wake_compute(
        &self,
        _ctx: &RequestContext,
-        user_info: &ComputeUserInfo,
+        _user_info: &ComputeUserInfo,
    ) -> Result<CachedNodeInfo, WakeComputeError> {
-        self.do_wake_compute(user_info)
-            .map_ok(Cached::new_uncached)
-            .await
+        self.do_wake_compute().map_ok(Cached::new_uncached).await
    }
 }

--- a/proxy/src/control_plane/mgmt.rs
+++ b/proxy/src/control_plane/mgmt.rs
@@ -4,10 +4,11 @@ use anyhow::Context;
 use once_cell::sync::Lazy;
 use postgres_backend::{AuthType, PostgresBackend, PostgresBackendTCP, QueryError};
 use pq_proto::{BeMessage, SINGLE_COL_ROWDESC};
-use tokio::net::{TcpListener, TcpStream};
+use tokio::net::TcpStream;
 use tokio_util::sync::CancellationToken;
 use tracing::{error, info, info_span, Instrument};

+use crate::conn::{Acceptor, TokioTcpAcceptor};
 use crate::control_plane::messages::{DatabaseInfo, KickSession};
 use crate::waiters::{self, Waiter, Waiters};

@@ -26,19 +27,15 @@ pub(crate) fn notify(psql_session_id: &str, msg: ComputeReady) -> Result<(), wai

 /// Management API listener task.
 /// It spawns management response handlers needed for the console redirect auth flow.
-pub async fn task_main(listener: TcpListener) -> anyhow::Result<Infallible> {
+pub async fn task_main(acceptor: TokioTcpAcceptor) -> anyhow::Result<Infallible> {
    scopeguard::defer! {
        info!("mgmt has shut down");
    }

    loop {
-        let (socket, peer_addr) = listener.accept().await?;
+        let (socket, peer_addr) = acceptor.accept().await?;
        info!("accepted connection from {peer_addr}");

-        socket
-            .set_nodelay(true)
-            .context("failed to set client socket option")?;
-
        let span = info_span!("mgmt", peer = %peer_addr);

        tokio::task::spawn(
--- a/proxy/src/control_plane/mod.rs
+++ b/proxy/src/control_plane/mod.rs
@@ -10,13 +10,13 @@ pub mod client;
 pub(crate) mod errors;

 use std::sync::Arc;
-use std::time::Duration;

 use crate::auth::backend::jwt::AuthRule;
 use crate::auth::backend::{ComputeCredentialKeys, ComputeUserInfo};
 use crate::auth::IpPattern;
 use crate::cache::project_info::ProjectInfoCacheImpl;
 use crate::cache::{Cached, TimedLru};
+use crate::config::ComputeConfig;
 use crate::context::RequestContext;
 use crate::control_plane::messages::{ControlPlaneErrorMessage, MetricsAuxInfo};
 use crate::intern::ProjectIdInt;
@@ -73,12 +73,9 @@ impl NodeInfo {
    pub(crate) async fn connect(
        &self,
        ctx: &RequestContext,
-        allow_self_signed_compute: bool,
-        timeout: Duration,
+        config: &ComputeConfig,
    ) -> Result<compute::PostgresConnection, compute::ConnectionError> {
-        self.config
-            .connect(ctx, allow_self_signed_compute, self.aux.clone(), timeout)
-            .await
+        self.config.connect(ctx, self.aux.clone(), config).await
    }

    pub(crate) fn reuse_settings(&mut self, other: Self) {
--- a/proxy/src/http/health_server.rs
+++ b/proxy/src/http/health_server.rs
@@ -1,5 +1,4 @@
 use std::convert::Infallible;
-use std::net::TcpListener;
 use std::sync::{Arc, Mutex};

 use anyhow::{anyhow, bail};
@@ -14,6 +13,7 @@ use utils::http::error::ApiError;
 use utils::http::json::json_response;
 use utils::http::{RouterBuilder, RouterService};

+use crate::conn::TokioTcpAcceptor;
 use crate::ext::{LockExt, TaskExt};
 use crate::jemalloc;

@@ -36,7 +36,7 @@ fn make_router(metrics: AppMetrics) -> RouterBuilder<hyper0::Body, ApiError> {
 }

 pub async fn task_main(
-    http_listener: TcpListener,
+    http_acceptor: TokioTcpAcceptor,
    metrics: AppMetrics,
 ) -> anyhow::Result<Infallible> {
    scopeguard::defer! {
@@ -45,7 +45,7 @@ pub async fn task_main(

    let service = || RouterService::new(make_router(metrics).build()?);

-    hyper0::Server::from_tcp(http_listener)?
+    hyper0::Server::from_tcp(http_acceptor.into_std()?)?
        .serve(service().map_err(|e| anyhow!(e))?)
        .await?;

--- a/proxy/src/lib.rs
+++ b/proxy/src/lib.rs
@@ -78,6 +78,7 @@ pub mod cancellation;
 pub mod compute;
 pub mod compute_ctl;
 pub mod config;
+pub mod conn;
 pub mod console_redirect_proxy;
 pub mod context;
 pub mod control_plane;
@@ -89,7 +90,6 @@ pub mod jemalloc;
 pub mod logging;
 pub mod metrics;
 pub mod parse;
-pub mod postgres_rustls;
 pub mod protocol2;
 pub mod proxy;
 pub mod rate_limiter;
@@ -99,6 +99,7 @@ pub mod scram;
 pub mod serverless;
 pub mod signals;
 pub mod stream;
+pub mod tls;
 pub mod types;
 pub mod url;
 pub mod usage_metrics;
--- a/proxy/src/proxy/connect_compute.rs
+++ b/proxy/src/proxy/connect_compute.rs
@@ -6,7 +6,7 @@ use tracing::{debug, info, warn};
 use super::retry::ShouldRetryWakeCompute;
 use crate::auth::backend::ComputeCredentialKeys;
 use crate::compute::{self, PostgresConnection, COULD_NOT_CONNECT};
-use crate::config::RetryConfig;
+use crate::config::{ComputeConfig, RetryConfig};
 use crate::context::RequestContext;
 use crate::control_plane::errors::WakeComputeError;
 use crate::control_plane::locks::ApiLocks;
@@ -19,8 +19,6 @@ use crate::proxy::retry::{retry_after, should_retry, CouldRetry};
 use crate::proxy::wake_compute::wake_compute;
 use crate::types::Host;

-const CONNECT_TIMEOUT: time::Duration = time::Duration::from_secs(2);
-
 /// 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.
@@ -49,7 +47,7 @@ pub(crate) trait ConnectMechanism {
        &self,
        ctx: &RequestContext,
        node_info: &control_plane::CachedNodeInfo,
-        timeout: time::Duration,
+        config: &ComputeConfig,
    ) -> Result<Self::Connection, Self::ConnectError>;

    fn update_connect_config(&self, conf: &mut compute::ConnCfg);
@@ -73,9 +71,6 @@ pub(crate) struct TcpMechanism<'a> {

    /// connect_to_compute concurrency lock
    pub(crate) locks: &'static ApiLocks<Host>,
-
-    /// Whether we should accept self-signed certificates (for testing)
-    pub(crate) allow_self_signed_compute: bool,
 }

 #[async_trait]
@@ -89,15 +84,11 @@ impl ConnectMechanism for TcpMechanism<'_> {
        &self,
        ctx: &RequestContext,
        node_info: &control_plane::CachedNodeInfo,
-        timeout: time::Duration,
+        config: &ComputeConfig,
    ) -> Result<PostgresConnection, Self::Error> {
        let host = node_info.config.get_host();
        let permit = self.locks.get_permit(&host).await?;
-        permit.release_result(
-            node_info
-                .connect(ctx, self.allow_self_signed_compute, timeout)
-                .await,
-        )
+        permit.release_result(node_info.connect(ctx, config).await)
    }

    fn update_connect_config(&self, config: &mut compute::ConnCfg) {
@@ -112,7 +103,7 @@ pub(crate) async fn connect_to_compute<M: ConnectMechanism, B: ComputeConnectBac
    mechanism: &M,
    user_info: &B,
    wake_compute_retry_config: RetryConfig,
-    connect_to_compute_retry_config: RetryConfig,
+    compute: &ComputeConfig,
 ) -> Result<M::Connection, M::Error>
 where
    M::ConnectError: CouldRetry + ShouldRetryWakeCompute + std::fmt::Debug,
@@ -126,10 +117,7 @@ where
    mechanism.update_connect_config(&mut node_info.config);

    // try once
-    let err = match mechanism
-        .connect_once(ctx, &node_info, CONNECT_TIMEOUT)
-        .await
-    {
+    let err = match mechanism.connect_once(ctx, &node_info, compute).await {
        Ok(res) => {
            ctx.success();
            Metrics::get().proxy.retries_metric.observe(
@@ -149,7 +137,7 @@ where
    let node_info = if !node_info.cached() || !err.should_retry_wake_compute() {
        // If we just recieved this from cplane and didn't get it from cache, we shouldn't retry.
        // Do not need to retrieve a new node_info, just return the old one.
-        if should_retry(&err, num_retries, connect_to_compute_retry_config) {
+        if should_retry(&err, num_retries, compute.retry) {
            Metrics::get().proxy.retries_metric.observe(
                RetriesMetricGroup {
                    outcome: ConnectOutcome::Failed,
@@ -179,10 +167,7 @@ where
    debug!("wake_compute success. attempting to connect");
    num_retries = 1;
    loop {
-        match mechanism
-            .connect_once(ctx, &node_info, CONNECT_TIMEOUT)
-            .await
-        {
+        match mechanism.connect_once(ctx, &node_info, compute).await {
            Ok(res) => {
                ctx.success();
                Metrics::get().proxy.retries_metric.observe(
@@ -197,7 +182,7 @@ where
                return Ok(res);
            }
            Err(e) => {
-                if !should_retry(&e, num_retries, connect_to_compute_retry_config) {
+                if !should_retry(&e, num_retries, compute.retry) {
                    // Don't log an error here, caller will print the error
                    Metrics::get().proxy.retries_metric.observe(
                        RetriesMetricGroup {
@@ -213,7 +198,7 @@ where
            }
        };

-        let wait_duration = retry_after(num_retries, connect_to_compute_retry_config);
+        let wait_duration = retry_after(num_retries, compute.retry);
        num_retries += 1;

        let pause = ctx.latency_timer_pause(crate::metrics::Waiting::RetryTimeout);
--- a/proxy/src/proxy/handshake.rs
+++ b/proxy/src/proxy/handshake.rs
@@ -8,12 +8,13 @@ use tokio::io::{AsyncRead, AsyncWrite};
 use tracing::{debug, info, warn};

 use crate::auth::endpoint_sni;
-use crate::config::{TlsConfig, PG_ALPN_PROTOCOL};
+use crate::config::TlsConfig;
 use crate::context::RequestContext;
 use crate::error::ReportableError;
 use crate::metrics::Metrics;
 use crate::proxy::ERR_INSECURE_CONNECTION;
 use crate::stream::{PqStream, Stream, StreamUpgradeError};
+use crate::tls::PG_ALPN_PROTOCOL;

 #[derive(Error, Debug)]
 pub(crate) enum HandshakeError {
--- a/proxy/src/proxy/mod.rs
+++ b/proxy/src/proxy/mod.rs
@@ -25,6 +25,7 @@ use self::connect_compute::{connect_to_compute, TcpMechanism};
 use self::passthrough::ProxyPassthrough;
 use crate::cancellation::{self, CancellationHandlerMain, CancellationHandlerMainInternal};
 use crate::config::{ProxyConfig, ProxyProtocolV2, TlsConfig};
+use crate::conn::{Acceptor, TokioTcpAcceptor};
 use crate::context::RequestContext;
 use crate::error::ReportableError;
 use crate::metrics::{Metrics, NumClientConnectionsGuard};
@@ -55,7 +56,7 @@ pub async fn run_until_cancelled<F: std::future::Future>(
 pub async fn task_main(
    config: &'static ProxyConfig,
    auth_backend: &'static auth::Backend<'static, ()>,
-    listener: tokio::net::TcpListener,
+    acceptor: TokioTcpAcceptor,
    cancellation_token: CancellationToken,
    cancellation_handler: Arc<CancellationHandlerMain>,
    endpoint_rate_limiter: Arc<EndpointRateLimiter>,
@@ -64,15 +65,11 @@ pub async fn task_main(
        info!("proxy has shut down");
    }

-    // When set for the server socket, the keepalive setting
-    // will be inherited by all accepted client sockets.
-    socket2::SockRef::from(&listener).set_keepalive(true)?;
-
    let connections = tokio_util::task::task_tracker::TaskTracker::new();
    let cancellations = tokio_util::task::task_tracker::TaskTracker::new();

    while let Some(accept_result) =
-        run_until_cancelled(listener.accept(), &cancellation_token).await
+        run_until_cancelled(acceptor.accept(), &cancellation_token).await
    {
        let (socket, peer_addr) = accept_result?;

@@ -152,7 +149,7 @@ pub async fn task_main(
                Ok(Some(p)) => {
                    ctx.set_success();
                    let _disconnect = ctx.log_connect();
-                    match p.proxy_pass().await {
+                    match p.proxy_pass(&config.connect_to_compute).await {
                        Ok(()) => {}
                        Err(ErrorSource::Client(e)) => {
                            warn!(?session_id, "per-client task finished with an IO error from the client: {e:#}");
@@ -168,7 +165,7 @@ pub async fn task_main(

    connections.close();
    cancellations.close();
-    drop(listener);
+    drop(acceptor);

    // Drain connections
    connections.wait().await;
@@ -348,12 +345,10 @@ pub(crate) async fn handle_client<S: AsyncRead + AsyncWrite + Unpin>(
            params_compat,
            params: &params,
            locks: &config.connect_compute_locks,
-            // only used for console redirect testing.
-            allow_self_signed_compute: false,
        },
        &user_info,
        config.wake_compute_retry_config,
-        config.connect_to_compute_retry_config,
+        &config.connect_to_compute,
    )
    .or_else(|e| stream.throw_error(e))
    .await?;
--- a/proxy/src/proxy/passthrough.rs
+++ b/proxy/src/proxy/passthrough.rs
@@ -5,6 +5,7 @@ use utils::measured_stream::MeasuredStream;
 use super::copy_bidirectional::ErrorSource;
 use crate::cancellation;
 use crate::compute::PostgresConnection;
+use crate::config::ComputeConfig;
 use crate::control_plane::messages::MetricsAuxInfo;
 use crate::metrics::{Direction, Metrics, NumClientConnectionsGuard, NumConnectionRequestsGuard};
 use crate::stream::Stream;
@@ -67,9 +68,17 @@ pub(crate) struct ProxyPassthrough<P, S> {
 }

 impl<P, S: AsyncRead + AsyncWrite + Unpin> ProxyPassthrough<P, S> {
-    pub(crate) async fn proxy_pass(self) -> Result<(), ErrorSource> {
+    pub(crate) async fn proxy_pass(
+        self,
+        compute_config: &ComputeConfig,
+    ) -> Result<(), ErrorSource> {
        let res = proxy_pass(self.client, self.compute.stream, self.aux).await;
-        if let Err(err) = self.compute.cancel_closure.try_cancel_query().await {
+        if let Err(err) = self
+            .compute
+            .cancel_closure
+            .try_cancel_query(compute_config)
+            .await
+        {
            tracing::warn!(session_id = ?self.session_id, ?err, "could not cancel the query in the database");
        }
        res
--- a/proxy/src/proxy/tests/mod.rs
+++ b/proxy/src/proxy/tests/mod.rs
@@ -22,14 +22,16 @@ use super::*;
 use crate::auth::backend::{
    ComputeCredentialKeys, ComputeCredentials, ComputeUserInfo, MaybeOwned,
 };
-use crate::config::{CertResolver, RetryConfig};
+use crate::config::{ComputeConfig, RetryConfig};
 use crate::control_plane::client::{ControlPlaneClient, TestControlPlaneClient};
 use crate::control_plane::messages::{ControlPlaneErrorMessage, Details, MetricsAuxInfo, Status};
 use crate::control_plane::{
    self, CachedAllowedIps, CachedNodeInfo, CachedRoleSecret, NodeInfo, NodeInfoCache,
 };
 use crate::error::ErrorKind;
-use crate::postgres_rustls::MakeRustlsConnect;
+use crate::tls::client_config::compute_client_config_with_certs;
+use crate::tls::postgres_rustls::MakeRustlsConnect;
+use crate::tls::server_config::CertResolver;
 use crate::types::{BranchId, EndpointId, ProjectId};
 use crate::{sasl, scram};

@@ -67,7 +69,7 @@ fn generate_certs(
 }

 struct ClientConfig<'a> {
-    config: rustls::ClientConfig,
+    config: Arc<rustls::ClientConfig>,
    hostname: &'a str,
 }

@@ -110,16 +112,7 @@ fn generate_tls_config<'a>(
    };

    let client_config = {
-        let config =
-            rustls::ClientConfig::builder_with_provider(Arc::new(ring::default_provider()))
-                .with_safe_default_protocol_versions()
-                .context("ring should support the default protocol versions")?
-                .with_root_certificates({
-                    let mut store = rustls::RootCertStore::empty();
-                    store.add(ca)?;
-                    store
-                })
-                .with_no_client_auth();
+        let config = Arc::new(compute_client_config_with_certs([ca]));

        ClientConfig { config, hostname }
    };
@@ -468,7 +461,7 @@ impl ConnectMechanism for TestConnectMechanism {
        &self,
        _ctx: &RequestContext,
        _node_info: &control_plane::CachedNodeInfo,
-        _timeout: std::time::Duration,
+        _config: &ComputeConfig,
    ) -> Result<Self::Connection, Self::ConnectError> {
        let mut counter = self.counter.lock().unwrap();
        let action = self.sequence[*counter];
@@ -576,6 +569,20 @@ fn helper_create_connect_info(
    user_info
 }

+fn config() -> ComputeConfig {
+    let retry = RetryConfig {
+        base_delay: Duration::from_secs(1),
+        max_retries: 5,
+        backoff_factor: 2.0,
+    };
+
+    ComputeConfig {
+        retry,
+        tls: Arc::new(compute_client_config_with_certs(std::iter::empty())),
+        timeout: Duration::from_secs(2),
+    }
+}
+
 #[tokio::test]
 async fn connect_to_compute_success() {
    let _ = env_logger::try_init();
@@ -583,12 +590,8 @@ async fn connect_to_compute_success() {
    let ctx = RequestContext::test();
    let mechanism = TestConnectMechanism::new(vec![Wake, Connect]);
    let user_info = helper_create_connect_info(&mechanism);
-    let config = RetryConfig {
-        base_delay: Duration::from_secs(1),
-        max_retries: 5,
-        backoff_factor: 2.0,
-    };
-    connect_to_compute(&ctx, &mechanism, &user_info, config, config)
+    let config = config();
+    connect_to_compute(&ctx, &mechanism, &user_info, config.retry, &config)
        .await
        .unwrap();
    mechanism.verify();
@@ -601,12 +604,8 @@ async fn connect_to_compute_retry() {
    let ctx = RequestContext::test();
    let mechanism = TestConnectMechanism::new(vec![Wake, Retry, Wake, Connect]);
    let user_info = helper_create_connect_info(&mechanism);
-    let config = RetryConfig {
-        base_delay: Duration::from_secs(1),
-        max_retries: 5,
-        backoff_factor: 2.0,
-    };
-    connect_to_compute(&ctx, &mechanism, &user_info, config, config)
+    let config = config();
+    connect_to_compute(&ctx, &mechanism, &user_info, config.retry, &config)
        .await
        .unwrap();
    mechanism.verify();
@@ -620,12 +619,8 @@ async fn connect_to_compute_non_retry_1() {
    let ctx = RequestContext::test();
    let mechanism = TestConnectMechanism::new(vec![Wake, Retry, Wake, Fail]);
    let user_info = helper_create_connect_info(&mechanism);
-    let config = RetryConfig {
-        base_delay: Duration::from_secs(1),
-        max_retries: 5,
-        backoff_factor: 2.0,
-    };
-    connect_to_compute(&ctx, &mechanism, &user_info, config, config)
+    let config = config();
+    connect_to_compute(&ctx, &mechanism, &user_info, config.retry, &config)
        .await
        .unwrap_err();
    mechanism.verify();
@@ -639,12 +634,8 @@ async fn connect_to_compute_non_retry_2() {
    let ctx = RequestContext::test();
    let mechanism = TestConnectMechanism::new(vec![Wake, Fail, Wake, Connect]);
    let user_info = helper_create_connect_info(&mechanism);
-    let config = RetryConfig {
-        base_delay: Duration::from_secs(1),
-        max_retries: 5,
-        backoff_factor: 2.0,
-    };
-    connect_to_compute(&ctx, &mechanism, &user_info, config, config)
+    let config = config();
+    connect_to_compute(&ctx, &mechanism, &user_info, config.retry, &config)
        .await
        .unwrap();
    mechanism.verify();
@@ -665,17 +656,13 @@ async fn connect_to_compute_non_retry_3() {
        max_retries: 1,
        backoff_factor: 2.0,
    };
-    let connect_to_compute_retry_config = RetryConfig {
-        base_delay: Duration::from_secs(1),
-        max_retries: 5,
-        backoff_factor: 2.0,
-    };
+    let config = config();
    connect_to_compute(
        &ctx,
        &mechanism,
        &user_info,
        wake_compute_retry_config,
-        connect_to_compute_retry_config,
+        &config,
    )
    .await
    .unwrap_err();
@@ -690,12 +677,8 @@ async fn wake_retry() {
    let ctx = RequestContext::test();
    let mechanism = TestConnectMechanism::new(vec![WakeRetry, Wake, Connect]);
    let user_info = helper_create_connect_info(&mechanism);
-    let config = RetryConfig {
-        base_delay: Duration::from_secs(1),
-        max_retries: 5,
-        backoff_factor: 2.0,
-    };
-    connect_to_compute(&ctx, &mechanism, &user_info, config, config)
+    let config = config();
+    connect_to_compute(&ctx, &mechanism, &user_info, config.retry, &config)
        .await
        .unwrap();
    mechanism.verify();
@@ -709,12 +692,8 @@ async fn wake_non_retry() {
    let ctx = RequestContext::test();
    let mechanism = TestConnectMechanism::new(vec![WakeRetry, WakeFail]);
    let user_info = helper_create_connect_info(&mechanism);
-    let config = RetryConfig {
-        base_delay: Duration::from_secs(1),
-        max_retries: 5,
-        backoff_factor: 2.0,
-    };
-    connect_to_compute(&ctx, &mechanism, &user_info, config, config)
+    let config = config();
+    connect_to_compute(&ctx, &mechanism, &user_info, config.retry, &config)
        .await
        .unwrap_err();
    mechanism.verify();
--- a/proxy/src/redis/notifications.rs
+++ b/proxy/src/redis/notifications.rs
@@ -12,6 +12,7 @@ use uuid::Uuid;
 use super::connection_with_credentials_provider::ConnectionWithCredentialsProvider;
 use crate::cache::project_info::ProjectInfoCache;
 use crate::cancellation::{CancelMap, CancellationHandler};
+use crate::config::ProxyConfig;
 use crate::intern::{ProjectIdInt, RoleNameInt};
 use crate::metrics::{Metrics, RedisErrors, RedisEventsCount};

@@ -39,6 +40,27 @@ pub(crate) enum Notification {
    AllowedIpsUpdate {
        allowed_ips_update: AllowedIpsUpdate,
    },
+    #[serde(
+        rename = "/block_public_or_vpc_access_updated",
+        deserialize_with = "deserialize_json_string"
+    )]
+    BlockPublicOrVpcAccessUpdated {
+        block_public_or_vpc_access_updated: BlockPublicOrVpcAccessUpdated,
+    },
+    #[serde(
+        rename = "/allowed_vpc_endpoints_updated_for_org",
+        deserialize_with = "deserialize_json_string"
+    )]
+    AllowedVpcEndpointsUpdatedForOrg {
+        allowed_vpc_endpoints_updated_for_org: AllowedVpcEndpointsUpdatedForOrg,
+    },
+    #[serde(
+        rename = "/allowed_vpc_endpoints_updated_for_projects",
+        deserialize_with = "deserialize_json_string"
+    )]
+    AllowedVpcEndpointsUpdatedForProjects {
+        allowed_vpc_endpoints_updated_for_projects: AllowedVpcEndpointsUpdatedForProjects,
+    },
    #[serde(
        rename = "/password_updated",
        deserialize_with = "deserialize_json_string"
@@ -51,6 +73,24 @@ pub(crate) enum Notification {
 pub(crate) struct AllowedIpsUpdate {
    project_id: ProjectIdInt,
 }
+
+#[derive(Clone, Debug, Serialize, Deserialize, Eq, PartialEq)]
+pub(crate) struct BlockPublicOrVpcAccessUpdated {
+    project_id: ProjectIdInt,
+}
+
+#[derive(Clone, Debug, Serialize, Deserialize, Eq, PartialEq)]
+pub(crate) struct AllowedVpcEndpointsUpdatedForOrg {
+    // TODO: change type once the implementation is more fully fledged.
+    // See e.g. https://github.com/neondatabase/neon/pull/10073.
+    account_id: ProjectIdInt,
+}
+
+#[derive(Clone, Debug, Serialize, Deserialize, Eq, PartialEq)]
+pub(crate) struct AllowedVpcEndpointsUpdatedForProjects {
+    project_ids: Vec<ProjectIdInt>,
+}
+
 #[derive(Clone, Debug, Serialize, Deserialize, Eq, PartialEq)]
 pub(crate) struct PasswordUpdate {
    project_id: ProjectIdInt,
@@ -164,7 +204,11 @@ impl<C: ProjectInfoCache + Send + Sync + 'static> MessageHandler<C> {
                    }
                }
            }
-            Notification::AllowedIpsUpdate { .. } | Notification::PasswordUpdate { .. } => {
+            Notification::AllowedIpsUpdate { .. }
+            | Notification::PasswordUpdate { .. }
+            | Notification::BlockPublicOrVpcAccessUpdated { .. }
+            | Notification::AllowedVpcEndpointsUpdatedForOrg { .. }
+            | Notification::AllowedVpcEndpointsUpdatedForProjects { .. } => {
                invalidate_cache(self.cache.clone(), msg.clone());
                if matches!(msg, Notification::AllowedIpsUpdate { .. }) {
                    Metrics::get()
@@ -177,6 +221,8 @@ impl<C: ProjectInfoCache + Send + Sync + 'static> MessageHandler<C> {
                        .redis_events_count
                        .inc(RedisEventsCount::PasswordUpdate);
                }
+                // TODO: add additional metrics for the other event types.
+
                // It might happen that the invalid entry is on the way to be cached.
                // To make sure that the entry is invalidated, let's repeat the invalidation in INVALIDATION_LAG seconds.
                // TODO: include the version (or the timestamp) in the message and invalidate only if the entry is cached before the message.
@@ -203,6 +249,15 @@ fn invalidate_cache<C: ProjectInfoCache>(cache: Arc<C>, msg: Notification) {
                password_update.role_name,
            ),
        Notification::Cancel(_) => unreachable!("cancel message should be handled separately"),
+        Notification::BlockPublicOrVpcAccessUpdated { .. } => {
+            // https://github.com/neondatabase/neon/pull/10073
+        }
+        Notification::AllowedVpcEndpointsUpdatedForOrg { .. } => {
+            // https://github.com/neondatabase/neon/pull/10073
+        }
+        Notification::AllowedVpcEndpointsUpdatedForProjects { .. } => {
+            // https://github.com/neondatabase/neon/pull/10073
+        }
    }
 }

@@ -249,6 +304,7 @@ async fn handle_messages<C: ProjectInfoCache + Send + Sync + 'static>(
 /// Handle console's invalidation messages.
 #[tracing::instrument(name = "redis_notifications", skip_all)]
 pub async fn task_main<C>(
+    config: &'static ProxyConfig,
    redis: ConnectionWithCredentialsProvider,
    cache: Arc<C>,
    cancel_map: CancelMap,
@@ -258,6 +314,7 @@ where
    C: ProjectInfoCache + Send + Sync + 'static,
 {
    let cancellation_handler = Arc::new(CancellationHandler::<()>::new(
+        &config.connect_to_compute,
        cancel_map,
        crate::metrics::CancellationSource::FromRedis,
    ));
--- a/proxy/src/sasl/stream.rs
+++ b/proxy/src/sasl/stream.rs
@@ -50,6 +50,12 @@ impl<S: AsyncWrite + Unpin> SaslStream<'_, S> {
        self.stream.write_message(&msg.to_reply()).await?;
        Ok(())
    }
+
+    // Queue a SASL message for the client.
+    fn send_noflush(&mut self, msg: &ServerMessage<&str>) -> io::Result<()> {
+        self.stream.write_message_noflush(&msg.to_reply())?;
+        Ok(())
+    }
 }

 /// SASL authentication outcome.
@@ -85,7 +91,7 @@ impl<S: AsyncRead + AsyncWrite + Unpin> SaslStream<'_, S> {
                    continue;
                }
                Step::Success(result, reply) => {
-                    self.send(&ServerMessage::Final(&reply)).await?;
+                    self.send_noflush(&ServerMessage::Final(&reply))?;
                    Outcome::Success(result)
                }
                Step::Failure(reason) => Outcome::Failure(reason),
--- a/proxy/src/scram/exchange.rs
+++ b/proxy/src/scram/exchange.rs
@@ -13,7 +13,6 @@ use super::secret::ServerSecret;
 use super::signature::SignatureBuilder;
 use super::threadpool::ThreadPool;
 use super::ScramKey;
-use crate::config;
 use crate::intern::EndpointIdInt;
 use crate::sasl::{self, ChannelBinding, Error as SaslError};

@@ -59,14 +58,14 @@ enum ExchangeState {
 pub(crate) struct Exchange<'a> {
    state: ExchangeState,
    secret: &'a ServerSecret,
-    tls_server_end_point: config::TlsServerEndPoint,
+    tls_server_end_point: crate::tls::TlsServerEndPoint,
 }

 impl<'a> Exchange<'a> {
    pub(crate) fn new(
        secret: &'a ServerSecret,
        nonce: fn() -> [u8; SCRAM_RAW_NONCE_LEN],
-        tls_server_end_point: config::TlsServerEndPoint,
+        tls_server_end_point: crate::tls::TlsServerEndPoint,
    ) -> Self {
        Self {
            state: ExchangeState::Initial(SaslInitial { nonce }),
@@ -120,7 +119,7 @@ impl SaslInitial {
    fn transition(
        &self,
        secret: &ServerSecret,
-        tls_server_end_point: &config::TlsServerEndPoint,
+        tls_server_end_point: &crate::tls::TlsServerEndPoint,
        input: &str,
    ) -> sasl::Result<sasl::Step<SaslSentInner, Infallible>> {
        let client_first_message = ClientFirstMessage::parse(input)
@@ -155,7 +154,7 @@ impl SaslSentInner {
    fn transition(
        &self,
        secret: &ServerSecret,
-        tls_server_end_point: &config::TlsServerEndPoint,
+        tls_server_end_point: &crate::tls::TlsServerEndPoint,
        input: &str,
    ) -> sasl::Result<sasl::Step<Infallible, super::ScramKey>> {
        let Self {
@@ -168,8 +167,8 @@ impl SaslSentInner {
            .ok_or(SaslError::BadClientMessage("invalid client-final-message"))?;

        let channel_binding = cbind_flag.encode(|_| match tls_server_end_point {
-            config::TlsServerEndPoint::Sha256(x) => Ok(x),
-            config::TlsServerEndPoint::Undefined => Err(SaslError::MissingBinding),
+            crate::tls::TlsServerEndPoint::Sha256(x) => Ok(x),
+            crate::tls::TlsServerEndPoint::Undefined => Err(SaslError::MissingBinding),
        })?;

        // This might've been caused by a MITM attack
--- a/proxy/src/scram/mod.rs
+++ b/proxy/src/scram/mod.rs
@@ -77,11 +77,8 @@ mod tests {
        const NONCE: [u8; 18] = [
            1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18,
        ];
-        let mut exchange = Exchange::new(
-            &secret,
-            || NONCE,
-            crate::config::TlsServerEndPoint::Undefined,
-        );
+        let mut exchange =
+            Exchange::new(&secret, || NONCE, crate::tls::TlsServerEndPoint::Undefined);

        let client_first = "n,,n=user,r=rOprNGfwEbeRWgbNEkqO";
        let client_final = "c=biws,r=rOprNGfwEbeRWgbNEkqOAQIDBAUGBwgJCgsMDQ4PEBES,p=rw1r5Kph5ThxmaUBC2GAQ6MfXbPnNkFiTIvdb/Rear0=";
--- a/proxy/src/serverless/backend.rs
+++ b/proxy/src/serverless/backend.rs
@@ -22,7 +22,7 @@ use crate::compute;
 use crate::compute_ctl::{
    ComputeCtlError, ExtensionInstallRequest, Privilege, SetRoleGrantsRequest,
 };
-use crate::config::ProxyConfig;
+use crate::config::{ComputeConfig, ProxyConfig};
 use crate::context::RequestContext;
 use crate::control_plane::client::ApiLockError;
 use crate::control_plane::errors::{GetAuthInfoError, WakeComputeError};
@@ -196,7 +196,7 @@ impl PoolingBackend {
            },
            &backend,
            self.config.wake_compute_retry_config,
-            self.config.connect_to_compute_retry_config,
+            &self.config.connect_to_compute,
        )
        .await
    }
@@ -237,7 +237,7 @@ impl PoolingBackend {
            },
            &backend,
            self.config.wake_compute_retry_config,
-            self.config.connect_to_compute_retry_config,
+            &self.config.connect_to_compute,
        )
        .await
    }
@@ -502,7 +502,7 @@ impl ConnectMechanism for TokioMechanism {
        &self,
        ctx: &RequestContext,
        node_info: &CachedNodeInfo,
-        timeout: Duration,
+        compute_config: &ComputeConfig,
    ) -> Result<Self::Connection, Self::ConnectError> {
        let host = node_info.config.get_host();
        let permit = self.locks.get_permit(&host).await?;
@@ -511,7 +511,7 @@ impl ConnectMechanism for TokioMechanism {
        let config = config
            .user(&self.conn_info.user_info.user)
            .dbname(&self.conn_info.dbname)
-            .connect_timeout(timeout);
+            .connect_timeout(compute_config.timeout);

        let pause = ctx.latency_timer_pause(crate::metrics::Waiting::Compute);
        let res = config.connect(postgres_client::NoTls).await;
@@ -552,7 +552,7 @@ impl ConnectMechanism for HyperMechanism {
        &self,
        ctx: &RequestContext,
        node_info: &CachedNodeInfo,
-        timeout: Duration,
+        config: &ComputeConfig,
    ) -> Result<Self::Connection, Self::ConnectError> {
        let host = node_info.config.get_host();
        let permit = self.locks.get_permit(&host).await?;
@@ -560,7 +560,7 @@ impl ConnectMechanism for HyperMechanism {
        let pause = ctx.latency_timer_pause(crate::metrics::Waiting::Compute);

        let port = node_info.config.get_port();
-        let res = connect_http2(&host, port, timeout).await;
+        let res = connect_http2(&host, port, config.timeout).await;
        drop(pause);
        let (client, connection) = permit.release_result(res)?;

--- a/proxy/src/serverless/mod.rs
+++ b/proxy/src/serverless/mod.rs
@@ -35,7 +35,7 @@ use rand::rngs::StdRng;
 use rand::SeedableRng;
 use sql_over_http::{uuid_to_header_value, NEON_REQUEST_ID};
 use tokio::io::{AsyncRead, AsyncWrite};
-use tokio::net::{TcpListener, TcpStream};
+use tokio::net::TcpStream;
 use tokio::time::timeout;
 use tokio_rustls::TlsAcceptor;
 use tokio_util::sync::CancellationToken;
@@ -45,6 +45,7 @@ use utils::http::error::ApiError;

 use crate::cancellation::CancellationHandlerMain;
 use crate::config::{ProxyConfig, ProxyProtocolV2};
+use crate::conn::{Acceptor, TokioTcpAcceptor};
 use crate::context::RequestContext;
 use crate::ext::TaskExt;
 use crate::metrics::Metrics;
@@ -59,7 +60,7 @@ pub(crate) const SERVERLESS_DRIVER_SNI: &str = "api";
 pub async fn task_main(
    config: &'static ProxyConfig,
    auth_backend: &'static crate::auth::Backend<'static, ()>,
-    ws_listener: TcpListener,
+    ws_acceptor: TokioTcpAcceptor,
    cancellation_token: CancellationToken,
    cancellation_handler: Arc<CancellationHandlerMain>,
    endpoint_rate_limiter: Arc<EndpointRateLimiter>,
@@ -134,7 +135,7 @@ pub async fn task_main(
    connections.close(); // allows `connections.wait to complete`

    let cancellations = tokio_util::task::task_tracker::TaskTracker::new();
-    while let Some(res) = run_until_cancelled(ws_listener.accept(), &cancellation_token).await {
+    while let Some(res) = run_until_cancelled(ws_acceptor.accept(), &cancellation_token).await {
        let (conn, peer_addr) = res.context("could not accept TCP stream")?;
        if let Err(e) = conn.set_nodelay(true) {
            tracing::error!("could not set nodelay: {e}");
--- a/proxy/src/serverless/websocket.rs
+++ b/proxy/src/serverless/websocket.rs
@@ -168,7 +168,7 @@ pub(crate) async fn serve_websocket(
        Ok(Some(p)) => {
            ctx.set_success();
            ctx.log_connect();
-            match p.proxy_pass().await {
+            match p.proxy_pass(&config.connect_to_compute).await {
                Ok(()) => Ok(()),
                Err(ErrorSource::Client(err)) => Err(err).context("client"),
                Err(ErrorSource::Compute(err)) => Err(err).context("compute"),
--- a/proxy/src/stream.rs
+++ b/proxy/src/stream.rs
@@ -11,9 +11,9 @@ use tokio::io::{AsyncRead, AsyncWrite, ReadBuf};
 use tokio_rustls::server::TlsStream;
 use tracing::debug;

-use crate::config::TlsServerEndPoint;
 use crate::error::{ErrorKind, ReportableError, UserFacingError};
 use crate::metrics::Metrics;
+use crate::tls::TlsServerEndPoint;

 /// Stream wrapper which implements libpq's protocol.
 ///
--- a/proxy/src/tls/client_config.rs
+++ b/proxy/src/tls/client_config.rs
@@ -0,0 +1,42 @@
+use std::sync::Arc;
+
+use anyhow::bail;
+use rustls::crypto::ring;
+
+pub(crate) fn load_certs() -> anyhow::Result<Arc<rustls::RootCertStore>> {
+    let der_certs = rustls_native_certs::load_native_certs();
+
+    if !der_certs.errors.is_empty() {
+        bail!("could not parse certificates: {:?}", der_certs.errors);
+    }
+
+    let mut store = rustls::RootCertStore::empty();
+    store.add_parsable_certificates(der_certs.certs);
+    Ok(Arc::new(store))
+}
+
+/// Loads the root certificates and constructs a client config suitable for connecting to the neon compute.
+/// This function is blocking.
+pub fn compute_client_config_with_root_certs() -> anyhow::Result<rustls::ClientConfig> {
+    Ok(
+        rustls::ClientConfig::builder_with_provider(Arc::new(ring::default_provider()))
+            .with_safe_default_protocol_versions()
+            .expect("ring should support the default protocol versions")
+            .with_root_certificates(load_certs()?)
+            .with_no_client_auth(),
+    )
+}
+
+#[cfg(test)]
+pub fn compute_client_config_with_certs(
+    certs: impl IntoIterator<Item = rustls::pki_types::CertificateDer<'static>>,
+) -> rustls::ClientConfig {
+    let mut store = rustls::RootCertStore::empty();
+    store.add_parsable_certificates(certs);
+
+    rustls::ClientConfig::builder_with_provider(Arc::new(ring::default_provider()))
+        .with_safe_default_protocol_versions()
+        .expect("ring should support the default protocol versions")
+        .with_root_certificates(store)
+        .with_no_client_auth()
+}
--- a/proxy/src/tls/mod.rs
+++ b/proxy/src/tls/mod.rs
@@ -0,0 +1,72 @@
+pub mod client_config;
+pub mod postgres_rustls;
+pub mod server_config;
+
+use anyhow::Context;
+use rustls::pki_types::CertificateDer;
+use sha2::{Digest, Sha256};
+use tracing::{error, info};
+use x509_parser::oid_registry;
+
+/// <https://github.com/postgres/postgres/blob/ca481d3c9ab7bf69ff0c8d71ad3951d407f6a33c/src/include/libpq/pqcomm.h#L159>
+pub const PG_ALPN_PROTOCOL: &[u8] = b"postgresql";
+
+/// Channel binding parameter
+///
+/// <https://www.rfc-editor.org/rfc/rfc5929#section-4>
+/// Description: The hash of the TLS server's certificate as it
+/// appears, octet for octet, in the server's Certificate message.  Note
+/// that the Certificate message contains a certificate_list, in which
+/// the first element is the server's certificate.
+///
+/// The hash function is to be selected as follows:
+///
+/// * if the certificate's signatureAlgorithm uses a single hash
+///   function, and that hash function is either MD5 or SHA-1, then use SHA-256;
+///
+/// * if the certificate's signatureAlgorithm uses a single hash
+///   function and that hash function neither MD5 nor SHA-1, then use
+///   the hash function associated with the certificate's
+///   signatureAlgorithm;
+///
+/// * if the certificate's signatureAlgorithm uses no hash functions or
+///   uses multiple hash functions, then this channel binding type's
+///   channel bindings are undefined at this time (updates to is channel
+///   binding type may occur to address this issue if it ever arises).
+#[derive(Debug, Clone, Copy)]
+pub enum TlsServerEndPoint {
+    Sha256([u8; 32]),
+    Undefined,
+}
+
+impl TlsServerEndPoint {
+    pub fn new(cert: &CertificateDer<'_>) -> anyhow::Result<Self> {
+        let sha256_oids = [
+            // I'm explicitly not adding MD5 or SHA1 here... They're bad.
+            oid_registry::OID_SIG_ECDSA_WITH_SHA256,
+            oid_registry::OID_PKCS1_SHA256WITHRSA,
+        ];
+
+        let pem = x509_parser::parse_x509_certificate(cert)
+            .context("Failed to parse PEM object from cerficiate")?
+            .1;
+
+        info!(subject = %pem.subject, "parsing TLS certificate");
+
+        let reg = oid_registry::OidRegistry::default().with_all_crypto();
+        let oid = pem.signature_algorithm.oid();
+        let alg = reg.get(oid);
+        if sha256_oids.contains(oid) {
+            let tls_server_end_point: [u8; 32] = Sha256::new().chain_update(cert).finalize().into();
+            info!(subject = %pem.subject, signature_algorithm = alg.map(|a| a.description()), tls_server_end_point = %base64::encode(tls_server_end_point), "determined channel binding");
+            Ok(Self::Sha256(tls_server_end_point))
+        } else {
+            error!(subject = %pem.subject, signature_algorithm = alg.map(|a| a.description()), "unknown channel binding");
+            Ok(Self::Undefined)
+        }
+    }
+
+    pub fn supported(&self) -> bool {
+        !matches!(self, TlsServerEndPoint::Undefined)
+    }
+}
--- a/proxy/src/tls/postgres_rustls.rs
+++ b/proxy/src/tls/postgres_rustls.rs
@@ -18,7 +18,7 @@ mod private {
    use tokio_rustls::client::TlsStream;
    use tokio_rustls::TlsConnector;

-    use crate::config::TlsServerEndPoint;
+    use crate::tls::TlsServerEndPoint;

    pub struct TlsConnectFuture<S> {
        inner: tokio_rustls::Connect<S>,
@@ -126,16 +126,14 @@ mod private {
 /// That way you can connect to PostgreSQL using `rustls` as the TLS stack.
 #[derive(Clone)]
 pub struct MakeRustlsConnect {
-    config: Arc<ClientConfig>,
+    pub config: Arc<ClientConfig>,
 }

 impl MakeRustlsConnect {
    /// Creates a new `MakeRustlsConnect` from the provided `ClientConfig`.
    #[must_use]
-    pub fn new(config: ClientConfig) -> Self {
-        Self {
-            config: Arc::new(config),
-        }
+    pub fn new(config: Arc<ClientConfig>) -> Self {
+        Self { config }
    }
 }

--- a/proxy/src/tls/server_config.rs
+++ b/proxy/src/tls/server_config.rs
@@ -0,0 +1,218 @@
+use std::collections::{HashMap, HashSet};
+use std::sync::Arc;
+
+use anyhow::{bail, Context};
+use itertools::Itertools;
+use rustls::crypto::ring::{self, sign};
+use rustls::pki_types::{CertificateDer, PrivateKeyDer};
+
+use super::{TlsServerEndPoint, PG_ALPN_PROTOCOL};
+
+pub struct TlsConfig {
+    pub config: Arc<rustls::ServerConfig>,
+    pub common_names: HashSet<String>,
+    pub cert_resolver: Arc<CertResolver>,
+}
+
+impl TlsConfig {
+    pub fn to_server_config(&self) -> Arc<rustls::ServerConfig> {
+        self.config.clone()
+    }
+}
+
+/// Configure TLS for the main endpoint.
+pub fn configure_tls(
+    key_path: &str,
+    cert_path: &str,
+    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)?;
+
+    // add extra certificates
+    if let Some(certs_dir) = certs_dir {
+        for entry in std::fs::read_dir(certs_dir)? {
+            let entry = entry?;
+            let path = entry.path();
+            if path.is_dir() {
+                // file names aligned with default cert-manager names
+                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,
+                    )?;
+                }
+            }
+        }
+    }
+
+    let common_names = cert_resolver.get_common_names();
+
+    let cert_resolver = Arc::new(cert_resolver);
+
+    // allow TLS 1.2 to be compatible with older client libraries
+    let mut config =
+        rustls::ServerConfig::builder_with_provider(Arc::new(ring::default_provider()))
+            .with_protocol_versions(&[&rustls::version::TLS13, &rustls::version::TLS12])
+            .context("ring should support TLS1.2 and TLS1.3")?
+            .with_no_client_auth()
+            .with_cert_resolver(cert_resolver.clone());
+
+    config.alpn_protocols = vec![PG_ALPN_PROTOCOL.to_vec()];
+
+    if allow_tls_keylogfile {
+        // KeyLogFile will check for the SSLKEYLOGFILE environment variable.
+        config.key_log = Arc::new(rustls::KeyLogFile::new());
+    }
+
+    Ok(TlsConfig {
+        config: Arc::new(config),
+        common_names,
+        cert_resolver,
+    })
+}
+
+#[derive(Default, Debug)]
+pub struct CertResolver {
+    certs: HashMap<String, (Arc<rustls::sign::CertifiedKey>, TlsServerEndPoint)>,
+    default: Option<(Arc<rustls::sign::CertifiedKey>, TlsServerEndPoint)>,
+}
+
+impl CertResolver {
+    pub fn new() -> Self {
+        Self::default()
+    }
+
+    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}'"))?
+        };
+
+        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)
+    }
+
+    pub 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 pem = x509_parser::parse_x509_certificate(first_cert)
+            .context("Failed to parse PEM object from cerficiate")?
+            .1;
+
+        let common_name = pem.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));
+        }
+
+        self.certs.insert(common_name, (cert, tls_server_end_point));
+
+        Ok(())
+    }
+
+    pub fn get_common_names(&self) -> HashSet<String> {
+        self.certs.keys().map(|s| s.to_string()).collect()
+    }
+}
+
+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)
+    }
+}
+
+impl CertResolver {
+    pub fn resolve(
+        &self,
+        server_name: Option<&str>,
+    ) -> Option<(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.
+        //
+        // With the current coding foo.com will match *.foo.com and that
+        // repeats behavior of the old code.
+        if let Some(mut sni_name) = server_name {
+            loop {
+                if let Some(cert) = self.certs.get(sni_name) {
+                    return Some(cert.clone());
+                }
+                if let Some((_, rest)) = sni_name.split_once('.') {
+                    sni_name = rest;
+                } else {
+                    return None;
+                }
+            }
+        } else {
+            // No SNI, use the default certificate, otherwise we can't get to
+            // options parameter which can be used to set endpoint name too.
+            // That means that non-SNI flow will not work for CNAME domains in
+            // verify-full mode.
+            //
+            // If that will be a problem we can:
+            //
+            // a) Instead of multi-cert approach use single cert with extra
+            //    domains listed in Subject Alternative Name (SAN).
+            // b) Deploy separate proxy instances for extra domains.
+            self.default.clone()
+        }
+    }
+}
--- a/safekeeper/Cargo.toml
+++ b/safekeeper/Cargo.toml
@@ -9,6 +9,7 @@ default = []
 # Enables test-only APIs, incuding failpoints. In particular, enables the `fail_point!` macro,
 # which adds some runtime cost to run tests on outage conditions
 testing = ["fail/failpoints"]
+benchmarking = []

 [dependencies]
 async-stream.workspace = true
@@ -77,3 +78,4 @@ tracing-subscriber = { workspace = true, features = ["json"] }
 [[bench]]
 name = "receive_wal"
 harness = false
+required-features = ["benchmarking"]
--- a/safekeeper/benches/receive_wal.rs
+++ b/safekeeper/benches/receive_wal.rs
@@ -1,11 +1,7 @@
 //! WAL ingestion benchmarks.

-#[path = "benchutils.rs"]
-mod benchutils;
-
 use std::io::Write as _;

-use benchutils::Env;
 use bytes::BytesMut;
 use camino_tempfile::tempfile;
 use criterion::{criterion_group, criterion_main, BatchSize, Bencher, Criterion};
@@ -16,6 +12,7 @@ use safekeeper::receive_wal::{self, WalAcceptor};
 use safekeeper::safekeeper::{
    AcceptorProposerMessage, AppendRequest, AppendRequestHeader, ProposerAcceptorMessage,
 };
+use safekeeper::test_utils::Env;
 use tokio::io::AsyncWriteExt as _;
 use utils::id::{NodeId, TenantTimelineId};
 use utils::lsn::Lsn;
@@ -76,12 +73,15 @@ fn bench_process_msg(c: &mut Criterion) {
        assert!(size >= prefixlen);
        let message = vec![0; size - prefixlen];

-        let walgen = &mut WalGenerator::new(LogicalMessageGenerator::new(prefix, &message));
+        let walgen = &mut WalGenerator::new(LogicalMessageGenerator::new(prefix, &message), Lsn(0));

        // Set up the Safekeeper.
        let env = Env::new(fsync)?;
-        let mut safekeeper =
-            runtime.block_on(env.make_safekeeper(NodeId(1), TenantTimelineId::generate()))?;
+        let mut safekeeper = runtime.block_on(env.make_safekeeper(
+            NodeId(1),
+            TenantTimelineId::generate(),
+            Lsn(0),
+        ))?;

        b.iter_batched_ref(
            // Pre-construct WAL records and requests. Criterion will batch them.
@@ -134,7 +134,8 @@ fn bench_wal_acceptor(c: &mut Criterion) {
        let runtime = tokio::runtime::Runtime::new()?; // needs multithreaded

        let env = Env::new(fsync)?;
-        let walgen = &mut WalGenerator::new(LogicalMessageGenerator::new(c"prefix", b"message"));
+        let walgen =
+            &mut WalGenerator::new(LogicalMessageGenerator::new(c"prefix", b"message"), Lsn(0));

        // Create buffered channels that can fit all requests, to avoid blocking on channels.
        let (msg_tx, msg_rx) = tokio::sync::mpsc::channel(n);
@@ -145,7 +146,7 @@ fn bench_wal_acceptor(c: &mut Criterion) {
            // TODO: WalAcceptor doesn't actually need a full timeline, only
            // Safekeeper::process_msg(). Consider decoupling them to simplify the setup.
            let tli = env
-                .make_timeline(NodeId(1), TenantTimelineId::generate())
+                .make_timeline(NodeId(1), TenantTimelineId::generate(), Lsn(0))
                .await?
                .wal_residence_guard()
                .await?;
@@ -239,7 +240,7 @@ fn bench_wal_acceptor_throughput(c: &mut Criterion) {
        assert!(size >= prefixlen);
        let message = vec![0; size - prefixlen];

-        let walgen = &mut WalGenerator::new(LogicalMessageGenerator::new(prefix, &message));
+        let walgen = &mut WalGenerator::new(LogicalMessageGenerator::new(prefix, &message), Lsn(0));

        // Construct and spawn the WalAcceptor task.
        let env = Env::new(fsync)?;
@@ -249,7 +250,7 @@ fn bench_wal_acceptor_throughput(c: &mut Criterion) {

        runtime.block_on(async {
            let tli = env
-                .make_timeline(NodeId(1), TenantTimelineId::generate())
+                .make_timeline(NodeId(1), TenantTimelineId::generate(), Lsn(0))
                .await?
                .wal_residence_guard()
                .await?;
--- a/safekeeper/src/http/routes.rs
+++ b/safekeeper/src/http/routes.rs
@@ -564,7 +564,7 @@ pub fn make_router(
    if conf.http_auth.is_some() {
        router = router.middleware(auth_middleware(|request| {
            const ALLOWLIST_ROUTES: &[&str] =
-                &["/v1/status", "/metrics", "/profile/cpu", "profile/heap"];
+                &["/v1/status", "/metrics", "/profile/cpu", "/profile/heap"];
            if ALLOWLIST_ROUTES.contains(&request.uri().path()) {
                None
            } else {
--- a/safekeeper/src/lib.rs
+++ b/safekeeper/src/lib.rs
@@ -43,6 +43,9 @@ pub mod wal_reader_stream;
 pub mod wal_service;
 pub mod wal_storage;

+#[cfg(any(test, feature = "benchmarking"))]
+pub mod test_utils;
+
 mod timelines_global_map;
 use std::sync::Arc;
 pub use timelines_global_map::GlobalTimelines;
--- a/safekeeper/src/send_interpreted_wal.rs
+++ b/safekeeper/src/send_interpreted_wal.rs
@@ -94,9 +94,14 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> InterpretedWalSender<'_, IO> {
                        }
                    }

+                    let max_next_record_lsn = match max_next_record_lsn {
+                        Some(lsn) => lsn,
+                        None => { continue; }
+                    };
+
                    let batch = InterpretedWalRecords {
                        records,
-                        next_record_lsn: max_next_record_lsn
+                        next_record_lsn: Some(max_next_record_lsn),
                    };

                    tx.send(Batch {wal_end_lsn, available_wal_end_lsn, records: batch}).await.unwrap();
--- a/safekeeper/benches/benchutils.rs
+++ b/safekeeper/benches/benchutils.rs
@@ -1,18 +1,18 @@
 use std::sync::Arc;

+use crate::rate_limit::RateLimiter;
+use crate::safekeeper::{ProposerAcceptorMessage, ProposerElected, SafeKeeper, TermHistory};
+use crate::state::{TimelinePersistentState, TimelineState};
+use crate::timeline::{get_timeline_dir, SharedState, StateSK, Timeline};
+use crate::timelines_set::TimelinesSet;
+use crate::wal_backup::remote_timeline_path;
+use crate::{control_file, wal_storage, SafeKeeperConf};
 use camino_tempfile::Utf8TempDir;
-use safekeeper::rate_limit::RateLimiter;
-use safekeeper::safekeeper::{ProposerAcceptorMessage, ProposerElected, SafeKeeper, TermHistory};
-use safekeeper::state::{TimelinePersistentState, TimelineState};
-use safekeeper::timeline::{get_timeline_dir, SharedState, StateSK, Timeline};
-use safekeeper::timelines_set::TimelinesSet;
-use safekeeper::wal_backup::remote_timeline_path;
-use safekeeper::{control_file, wal_storage, SafeKeeperConf};
 use tokio::fs::create_dir_all;
 use utils::id::{NodeId, TenantTimelineId};
 use utils::lsn::Lsn;

-/// A Safekeeper benchmarking environment. Uses a tempdir for storage, removed on drop.
+/// A Safekeeper testing or benchmarking environment. Uses a tempdir for storage, removed on drop.
 pub struct Env {
    /// Whether to enable fsync.
    pub fsync: bool,
@@ -21,7 +21,7 @@ pub struct Env {
 }

 impl Env {
-    /// Creates a new benchmarking environment in a temporary directory. fsync controls whether to
+    /// Creates a new test or benchmarking environment in a temporary directory. fsync controls whether to
    /// enable fsyncing.
    pub fn new(fsync: bool) -> anyhow::Result<Self> {
        let tempdir = camino_tempfile::tempdir()?;
@@ -47,6 +47,7 @@ impl Env {
        &self,
        node_id: NodeId,
        ttid: TenantTimelineId,
+        start_lsn: Lsn,
    ) -> anyhow::Result<SafeKeeper<control_file::FileStorage, wal_storage::PhysicalStorage>> {
        let conf = self.make_conf(node_id);

@@ -67,9 +68,9 @@ impl Env {
        safekeeper
            .process_msg(&ProposerAcceptorMessage::Elected(ProposerElected {
                term: 1,
-                start_streaming_at: Lsn(0),
-                term_history: TermHistory(vec![(1, Lsn(0)).into()]),
-                timeline_start_lsn: Lsn(0),
+                start_streaming_at: start_lsn,
+                term_history: TermHistory(vec![(1, start_lsn).into()]),
+                timeline_start_lsn: start_lsn,
            }))
            .await?;

@@ -82,12 +83,13 @@ impl Env {
        &self,
        node_id: NodeId,
        ttid: TenantTimelineId,
+        start_lsn: Lsn,
    ) -> anyhow::Result<Arc<Timeline>> {
        let conf = Arc::new(self.make_conf(node_id));
        let timeline_dir = get_timeline_dir(&conf, &ttid);
        let remote_path = remote_timeline_path(&ttid)?;

-        let safekeeper = self.make_safekeeper(node_id, ttid).await?;
+        let safekeeper = self.make_safekeeper(node_id, ttid, start_lsn).await?;
        let shared_state = SharedState::new(StateSK::Loaded(safekeeper));

        let timeline = Timeline::new(
--- a/safekeeper/tests/walproposer_sim/walproposer_disk.rs
+++ b/safekeeper/tests/walproposer_sim/walproposer_disk.rs
@@ -18,7 +18,7 @@ impl DiskWalProposer {
                internal_available_lsn: Lsn(0),
                prev_lsn: Lsn(0),
                disk: BlockStorage::new(),
-                wal_generator: WalGenerator::new(LogicalMessageGenerator::new(c"", &[])),
+                wal_generator: WalGenerator::new(LogicalMessageGenerator::new(c"", &[]), Lsn(0)),
            }),
        })
    }
--- a/storage_controller/Cargo.toml
+++ b/storage_controller/Cargo.toml
@@ -43,13 +43,13 @@ scopeguard.workspace = true
 strum.workspace = true
 strum_macros.workspace = true

-diesel = { version = "2.1.4", features = [
+diesel = { version = "2.2.6", features = [
    "serde_json",
    "postgres",
    "r2d2",
    "chrono",
 ] }
-diesel_migrations = { version = "2.1.0" }
+diesel_migrations = { version = "2.2.0" }
 r2d2 = { version = "0.8.10" }

 utils = { path = "../libs/utils/" }
--- a/storage_controller/src/compute_hook.rs
+++ b/storage_controller/src/compute_hook.rs
@@ -1,3 +1,4 @@
+use std::borrow::Cow;
 use std::error::Error as _;
 use std::sync::Arc;
 use std::{collections::HashMap, time::Duration};
@@ -6,6 +7,7 @@ use control_plane::endpoint::{ComputeControlPlane, EndpointStatus};
 use control_plane::local_env::LocalEnv;
 use futures::StreamExt;
 use hyper::StatusCode;
+use pageserver_api::controller_api::AvailabilityZone;
 use pageserver_api::shard::{ShardCount, ShardNumber, ShardStripeSize, TenantShardId};
 use postgres_connection::parse_host_port;
 use serde::{Deserialize, Serialize};
@@ -28,6 +30,9 @@ struct UnshardedComputeHookTenant {
    // Which node is this tenant attached to
    node_id: NodeId,

+    // The tenant's preferred AZ, so that we may pass this on to the control plane
+    preferred_az: Option<AvailabilityZone>,
+
    // Must hold this lock to send a notification.
    send_lock: Arc<tokio::sync::Mutex<Option<ComputeRemoteState>>>,
 }
@@ -36,6 +41,9 @@ struct ShardedComputeHookTenant {
    shard_count: ShardCount,
    shards: Vec<(ShardNumber, NodeId)>,

+    // The tenant's preferred AZ, so that we may pass this on to the control plane
+    preferred_az: Option<AvailabilityZone>,
+
    // Must hold this lock to send a notification.  The contents represent
    // the last successfully sent notification, and are used to coalesce multiple
    // updates by only sending when there is a chance since our last successful send.
@@ -64,17 +72,24 @@ enum ComputeHookTenant {

 impl ComputeHookTenant {
    /// Construct with at least one shard's information
-    fn new(tenant_shard_id: TenantShardId, stripe_size: ShardStripeSize, node_id: NodeId) -> Self {
+    fn new(
+        tenant_shard_id: TenantShardId,
+        stripe_size: ShardStripeSize,
+        preferred_az: Option<AvailabilityZone>,
+        node_id: NodeId,
+    ) -> Self {
        if tenant_shard_id.shard_count.count() > 1 {
            Self::Sharded(ShardedComputeHookTenant {
                shards: vec![(tenant_shard_id.shard_number, node_id)],
                stripe_size,
                shard_count: tenant_shard_id.shard_count,
+                preferred_az,
                send_lock: Arc::default(),
            })
        } else {
            Self::Unsharded(UnshardedComputeHookTenant {
                node_id,
+                preferred_az,
                send_lock: Arc::default(),
            })
        }
@@ -120,15 +135,20 @@ impl ComputeHookTenant {

    /// Set one shard's location.  If stripe size or shard count have changed, Self is reset
    /// and drops existing content.
-    fn update(
-        &mut self,
-        tenant_shard_id: TenantShardId,
-        stripe_size: ShardStripeSize,
-        node_id: NodeId,
-    ) {
+    fn update(&mut self, shard_update: ShardUpdate) {
+        let tenant_shard_id = shard_update.tenant_shard_id;
+        let node_id = shard_update.node_id;
+        let stripe_size = shard_update.stripe_size;
+        let preferred_az = shard_update.preferred_az;
+
        match self {
            Self::Unsharded(unsharded_tenant) if tenant_shard_id.shard_count.count() == 1 => {
-                unsharded_tenant.node_id = node_id
+                unsharded_tenant.node_id = node_id;
+                if unsharded_tenant.preferred_az.as_ref()
+                    != preferred_az.as_ref().map(|az| az.as_ref())
+                {
+                    unsharded_tenant.preferred_az = preferred_az.map(|az| az.as_ref().clone());
+                }
            }
            Self::Sharded(sharded_tenant)
                if sharded_tenant.stripe_size == stripe_size
@@ -146,10 +166,21 @@ impl ComputeHookTenant {
                        .push((tenant_shard_id.shard_number, node_id));
                    sharded_tenant.shards.sort_by_key(|s| s.0)
                }
+
+                if sharded_tenant.preferred_az.as_ref()
+                    != preferred_az.as_ref().map(|az| az.as_ref())
+                {
+                    sharded_tenant.preferred_az = preferred_az.map(|az| az.as_ref().clone());
+                }
            }
            _ => {
                // Shard count changed: reset struct.
-                *self = Self::new(tenant_shard_id, stripe_size, node_id);
+                *self = Self::new(
+                    tenant_shard_id,
+                    stripe_size,
+                    preferred_az.map(|az| az.into_owned()),
+                    node_id,
+                );
            }
        }
    }
@@ -165,6 +196,7 @@ struct ComputeHookNotifyRequestShard {
 #[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
 struct ComputeHookNotifyRequest {
    tenant_id: TenantId,
+    preferred_az: Option<String>,
    stripe_size: Option<ShardStripeSize>,
    shards: Vec<ComputeHookNotifyRequestShard>,
 }
@@ -238,6 +270,10 @@ impl ComputeHookTenant {
                    node_id: unsharded_tenant.node_id,
                }],
                stripe_size: None,
+                preferred_az: unsharded_tenant
+                    .preferred_az
+                    .as_ref()
+                    .map(|az| az.0.clone()),
            }),
            Self::Sharded(sharded_tenant)
                if sharded_tenant.shards.len() == sharded_tenant.shard_count.count() as usize =>
@@ -253,6 +289,7 @@ impl ComputeHookTenant {
                        })
                        .collect(),
                    stripe_size: Some(sharded_tenant.stripe_size),
+                    preferred_az: sharded_tenant.preferred_az.as_ref().map(|az| az.0.clone()),
                })
            }
            Self::Sharded(sharded_tenant) => {
@@ -313,6 +350,17 @@ pub(super) struct ComputeHook {
    client: reqwest::Client,
 }

+/// Callers may give us a list of these when asking us to send a bulk batch
+/// of notifications in the background.  This is a 'notification' in the sense of
+/// other code notifying us of a shard's status, rather than being the final notification
+/// that we send upwards to the control plane for the whole tenant.
+pub(crate) struct ShardUpdate<'a> {
+    pub(crate) tenant_shard_id: TenantShardId,
+    pub(crate) node_id: NodeId,
+    pub(crate) stripe_size: ShardStripeSize,
+    pub(crate) preferred_az: Option<Cow<'a, AvailabilityZone>>,
+}
+
 impl ComputeHook {
    pub(super) fn new(config: Config) -> Self {
        let authorization_header = config
@@ -363,6 +411,7 @@ impl ComputeHook {
            tenant_id,
            shards,
            stripe_size,
+            preferred_az: _preferred_az,
        } = reconfigure_request;

        let compute_pageservers = shards
@@ -503,24 +552,30 @@ impl ComputeHook {
    }

    /// Synchronous phase: update the per-tenant state for the next intended notification
-    fn notify_prepare(
-        &self,
-        tenant_shard_id: TenantShardId,
-        node_id: NodeId,
-        stripe_size: ShardStripeSize,
-    ) -> MaybeSendResult {
+    fn notify_prepare(&self, shard_update: ShardUpdate) -> MaybeSendResult {
        let mut state_locked = self.state.lock().unwrap();

        use std::collections::hash_map::Entry;
+        let tenant_shard_id = shard_update.tenant_shard_id;
+
        let tenant = match state_locked.entry(tenant_shard_id.tenant_id) {
-            Entry::Vacant(e) => e.insert(ComputeHookTenant::new(
-                tenant_shard_id,
-                stripe_size,
-                node_id,
-            )),
+            Entry::Vacant(e) => {
+                let ShardUpdate {
+                    tenant_shard_id,
+                    node_id,
+                    stripe_size,
+                    preferred_az,
+                } = shard_update;
+                e.insert(ComputeHookTenant::new(
+                    tenant_shard_id,
+                    stripe_size,
+                    preferred_az.map(|az| az.into_owned()),
+                    node_id,
+                ))
+            }
            Entry::Occupied(e) => {
                let tenant = e.into_mut();
-                tenant.update(tenant_shard_id, stripe_size, node_id);
+                tenant.update(shard_update);
                tenant
            }
        };
@@ -608,13 +663,14 @@ impl ComputeHook {
    /// if something failed.
    pub(super) fn notify_background(
        self: &Arc<Self>,
-        notifications: Vec<(TenantShardId, NodeId, ShardStripeSize)>,
+        notifications: Vec<ShardUpdate>,
        result_tx: tokio::sync::mpsc::Sender<Result<(), (TenantShardId, NotifyError)>>,
        cancel: &CancellationToken,
    ) {
        let mut maybe_sends = Vec::new();
-        for (tenant_shard_id, node_id, stripe_size) in notifications {
-            let maybe_send_result = self.notify_prepare(tenant_shard_id, node_id, stripe_size);
+        for shard_update in notifications {
+            let tenant_shard_id = shard_update.tenant_shard_id;
+            let maybe_send_result = self.notify_prepare(shard_update);
            maybe_sends.push((tenant_shard_id, maybe_send_result))
        }

@@ -678,15 +734,14 @@ impl ComputeHook {
    /// periods, but we don't retry forever.  The **caller** is responsible for handling failures and
    /// ensuring that they eventually call again to ensure that the compute is eventually notified of
    /// the proper pageserver nodes for a tenant.
-    #[tracing::instrument(skip_all, fields(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), node_id))]
-    pub(super) async fn notify(
+    #[tracing::instrument(skip_all, fields(tenant_id=%shard_update.tenant_shard_id.tenant_id, shard_id=%shard_update.tenant_shard_id.shard_slug(), node_id))]
+    pub(super) async fn notify<'a>(
        &self,
-        tenant_shard_id: TenantShardId,
-        node_id: NodeId,
-        stripe_size: ShardStripeSize,
+        shard_update: ShardUpdate<'a>,
        cancel: &CancellationToken,
    ) -> Result<(), NotifyError> {
-        let maybe_send_result = self.notify_prepare(tenant_shard_id, node_id, stripe_size);
+        let tenant_shard_id = shard_update.tenant_shard_id;
+        let maybe_send_result = self.notify_prepare(shard_update);
        self.notify_execute(maybe_send_result, tenant_shard_id, cancel)
            .await
    }
@@ -739,6 +794,7 @@ pub(crate) mod tests {
                shard_number: ShardNumber(0),
            },
            ShardStripeSize(12345),
+            None,
            NodeId(1),
        );

@@ -765,30 +821,32 @@ pub(crate) mod tests {
        // Writing the first shard of a multi-sharded situation (i.e. in a split)
        // resets the tenant state and puts it in an non-notifying state (need to
        // see all shards)
-        tenant_state.update(
-            TenantShardId {
+        tenant_state.update(ShardUpdate {
+            tenant_shard_id: TenantShardId {
                tenant_id,
                shard_count: ShardCount::new(2),
                shard_number: ShardNumber(1),
            },
-            ShardStripeSize(32768),
-            NodeId(1),
-        );
+            stripe_size: ShardStripeSize(32768),
+            preferred_az: None,
+            node_id: NodeId(1),
+        });
        assert!(matches!(
            tenant_state.maybe_send(tenant_id, None),
            MaybeSendResult::Noop
        ));

        // Writing the second shard makes it ready to notify
-        tenant_state.update(
-            TenantShardId {
+        tenant_state.update(ShardUpdate {
+            tenant_shard_id: TenantShardId {
                tenant_id,
                shard_count: ShardCount::new(2),
                shard_number: ShardNumber(0),
            },
-            ShardStripeSize(32768),
-            NodeId(1),
-        );
+            stripe_size: ShardStripeSize(32768),
+            preferred_az: None,
+            node_id: NodeId(1),
+        });

        let send_result = tenant_state.maybe_send(tenant_id, None);
        let MaybeSendResult::Transmit((request, mut guard)) = send_result else {
--- a/storage_controller/src/persistence.rs
+++ b/storage_controller/src/persistence.rs
@@ -11,6 +11,7 @@ use diesel::Connection;
 use itertools::Itertools;
 use pageserver_api::controller_api::AvailabilityZone;
 use pageserver_api::controller_api::MetadataHealthRecord;
+use pageserver_api::controller_api::SafekeeperDescribeResponse;
 use pageserver_api::controller_api::ShardSchedulingPolicy;
 use pageserver_api::controller_api::{NodeSchedulingPolicy, PlacementPolicy};
 use pageserver_api::models::TenantConfig;
@@ -1241,6 +1242,18 @@ impl SafekeeperPersistence {
            availability_zone_id: &self.availability_zone_id,
        }
    }
+    pub(crate) fn as_describe_response(&self) -> SafekeeperDescribeResponse {
+        // omit the `active` flag on purpose: it is deprecated.
+        SafekeeperDescribeResponse {
+            id: NodeId(self.id as u64),
+            region_id: self.region_id.clone(),
+            version: self.version,
+            host: self.host.clone(),
+            port: self.port,
+            http_port: self.http_port,
+            availability_zone_id: self.availability_zone_id.clone(),
+        }
+    }
 }

 #[derive(Insertable, AsChangeset)]
--- a/Show More
+++ b/Show More