Remove assert for non-zero number of shards from AssignPageserverConnstring

Bump postgres version
Bump Postgres version
2026-08-14 10:11:33 +00:00 · 2023-12-20 09:16:54 +02:00 · 2023-12-19 22:17:07 +02:00 · 2023-12-19 16:21:15 +02:00 · 2023-12-19 14:45:25 +02:00 · 2023-12-19 10:20:56 +02:00
36 changed files with 1153 additions and 1309 deletions
@@ -11,7 +11,7 @@ on:
    #          │ │ ┌───────────── day of the month (1 - 31)
    #          │ │ │ ┌───────────── month (1 - 12 or JAN-DEC)
    #          │ │ │ │ ┌───────────── day of the week (0 - 6 or SUN-SAT)
-    - cron:   '0 3 * * *' # run once a day, timezone is utc
+    - cron:  '0 3 * * *' # run once a day, timezone is utc

  workflow_dispatch: # adds ability to run this manually
    inputs:
@@ -23,21 +23,6 @@ on:
        type: boolean
        description: 'Publish perf report. If not set, the report will be published only for the main branch'
        required: false
-      collect_olap_explain:
-        type: boolean
-        description: 'Collect EXPLAIN ANALYZE for OLAP queries. If not set, EXPLAIN ANALYZE will not be collected'
-        required: false
-        default: false
-      collect_pg_stat_statements:
-        type: boolean
-        description: 'Collect pg_stat_statements for OLAP queries. If not set, pg_stat_statements will not be collected'
-        required: false
-        default: false
-      run_AWS_RDS_AND_AURORA:
-        type: boolean
-        description: 'AWS-RDS and AWS-AURORA normally only run on Saturday. Set this to true to run them on every workflow_dispatch'
-        required: false
-        default: false

 defaults:
  run:
@@ -128,8 +113,6 @@ jobs:
    # - neon-captest-reuse: Reusing existing project
    # - rds-aurora: Aurora Postgres Serverless v2 with autoscaling from 0.5 to 2 ACUs
    # - rds-postgres: RDS Postgres db.m5.large instance (2 vCPU, 8 GiB) with gp3 EBS storage
-    env:
-      RUN_AWS_RDS_AND_AURORA: ${{ github.event.inputs.run_AWS_RDS_AND_AURORA || 'false' }}
    runs-on: ubuntu-latest
    outputs:
      pgbench-compare-matrix: ${{ steps.pgbench-compare-matrix.outputs.matrix }}
@@ -169,7 +152,7 @@ jobs:
          ]
        }'

-        if [ "$(date +%A)" = "Saturday" ] || [ ${RUN_AWS_RDS_AND_AURORA} = "true" ]; then
+        if [ "$(date +%A)" = "Saturday" ]; then
          matrix=$(echo "$matrix" | jq '.include += [{ "platform": "rds-postgres" },
                                                   { "platform": "rds-aurora"   }]')
        fi
@@ -188,9 +171,9 @@ jobs:
          ]
        }'

-        if [ "$(date +%A)" = "Saturday" ] || [ ${RUN_AWS_RDS_AND_AURORA} = "true" ]; then
+        if [ "$(date +%A)" = "Saturday" ]; then
          matrix=$(echo "$matrix" | jq '.include += [{ "platform": "rds-postgres", "scale": "10" },
-                                                    { "platform": "rds-aurora",   "scale": "10" }]')
+                                                   { "platform": "rds-aurora",   "scale": "10" }]')
        fi

        echo "matrix=$(echo "$matrix" | jq --compact-output '.')" >> $GITHUB_OUTPUT
@@ -354,8 +337,6 @@ jobs:
      POSTGRES_DISTRIB_DIR: /tmp/neon/pg_install
      DEFAULT_PG_VERSION: 14
      TEST_OUTPUT: /tmp/test_output
-      TEST_OLAP_COLLECT_EXPLAIN: ${{ github.event.inputs.collect_olap_explain }}
-      TEST_OLAP_COLLECT_PG_STAT_STATEMENTS: ${{ github.event.inputs.collect_pg_stat_statements }}
      BUILD_TYPE: remote
      SAVE_PERF_REPORT: ${{ github.event.inputs.save_perf_report || ( github.ref_name == 'main' ) }}
      PLATFORM: ${{ matrix.platform }}
@@ -418,8 +399,6 @@ jobs:
      env:
        VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
        PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
-        TEST_OLAP_COLLECT_EXPLAIN: ${{ github.event.inputs.collect_olap_explain || 'false' }}
-        TEST_OLAP_COLLECT_PG_STAT_STATEMENTS: ${{ github.event.inputs.collect_pg_stat_statements || 'false' }}
        BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
        TEST_OLAP_SCALE: 10

@@ -190,9 +190,9 @@ dependencies = [

 [[package]]
 name = "async-compression"
-version = "0.4.5"
+version = "0.4.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "bc2d0cfb2a7388d34f590e76686704c494ed7aaceed62ee1ba35cbf363abc2a5"
+checksum = "5b0122885821398cc923ece939e24d1056a2384ee719432397fa9db87230ff11"
 dependencies = [
 "flate2",
 "futures-core",
@@ -572,7 +572,7 @@ dependencies = [
 "once_cell",
 "pin-project-lite",
 "pin-utils",
- "rustls 0.21.9",
+ "rustls",
 "tokio",
 "tracing",
 ]
@@ -2278,10 +2278,10 @@ dependencies = [
 "http",
 "hyper",
 "log",
- "rustls 0.21.9",
+ "rustls",
 "rustls-native-certs",
 "tokio",
- "tokio-rustls 0.24.0",
+ "tokio-rustls",
 ]

 [[package]]
@@ -2487,14 +2487,13 @@ dependencies = [

 [[package]]
 name = "jsonwebtoken"
-version = "9.2.0"
+version = "8.3.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "5c7ea04a7c5c055c175f189b6dc6ba036fd62306b58c66c9f6389036c503a3f4"
+checksum = "6971da4d9c3aa03c3d8f3ff0f4155b534aad021292003895a469716b2a230378"
 dependencies = [
 "base64 0.21.1",
- "js-sys",
- "pem",
- "ring 0.17.6",
+ "pem 1.1.1",
+ "ring 0.16.20",
 "serde",
 "serde_json",
 "simple_asn1",
@@ -3292,9 +3291,18 @@ checksum = "19b17cddbe7ec3f8bc800887bab5e717348c95ea2ca0b1bf0837fb964dc67099"

 [[package]]
 name = "pem"
-version = "3.0.3"
+version = "1.1.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "1b8fcc794035347fb64beda2d3b462595dd2753e3f268d89c5aae77e8cf2c310"
+checksum = "a8835c273a76a90455d7344889b0964598e3316e2a79ede8e36f16bdcf2228b8"
+dependencies = [
+ "base64 0.13.1",
+]
+
+[[package]]
+name = "pem"
+version = "2.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6b13fe415cdf3c8e44518e18a7c95a13431d9bdf6d15367d82b23c377fdd441a"
 dependencies = [
 "base64 0.21.1",
 "serde",
@@ -3472,14 +3480,14 @@ dependencies = [
 "futures",
 "once_cell",
 "pq_proto",
- "ring 0.17.6",
- "rustls 0.22.1",
- "rustls-pemfile 2.0.0",
+ "rustls",
+ "rustls-pemfile",
 "serde",
 "thiserror",
 "tokio",
 "tokio-postgres",
- "tokio-rustls 0.25.0",
+ "tokio-postgres-rustls",
+ "tokio-rustls",
 "tracing",
 "workspace_hack",
 ]
@@ -3707,8 +3715,8 @@ dependencies = [
 "routerify",
 "rstest",
 "rustc-hash",
- "rustls 0.22.1",
- "rustls-pemfile 2.0.0",
+ "rustls",
+ "rustls-pemfile",
 "scopeguard",
 "serde",
 "serde_json",
@@ -3722,7 +3730,7 @@ dependencies = [
 "tokio",
 "tokio-postgres",
 "tokio-postgres-rustls",
- "tokio-rustls 0.25.0",
+ "tokio-rustls",
 "tokio-util",
 "tracing",
 "tracing-opentelemetry",
@@ -3731,7 +3739,7 @@ dependencies = [
 "url",
 "utils",
 "uuid",
- "webpki-roots",
+ "webpki-roots 0.25.2",
 "workspace_hack",
 "x509-parser",
 ]
@@ -3850,12 +3858,12 @@ dependencies = [

 [[package]]
 name = "rcgen"
-version = "0.12.0"
+version = "0.11.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "5d918c80c5a4c7560db726763020bd16db179e4d5b828078842274a443addb5d"
+checksum = "4954fbc00dcd4d8282c987710e50ba513d351400dbdd00e803a05172a90d8976"
 dependencies = [
- "pem",
- "ring 0.17.6",
+ "pem 2.0.1",
+ "ring 0.16.20",
 "time",
 "yasna",
 ]
@@ -3993,14 +4001,14 @@ dependencies = [
 "once_cell",
 "percent-encoding",
 "pin-project-lite",
- "rustls 0.21.9",
- "rustls-pemfile 1.0.2",
+ "rustls",
+ "rustls-pemfile",
 "serde",
 "serde_json",
 "serde_urlencoded",
 "tokio",
 "tokio-native-tls",
- "tokio-rustls 0.24.0",
+ "tokio-rustls",
 "tokio-util",
 "tower-service",
 "url",
@@ -4008,7 +4016,7 @@ dependencies = [
 "wasm-bindgen-futures",
 "wasm-streams",
 "web-sys",
- "webpki-roots",
+ "webpki-roots 0.25.2",
 "winreg",
 ]

@@ -4240,20 +4248,6 @@ dependencies = [
 "sct",
 ]

-[[package]]
-name = "rustls"
-version = "0.22.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "fe6b63262c9fcac8659abfaa96cac103d28166d3ff3eaf8f412e19f3ae9e5a48"
-dependencies = [
- "log",
- "ring 0.17.6",
- "rustls-pki-types",
- "rustls-webpki 0.102.0",
- "subtle",
- "zeroize",
-]
-
 [[package]]
 name = "rustls-native-certs"
 version = "0.6.2"
@@ -4261,7 +4255,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "0167bac7a9f490495f3c33013e7722b53cb087ecbe082fb0c6387c96f634ea50"
 dependencies = [
 "openssl-probe",
- "rustls-pemfile 1.0.2",
+ "rustls-pemfile",
 "schannel",
 "security-framework",
 ]
@@ -4276,21 +4270,15 @@ dependencies = [
 ]

 [[package]]
-name = "rustls-pemfile"
-version = "2.0.0"
+name = "rustls-webpki"
+version = "0.100.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "35e4980fa29e4c4b212ffb3db068a564cbf560e51d3944b7c88bd8bf5bec64f4"
+checksum = "e98ff011474fa39949b7e5c0428f9b4937eda7da7848bbb947786b7be0b27dab"
 dependencies = [
- "base64 0.21.1",
- "rustls-pki-types",
+ "ring 0.16.20",
+ "untrusted 0.7.1",
 ]

-[[package]]
-name = "rustls-pki-types"
-version = "1.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e7673e0aa20ee4937c6aacfc12bb8341cfbf054cdd21df6bec5fd0629fe9339b"
-
 [[package]]
 name = "rustls-webpki"
 version = "0.101.7"
@@ -4301,17 +4289,6 @@ dependencies = [
 "untrusted 0.9.0",
 ]

-[[package]]
-name = "rustls-webpki"
-version = "0.102.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "de2635c8bc2b88d367767c5de8ea1d8db9af3f6219eba28442242d9ab81d1b89"
-dependencies = [
- "ring 0.17.6",
- "rustls-pki-types",
- "untrusted 0.9.0",
-]
-
 [[package]]
 name = "rustversion"
 version = "1.0.12"
@@ -4352,7 +4329,7 @@ dependencies = [
 "serde_with",
 "thiserror",
 "tokio",
- "tokio-rustls 0.25.0",
+ "tokio-rustls",
 "tokio-stream",
 "tracing",
 "tracing-appender",
@@ -4451,12 +4428,12 @@ checksum = "d29ab0c6d3fc0ee92fe66e2d99f700eab17a8d57d1c1d3b748380fb20baa78cd"

 [[package]]
 name = "sct"
-version = "0.7.1"
+version = "0.7.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "da046153aa2352493d6cb7da4b6e5c0c057d8a1d0a9aa8560baffdd945acd414"
+checksum = "d53dcdb7c9f8158937a7981b48accfd39a43af418591a5d008c7b22b5e1b7ca4"
 dependencies = [
- "ring 0.17.6",
- "untrusted 0.9.0",
+ "ring 0.16.20",
+ "untrusted 0.7.1",
 ]

 [[package]]
@@ -4516,7 +4493,7 @@ checksum = "2e95efd0cefa32028cdb9766c96de71d96671072f9fb494dc9fb84c0ef93e52b"
 dependencies = [
 "httpdate",
 "reqwest",
- "rustls 0.21.9",
+ "rustls",
 "sentry-backtrace",
 "sentry-contexts",
 "sentry-core",
@@ -4524,7 +4501,7 @@ dependencies = [
 "sentry-tracing",
 "tokio",
 "ureq",
- "webpki-roots",
+ "webpki-roots 0.25.2",
 ]

 [[package]]
@@ -5182,14 +5159,16 @@ checksum = "1f3ccbac311fea05f86f61904b462b55fb3df8837a366dfc601a0161d0532f20"

 [[package]]
 name = "tls-listener"
-version = "0.9.0"
-source = "git+https://github.com/conradludgate/tls-listener?branch=main#4801141b5660613e77816044da6540aa64f388ec"
+version = "0.7.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "81294c017957a1a69794f506723519255879e15a870507faf45dfed288b763dd"
 dependencies = [
 "futures-util",
+ "hyper",
 "pin-project-lite",
 "thiserror",
 "tokio",
- "tokio-rustls 0.25.0",
+ "tokio-rustls",
 ]

 [[package]]
@@ -5272,10 +5251,10 @@ checksum = "dd5831152cb0d3f79ef5523b357319ba154795d64c7078b2daa95a803b54057f"
 dependencies = [
 "futures",
 "ring 0.16.20",
- "rustls 0.21.9",
+ "rustls",
 "tokio",
 "tokio-postgres",
- "tokio-rustls 0.24.0",
+ "tokio-rustls",
 ]

 [[package]]
@@ -5284,18 +5263,7 @@ version = "0.24.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e0d409377ff5b1e3ca6437aa86c1eb7d40c134bfec254e44c830defa92669db5"
 dependencies = [
- "rustls 0.21.9",
- "tokio",
-]
-
-[[package]]
-name = "tokio-rustls"
-version = "0.25.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "775e0c0f0adb3a2f22a00c4745d728b479985fc15ee7ca6a2608388c5569860f"
-dependencies = [
- "rustls 0.22.1",
- "rustls-pki-types",
+ "rustls",
 "tokio",
 ]

@@ -5442,9 +5410,9 @@ dependencies = [
 "pin-project",
 "prost",
 "rustls-native-certs",
- "rustls-pemfile 1.0.2",
+ "rustls-pemfile",
 "tokio",
- "tokio-rustls 0.24.0",
+ "tokio-rustls",
 "tokio-stream",
 "tower",
 "tower-layer",
@@ -5749,17 +5717,17 @@ checksum = "8ecb6da28b8a351d773b68d5825ac39017e680750f980f3a1a85cd8dd28a47c1"

 [[package]]
 name = "ureq"
-version = "2.9.1"
+version = "2.7.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f8cdd25c339e200129fe4de81451814e5228c9b771d57378817d6117cc2b3f97"
+checksum = "0b11c96ac7ee530603dcdf68ed1557050f374ce55a5a07193ebf8cbc9f8927e9"
 dependencies = [
 "base64 0.21.1",
 "log",
 "once_cell",
- "rustls 0.21.9",
- "rustls-webpki 0.101.7",
+ "rustls",
+ "rustls-webpki 0.100.2",
 "url",
- "webpki-roots",
+ "webpki-roots 0.23.1",
 ]

 [[package]]
@@ -6068,6 +6036,15 @@ dependencies = [
 "wasm-bindgen",
 ]

+[[package]]
+name = "webpki-roots"
+version = "0.23.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b03058f88386e5ff5310d9111d53f48b17d732b401aeb83a8d5190f2ac459338"
+dependencies = [
+ "rustls-webpki 0.100.2",
+]
+
 [[package]]
 name = "webpki-roots"
 version = "0.25.2"
@@ -6316,8 +6293,11 @@ dependencies = [
 "either",
 "fail",
 "futures",
+ "futures-channel",
+ "futures-core",
 "futures-executor",
 "futures-io",
+ "futures-sink",
 "futures-util",
 "hex",
 "hmac",
@@ -6336,7 +6316,8 @@ dependencies = [
 "regex-automata 0.4.3",
 "regex-syntax 0.8.2",
 "reqwest",
- "rustls 0.21.9",
+ "ring 0.16.20",
+ "rustls",
 "scopeguard",
 "serde",
 "serde_json",
@@ -6347,7 +6328,7 @@ dependencies = [
 "time",
 "time-macros",
 "tokio",
- "tokio-rustls 0.24.0",
+ "tokio-rustls",
 "tokio-util",
 "toml_datetime",
 "toml_edit",
@@ -6431,28 +6412,30 @@ checksum = "2a0956f1ba7c7909bfb66c2e9e4124ab6f6482560f6628b5aaeba39207c9aad9"

 [[package]]
 name = "zstd"
-version = "0.13.0"
+version = "0.12.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "bffb3309596d527cfcba7dfc6ed6052f1d39dfbd7c867aa2e865e4a449c10110"
+checksum = "1a27595e173641171fc74a1232b7b1c7a7cb6e18222c11e9dfb9888fa424c53c"
 dependencies = [
 "zstd-safe",
 ]

 [[package]]
 name = "zstd-safe"
-version = "7.0.0"
+version = "6.0.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "43747c7422e2924c11144d5229878b98180ef8b06cca4ab5af37afc8a8d8ea3e"
+checksum = "ee98ffd0b48ee95e6c5168188e44a54550b1564d9d530ee21d5f0eaed1069581"
 dependencies = [
+ "libc",
 "zstd-sys",
 ]

 [[package]]
 name = "zstd-sys"
-version = "2.0.9+zstd.1.5.5"
+version = "2.0.8+zstd.1.5.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9e16efa8a874a0481a574084d34cc26fdb3b99627480f785888deb6386506656"
+checksum = "5556e6ee25d32df2586c098bbfa278803692a20d0ab9565e049480d52707ec8c"
 dependencies = [
 "cc",
+ "libc",
 "pkg-config",
 ]
@@ -91,7 +91,7 @@ hyper-tungstenite = "0.11"
 inotify = "0.10.2"
 ipnet = "2.9.0"
 itertools = "0.10"
-jsonwebtoken = "9"
+jsonwebtoken = "8"
 libc = "0.2"
 md5 = "0.7.0"
 memoffset = "0.8"
@@ -115,12 +115,11 @@ reqwest = { version = "0.11", default-features = false, features = ["rustls-tls"
 reqwest-tracing = { version = "0.4.0", features = ["opentelemetry_0_19"] }
 reqwest-middleware = "0.2.0"
 reqwest-retry = "0.2.2"
-ring = "0.17"
 routerify = "3"
 rpds = "0.13"
 rustc-hash = "1.1.0"
-rustls = "0.22.1"
-rustls-pemfile = "2.0.0"
+rustls = "0.21"
+rustls-pemfile = "1"
 rustls-split = "0.3"
 scopeguard = "1.1"
 sysinfo = "0.29.2"
@@ -144,11 +143,11 @@ tar = "0.4"
 task-local-extensions = "0.1.4"
 test-context = "0.1"
 thiserror = "1.0"
-tls-listener = { version = "0.9.0", features = ["rustls"] }
+tls-listener = { version = "0.7", features = ["rustls", "hyper-h1"] }
 tokio = { version = "1.17", features = ["macros"] }
 tokio-io-timeout = "1.2.0"
 tokio-postgres-rustls = "0.10.0"
-tokio-rustls = "0.25.0"
+tokio-rustls = "0.24"
 tokio-stream = "0.1"
 tokio-tar = "0.3"
 tokio-util = { version = "0.7.10", features = ["io", "rt"] }
@@ -203,7 +202,7 @@ workspace_hack = { version = "0.1", path = "./workspace_hack/" }

 ## Build dependencies
 criterion = "0.5.1"
-rcgen = "0.12"
+rcgen = "0.11"
 rstest = "0.18"
 camino-tempfile = "1.0.2"
 tonic-build = "0.9"
@@ -214,8 +213,6 @@ tonic-build = "0.9"
 # TODO: we should probably fork `tokio-postgres-rustls` instead.
 tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", branch="neon" }

-tls-listener = { git = "https://github.com/conradludgate/tls-listener", branch="main" }
-
 ################# Binary contents sections

 [profile.release]
@@ -569,23 +569,6 @@ RUN wget https://github.com/ChenHuajun/pg_roaringbitmap/archive/refs/tags/v0.5.4
    make -j $(getconf _NPROCESSORS_ONLN) install && \
    echo 'trusted = true' >> /usr/local/pgsql/share/extension/roaringbitmap.control

-#########################################################################################
-#
-# Layer "pg-semver-pg-build"
-# compile pg_semver extension
-#
-#########################################################################################
-FROM build-deps AS pg-semver-pg-build
-COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
-
-ENV PATH "/usr/local/pgsql/bin/:$PATH"
-RUN wget https://github.com/theory/pg-semver/archive/refs/tags/v0.32.1.tar.gz -O pg_semver.tar.gz && \
-    echo "fbdaf7512026d62eec03fad8687c15ed509b6ba395bff140acd63d2e4fbe25d7 pg_semver.tar.gz" | sha256sum --check && \
-    mkdir pg_semver-src && cd pg_semver-src && tar xvzf ../pg_semver.tar.gz --strip-components=1 -C . && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/semver.control
-
 #########################################################################################
 #
 # Layer "pg-embedding-pg-build"
@@ -785,7 +768,6 @@ COPY --from=pg-pgx-ulid-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=rdkit-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-uuidv7-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-roaringbitmap-pg-build /usr/local/pgsql/ /usr/local/pgsql/
-COPY --from=pg-semver-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-embedding-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=wal2json-pg-build /usr/local/pgsql /usr/local/pgsql
 COPY pgxn/ pgxn/
@@ -37,5 +37,5 @@ workspace_hack.workspace = true
 toml_edit.workspace = true
 remote_storage = { version = "0.1", path = "../libs/remote_storage/" }
 vm_monitor = { version = "0.1", path = "../libs/vm_monitor/" }
-zstd = "0.13"
+zstd = "0.12.4"
 bytes = "1.0"
@@ -1,197 +0,0 @@
-# Per-Tenant GetPage@LSN Throttling
-
-Author: Christian Schwarz
-Date: Oct 24, 2023
-
-## Summary
-
-This RFC proposes per-tenant throttling of GetPage@LSN requests inside Pageserver
-and the interactions with its client, i.e., the neon_smgr component in Compute.
-
-The result of implementing & executing this RFC will be a fleet-wide upper limit for
-**"the highest GetPage/second that Pageserver can support for a single tenant/shard"**.
-
-## Background
-
-### GetPage@LSN Request Flow
-
-Pageserver exposes its `page_service.rs` as a libpq listener.
-The Computes' `neon_smgr` module connects to that libpq listener.
-Once a connection is established, the protocol allows Compute to request page images at a given LSN.
-We call these requests GetPage@LSN requests, or GetPage requests for short.
-Other request types can be sent, but these are low traffic compared to GetPage requests
-and are not the concern of this RFC.
-
-Pageserver associates one libpq connection with one tokio task.
-
-Per connection/task, the pq protocol is handled by the common `postgres_backend` crate.
-Its `run_message_loop` function invokes the `page_service` specific `impl<IO> postgres_backend::Handler<IO> for PageServerHandler`.
-Requests are processed in the order in which they arrive via the TCP-based pq protocol.
-So, there is no concurrent request processing within one connection/task.
-
-There is a degree of natural pipelining:
-Compute can "fill the pipe" by sending more than one GetPage request into the libpq TCP stream.
-And Pageserver can fill the pipe with responses in the other direction.
-Both directions are subject to the limit of tx/rx buffers, nodelay, TCP flow control, etc.
-
-### GetPage@LSN Access Pattern
-
-The Compute has its own hierarchy of caches, specifically `shared_buffers` and the `local file cache` (LFC).
-Compute only issues GetPage requests to Pageserver if it encounters a miss in these caches.
-
-If the working set stops fitting into Compute's caches, requests to Pageserver increase sharply -- the Compute starts *thrashing*.
-
-## Motivation
-
-In INC-69, a tenant issued 155k GetPage/second for a period of 10 minutes and 60k GetPage/second for a period of 3h,
-then dropping to ca 18k GetPage/second for a period of 9h.
-
-We noticed this because of an internal GetPage latency SLO burn rate alert, i.e.,
-the request latency profile during this period significantly exceeded what was acceptable according to the internal SLO.
-
-Sadly, we do not have the observability data to determine the impact of this tenant on other tenants on the same tenants.
-
-However, here are some illustrative data points for the 155k period:
-The tenant was responsible for >= 99% of the GetPage traffic and, frankly, the overall activity on this Pageserver instance.
-We were serving pages at 10 Gb/s (`155k x 8 kbyte (PAGE_SZ) per second is 1.12GiB/s = 9.4Gb/s.`)
-The CPU utilization of the instance was 75% user+system.
-Pageserver page cache served 1.75M accesses/second at a hit rate of ca 90%.
-The hit rate for materialized pages was ca. 40%.
-Curiously, IOPS to the Instance Store NVMe were very low, rarely exceeding 100.
-
-The fact that the IOPS were so low / the materialized page cache hit rate was so high suggests that **this tenant's compute's caches were thrashing**.
-The compute was of type `k8s-pod`; hence, auto-scaling could/would not have helped remediate the thrashing by provisioning more RAM.
-The consequence was that the **thrashing translated into excessive GetPage requests against Pageserver**.
-
-My claim is that it was **unhealthy to serve this workload at the pace we did**:
-* it is likely that other tenants were/would have experienced high latencies (again, we sadly don't have per-tenant latency data to confirm this)
-* more importantly, it was **unsustainable** to serve traffic at this pace for multiple reasons:
-    * **predictability of performance**: when the working set grows, the pageserver materialized page cache hit rate drops.
-      At some point, we're bound by the EC2 Instance Store NVMe drive's IOPS limit.
-      The result is an **uneven** performance profile from the Compute perspective.
-
-    * **economics**: Neon currently does not charge for IOPS, only capacity.
-      **We cannot afford to undercut the market in IOPS/$ this drastically; it leads to adverse selection and perverse incentives.**
-      For example, the 155k IOPS, which we served for 10min, would cost ca. 6.5k$/month when provisioned as an io2 EBS volume.
-      Even the 18k IOPS, which we served for 9h, would cost ca. 1.1k$/month when provisioned as an io2 EBS volume.
-      We charge 0$.
-      It could be economically advantageous to keep using a low-DRAM compute because Pageserver IOPS are fast enough and free.
-
-
-Note: It is helpful to think of Pageserver as a disk, because it's precisely where `neon_smgr` sits:
-vanilla Postgres gets its pages from disk, Neon Postgres gets them from Pageserver.
-So, regarding the above performance & economic arguments, it is fair to say that we currently provide an "as-fast-as-possible-IOPS" disk that we charge for only by capacity.
-
-## Solution: Throttling GetPage Requests
-
-**The consequence of the above analysis must be that Pageserver throttles GetPage@LSN requests**.
-That is, unless we want to start charging for provisioned GetPage@LSN/second.
-Throttling sets the correct incentive for a thrashing Compute to scale up its DRAM to the working set size.
-Neon Autoscaling will make this easy, [eventually](https://github.com/neondatabase/neon/pull/3913).
-
-## The Design Space
-
-What that remains is the question about *policy* and *mechanism*:
-
-**Policy** concerns itself with the question of what limit applies to a given connection|timeline|tenant.
-Candidates are:
-
-* hard limit, same limit value per connection|timeline|tenant
-    * Per-tenant will provide an upper bound for the impact of a tenant on a given Pageserver instance.
-      This is a major operational pain point / risk right now.
-* hard limit, configurable per connection|timeline|tenant
-    * This outsources policy to console/control plane, with obvious advantages for flexible structuring of what service we offer to customers.
-    * Note that this is not a mechanism to guarantee a minium provisioned rate, i.e., this is not a mechanism to guarantee a certain QoS for a tenant.
-* fair share among active connections|timelines|tenants per instance
-    * example: each connection|timeline|tenant gets a fair fraction of the machine's GetPage/second capacity
-    * NB: needs definition of "active", and knowledge of available GetPage/second capacity in advance
-* ...
-
-
-Regarding **mechanism**, it's clear that **backpressure** is the way to go.
-However, we must choose between
-* **implicit** backpressure through pq/TCP and
-* **explicit** rejection of requests + retries with exponential backoff
-
-Further, there is the question of how throttling GetPage@LSN will affect the **internal GetPage latency SLO**:
-where do we measure the SLI for Pageserver's internal getpage latency SLO? Before or after the throttling?
-
-And when we eventually move the measurement point into the Computes (to avoid coordinated omission),
-how do we avoid counting throttling-induced latency toward the internal getpage latency SLI/SLO?
-
-## Scope Of This RFC
-
-**This RFC proposes introducing a hard GetPage@LSN/second limit per tenant, with the same value applying to each tenant on a Pageserver**.
-
-This proposal is easy to implement and significantly de-risks operating large Pageservers,
-based on the assumption that extremely-high-GetPage-rate-episodes like the one from the "Motivation" section are uncorrelated between tenants.
-
-For example, suppose we pick a limit that allows up to 10 tenants to go at limit rate.
-Suppose our Pageserver can serve 100k GetPage/second total at a 100% page cache miss rate.
-If each tenant gets a hard limit of 10k GetPage/second, we can serve up to 10 tenants at limit speed without latency degradation.
-
-The mechanism for backpressure will be TCP-based implicit backpressure.
-The compute team isn't concerned about prefetch queue depth.
-Pageserver will implement it by delaying the reading of requests from the libpq connection(s).
-
-The rate limit will be implemented using a per-tenant token bucket.
-The bucket will be be shared among all connections to the tenant.
-The bucket implementation supports starvation-preventing `await`ing.
-The current candidate for the implementation is [`leaky_bucket`](https://docs.rs/leaky-bucket/).
-The getpage@lsn benchmark that's being added in https://github.com/neondatabase/neon/issues/5771
-can be used to evaluate the overhead of sharing the bucket among connections of a tenant.
-A possible technique to mitigate the impact of sharing the bucket would be to maintain a buffer of a few tokens per connection handler.
-
-Regarding metrics / the internal GetPage latency SLO:
-we will measure the GetPage latency SLO _after_ the throttler and introduce a new metric to measure the amount of throttling, quantified by:
- histogram that records the tenants' observations of queue depth before they start waiting (one such histogram per pageserver)
- histogram that records the tenants' observations of time spent waiting (one such histogram per pageserver)
-
-Further observability measures:
- an INFO log message at frequency 1/min if the tenant/timeline/connection was throttled in that last minute.
-  The message will identify the tenant/timeline/connection to allow correlation with compute logs/stats.
-
-Rollout will happen as follows:
- deploy 1: implementation + config: disabled by default, ability to enable it per tenant through tenant_conf
- experimentation in staging and later production to study impact & interaction with auto-scaling
- determination of a sensible global default value
-  - the value will be chosen as high as possible ...
-  - ... but low enough to work towards this RFC's goal that one tenant should not be able to dominate a pageserver instance.
- deploy 2: implementation fixes if any + config: enabled by default with the aforementioned global default
- reset of the experimental per-tenant overrides
- gain experience & lower the limit over time
-  - we stop lowering the limit as soon as this RFC's goal is achieved, i.e.,
-    once we decide that in practice the chosen value sufficiently de-risks operating large pageservers
-
-The per-tenant override will remain for emergencies and testing.
-But since Console doesn't preserve it during tenant migrations, it isn't durably configurable for the tenant.
-
-Toward the upper layers of the Neon stack, the resulting limit will be
-**"the highest GetPage/second that Pageserver can support for a single tenant"**.
-
-### Rationale
-
-We decided against error + retry because of worries about starvation.
-
-## Future Work
-
-Enable per-tenant emergency override of the limit via Console.
-Should be part of a more general framework to specify tenant config overrides.
-**NB:** this is **not** the right mechanism to _sell_ different max GetPage/second levels to users,
-or _auto-scale_ the GetPage/second levels. Such functionality will require a separate RFC that
-concerns itself with GetPage/second capacity planning.
-
-Compute-side metrics for GetPage latency.
-
-Back-channel to inform Compute/Autoscaling/ControlPlane that the project is being throttled.
-
-Compute-side neon_smgr improvements to avoid sending the same GetPage request multiple times if multiple backends experience a cache miss.
-
-Dealing with read-only endpoints: users use read-only endpoints to scale reads for a single tenant.
-Possibly there are also assumptions around read-only endpoints not affecting the primary read-write endpoint's performance.
-With per-tenant rate limiting, we will not meet that expectation.
-However, we can currently only scale per tenant.
-Soon, we will have sharding (#5505), which will apply the throttling on a per-shard basis.
-But, that's orthogonal to scaling reads: if many endpoints hit one shard, they share the same throttling limit.
-To solve this properly, I think we'll need replicas for tenants / shard.
-To performance-isolate a tenant's endpoints from each other, we'd then route them to different replicas.
@@ -370,14 +370,6 @@ pub struct TenantInfo {
    pub attachment_status: TenantAttachmentStatus,
 }

-#[derive(Serialize, Deserialize, Clone)]
-pub struct TenantDetails {
-    #[serde(flatten)]
-    pub tenant_info: TenantInfo,
-
-    pub timelines: Vec<TimelineId>,
-}
-
 /// This represents the output of the "timeline_detail" and "timeline_list" API calls.
 #[derive(Debug, Serialize, Deserialize, Clone)]
 pub struct TimelineInfo {
@@ -9,12 +9,10 @@ async-trait.workspace = true
 anyhow.workspace = true
 bytes.workspace = true
 futures.workspace = true
-ring.workspace = true
 rustls.workspace = true
 serde.workspace = true
 thiserror.workspace = true
 tokio.workspace = true
-tokio-postgres.workspace = true
 tokio-rustls.workspace = true
 tracing.workspace = true

@@ -24,4 +22,5 @@ workspace_hack.workspace = true
 [dev-dependencies]
 once_cell.workspace = true
 rustls-pemfile.workspace = true
-# tokio-postgres-rustls.workspace = true
+tokio-postgres.workspace = true
+tokio-postgres-rustls.workspace = true
@@ -6,7 +6,7 @@
 #![deny(clippy::undocumented_unsafe_blocks)]
 use anyhow::Context;
 use bytes::Bytes;
-use futures::{pin_mut, TryFutureExt, FutureExt};
+use futures::pin_mut;
 use serde::{Deserialize, Serialize};
 use std::io::ErrorKind;
 use std::net::SocketAddr;
@@ -1030,115 +1030,3 @@ pub enum CopyStreamHandlerEnd {
    #[error(transparent)]
    Other(#[from] anyhow::Error),
 }
-
-#[derive(Clone)]
-pub struct MakeRustlsConnect {
-    config: Arc<rustls::ClientConfig>,
-}
-
-impl MakeRustlsConnect {
-    pub fn new(config: rustls::ClientConfig) -> Self {
-        Self {
-            config: Arc::new(config),
-        }
-    }
-}
-
-impl<S> tokio_postgres::tls::MakeTlsConnect<S> for MakeRustlsConnect
-where
-    S: AsyncRead + AsyncWrite + Unpin + Send + 'static,
-{
-    type Stream = RustlsStream<S>;
-    type TlsConnect = RustlsConnect;
-    type Error = io::Error;
-
-    fn make_tls_connect(&mut self, hostname: &str) -> io::Result<RustlsConnect> {
-        rustls::pki_types::ServerName::try_from(hostname)
-            .map(|dns_name| {
-                RustlsConnect(Some(RustlsConnectData {
-                    hostname: dns_name.to_owned(),
-                    connector: Arc::clone(&self.config).into(),
-                }))
-            })
-            .or(Ok(RustlsConnect(None)))
-    }
-}
-
-pub struct RustlsConnect(Option<RustlsConnectData>);
-
-struct RustlsConnectData {
-    hostname: rustls::pki_types::ServerName<'static>,
-    connector: tokio_rustls::TlsConnector,
-}
-
-impl<S> tokio_postgres::tls::TlsConnect<S> for RustlsConnect
-where
-    S: AsyncRead + AsyncWrite + Unpin + Send + 'static,
-{
-    type Stream = RustlsStream<S>;
-    type Error = io::Error;
-    type Future = Pin<Box<dyn Future<Output = io::Result<RustlsStream<S>>> + Send>>;
-
-    fn connect(self, stream: S) -> Self::Future {
-        match self.0 {
-            None => Box::pin(core::future::ready(Err(io::ErrorKind::InvalidInput.into()))),
-            Some(c) => c
-                .connector
-                .connect(c.hostname, stream)
-                .map_ok(|s| RustlsStream(Box::pin(s)))
-                .boxed(),
-        }
-    }
-}
-
-pub struct RustlsStream<S>(Pin<Box<tokio_rustls::client:: TlsStream<S>>>);
-
-impl<S> tokio_postgres::tls::TlsStream for RustlsStream<S>
-where
-    S: AsyncRead + AsyncWrite + Unpin,
-{
-    fn channel_binding(&self) -> tokio_postgres::tls::ChannelBinding {
-        let (_, session) = self.0.get_ref();
-        match session.peer_certificates() {
-            Some(certs) if !certs.is_empty() => {
-                let sha256 = ring::digest::digest(&ring::digest::SHA256, certs[0].as_ref());
-                tokio_postgres::tls::ChannelBinding::tls_server_end_point(sha256.as_ref().into())
-            }
-            _ => tokio_postgres::tls::ChannelBinding::none(),
-        }
-    }
-}
-
-impl<S> AsyncRead for RustlsStream<S>
-where
-    S: AsyncRead + AsyncWrite + Unpin,
-{
-    fn poll_read(
-        mut self: Pin<&mut Self>,
-        cx: &mut std::task:: Context,
-        buf: &mut tokio::io::ReadBuf<'_>,
-    ) -> Poll<tokio::io::Result<()>> {
-        self.0.as_mut().poll_read(cx, buf)
-    }
-}
-
-impl<S> AsyncWrite for RustlsStream<S>
-where
-    S: AsyncRead + AsyncWrite + Unpin,
-{
-    fn poll_write(
-        mut self: Pin<&mut Self>,
-        cx: &mut std::task:: Context,
-        buf: &[u8],
-    ) -> Poll<tokio::io::Result<usize>> {
-        self.0.as_mut().poll_write(cx, buf)
-    }
-
-    fn poll_flush(mut self: Pin<&mut Self>, cx: &mut std::task:: Context) -> Poll<tokio::io::Result<()>> {
-        self.0.as_mut().poll_flush(cx)
-    }
-
-    fn poll_shutdown(mut self: Pin<&mut Self>, cx: &mut std::task:: Context) -> Poll<tokio::io::Result<()>> {
-        self.0.as_mut().poll_shutdown(cx)
-    }
-}
@@ -1,6 +1,5 @@
 /// Test postgres_backend_async with tokio_postgres
 use once_cell::sync::Lazy;
-use postgres_backend::MakeRustlsConnect;
 use postgres_backend::{AuthType, Handler, PostgresBackend, QueryError};
 use pq_proto::{BeMessage, RowDescriptor};
 use std::io::Cursor;
@@ -10,6 +9,7 @@ use tokio::net::{TcpListener, TcpStream};
 use tokio_postgres::config::SslMode;
 use tokio_postgres::tls::MakeTlsConnect;
 use tokio_postgres::{Config, NoTls, SimpleQueryMessage};
+use tokio_postgres_rustls::MakeRustlsConnect;

 // generate client, server test streams
 async fn make_tcp_pair() -> (TcpStream, TcpStream) {
@@ -72,21 +72,14 @@ async fn simple_select() {
    }
 }

-static KEY: Lazy<rustls::pki_types::PrivatePkcs1KeyDer<'static>> = Lazy::new(|| {
+static KEY: Lazy<rustls::PrivateKey> = Lazy::new(|| {
    let mut cursor = Cursor::new(include_bytes!("key.pem"));
-
-    let key = rustls_pemfile::rsa_private_keys(&mut cursor)
-        .next()
-        .unwrap()
-        .unwrap();
-    key.secret_pkcs1_der().to_owned().into()
+    rustls::PrivateKey(rustls_pemfile::rsa_private_keys(&mut cursor).unwrap()[0].clone())
 });

-static CERT: Lazy<rustls::pki_types::CertificateDer<'static>> = Lazy::new(|| {
+static CERT: Lazy<rustls::Certificate> = Lazy::new(|| {
    let mut cursor = Cursor::new(include_bytes!("cert.pem"));
-    let cert = rustls_pemfile::certs(&mut cursor).next().unwrap().unwrap();
-
-    cert.into_owned()
+    rustls::Certificate(rustls_pemfile::certs(&mut cursor).unwrap()[0].clone())
 });

 // test that basic select with ssl works
@@ -94,10 +87,10 @@ static CERT: Lazy<rustls::pki_types::CertificateDer<'static>> = Lazy::new(|| {
 async fn simple_select_ssl() {
    let (client_sock, server_sock) = make_tcp_pair().await;

-    let key = rustls::pki_types::PrivateKeyDer::Pkcs1(KEY.secret_pkcs1_der().to_owned().into());
    let server_cfg = rustls::ServerConfig::builder()
+        .with_safe_defaults()
        .with_no_client_auth()
-        .with_single_cert(vec![CERT.clone()], key)
+        .with_single_cert(vec![CERT.clone()], KEY.clone())
        .unwrap();
    let tls_config = Some(Arc::new(server_cfg));
    let pgbackend =
@@ -109,13 +102,14 @@ async fn simple_select_ssl() {
    });

    let client_cfg = rustls::ClientConfig::builder()
+        .with_safe_defaults()
        .with_root_certificates({
            let mut store = rustls::RootCertStore::empty();
-            store.add(CERT.clone()).unwrap();
+            store.add(&CERT).unwrap();
            store
        })
        .with_no_client_auth();
-    let mut make_tls_connect = MakeRustlsConnect::new(client_cfg);
+    let mut make_tls_connect = tokio_postgres_rustls::MakeRustlsConnect::new(client_cfg);
    let tls_connect = <MakeRustlsConnect as MakeTlsConnect<TcpStream>>::make_tls_connect(
        &mut make_tls_connect,
        "localhost",
@@ -1,200 +0,0 @@
-use std::collections::HashSet;
-use std::ops::ControlFlow;
-use std::path::PathBuf;
-use std::sync::Arc;
-
-use anyhow::Context;
-use bytes::Bytes;
-use camino::Utf8Path;
-use futures::stream::Stream;
-use once_cell::sync::OnceCell;
-use remote_storage::{Download, GenericRemoteStorage, RemotePath};
-use tokio::task::JoinSet;
-use tracing::{debug, error, info};
-
-static LOGGING_DONE: OnceCell<()> = OnceCell::new();
-
-pub(crate) fn upload_stream(
-    content: std::borrow::Cow<'static, [u8]>,
-) -> (
-    impl Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static,
-    usize,
-) {
-    use std::borrow::Cow;
-
-    let content = match content {
-        Cow::Borrowed(x) => Bytes::from_static(x),
-        Cow::Owned(vec) => Bytes::from(vec),
-    };
-    wrap_stream(content)
-}
-
-pub(crate) fn wrap_stream(
-    content: bytes::Bytes,
-) -> (
-    impl Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static,
-    usize,
-) {
-    let len = content.len();
-    let content = futures::future::ready(Ok(content));
-
-    (futures::stream::once(content), len)
-}
-
-pub(crate) async fn download_to_vec(dl: Download) -> anyhow::Result<Vec<u8>> {
-    let mut buf = Vec::new();
-    tokio::io::copy_buf(
-        &mut tokio_util::io::StreamReader::new(dl.download_stream),
-        &mut buf,
-    )
-    .await?;
-    Ok(buf)
-}
-
-// Uploads files `folder{j}/blob{i}.txt`. See test description for more details.
-pub(crate) async fn upload_simple_remote_data(
-    client: &Arc<GenericRemoteStorage>,
-    upload_tasks_count: usize,
-) -> ControlFlow<HashSet<RemotePath>, HashSet<RemotePath>> {
-    info!("Creating {upload_tasks_count} remote files");
-    let mut upload_tasks = JoinSet::new();
-    for i in 1..upload_tasks_count + 1 {
-        let task_client = Arc::clone(client);
-        upload_tasks.spawn(async move {
-            let blob_path = PathBuf::from(format!("folder{}/blob_{}.txt", i / 7, i));
-            let blob_path = RemotePath::new(
-                Utf8Path::from_path(blob_path.as_path()).expect("must be valid blob path"),
-            )
-            .with_context(|| format!("{blob_path:?} to RemotePath conversion"))?;
-            debug!("Creating remote item {i} at path {blob_path:?}");
-
-            let (data, len) = upload_stream(format!("remote blob data {i}").into_bytes().into());
-            task_client.upload(data, len, &blob_path, None).await?;
-
-            Ok::<_, anyhow::Error>(blob_path)
-        });
-    }
-
-    let mut upload_tasks_failed = false;
-    let mut uploaded_blobs = HashSet::with_capacity(upload_tasks_count);
-    while let Some(task_run_result) = upload_tasks.join_next().await {
-        match task_run_result
-            .context("task join failed")
-            .and_then(|task_result| task_result.context("upload task failed"))
-        {
-            Ok(upload_path) => {
-                uploaded_blobs.insert(upload_path);
-            }
-            Err(e) => {
-                error!("Upload task failed: {e:?}");
-                upload_tasks_failed = true;
-            }
-        }
-    }
-
-    if upload_tasks_failed {
-        ControlFlow::Break(uploaded_blobs)
-    } else {
-        ControlFlow::Continue(uploaded_blobs)
-    }
-}
-
-pub(crate) async fn cleanup(
-    client: &Arc<GenericRemoteStorage>,
-    objects_to_delete: HashSet<RemotePath>,
-) {
-    info!(
-        "Removing {} objects from the remote storage during cleanup",
-        objects_to_delete.len()
-    );
-    let mut delete_tasks = JoinSet::new();
-    for object_to_delete in objects_to_delete {
-        let task_client = Arc::clone(client);
-        delete_tasks.spawn(async move {
-            debug!("Deleting remote item at path {object_to_delete:?}");
-            task_client
-                .delete(&object_to_delete)
-                .await
-                .with_context(|| format!("{object_to_delete:?} removal"))
-        });
-    }
-
-    while let Some(task_run_result) = delete_tasks.join_next().await {
-        match task_run_result {
-            Ok(task_result) => match task_result {
-                Ok(()) => {}
-                Err(e) => error!("Delete task failed: {e:?}"),
-            },
-            Err(join_err) => error!("Delete task did not finish correctly: {join_err}"),
-        }
-    }
-}
-pub(crate) struct Uploads {
-    pub(crate) prefixes: HashSet<RemotePath>,
-    pub(crate) blobs: HashSet<RemotePath>,
-}
-
-pub(crate) async fn upload_remote_data(
-    client: &Arc<GenericRemoteStorage>,
-    base_prefix_str: &'static str,
-    upload_tasks_count: usize,
-) -> ControlFlow<Uploads, Uploads> {
-    info!("Creating {upload_tasks_count} remote files");
-    let mut upload_tasks = JoinSet::new();
-    for i in 1..upload_tasks_count + 1 {
-        let task_client = Arc::clone(client);
-        upload_tasks.spawn(async move {
-            let prefix = format!("{base_prefix_str}/sub_prefix_{i}/");
-            let blob_prefix = RemotePath::new(Utf8Path::new(&prefix))
-                .with_context(|| format!("{prefix:?} to RemotePath conversion"))?;
-            let blob_path = blob_prefix.join(Utf8Path::new(&format!("blob_{i}")));
-            debug!("Creating remote item {i} at path {blob_path:?}");
-
-            let (data, data_len) =
-                upload_stream(format!("remote blob data {i}").into_bytes().into());
-            task_client.upload(data, data_len, &blob_path, None).await?;
-
-            Ok::<_, anyhow::Error>((blob_prefix, blob_path))
-        });
-    }
-
-    let mut upload_tasks_failed = false;
-    let mut uploaded_prefixes = HashSet::with_capacity(upload_tasks_count);
-    let mut uploaded_blobs = HashSet::with_capacity(upload_tasks_count);
-    while let Some(task_run_result) = upload_tasks.join_next().await {
-        match task_run_result
-            .context("task join failed")
-            .and_then(|task_result| task_result.context("upload task failed"))
-        {
-            Ok((upload_prefix, upload_path)) => {
-                uploaded_prefixes.insert(upload_prefix);
-                uploaded_blobs.insert(upload_path);
-            }
-            Err(e) => {
-                error!("Upload task failed: {e:?}");
-                upload_tasks_failed = true;
-            }
-        }
-    }
-
-    let uploads = Uploads {
-        prefixes: uploaded_prefixes,
-        blobs: uploaded_blobs,
-    };
-    if upload_tasks_failed {
-        ControlFlow::Break(uploads)
-    } else {
-        ControlFlow::Continue(uploads)
-    }
-}
-
-pub(crate) fn ensure_logging_ready() {
-    LOGGING_DONE.get_or_init(|| {
-        utils::logging::init(
-            utils::logging::LogFormat::Test,
-            utils::logging::TracingErrorLayerEnablement::Disabled,
-            utils::logging::Output::Stdout,
-        )
-        .expect("logging init failed");
-    });
-}
@@ -2,23 +2,23 @@ use std::collections::HashSet;
 use std::env;
 use std::num::NonZeroUsize;
 use std::ops::ControlFlow;
+use std::path::PathBuf;
 use std::sync::Arc;
 use std::time::UNIX_EPOCH;

 use anyhow::Context;
+use bytes::Bytes;
 use camino::Utf8Path;
+use futures::stream::Stream;
+use once_cell::sync::OnceCell;
 use remote_storage::{
-    AzureConfig, GenericRemoteStorage, RemotePath, RemoteStorageConfig, RemoteStorageKind,
+    AzureConfig, Download, GenericRemoteStorage, RemotePath, RemoteStorageConfig, RemoteStorageKind,
 };
 use test_context::{test_context, AsyncTestContext};
-use tracing::{debug, info};
+use tokio::task::JoinSet;
+use tracing::{debug, error, info};

-mod common;
-
-use common::{
-    cleanup, download_to_vec, ensure_logging_ready, upload_remote_data, upload_simple_remote_data,
-    upload_stream, wrap_stream,
-};
+static LOGGING_DONE: OnceCell<()> = OnceCell::new();

 const ENABLE_REAL_AZURE_REMOTE_STORAGE_ENV_VAR_NAME: &str = "ENABLE_REAL_AZURE_REMOTE_STORAGE";

@@ -30,7 +30,7 @@ const BASE_PREFIX: &str = "test";
 /// If real Azure tests are disabled, the test passes, skipping any real test run: currently, there's no way to mark the test ignored in runtime with the
 /// deafult test framework, see https://github.com/rust-lang/rust/issues/68007 for details.
 ///
-/// First, the test creates a set of Azure blobs with keys `/${random_prefix_part}/${base_prefix_str}/sub_prefix_${i}/blob_${i}` in [`upload_remote_data`]
+/// First, the test creates a set of Azure blobs with keys `/${random_prefix_part}/${base_prefix_str}/sub_prefix_${i}/blob_${i}` in [`upload_azure_data`]
 /// where
 /// * `random_prefix_part` is set for the entire Azure client during the Azure client creation in [`create_azure_client`], to avoid multiple test runs interference
 /// * `base_prefix_str` is a common prefix to use in the client requests: we would want to ensure that the client is able to list nested prefixes inside the bucket
@@ -97,7 +97,7 @@ async fn azure_pagination_should_work(
 /// Uses real Azure and requires [`ENABLE_REAL_AZURE_REMOTE_STORAGE_ENV_VAR_NAME`] and related Azure cred env vars specified. Test will skip real code and pass if env vars not set.
 /// See `Azure_pagination_should_work` for more information.
 ///
-/// First, create a set of Azure objects with keys `random_prefix/folder{j}/blob_{i}.txt` in [`upload_remote_data`]
+/// First, create a set of Azure objects with keys `random_prefix/folder{j}/blob_{i}.txt` in [`upload_azure_data`]
 /// Then performs the following queries:
 ///    1. `list_files(None)`. This should return all files `random_prefix/folder{j}/blob_{i}.txt`
 ///    2. `list_files("folder1")`.  This  should return all files `random_prefix/folder1/blob_{i}.txt`
@@ -218,9 +218,18 @@ async fn azure_upload_download_works(ctx: &mut MaybeEnabledAzure) -> anyhow::Res

    ctx.client.upload(data, len, &path, None).await?;

+    async fn download_and_compare(dl: Download) -> anyhow::Result<Vec<u8>> {
+        let mut buf = Vec::new();
+        tokio::io::copy_buf(
+            &mut tokio_util::io::StreamReader::new(dl.download_stream),
+            &mut buf,
+        )
+        .await?;
+        Ok(buf)
+    }
    // Normal download request
    let dl = ctx.client.download(&path).await?;
-    let buf = download_to_vec(dl).await?;
+    let buf = download_and_compare(dl).await?;
    assert_eq!(&buf, &orig);

    // Full range (end specified)
@@ -228,12 +237,12 @@ async fn azure_upload_download_works(ctx: &mut MaybeEnabledAzure) -> anyhow::Res
        .client
        .download_byte_range(&path, 0, Some(len as u64))
        .await?;
-    let buf = download_to_vec(dl).await?;
+    let buf = download_and_compare(dl).await?;
    assert_eq!(&buf, &orig);

    // partial range (end specified)
    let dl = ctx.client.download_byte_range(&path, 4, Some(10)).await?;
-    let buf = download_to_vec(dl).await?;
+    let buf = download_and_compare(dl).await?;
    assert_eq!(&buf, &orig[4..10]);

    // partial range (end beyond real end)
@@ -241,17 +250,17 @@ async fn azure_upload_download_works(ctx: &mut MaybeEnabledAzure) -> anyhow::Res
        .client
        .download_byte_range(&path, 8, Some(len as u64 * 100))
        .await?;
-    let buf = download_to_vec(dl).await?;
+    let buf = download_and_compare(dl).await?;
    assert_eq!(&buf, &orig[8..]);

    // Partial range (end unspecified)
    let dl = ctx.client.download_byte_range(&path, 4, None).await?;
-    let buf = download_to_vec(dl).await?;
+    let buf = download_and_compare(dl).await?;
    assert_eq!(&buf, &orig[4..]);

    // Full range (end unspecified)
    let dl = ctx.client.download_byte_range(&path, 0, None).await?;
-    let buf = download_to_vec(dl).await?;
+    let buf = download_and_compare(dl).await?;
    assert_eq!(&buf, &orig);

    debug!("Cleanup: deleting file at path {path:?}");
@@ -263,6 +272,17 @@ async fn azure_upload_download_works(ctx: &mut MaybeEnabledAzure) -> anyhow::Res
    Ok(())
 }

+fn ensure_logging_ready() {
+    LOGGING_DONE.get_or_init(|| {
+        utils::logging::init(
+            utils::logging::LogFormat::Test,
+            utils::logging::TracingErrorLayerEnablement::Disabled,
+            utils::logging::Output::Stdout,
+        )
+        .expect("logging init failed");
+    });
+}
+
 struct EnabledAzure {
    client: Arc<GenericRemoteStorage>,
    base_prefix: &'static str,
@@ -332,7 +352,7 @@ impl AsyncTestContext for MaybeEnabledAzureWithTestBlobs {

        let enabled = EnabledAzure::setup(Some(max_keys_in_list_response)).await;

-        match upload_remote_data(&enabled.client, enabled.base_prefix, upload_tasks_count).await {
+        match upload_azure_data(&enabled.client, enabled.base_prefix, upload_tasks_count).await {
            ControlFlow::Continue(uploads) => {
                info!("Remote objects created successfully");

@@ -394,7 +414,7 @@ impl AsyncTestContext for MaybeEnabledAzureWithSimpleTestBlobs {

        let enabled = EnabledAzure::setup(Some(max_keys_in_list_response)).await;

-        match upload_simple_remote_data(&enabled.client, upload_tasks_count).await {
+        match upload_simple_azure_data(&enabled.client, upload_tasks_count).await {
            ControlFlow::Continue(uploads) => {
                info!("Remote objects created successfully");

@@ -458,3 +478,166 @@ fn create_azure_client(
        GenericRemoteStorage::from_config(&remote_storage_config).context("remote storage init")?,
    ))
 }
+
+struct Uploads {
+    prefixes: HashSet<RemotePath>,
+    blobs: HashSet<RemotePath>,
+}
+
+async fn upload_azure_data(
+    client: &Arc<GenericRemoteStorage>,
+    base_prefix_str: &'static str,
+    upload_tasks_count: usize,
+) -> ControlFlow<Uploads, Uploads> {
+    info!("Creating {upload_tasks_count} Azure files");
+    let mut upload_tasks = JoinSet::new();
+    for i in 1..upload_tasks_count + 1 {
+        let task_client = Arc::clone(client);
+        upload_tasks.spawn(async move {
+            let prefix = format!("{base_prefix_str}/sub_prefix_{i}/");
+            let blob_prefix = RemotePath::new(Utf8Path::new(&prefix))
+                .with_context(|| format!("{prefix:?} to RemotePath conversion"))?;
+            let blob_path = blob_prefix.join(Utf8Path::new(&format!("blob_{i}")));
+            debug!("Creating remote item {i} at path {blob_path:?}");
+
+            let (data, len) = upload_stream(format!("remote blob data {i}").into_bytes().into());
+            task_client.upload(data, len, &blob_path, None).await?;
+
+            Ok::<_, anyhow::Error>((blob_prefix, blob_path))
+        });
+    }
+
+    let mut upload_tasks_failed = false;
+    let mut uploaded_prefixes = HashSet::with_capacity(upload_tasks_count);
+    let mut uploaded_blobs = HashSet::with_capacity(upload_tasks_count);
+    while let Some(task_run_result) = upload_tasks.join_next().await {
+        match task_run_result
+            .context("task join failed")
+            .and_then(|task_result| task_result.context("upload task failed"))
+        {
+            Ok((upload_prefix, upload_path)) => {
+                uploaded_prefixes.insert(upload_prefix);
+                uploaded_blobs.insert(upload_path);
+            }
+            Err(e) => {
+                error!("Upload task failed: {e:?}");
+                upload_tasks_failed = true;
+            }
+        }
+    }
+
+    let uploads = Uploads {
+        prefixes: uploaded_prefixes,
+        blobs: uploaded_blobs,
+    };
+    if upload_tasks_failed {
+        ControlFlow::Break(uploads)
+    } else {
+        ControlFlow::Continue(uploads)
+    }
+}
+
+async fn cleanup(client: &Arc<GenericRemoteStorage>, objects_to_delete: HashSet<RemotePath>) {
+    info!(
+        "Removing {} objects from the remote storage during cleanup",
+        objects_to_delete.len()
+    );
+    let mut delete_tasks = JoinSet::new();
+    for object_to_delete in objects_to_delete {
+        let task_client = Arc::clone(client);
+        delete_tasks.spawn(async move {
+            debug!("Deleting remote item at path {object_to_delete:?}");
+            task_client
+                .delete(&object_to_delete)
+                .await
+                .with_context(|| format!("{object_to_delete:?} removal"))
+        });
+    }
+
+    while let Some(task_run_result) = delete_tasks.join_next().await {
+        match task_run_result {
+            Ok(task_result) => match task_result {
+                Ok(()) => {}
+                Err(e) => error!("Delete task failed: {e:?}"),
+            },
+            Err(join_err) => error!("Delete task did not finish correctly: {join_err}"),
+        }
+    }
+}
+
+// Uploads files `folder{j}/blob{i}.txt`. See test description for more details.
+async fn upload_simple_azure_data(
+    client: &Arc<GenericRemoteStorage>,
+    upload_tasks_count: usize,
+) -> ControlFlow<HashSet<RemotePath>, HashSet<RemotePath>> {
+    info!("Creating {upload_tasks_count} Azure files");
+    let mut upload_tasks = JoinSet::new();
+    for i in 1..upload_tasks_count + 1 {
+        let task_client = Arc::clone(client);
+        upload_tasks.spawn(async move {
+            let blob_path = PathBuf::from(format!("folder{}/blob_{}.txt", i / 7, i));
+            let blob_path = RemotePath::new(
+                Utf8Path::from_path(blob_path.as_path()).expect("must be valid blob path"),
+            )
+            .with_context(|| format!("{blob_path:?} to RemotePath conversion"))?;
+            debug!("Creating remote item {i} at path {blob_path:?}");
+
+            let (data, len) = upload_stream(format!("remote blob data {i}").into_bytes().into());
+            task_client.upload(data, len, &blob_path, None).await?;
+
+            Ok::<_, anyhow::Error>(blob_path)
+        });
+    }
+
+    let mut upload_tasks_failed = false;
+    let mut uploaded_blobs = HashSet::with_capacity(upload_tasks_count);
+    while let Some(task_run_result) = upload_tasks.join_next().await {
+        match task_run_result
+            .context("task join failed")
+            .and_then(|task_result| task_result.context("upload task failed"))
+        {
+            Ok(upload_path) => {
+                uploaded_blobs.insert(upload_path);
+            }
+            Err(e) => {
+                error!("Upload task failed: {e:?}");
+                upload_tasks_failed = true;
+            }
+        }
+    }
+
+    if upload_tasks_failed {
+        ControlFlow::Break(uploaded_blobs)
+    } else {
+        ControlFlow::Continue(uploaded_blobs)
+    }
+}
+
+// FIXME: copypasted from test_real_s3, can't remember how to share a module which is not compiled
+// to binary
+fn upload_stream(
+    content: std::borrow::Cow<'static, [u8]>,
+) -> (
+    impl Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static,
+    usize,
+) {
+    use std::borrow::Cow;
+
+    let content = match content {
+        Cow::Borrowed(x) => Bytes::from_static(x),
+        Cow::Owned(vec) => Bytes::from(vec),
+    };
+    wrap_stream(content)
+}
+
+fn wrap_stream(
+    content: bytes::Bytes,
+) -> (
+    impl Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static,
+    usize,
+) {
+    let len = content.len();
+    let content = futures::future::ready(Ok(content));
+
+    (futures::stream::once(content), len)
+}
@@ -2,23 +2,23 @@ use std::collections::HashSet;
 use std::env;
 use std::num::NonZeroUsize;
 use std::ops::ControlFlow;
+use std::path::PathBuf;
 use std::sync::Arc;
 use std::time::UNIX_EPOCH;

 use anyhow::Context;
+use bytes::Bytes;
 use camino::Utf8Path;
+use futures::stream::Stream;
+use once_cell::sync::OnceCell;
 use remote_storage::{
    GenericRemoteStorage, RemotePath, RemoteStorageConfig, RemoteStorageKind, S3Config,
 };
 use test_context::{test_context, AsyncTestContext};
-use tracing::{debug, info};
+use tokio::task::JoinSet;
+use tracing::{debug, error, info};

-mod common;
-
-use common::{
-    cleanup, download_to_vec, ensure_logging_ready, upload_remote_data, upload_simple_remote_data,
-    upload_stream, wrap_stream,
-};
+static LOGGING_DONE: OnceCell<()> = OnceCell::new();

 const ENABLE_REAL_S3_REMOTE_STORAGE_ENV_VAR_NAME: &str = "ENABLE_REAL_S3_REMOTE_STORAGE";

@@ -30,7 +30,7 @@ const BASE_PREFIX: &str = "test";
 /// If real S3 tests are disabled, the test passes, skipping any real test run: currently, there's no way to mark the test ignored in runtime with the
 /// deafult test framework, see https://github.com/rust-lang/rust/issues/68007 for details.
 ///
-/// First, the test creates a set of S3 objects with keys `/${random_prefix_part}/${base_prefix_str}/sub_prefix_${i}/blob_${i}` in [`upload_remote_data`]
+/// First, the test creates a set of S3 objects with keys `/${random_prefix_part}/${base_prefix_str}/sub_prefix_${i}/blob_${i}` in [`upload_s3_data`]
 /// where
 /// * `random_prefix_part` is set for the entire S3 client during the S3 client creation in [`create_s3_client`], to avoid multiple test runs interference
 /// * `base_prefix_str` is a common prefix to use in the client requests: we would want to ensure that the client is able to list nested prefixes inside the bucket
@@ -95,7 +95,7 @@ async fn s3_pagination_should_work(ctx: &mut MaybeEnabledS3WithTestBlobs) -> any
 /// Uses real S3 and requires [`ENABLE_REAL_S3_REMOTE_STORAGE_ENV_VAR_NAME`] and related S3 cred env vars specified. Test will skip real code and pass if env vars not set.
 /// See `s3_pagination_should_work` for more information.
 ///
-/// First, create a set of S3 objects with keys `random_prefix/folder{j}/blob_{i}.txt` in [`upload_remote_data`]
+/// First, create a set of S3 objects with keys `random_prefix/folder{j}/blob_{i}.txt` in [`upload_s3_data`]
 /// Then performs the following queries:
 ///    1. `list_files(None)`. This should return all files `random_prefix/folder{j}/blob_{i}.txt`
 ///    2. `list_files("folder1")`.  This  should return all files `random_prefix/folder1/blob_{i}.txt`
@@ -198,65 +198,15 @@ async fn s3_delete_objects_works(ctx: &mut MaybeEnabledS3) -> anyhow::Result<()>
    Ok(())
 }

-#[test_context(MaybeEnabledS3)]
-#[tokio::test]
-async fn s3_upload_download_works(ctx: &mut MaybeEnabledS3) -> anyhow::Result<()> {
-    let MaybeEnabledS3::Enabled(ctx) = ctx else {
-        return Ok(());
-    };
-
-    let path = RemotePath::new(Utf8Path::new(format!("{}/file", ctx.base_prefix).as_str()))
-        .with_context(|| "RemotePath conversion")?;
-
-    let orig = bytes::Bytes::from_static("remote blob data here".as_bytes());
-
-    let (data, len) = wrap_stream(orig.clone());
-
-    ctx.client.upload(data, len, &path, None).await?;
-
-    // Normal download request
-    let dl = ctx.client.download(&path).await?;
-    let buf = download_to_vec(dl).await?;
-    assert_eq!(&buf, &orig);
-
-    // Full range (end specified)
-    let dl = ctx
-        .client
-        .download_byte_range(&path, 0, Some(len as u64))
-        .await?;
-    let buf = download_to_vec(dl).await?;
-    assert_eq!(&buf, &orig);
-
-    // partial range (end specified)
-    let dl = ctx.client.download_byte_range(&path, 4, Some(10)).await?;
-    let buf = download_to_vec(dl).await?;
-    assert_eq!(&buf, &orig[4..10]);
-
-    // partial range (end beyond real end)
-    let dl = ctx
-        .client
-        .download_byte_range(&path, 8, Some(len as u64 * 100))
-        .await?;
-    let buf = download_to_vec(dl).await?;
-    assert_eq!(&buf, &orig[8..]);
-
-    // Partial range (end unspecified)
-    let dl = ctx.client.download_byte_range(&path, 4, None).await?;
-    let buf = download_to_vec(dl).await?;
-    assert_eq!(&buf, &orig[4..]);
-
-    // Full range (end unspecified)
-    let dl = ctx.client.download_byte_range(&path, 0, None).await?;
-    let buf = download_to_vec(dl).await?;
-    assert_eq!(&buf, &orig);
-
-    debug!("Cleanup: deleting file at path {path:?}");
-    ctx.client
-        .delete(&path)
-        .await
-        .with_context(|| format!("{path:?} removal"))?;
-
-    Ok(())
+fn ensure_logging_ready() {
+    LOGGING_DONE.get_or_init(|| {
+        utils::logging::init(
+            utils::logging::LogFormat::Test,
+            utils::logging::TracingErrorLayerEnablement::Disabled,
+            utils::logging::Output::Stdout,
+        )
+        .expect("logging init failed");
+    });
 }

 struct EnabledS3 {
@@ -328,7 +278,7 @@ impl AsyncTestContext for MaybeEnabledS3WithTestBlobs {

        let enabled = EnabledS3::setup(Some(max_keys_in_list_response)).await;

-        match upload_remote_data(&enabled.client, enabled.base_prefix, upload_tasks_count).await {
+        match upload_s3_data(&enabled.client, enabled.base_prefix, upload_tasks_count).await {
            ControlFlow::Continue(uploads) => {
                info!("Remote objects created successfully");

@@ -390,7 +340,7 @@ impl AsyncTestContext for MaybeEnabledS3WithSimpleTestBlobs {

        let enabled = EnabledS3::setup(Some(max_keys_in_list_response)).await;

-        match upload_simple_remote_data(&enabled.client, upload_tasks_count).await {
+        match upload_simple_s3_data(&enabled.client, upload_tasks_count).await {
            ControlFlow::Continue(uploads) => {
                info!("Remote objects created successfully");

@@ -453,3 +403,166 @@ fn create_s3_client(
        GenericRemoteStorage::from_config(&remote_storage_config).context("remote storage init")?,
    ))
 }
+
+struct Uploads {
+    prefixes: HashSet<RemotePath>,
+    blobs: HashSet<RemotePath>,
+}
+
+async fn upload_s3_data(
+    client: &Arc<GenericRemoteStorage>,
+    base_prefix_str: &'static str,
+    upload_tasks_count: usize,
+) -> ControlFlow<Uploads, Uploads> {
+    info!("Creating {upload_tasks_count} S3 files");
+    let mut upload_tasks = JoinSet::new();
+    for i in 1..upload_tasks_count + 1 {
+        let task_client = Arc::clone(client);
+        upload_tasks.spawn(async move {
+            let prefix = format!("{base_prefix_str}/sub_prefix_{i}/");
+            let blob_prefix = RemotePath::new(Utf8Path::new(&prefix))
+                .with_context(|| format!("{prefix:?} to RemotePath conversion"))?;
+            let blob_path = blob_prefix.join(Utf8Path::new(&format!("blob_{i}")));
+            debug!("Creating remote item {i} at path {blob_path:?}");
+
+            let (data, data_len) =
+                upload_stream(format!("remote blob data {i}").into_bytes().into());
+            task_client.upload(data, data_len, &blob_path, None).await?;
+
+            Ok::<_, anyhow::Error>((blob_prefix, blob_path))
+        });
+    }
+
+    let mut upload_tasks_failed = false;
+    let mut uploaded_prefixes = HashSet::with_capacity(upload_tasks_count);
+    let mut uploaded_blobs = HashSet::with_capacity(upload_tasks_count);
+    while let Some(task_run_result) = upload_tasks.join_next().await {
+        match task_run_result
+            .context("task join failed")
+            .and_then(|task_result| task_result.context("upload task failed"))
+        {
+            Ok((upload_prefix, upload_path)) => {
+                uploaded_prefixes.insert(upload_prefix);
+                uploaded_blobs.insert(upload_path);
+            }
+            Err(e) => {
+                error!("Upload task failed: {e:?}");
+                upload_tasks_failed = true;
+            }
+        }
+    }
+
+    let uploads = Uploads {
+        prefixes: uploaded_prefixes,
+        blobs: uploaded_blobs,
+    };
+    if upload_tasks_failed {
+        ControlFlow::Break(uploads)
+    } else {
+        ControlFlow::Continue(uploads)
+    }
+}
+
+async fn cleanup(client: &Arc<GenericRemoteStorage>, objects_to_delete: HashSet<RemotePath>) {
+    info!(
+        "Removing {} objects from the remote storage during cleanup",
+        objects_to_delete.len()
+    );
+    let mut delete_tasks = JoinSet::new();
+    for object_to_delete in objects_to_delete {
+        let task_client = Arc::clone(client);
+        delete_tasks.spawn(async move {
+            debug!("Deleting remote item at path {object_to_delete:?}");
+            task_client
+                .delete(&object_to_delete)
+                .await
+                .with_context(|| format!("{object_to_delete:?} removal"))
+        });
+    }
+
+    while let Some(task_run_result) = delete_tasks.join_next().await {
+        match task_run_result {
+            Ok(task_result) => match task_result {
+                Ok(()) => {}
+                Err(e) => error!("Delete task failed: {e:?}"),
+            },
+            Err(join_err) => error!("Delete task did not finish correctly: {join_err}"),
+        }
+    }
+}
+
+// Uploads files `folder{j}/blob{i}.txt`. See test description for more details.
+async fn upload_simple_s3_data(
+    client: &Arc<GenericRemoteStorage>,
+    upload_tasks_count: usize,
+) -> ControlFlow<HashSet<RemotePath>, HashSet<RemotePath>> {
+    info!("Creating {upload_tasks_count} S3 files");
+    let mut upload_tasks = JoinSet::new();
+    for i in 1..upload_tasks_count + 1 {
+        let task_client = Arc::clone(client);
+        upload_tasks.spawn(async move {
+            let blob_path = PathBuf::from(format!("folder{}/blob_{}.txt", i / 7, i));
+            let blob_path = RemotePath::new(
+                Utf8Path::from_path(blob_path.as_path()).expect("must be valid blob path"),
+            )
+            .with_context(|| format!("{blob_path:?} to RemotePath conversion"))?;
+            debug!("Creating remote item {i} at path {blob_path:?}");
+
+            let (data, data_len) =
+                upload_stream(format!("remote blob data {i}").into_bytes().into());
+            task_client.upload(data, data_len, &blob_path, None).await?;
+
+            Ok::<_, anyhow::Error>(blob_path)
+        });
+    }
+
+    let mut upload_tasks_failed = false;
+    let mut uploaded_blobs = HashSet::with_capacity(upload_tasks_count);
+    while let Some(task_run_result) = upload_tasks.join_next().await {
+        match task_run_result
+            .context("task join failed")
+            .and_then(|task_result| task_result.context("upload task failed"))
+        {
+            Ok(upload_path) => {
+                uploaded_blobs.insert(upload_path);
+            }
+            Err(e) => {
+                error!("Upload task failed: {e:?}");
+                upload_tasks_failed = true;
+            }
+        }
+    }
+
+    if upload_tasks_failed {
+        ControlFlow::Break(uploaded_blobs)
+    } else {
+        ControlFlow::Continue(uploaded_blobs)
+    }
+}
+
+fn upload_stream(
+    content: std::borrow::Cow<'static, [u8]>,
+) -> (
+    impl Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static,
+    usize,
+) {
+    use std::borrow::Cow;
+
+    let content = match content {
+        Cow::Borrowed(x) => Bytes::from_static(x),
+        Cow::Owned(vec) => Bytes::from(vec),
+    };
+    wrap_stream(content)
+}
+
+fn wrap_stream(
+    content: bytes::Bytes,
+) -> (
+    impl Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static,
+    usize,
+) {
+    let len = content.len();
+    let content = futures::future::ready(Ok(content));
+
+    (futures::stream::once(content), len)
+}
@@ -64,18 +64,6 @@ impl Client {
        resp.json().await.map_err(Error::ReceiveBody)
    }

-    pub async fn tenant_details(
-        &self,
-        tenant_id: TenantId,
-    ) -> Result<pageserver_api::models::TenantDetails> {
-        let uri = format!("{}/v1/tenant/{tenant_id}", self.mgmt_api_endpoint);
-        self.get(uri)
-            .await?
-            .json()
-            .await
-            .map_err(Error::ReceiveBody)
-    }
-
    pub async fn list_timelines(
        &self,
        tenant_id: TenantId,
@@ -14,7 +14,6 @@ use hyper::header;
 use hyper::StatusCode;
 use hyper::{Body, Request, Response, Uri};
 use metrics::launch_timestamp::LaunchTimestamp;
-use pageserver_api::models::TenantDetails;
 use pageserver_api::models::{
    DownloadRemoteLayersTaskSpawnRequest, LocationConfigMode, TenantAttachRequest,
    TenantLoadRequest, TenantLocationConfigRequest,
@@ -593,6 +592,8 @@ async fn get_lsn_by_timestamp_handler(
        )));
    }

+    let version: Option<u8> = parse_query_param(&request, "version")?;
+
    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
    let timestamp_raw = must_get_query_param(&request, "timestamp")?;
    let timestamp = humantime::parse_rfc3339(&timestamp_raw)
@@ -605,18 +606,31 @@ async fn get_lsn_by_timestamp_handler(
    let result = timeline
        .find_lsn_for_timestamp(timestamp_pg, &cancel, &ctx)
        .await?;
-    #[derive(serde::Serialize)]
-    struct Result {
-        lsn: Lsn,
-        kind: &'static str,
+
+    if version.unwrap_or(0) > 1 {
+        #[derive(serde::Serialize)]
+        struct Result {
+            lsn: Lsn,
+            kind: &'static str,
+        }
+        let (lsn, kind) = match result {
+            LsnForTimestamp::Present(lsn) => (lsn, "present"),
+            LsnForTimestamp::Future(lsn) => (lsn, "future"),
+            LsnForTimestamp::Past(lsn) => (lsn, "past"),
+            LsnForTimestamp::NoData(lsn) => (lsn, "nodata"),
+        };
+        json_response(StatusCode::OK, Result { lsn, kind })
+    } else {
+        // FIXME: this is a temporary crutch not to break backwards compatibility
+        // See https://github.com/neondatabase/neon/pull/5608
+        let result = match result {
+            LsnForTimestamp::Present(lsn) => format!("{lsn}"),
+            LsnForTimestamp::Future(_lsn) => "future".into(),
+            LsnForTimestamp::Past(_lsn) => "past".into(),
+            LsnForTimestamp::NoData(_lsn) => "nodata".into(),
+        };
+        json_response(StatusCode::OK, result)
    }
-    let (lsn, kind) = match result {
-        LsnForTimestamp::Present(lsn) => (lsn, "present"),
-        LsnForTimestamp::Future(lsn) => (lsn, "future"),
-        LsnForTimestamp::Past(lsn) => (lsn, "past"),
-        LsnForTimestamp::NoData(lsn) => (lsn, "nodata"),
-    };
-    json_response(StatusCode::OK, Result { lsn, kind })
 }

 async fn get_timestamp_of_lsn_handler(
@@ -858,14 +872,11 @@ async fn tenant_status(
        }

        let state = tenant.current_state();
-        Result::<_, ApiError>::Ok(TenantDetails {
-            tenant_info: TenantInfo {
-                id: tenant_shard_id,
-                state: state.clone(),
-                current_physical_size: Some(current_physical_size),
-                attachment_status: state.attachment_status(),
-            },
-            timelines: tenant.list_timeline_ids(),
+        Result::<_, ApiError>::Ok(TenantInfo {
+            id: tenant_shard_id,
+            state: state.clone(),
+            current_physical_size: Some(current_physical_size),
+            attachment_status: state.attachment_status(),
        })
    }
    .instrument(info_span!("tenant_status_handler",
@@ -1019,62 +1019,12 @@ static SMGR_QUERY_TIME_PER_TENANT_TIMELINE: Lazy<HistogramVec> = Lazy::new(|| {
    .expect("failed to define a metric")
 });

-static SMGR_QUERY_TIME_GLOBAL_BUCKETS: Lazy<Vec<f64>> = Lazy::new(|| {
-    [
-        1,
-        10,
-        20,
-        40,
-        60,
-        80,
-        100,
-        200,
-        300,
-        400,
-        500,
-        600,
-        700,
-        800,
-        900,
-        1_000, // 1ms
-        2_000,
-        4_000,
-        6_000,
-        8_000,
-        10_000, // 10ms
-        20_000,
-        40_000,
-        60_000,
-        80_000,
-        100_000,
-        200_000,
-        400_000,
-        600_000,
-        800_000,
-        1_000_000, // 1s
-        2_000_000,
-        4_000_000,
-        6_000_000,
-        8_000_000,
-        10_000_000, // 10s
-        20_000_000,
-        50_000_000,
-        100_000_000,
-        200_000_000,
-        1_000_000_000, // 1000s
-    ]
-    .into_iter()
-    .map(Duration::from_micros)
-    .map(|d| d.as_secs_f64())
-    .collect()
-});
-
 static SMGR_QUERY_TIME_GLOBAL: Lazy<HistogramVec> = Lazy::new(|| {
    register_histogram_vec!(
        "pageserver_smgr_query_seconds_global",
        "Time spent on smgr query handling, aggregated by query type.",
        &["smgr_query_type"],
-        SMGR_QUERY_TIME_GLOBAL_BUCKETS.clone(),
+        CRITICAL_OP_BUCKETS.into(),
    )
    .expect("failed to define a metric")
 });
@@ -1552,10 +1552,6 @@ impl Tenant {
            .collect()
    }

-    pub fn list_timeline_ids(&self) -> Vec<TimelineId> {
-        self.timelines.lock().unwrap().keys().cloned().collect()
-    }
-
    /// This is used to create the initial 'main' timeline during bootstrapping,
    /// or when importing a new base backup. The caller is expected to load an
    /// initial image of the datadir to the new timeline after this.
@@ -138,7 +138,7 @@ pub(super) async fn connection_manager_loop_step(
                    Ok(Some(broker_update)) => connection_manager_state.register_timeline_update(broker_update),
                    Err(status) => {
                        match status.code() {
-                            Code::Unknown if status.message().contains("stream closed because of a broken pipe") || status.message().contains("connection reset") => {
+                            Code::Unknown if status.message().contains("stream closed because of a broken pipe") => {
                                // tonic's error handling doesn't provide a clear code for disconnections: we get
                                // "h2 protocol error: error reading a body from connection: stream closed because of a broken pipe"
                                info!("broker disconnected: {status}");
@@ -308,13 +308,13 @@ lfc_change_limit_hook(int newval, void *extra)
 		Assert(victim->access_count == 0);
 #ifdef FALLOC_FL_PUNCH_HOLE
 		if (fallocate(lfc_desc, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, (off_t) victim->offset * BLOCKS_PER_CHUNK * BLCKSZ, BLOCKS_PER_CHUNK * BLCKSZ) < 0)
-			elog(LOG, "Failed to punch hole in file: %m");
+			neon_log(LOG, "Failed to punch hole in file: %m");
 #endif
 		hash_search_with_hash_value(lfc_hash, &victim->key, victim->hash, HASH_REMOVE, NULL);
 		lfc_ctl->used -= 1;
 	}
 	lfc_ctl->limit = new_size;
-	elog(DEBUG1, "set local file cache limit to %d", new_size);
+	neon_log(DEBUG1, "set local file cache limit to %d", new_size);

 	LWLockRelease(lfc_lock);
 }
@@ -327,7 +327,7 @@ lfc_init(void)
 	 * shared_preload_libraries.
 	 */
 	if (!process_shared_preload_libraries_in_progress)
-		elog(ERROR, "Neon module should be loaded via shared_preload_libraries");
+		neon_log(ERROR, "Neon module should be loaded via shared_preload_libraries");


 	DefineCustomIntVariable("neon.max_file_cache_size",
@@ -643,7 +643,7 @@ lfc_write(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno, const void
 			Assert(victim->access_count == 0);
 			entry->offset = victim->offset; /* grab victim's chunk */
 			hash_search_with_hash_value(lfc_hash, &victim->key, victim->hash, HASH_REMOVE, NULL);
-			elog(DEBUG2, "Swap file cache page");
+			neon_log(DEBUG2, "Swap file cache page");
 		}
 		else
 		{
@@ -846,10 +846,10 @@ local_cache_pages(PG_FUNCTION_ARGS)
 		 * wrong) function definition though.
 		 */
 		if (get_call_result_type(fcinfo, NULL, &expected_tupledesc) != TYPEFUNC_COMPOSITE)
-			elog(ERROR, "return type must be a row type");
+			neon_log(ERROR, "return type must be a row type");

 		if (expected_tupledesc->natts != NUM_LOCALCACHE_PAGES_ELEM)
-			elog(ERROR, "incorrect number of output arguments");
+			neon_log(ERROR, "incorrect number of output arguments");

 		/* Construct a tuple descriptor for the result rows. */
 		tupledesc = CreateTemplateTupleDesc(expected_tupledesc->natts);
@@ -15,6 +15,7 @@
 #include "postgres.h"

 #include "access/xlog.h"
+#include "common/hashfn.h"
 #include "fmgr.h"
 #include "libpq-fe.h"
 #include "libpq/libpq.h"
@@ -37,17 +38,6 @@

 #define RECONNECT_INTERVAL_USEC 1000000

-bool		connected = false;
-PGconn	   *pageserver_conn = NULL;
-
-/*
- * WaitEventSet containing:
- * - WL_SOCKET_READABLE on pageserver_conn,
- * - WL_LATCH_SET on MyLatch, and
- * - WL_EXIT_ON_PM_DEATH.
- */
-WaitEventSet *pageserver_conn_wes = NULL;
-
 /* GUCs */
 char	   *neon_timeline;
 char	   *neon_tenant;
@@ -58,87 +48,206 @@ char	   *neon_auth_token;
 int			readahead_buffer_size = 128;
 int			flush_every_n_requests = 8;

-static int n_reconnect_attempts = 0;
-static int max_reconnect_attempts = 60;
+static int	n_reconnect_attempts = 0;
+static int	max_reconnect_attempts = 60;
+static int	stripe_size;

-#define MAX_PAGESERVER_CONNSTRING_SIZE 256
+bool	(*old_redo_read_buffer_filter) (XLogReaderState *record, uint8 block_id) = NULL;
+
+static bool pageserver_flush(shardno_t shard_no);
+static void pageserver_disconnect(shardno_t shard_no);
+static void AssignPageserverConnstring(const char *newval, void *extra);
+static bool CheckPageserverConnstring(char **newval, void **extra, GucSource source);
+
+static shmem_startup_hook_type prev_shmem_startup_hook;
+#if PG_VERSION_NUM>=150000
+static shmem_request_hook_type prev_shmem_request_hook;
+#endif
+
+/*
+ * ShardMap is kept in shared memory. It contains the connection strings for
+ * each shard.
+ *
+ * There is "neon.pageserver_connstring" GUC with PGC_SIGHUP option, allowing to change it using
+ * pg_reload_conf(). It is used by control plane to update shards information if page server is crashed,
+ * relocated or new shards are added. This GUC variable contains comma separated list of connection strings.
+ * It is copied to shared memory because config can not be loaded during query execution and we need to
+ * reestablish connection to page server.
+ *
+ * So usually copying connection string to shared memory is done by postmaster. And other backends
+ * should check update counter to determine of connection URL is changed and connection needs to be reestablished.
+ *
+ * But at startup shared memory is not yet initialized and so we need to copy in some other process.
+ * Moreover, we can not use standard Postgres LW-locks, because postmaster has proc entry and so can not wait
+ * on this primitive. This is why lockless access algorithm is implemented using two atomic counters to enforce
+ * consistent reading of connection string value from shared memory.
+ */
+typedef struct
+{
+	size_t n_shards;
+	pg_atomic_uint64 begin_update_counter;
+	pg_atomic_uint64 end_update_counter;
+	char   shard_connstr[MAX_SHARDS][MAX_PS_CONNSTR_LEN];
+} ShardMap;
+
+
+static ShardMap* shard_map;
+static uint64    shard_map_update_counter;

 typedef struct
 {
-	LWLockId	lock;
-	pg_atomic_uint64 update_counter;
-	char		pageserver_connstring[MAX_PAGESERVER_CONNSTRING_SIZE];
-} PagestoreShmemState;
+	/*
+	 * Connection for each shard
+	 */
+	PGconn	   *conn;
+    /*
+	 * WaitEventSet containing:
+	 * - WL_SOCKET_READABLE on 'conn'
+	 * - WL_LATCH_SET on MyLatch, and
+	 * - WL_EXIT_ON_PM_DEATH.
+	 */
+	WaitEventSet    *wes;
+} PageServer;

-#if PG_VERSION_NUM >= 150000
-static shmem_request_hook_type prev_shmem_request_hook = NULL;
-static void walproposer_shmem_request(void);
+static PageServer page_servers[MAX_SHARDS];
+static shardno_t  max_attached_shard_no;
+
+static void
+psm_shmem_startup(void)
+{
+	bool found;
+	if (prev_shmem_startup_hook)
+	{
+		prev_shmem_startup_hook();
+	}
+
+	LWLockAcquire(AddinShmemInitLock, LW_EXCLUSIVE);
+
+	shard_map = (ShardMap*)ShmemInitStruct("shard_map", sizeof(ShardMap), &found);
+	if (!found)
+	{
+		shard_map->n_shards = 0;
+		pg_atomic_init_u64(&shard_map->begin_update_counter, 0);
+		pg_atomic_init_u64(&shard_map->end_update_counter, 0);
+		AssignPageserverConnstring(page_server_connstring, NULL);
+	}
+	LWLockRelease(AddinShmemInitLock);
+}
+
+static void
+psm_shmem_request(void)
+{
+#if PG_VERSION_NUM>=150000
+	if (prev_shmem_request_hook)
+		prev_shmem_request_hook();
 #endif
-static shmem_startup_hook_type prev_shmem_startup_hook;
-static PagestoreShmemState *pagestore_shared;
-static uint64 pagestore_local_counter = 0;
-static char local_pageserver_connstring[MAX_PAGESERVER_CONNSTRING_SIZE];

-static bool pageserver_flush(void);
-static void pageserver_disconnect(void);
-
-static bool
-PagestoreShmemIsValid()
-{
-	return pagestore_shared && UsedShmemSegAddr;
-}
-
-static bool
-CheckPageserverConnstring(char **newval, void **extra, GucSource source)
-{
-	return strlen(*newval) < MAX_PAGESERVER_CONNSTRING_SIZE;
+	RequestAddinShmemSpace(sizeof(ShardMap));
 }

 static void
-AssignPageserverConnstring(const char *newval, void *extra)
+psm_init(void)
 {
-	if (!PagestoreShmemIsValid())
-		return;
-	LWLockAcquire(pagestore_shared->lock, LW_EXCLUSIVE);
-	strlcpy(pagestore_shared->pageserver_connstring, newval, MAX_PAGESERVER_CONNSTRING_SIZE);
-	pg_atomic_fetch_add_u64(&pagestore_shared->update_counter, 1);
-	LWLockRelease(pagestore_shared->lock);
+	prev_shmem_startup_hook = shmem_startup_hook;
+	shmem_startup_hook = psm_shmem_startup;
+#if PG_VERSION_NUM>=150000
+	prev_shmem_request_hook = shmem_request_hook;
+	shmem_request_hook = psm_shmem_request;
+#else
+	psm_shmem_request();
+#endif
+}
+
+/*
+ * Reload page map if needed and return number of shards and connection string for the specified shard
+ * 'connstr' is an output buffer. If not NULL, it must point to a buffer at least MAX_PS_CONNSTR_LEN bytes
+ * long. The connection string for the gven shard is copied to it.
+ */
+static shardno_t
+load_shard_map(shardno_t shard_no, char* connstr)
+{
+	shardno_t n_shards;
+	uint64 begin_update_counter;
+	uint64 end_update_counter;
+
+	/*
+	 * There is race condition here between backend and postmaster which can update shard map.
+	 * We recheck update counter after copying shard map to check that configuration was not changed.
+	 */
+	do
+	{
+		begin_update_counter = pg_atomic_read_u64(&shard_map->begin_update_counter);
+		end_update_counter = pg_atomic_read_u64(&shard_map->end_update_counter);
+
+		n_shards = shard_map->n_shards;
+		if (shard_no >= n_shards)
+			neon_log(ERROR, "Shard %d is greater or equal than number of shards %d", shard_no, n_shards);
+
+		if (connstr)
+		{
+			/*
+			 * We need to use strlcpy here because due to race condition string oin shared memory
+			 * may be not zero terminated.
+			 */
+			strlcpy(connstr, shard_map->shard_connstr[shard_no], MAX_PS_CONNSTR_LEN);
+			pg_memory_barrier();
+		}
+
+	}
+	while (begin_update_counter != end_update_counter
+		   || begin_update_counter != pg_atomic_read_u64(&shard_map->begin_update_counter)
+		   || end_update_counter != pg_atomic_read_u64(&shard_map->end_update_counter));
+
+
+	if (shard_map_update_counter != end_update_counter)
+	{
+		/* Reset all connections if connection strings are changed */
+		for (shardno_t i = 0; i < max_attached_shard_no; i++)
+		{
+			if (page_servers[i].conn)
+				pageserver_disconnect(i);
+		}
+		max_attached_shard_no = 0;
+		shard_map_update_counter = end_update_counter;
+	}
+
+	return n_shards;
+}
+
+#define MB (1024*1024)
+
+shardno_t
+get_shard_number(BufferTag* tag)
+{
+	shardno_t n_shards = load_shard_map(0, NULL);
+	uint32	  hash;
+
+#if PG_MAJORVERSION_NUM < 16
+	hash = murmurhash32(tag->rnode.relNode);
+	hash = hash_combine(hash, murmurhash32(tag->blockNum/(MB/BLCKSZ)/stripe_size));
+#else
+	hash = murmurhash32(tag->relNumber);
+	hash = hash_combine(hash, murmurhash32(tag->blockNum/(MB/BLCKSZ)/stripe_size));
+#endif
+
+	return hash % n_shards;
 }

 static bool
-CheckConnstringUpdated()
-{
-	if (!PagestoreShmemIsValid())
-		return false;
-	return pagestore_local_counter < pg_atomic_read_u64(&pagestore_shared->update_counter);
-}
-
-static void
-ReloadConnstring()
-{
-	if (!PagestoreShmemIsValid())
-		return;
-	LWLockAcquire(pagestore_shared->lock, LW_SHARED);
-	strlcpy(local_pageserver_connstring, pagestore_shared->pageserver_connstring, sizeof(local_pageserver_connstring));
-	pagestore_local_counter = pg_atomic_read_u64(&pagestore_shared->update_counter);
-	LWLockRelease(pagestore_shared->lock);
-}
-
-static bool
-pageserver_connect(int elevel)
+pageserver_connect(shardno_t shard_no, int elevel)
 {
 	char	   *query;
 	int			ret;
 	const char *keywords[3];
 	const char *values[3];
 	int			n;
+	PGconn*		conn;
+	WaitEventSet *wes;
+	char        connstr[MAX_PS_CONNSTR_LEN];

-	Assert(!connected);
+	Assert(page_servers[shard_no].conn == NULL);

-	if (CheckConnstringUpdated())
-	{
-		ReloadConnstring();
-	}
+	(void)load_shard_map(shard_no, connstr); /* refresh page map if needed */

 	/*
 	 * Connect using the connection string we got from the
@@ -158,50 +267,47 @@ pageserver_connect(int elevel)
 		n++;
 	}
 	keywords[n] = "dbname";
-	values[n] = local_pageserver_connstring;
+	values[n] = connstr;
 	n++;
 	keywords[n] = NULL;
 	values[n] = NULL;
 	n++;
-	pageserver_conn = PQconnectdbParams(keywords, values, 1);
+	conn = PQconnectdbParams(keywords, values, 1);

-	if (PQstatus(pageserver_conn) == CONNECTION_BAD)
+	if (PQstatus(conn) == CONNECTION_BAD)
 	{
-		char	   *msg = pchomp(PQerrorMessage(pageserver_conn));
+		char	   *msg = pchomp(PQerrorMessage(conn));

-		PQfinish(pageserver_conn);
-		pageserver_conn = NULL;
+		PQfinish(conn);

 		ereport(elevel,
 				(errcode(ERRCODE_SQLCLIENT_UNABLE_TO_ESTABLISH_SQLCONNECTION),
-				 errmsg(NEON_TAG "could not establish connection to pageserver"),
+				 errmsg(NEON_TAG "[shard %d] could not establish connection to pageserver", shard_no),
 				 errdetail_internal("%s", msg)));
 		return false;
 	}
-
 	query = psprintf("pagestream %s %s", neon_tenant, neon_timeline);
-	ret = PQsendQuery(pageserver_conn, query);
+	ret = PQsendQuery(conn, query);
 	if (ret != 1)
 	{
-		PQfinish(pageserver_conn);
-		pageserver_conn = NULL;
-		neon_log(elevel, "could not send pagestream command to pageserver");
+		PQfinish(conn);
+		neon_shard_log(shard_no, elevel, "could not send pagestream command to pageserver");
 		return false;
 	}

-	pageserver_conn_wes = CreateWaitEventSet(TopMemoryContext, 3);
-	AddWaitEventToSet(pageserver_conn_wes, WL_LATCH_SET, PGINVALID_SOCKET,
+	wes = CreateWaitEventSet(TopMemoryContext, 3);
+	AddWaitEventToSet(wes, WL_LATCH_SET, PGINVALID_SOCKET,
 					  MyLatch, NULL);
-	AddWaitEventToSet(pageserver_conn_wes, WL_EXIT_ON_PM_DEATH, PGINVALID_SOCKET,
+	AddWaitEventToSet(wes, WL_EXIT_ON_PM_DEATH, PGINVALID_SOCKET,
 					  NULL, NULL);
-	AddWaitEventToSet(pageserver_conn_wes, WL_SOCKET_READABLE, PQsocket(pageserver_conn), NULL, NULL);
+	AddWaitEventToSet(wes, WL_SOCKET_READABLE, PQsocket(conn), NULL, NULL);

-	while (PQisBusy(pageserver_conn))
+	while (PQisBusy(conn))
 	{
 		WaitEvent	event;

 		/* Sleep until there's something to do */
-		(void) WaitEventSetWait(pageserver_conn_wes, -1L, &event, 1, PG_WAIT_EXTENSION);
+		(void) WaitEventSetWait(wes, -1L, &event, 1, PG_WAIT_EXTENSION);
 		ResetLatch(MyLatch);

 		CHECK_FOR_INTERRUPTS();
@@ -209,25 +315,25 @@ pageserver_connect(int elevel)
 		/* Data available in socket? */
 		if (event.events & WL_SOCKET_READABLE)
 		{
-			if (!PQconsumeInput(pageserver_conn))
+			if (!PQconsumeInput(conn))
 			{
-				char	   *msg = pchomp(PQerrorMessage(pageserver_conn));
+				char	   *msg = pchomp(PQerrorMessage(conn));

-				PQfinish(pageserver_conn);
-				pageserver_conn = NULL;
-				FreeWaitEventSet(pageserver_conn_wes);
-				pageserver_conn_wes = NULL;
+				PQfinish(conn);
+				FreeWaitEventSet(wes);

-				neon_log(elevel, "could not complete handshake with pageserver: %s",
-						 msg);
+				neon_shard_log(shard_no, elevel, "could not complete handshake with pageserver: %s",
+							   msg);
 				return false;
 			}
 		}
 	}

-	neon_log(LOG, "libpagestore: connected to '%s'", page_server_connstring);
+	neon_shard_log(shard_no, LOG, "libpagestore: connected to '%s'", connstr);
+	page_servers[shard_no].conn = conn;
+	page_servers[shard_no].wes = wes;
+	max_attached_shard_no = Max(shard_no+1, max_attached_shard_no);

-	connected = true;
 	return true;
 }

@@ -235,10 +341,10 @@ pageserver_connect(int elevel)
 * A wrapper around PQgetCopyData that checks for interrupts while sleeping.
 */
 static int
-call_PQgetCopyData(char **buffer)
+call_PQgetCopyData(shardno_t shard_no, char **buffer)
 {
 	int			ret;
-
+	PGconn*     pageserver_conn = page_servers[shard_no].conn;
 retry:
 	ret = PQgetCopyData(pageserver_conn, buffer, 1 /* async */ );

@@ -247,7 +353,7 @@ retry:
 		WaitEvent	event;

 		/* Sleep until there's something to do */
-		(void) WaitEventSetWait(pageserver_conn_wes, -1L, &event, 1, PG_WAIT_EXTENSION);
+		(void) WaitEventSetWait(page_servers[shard_no].wes, -1L, &event, 1, PG_WAIT_EXTENSION);
 		ResetLatch(MyLatch);

 		CHECK_FOR_INTERRUPTS();
@@ -259,7 +365,7 @@ retry:
 			{
 				char	   *msg = pchomp(PQerrorMessage(pageserver_conn));

-				neon_log(LOG, "could not get response from pageserver: %s", msg);
+				neon_shard_log(shard_no, LOG, "could not get response from pageserver: %s", msg);
 				pfree(msg);
 				return -1;
 			}
@@ -273,7 +379,7 @@ retry:


 static void
-pageserver_disconnect(void)
+pageserver_disconnect(shardno_t shard_no)
 {
 	/*
 	 * If anything goes wrong while we were sending a request, it's not clear
@@ -282,38 +388,32 @@ pageserver_disconnect(void)
 	 * time later after we have already sent a new unrelated request. Close
 	 * the connection to avoid getting confused.
 	 */
-	if (connected)
+	if (page_servers[shard_no].conn)
 	{
-		neon_log(LOG, "dropping connection to page server due to error");
-		PQfinish(pageserver_conn);
-		pageserver_conn = NULL;
-		connected = false;
+		neon_shard_log(shard_no, LOG, "dropping connection to page server due to error");
+		PQfinish(page_servers[shard_no].conn);
+		page_servers[shard_no].conn = NULL;

 		prefetch_on_ps_disconnect();
 	}
-	if (pageserver_conn_wes != NULL)
+	if (page_servers[shard_no].wes != NULL)
 	{
-		FreeWaitEventSet(pageserver_conn_wes);
-		pageserver_conn_wes = NULL;
+		FreeWaitEventSet(page_servers[shard_no].wes);
+		page_servers[shard_no].wes = NULL;
 	}
 }

 static bool
-pageserver_send(NeonRequest *request)
+pageserver_send(shardno_t shard_no, NeonRequest *request)
 {
 	StringInfoData req_buff;
-
-	if (CheckConnstringUpdated())
-	{
-		pageserver_disconnect();
-		ReloadConnstring();
-	}
+	PGconn* pageserver_conn = page_servers[shard_no].conn;

 	/* If the connection was lost for some reason, reconnect */
-	if (connected && PQstatus(pageserver_conn) == CONNECTION_BAD)
+	if (pageserver_conn && PQstatus(pageserver_conn) == CONNECTION_BAD)
 	{
-		neon_log(LOG, "pageserver_send disconnect bad connection");
-		pageserver_disconnect();
+		neon_shard_log(shard_no, LOG, "pageserver_send disconnect bad connection");
+		pageserver_disconnect(shard_no);
 	}

 	req_buff = nm_pack_request(request);
@@ -327,9 +427,9 @@ pageserver_send(NeonRequest *request)
 	 * https://github.com/neondatabase/neon/issues/1138 So try to reestablish
 	 * connection in case of failure.
 	 */
-	if (!connected)
+	if (!page_servers[shard_no].conn)
 	{
-		while (!pageserver_connect(n_reconnect_attempts < max_reconnect_attempts ? LOG : ERROR))
+		while (!pageserver_connect(shard_no, n_reconnect_attempts < max_reconnect_attempts ? LOG : ERROR))
 		{
 			HandleMainLoopInterrupts();
 			n_reconnect_attempts += 1;
@@ -338,7 +438,9 @@ pageserver_send(NeonRequest *request)
 		n_reconnect_attempts = 0;
 	}

-	/*
+	pageserver_conn = page_servers[shard_no].conn;
+
+    /*
 	 * Send request.
 	 *
 	 * In principle, this could block if the output buffer is full, and we
@@ -349,9 +451,8 @@ pageserver_send(NeonRequest *request)
 	if (PQputCopyData(pageserver_conn, req_buff.data, req_buff.len) <= 0)
 	{
 		char	   *msg = pchomp(PQerrorMessage(pageserver_conn));
-
-		pageserver_disconnect();
-		neon_log(LOG, "pageserver_send disconnect because failed to send page request (try to reconnect): %s", msg);
+		pageserver_disconnect(shard_no);
+		neon_shard_log(shard_no, LOG, "pageserver_send disconnect because failed to send page request (try to reconnect): %s", msg);
 		pfree(msg);
 		pfree(req_buff.data);
 		return false;
@@ -363,19 +464,19 @@ pageserver_send(NeonRequest *request)
 	{
 		char	   *msg = nm_to_string((NeonMessage *) request);

-		neon_log(PageStoreTrace, "sent request: %s", msg);
+		neon_shard_log(shard_no, PageStoreTrace, "sent request: %s", msg);
 		pfree(msg);
 	}
 	return true;
 }

 static NeonResponse *
-pageserver_receive(void)
+pageserver_receive(shardno_t shard_no)
 {
 	StringInfoData resp_buff;
 	NeonResponse *resp;
-
-	if (!connected)
+	PGconn* pageserver_conn = page_servers[shard_no].conn;
+	if (!pageserver_conn)
 		return NULL;

 	PG_TRY();
@@ -383,7 +484,7 @@ pageserver_receive(void)
 		/* read response */
 		int			rc;

-		rc = call_PQgetCopyData(&resp_buff.data);
+		rc = call_PQgetCopyData(shard_no, &resp_buff.data);
 		if (rc >= 0)
 		{
 			resp_buff.len = rc;
@@ -395,33 +496,33 @@ pageserver_receive(void)
 			{
 				char	   *msg = nm_to_string((NeonMessage *) resp);

-				neon_log(PageStoreTrace, "got response: %s", msg);
+				neon_shard_log(shard_no, PageStoreTrace, "got response: %s", msg);
 				pfree(msg);
 			}
 		}
 		else if (rc == -1)
 		{
-			neon_log(LOG, "pageserver_receive disconnect because call_PQgetCopyData returns -1: %s", pchomp(PQerrorMessage(pageserver_conn)));
-			pageserver_disconnect();
+			neon_shard_log(shard_no, LOG, "pageserver_receive disconnect because call_PQgetCopyData returns -1: %s", pchomp(PQerrorMessage(pageserver_conn)));
+			pageserver_disconnect(shard_no);
 			resp = NULL;
 		}
 		else if (rc == -2)
 		{
 			char	   *msg = pchomp(PQerrorMessage(pageserver_conn));

-			pageserver_disconnect();
-			neon_log(ERROR, "pageserver_receive disconnect because could not read COPY data: %s", msg);
+			pageserver_disconnect(shard_no);
+			neon_shard_log(shard_no, ERROR, "pageserver_receive disconnect because could not read COPY data: %s", msg);
 		}
 		else
 		{
-			pageserver_disconnect();
-			neon_log(ERROR, "pageserver_receive disconnect because unexpected PQgetCopyData return value: %d", rc);
+			pageserver_disconnect(shard_no);
+			neon_shard_log(shard_no, ERROR, "pageserver_receive disconnect because unexpected PQgetCopyData return value: %d", rc);
 		}
 	}
 	PG_CATCH();
 	{
-		neon_log(LOG, "pageserver_receive disconnect due to caught exception");
-		pageserver_disconnect();
+		neon_shard_log(shard_no, LOG, "pageserver_receive disconnect due to caught exception");
+		pageserver_disconnect(shard_no);
 		PG_RE_THROW();
 	}
 	PG_END_TRY();
@@ -431,11 +532,12 @@ pageserver_receive(void)


 static bool
-pageserver_flush(void)
+pageserver_flush(shardno_t shard_no)
 {
-	if (!connected)
+	PGconn* pageserver_conn = page_servers[shard_no].conn;
+	if (!pageserver_conn)
 	{
-		neon_log(WARNING, "Tried to flush while disconnected");
+		neon_shard_log(shard_no, WARNING, "Tried to flush while disconnected");
 	}
 	else
 	{
@@ -443,8 +545,8 @@ pageserver_flush(void)
 		{
 			char	   *msg = pchomp(PQerrorMessage(pageserver_conn));

-			pageserver_disconnect();
-			neon_log(LOG, "pageserver_flush disconnect because failed to flush page requests: %s", msg);
+			pageserver_disconnect(shard_no);
+			neon_shard_log(shard_no, LOG, "pageserver_flush disconnect because failed to flush page requests: %s", msg);
 			pfree(msg);
 			return false;
 		}
@@ -467,63 +569,83 @@ check_neon_id(char **newval, void **extra, GucSource source)
 	return **newval == '\0' || HexDecodeString(id, *newval, 16);
 }

-static Size
-PagestoreShmemSize(void)
-{
-	return sizeof(PagestoreShmemState);
-}
-
 static bool
-PagestoreShmemInit(void)
+CheckPageserverConnstring(char **newval, void **extra, GucSource source)
 {
-	bool		found;
-
-	LWLockAcquire(AddinShmemInitLock, LW_EXCLUSIVE);
-	pagestore_shared = ShmemInitStruct("libpagestore shared state",
-									   PagestoreShmemSize(),
-									   &found);
-	if (!found)
+	const char* shard_connstr = *newval;
+	const char* sep;
+	size_t connstr_len;
+	int i = 0;
+	do
 	{
-		pagestore_shared->lock = &(GetNamedLWLockTranche("neon_libpagestore")->lock);
-		pg_atomic_init_u64(&pagestore_shared->update_counter, 0);
-		AssignPageserverConnstring(page_server_connstring, NULL);
+		sep = strchr(shard_connstr, ',');
+		connstr_len = sep != NULL ? sep - shard_connstr : strlen(shard_connstr);
+		if (connstr_len == 0)
+			break; /* trailing comma */
+		if (i >= MAX_SHARDS)
+		{
+			neon_log(LOG, "Too many shards");
+			return false;
+		}
+		if (connstr_len >= MAX_PS_CONNSTR_LEN)
+		{
+			neon_log(LOG, "Connection string too long");
+			return false;
+		}
+		shard_connstr = sep + 1;
+		i += 1;
+	} while (sep != NULL);
+
+	return true;
+}
+
+static void
+AssignPageserverConnstring(const char *newval, void *extra)
+{
+	/*
+	 * Load shard map only at Postmaster.
+	 * If old page server is not available, then backends can be blocked in attempts to reconnect to it and do not reload config in this loop
+	 *
+	 * Copying GUC value to shared memory is usually performed by postmaster. But in case of startup,
+	 * shared memory is not yet initialized. So it has to be performed by any other process.
+	 * It is not a problem if more than one process do this initialization.
+	 */
+	if (shard_map != NULL && UsedShmemSegAddr != NULL && (MyProcPid == PostmasterPid || shard_map->n_shards == 0))
+	{
+		const char* shard_connstr = newval;
+		const char* sep;
+		size_t connstr_len;
+		int i = 0;
+		bool shard_map_changed = false;
+		do
+		{
+			sep = strchr(shard_connstr, ',');
+			connstr_len = sep != NULL ? sep - shard_connstr : strlen(shard_connstr);
+			if (connstr_len == 0)
+				break; /* trailing comma */
+			Assert(i < MAX_SHARDS);
+			Assert(connstr_len < MAX_PS_CONNSTR_LEN);
+			if (i >= shard_map->n_shards ||
+				strcmp(shard_map->shard_connstr[i], shard_connstr) != 0)
+			{
+				if (!shard_map_changed)
+				{
+					pg_atomic_add_fetch_u64(&shard_map->begin_update_counter, 1);
+					shard_map_changed = true;
+				}
+				memcpy(shard_map->shard_connstr[i], shard_connstr, connstr_len+1);
+			}
+			shard_connstr = sep + 1;
+			i += 1;
+		} while (sep != NULL);
+
+		if (shard_map_changed)
+		{
+			shard_map->n_shards = i;
+			pg_memory_barrier();
+			pg_atomic_add_fetch_u64(&shard_map->end_update_counter, 1);
+		}
 	}
-	LWLockRelease(AddinShmemInitLock);
-	return found;
-}
-
-static void
-pagestore_shmem_startup_hook(void)
-{
-	if (prev_shmem_startup_hook)
-		prev_shmem_startup_hook();
-
-	PagestoreShmemInit();
-}
-
-static void
-pagestore_shmem_request(void)
-{
-#if PG_VERSION_NUM >= 150000
-	if (prev_shmem_request_hook)
-		prev_shmem_request_hook();
-#endif
-
-	RequestAddinShmemSpace(PagestoreShmemSize());
-	RequestNamedLWLockTranche("neon_libpagestore", 1);
-}
-
-static void
-pagestore_prepare_shmem(void)
-{
-#if PG_VERSION_NUM >= 150000
-	prev_shmem_request_hook = shmem_request_hook;
-	shmem_request_hook = pagestore_shmem_request;
-#else
-	pagestore_shmem_request();
-#endif
-	prev_shmem_startup_hook = shmem_startup_hook;
-	shmem_startup_hook = pagestore_shmem_startup_hook;
 }

 /*
@@ -532,8 +654,6 @@ pagestore_prepare_shmem(void)
 void
 pg_init_libpagestore(void)
 {
-	pagestore_prepare_shmem();
-
 	DefineCustomStringVariable("neon.pageserver_connstring",
 							   "connection string to the page server",
 							   NULL,
@@ -561,6 +681,15 @@ pg_init_libpagestore(void)
 							   0,	/* no flags required */
 							   check_neon_id, NULL, NULL);

+	DefineCustomIntVariable("neon.stripe_size",
+							"sharding stripe size",
+							NULL,
+							&stripe_size,
+							32768, 1, INT_MAX,
+							PGC_SIGHUP,
+							GUC_UNIT_BLOCKS,
+							NULL, NULL, NULL);
+
 	DefineCustomIntVariable("neon.max_cluster_size",
 							"cluster size limit",
 							NULL,
@@ -624,4 +753,5 @@ pg_init_libpagestore(void)
 	}

 	lfc_init();
+	psm_init();
 }
@@ -17,12 +17,20 @@

 #include "access/xlogdefs.h"
 #include RELFILEINFO_HDR
+#include "storage/block.h"
+#include "storage/smgr.h"
+#include "storage/buf_internals.h"
 #include "lib/stringinfo.h"
 #include "libpq/pqformat.h"
 #include "storage/block.h"
 #include "storage/smgr.h"
 #include "utils/memutils.h"

+#include "pg_config.h"
+
+#define MAX_SHARDS 128
+#define MAX_PS_CONNSTR_LEN 128
+
 typedef enum
 {
 	/* pagestore_client -> pagestore */
@@ -51,6 +59,9 @@ typedef struct
 #define neon_log(tag, fmt, ...) ereport(tag,                                  \
 										(errmsg(NEON_TAG fmt, ##__VA_ARGS__), \
 										 errhidestmt(true), errhidecontext(true), errposition(0), internalerrposition(0)))
+#define neon_shard_log(shard_no, tag, fmt, ...) ereport(tag,	\
+														(errmsg(NEON_TAG "[shard %d] " fmt, shard_no, ##__VA_ARGS__), \
+														 errhidestmt(true), errhidecontext(true), errposition(0), internalerrposition(0)))

 /*
 * supertype of all the Neon*Request structs below
@@ -141,11 +152,13 @@ extern char *nm_to_string(NeonMessage *msg);
 * API
 */

+typedef unsigned shardno_t;
+
 typedef struct
 {
-	bool		(*send) (NeonRequest *request);
-	NeonResponse *(*receive) (void);
-	bool		(*flush) (void);
+	bool		(*send) (shardno_t  shard_no, NeonRequest * request);
+	NeonResponse *(*receive) (shardno_t shard_no);
+	bool		(*flush) (shardno_t shard_no);
 } page_server_api;

 extern void prefetch_on_ps_disconnect(void);
@@ -159,6 +172,8 @@ extern char *neon_timeline;
 extern char *neon_tenant;
 extern int32 max_cluster_size;

+extern shardno_t get_shard_number(BufferTag* tag);
+
 extern const f_smgr *smgr_neon(BackendId backend, NRelFileInfo rinfo);
 extern void smgr_init_neon(void);
 extern void readahead_buffer_resize(int newsize, void *extra);
@@ -172,6 +172,7 @@ typedef struct PrefetchRequest
 	XLogRecPtr	actual_request_lsn;
 	NeonResponse *response;		/* may be null */
 	PrefetchStatus status;
+	shardno_t   shard_no;
 	uint64		my_ring_index;
 } PrefetchRequest;

@@ -239,7 +240,9 @@ typedef struct PrefetchState
 								 * also unused */

 	/* the buffers */
-	prfh_hash  *prf_hash;
+	prfh_hash	*prf_hash;
+	int			max_shard_no;
+	uint8		shard_bitmap[(MAX_SHARDS + 7)/8];
 	PrefetchRequest prf_buffer[];	/* prefetch buffers */
 } PrefetchState;

@@ -327,6 +330,7 @@ compact_prefetch_buffers(void)
 		Assert(target_slot->status == PRFS_UNUSED);

 		target_slot->buftag = source_slot->buftag;
+		target_slot->shard_no = source_slot->shard_no;
 		target_slot->status = source_slot->status;
 		target_slot->response = source_slot->response;
 		target_slot->effective_request_lsn = source_slot->effective_request_lsn;
@@ -494,6 +498,23 @@ prefetch_cleanup_trailing_unused(void)
 	}
 }

+
+static bool
+prefetch_flush_requests(void)
+{
+	for (shardno_t shard_no = 0; shard_no < MyPState->max_shard_no; shard_no++)
+	{
+		if (MyPState->shard_bitmap[shard_no >> 3] & (1 << (shard_no & 7)))
+		{
+			if (!page_server->flush(shard_no))
+				return false;
+			MyPState->shard_bitmap[shard_no >> 3] &= ~(1 << (shard_no & 7));
+		}
+	}
+	MyPState->max_shard_no = 0;
+	return true;
+}
+
 /*
 * Wait for slot of ring_index to have received its response.
 * The caller is responsible for making sure the request buffer is flushed.
@@ -509,7 +530,7 @@ prefetch_wait_for(uint64 ring_index)
 	if (MyPState->ring_flush <= ring_index &&
 		MyPState->ring_unused > MyPState->ring_flush)
 	{
-		if (!page_server->flush())
+		if (!prefetch_flush_requests())
 			return false;
 		MyPState->ring_flush = MyPState->ring_unused;
 	}
@@ -547,7 +568,7 @@ prefetch_read(PrefetchRequest *slot)
 	Assert(slot->my_ring_index == MyPState->ring_receive);

 	old = MemoryContextSwitchTo(MyPState->errctx);
-	response = (NeonResponse *) page_server->receive();
+	response = (NeonResponse *) page_server->receive(slot->shard_no);
 	MemoryContextSwitchTo(old);
 	if (response)
 	{
@@ -704,12 +725,14 @@ prefetch_do_request(PrefetchRequest *slot, bool *force_latest, XLogRecPtr *force
 	Assert(slot->response == NULL);
 	Assert(slot->my_ring_index == MyPState->ring_unused);

-	while (!page_server->send((NeonRequest *) &request));
+	while (!page_server->send(slot->shard_no, (NeonRequest *) &request));

 	/* update prefetch state */
 	MyPState->n_requests_inflight += 1;
 	MyPState->n_unused -= 1;
 	MyPState->ring_unused += 1;
+	MyPState->shard_bitmap[slot->shard_no >> 3] |= 1 << (slot->shard_no & 7);
+	MyPState->max_shard_no = Max(slot->shard_no+1, MyPState->max_shard_no);

 	/* update slot state */
 	slot->status = PRFS_REQUESTED;
@@ -880,6 +903,7 @@ Retry:
 	 * function reads the buffer tag from the slot.
 	 */
 	slot->buftag = tag;
+	slot->shard_no = get_shard_number(&tag);
 	slot->my_ring_index = ring_index;

 	prefetch_do_request(slot, force_latest, force_lsn);
@@ -890,7 +914,7 @@ Retry:
 	if (flush_every_n_requests > 0 &&
 		MyPState->ring_unused - MyPState->ring_flush >= flush_every_n_requests)
 	{
-		if (!page_server->flush())
+		if (!prefetch_flush_requests())
 		{
 			/*
 			 * Prefetch set is reset in case of error, so we should try to
@@ -908,13 +932,44 @@ static NeonResponse *
 page_server_request(void const *req)
 {
 	NeonResponse *resp;
+	BufferTag tag = {0};
+	shardno_t shard_no;
+
+	switch (((NeonRequest *) req)->tag)
+	{
+		case T_NeonExistsRequest:
+			CopyNRelFileInfoToBufTag(tag, ((NeonExistsRequest *) req)->rinfo);
+			break;
+		case T_NeonNblocksRequest:
+			CopyNRelFileInfoToBufTag(tag, ((NeonNblocksRequest *) req)->rinfo);
+			break;
+		case T_NeonDbSizeRequest:
+			NInfoGetDbOid(BufTagGetNRelFileInfo(tag)) = ((NeonDbSizeRequest *) req)->dbNode;
+			break;
+		case T_NeonGetPageRequest:
+			CopyNRelFileInfoToBufTag(tag, ((NeonGetPageRequest *) req)->rinfo);
+			tag.blockNum = ((NeonGetPageRequest *) req)->blkno;
+			break;
+		default:
+			neon_log(ERROR, "Unexpected request tag: %d", ((NeonRequest *) req)->tag);
+	}
+	shard_no = get_shard_number(&tag);
+
+
+	/*
+	 * Current sharding model assumes that all metadata is present only at shard 0.
+	 * We still need to call get_shard_no() to check if shard map is up-to-date.
+	 */
+	if (((NeonRequest *) req)->tag != T_NeonGetPageRequest || ((NeonGetPageRequest *) req)->forknum != MAIN_FORKNUM)
+	{
+		shard_no = 0;
+	}

 	do
 	{
-		while (!page_server->send((NeonRequest *) req) || !page_server->flush());
-		MyPState->ring_flush = MyPState->ring_unused;
+		while (!page_server->send(shard_no, (NeonRequest *) req) || !page_server->flush(shard_no));
 		consume_prefetch_responses();
-		resp = page_server->receive();
+		resp = page_server->receive(shard_no);
 	} while (resp == NULL);
 	return resp;

@@ -990,7 +1045,7 @@ nm_pack_request(NeonRequest *msg)
 		case T_NeonErrorResponse:
 		case T_NeonDbSizeResponse:
 		default:
-			elog(ERROR, "unexpected neon message tag 0x%02x", msg->tag);
+			neon_log(ERROR, "unexpected neon message tag 0x%02x", msg->tag);
 			break;
 	}
 	return s;
@@ -1085,7 +1140,7 @@ nm_unpack_response(StringInfo s)
 		case T_NeonGetPageRequest:
 		case T_NeonDbSizeRequest:
 		default:
-			elog(ERROR, "unexpected neon message tag 0x%02x", tag);
+			neon_log(ERROR, "unexpected neon message tag 0x%02x", tag);
 			break;
 	}

@@ -1277,7 +1332,7 @@ neon_wallog_page(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, co
 		XLogFlush(recptr);
 		lsn = recptr;
 		ereport(SmgrTrace,
-				(errmsg("Page %u of relation %u/%u/%u.%u was force logged. Evicted at lsn=%X/%X",
+				(errmsg(NEON_TAG "Page %u of relation %u/%u/%u.%u was force logged. Evicted at lsn=%X/%X",
 						blocknum,
 						RelFileInfoFmt(InfoFromSMgrRel(reln)),
 						forknum, LSN_FORMAT_ARGS(lsn))));
@@ -1305,7 +1360,7 @@ neon_wallog_page(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, co
 		if (PageIsNew((Page) buffer))
 		{
 			ereport(SmgrTrace,
-					(errmsg("Page %u of relation %u/%u/%u.%u is all-zeros",
+					(errmsg(NEON_TAG "Page %u of relation %u/%u/%u.%u is all-zeros",
 							blocknum,
 							RelFileInfoFmt(InfoFromSMgrRel(reln)),
 							forknum)));
@@ -1313,7 +1368,7 @@ neon_wallog_page(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, co
 		else if (PageIsEmptyHeapPage((Page) buffer))
 		{
 			ereport(SmgrTrace,
-					(errmsg("Page %u of relation %u/%u/%u.%u is an empty heap page with no LSN",
+					(errmsg(NEON_TAG "Page %u of relation %u/%u/%u.%u is an empty heap page with no LSN",
 							blocknum,
 							RelFileInfoFmt(InfoFromSMgrRel(reln)),
 							forknum)));
@@ -1321,7 +1376,7 @@ neon_wallog_page(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, co
 		else
 		{
 			ereport(PANIC,
-					(errmsg("Page %u of relation %u/%u/%u.%u is evicted with zero LSN",
+					(errmsg(NEON_TAG "Page %u of relation %u/%u/%u.%u is evicted with zero LSN",
 							blocknum,
 							RelFileInfoFmt(InfoFromSMgrRel(reln)),
 							forknum)));
@@ -1330,7 +1385,7 @@ neon_wallog_page(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, co
 	else
 	{
 		ereport(SmgrTrace,
-				(errmsg("Page %u of relation %u/%u/%u.%u is already wal logged at lsn=%X/%X",
+				(errmsg(NEON_TAG "Page %u of relation %u/%u/%u.%u is already wal logged at lsn=%X/%X",
 						blocknum,
 						RelFileInfoFmt(InfoFromSMgrRel(reln)),
 						forknum, LSN_FORMAT_ARGS(lsn))));
@@ -1430,7 +1485,7 @@ neon_get_request_lsn(bool *latest, NRelFileInfo rinfo, ForkNumber forknum, Block
 		lsn = GetLastWrittenLSN(rinfo, forknum, blkno);
 		lsn = nm_adjust_lsn(lsn);

-		elog(DEBUG1, "neon_get_request_lsn GetXLogReplayRecPtr %X/%X request lsn 0 ",
+		neon_log(DEBUG1, "neon_get_request_lsn GetXLogReplayRecPtr %X/%X request lsn 0 ",
 			 (uint32) ((lsn) >> 32), (uint32) (lsn));
 	}
 	else
@@ -1445,7 +1500,7 @@ neon_get_request_lsn(bool *latest, NRelFileInfo rinfo, ForkNumber forknum, Block
 		*latest = true;
 		lsn = GetLastWrittenLSN(rinfo, forknum, blkno);
 		Assert(lsn != InvalidXLogRecPtr);
-		elog(DEBUG1, "neon_get_request_lsn GetLastWrittenLSN lsn %X/%X ",
+		neon_log(DEBUG1, "neon_get_request_lsn GetLastWrittenLSN lsn %X/%X ",
 			 (uint32) ((lsn) >> 32), (uint32) (lsn));

 		lsn = nm_adjust_lsn(lsn);
@@ -1465,7 +1520,7 @@ neon_get_request_lsn(bool *latest, NRelFileInfo rinfo, ForkNumber forknum, Block
 #endif
 		if (lsn > flushlsn)
 		{
-			elog(DEBUG5, "last-written LSN %X/%X is ahead of last flushed LSN %X/%X",
+			neon_log(DEBUG5, "last-written LSN %X/%X is ahead of last flushed LSN %X/%X",
 				 (uint32) (lsn >> 32), (uint32) lsn,
 				 (uint32) (flushlsn >> 32), (uint32) flushlsn);
 			XLogFlush(lsn);
@@ -1509,7 +1564,7 @@ neon_exists(SMgrRelation reln, ForkNumber forkNum)
 			return mdexists(reln, forkNum);

 		default:
-			elog(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
+			neon_log(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
 	}

 	if (get_cached_relsize(InfoFromSMgrRel(reln), forkNum, &n_blocks))
@@ -1561,7 +1616,7 @@ neon_exists(SMgrRelation reln, ForkNumber forkNum)
 		case T_NeonErrorResponse:
 			ereport(ERROR,
 					(errcode(ERRCODE_IO_ERROR),
-					 errmsg("could not read relation existence of rel %u/%u/%u.%u from page server at lsn %X/%08X",
+					 errmsg(NEON_TAG "could not read relation existence of rel %u/%u/%u.%u from page server at lsn %X/%08X",
 							RelFileInfoFmt(InfoFromSMgrRel(reln)),
 							forkNum,
 							(uint32) (request_lsn >> 32), (uint32) request_lsn),
@@ -1570,7 +1625,7 @@ neon_exists(SMgrRelation reln, ForkNumber forkNum)
 			break;

 		default:
-			elog(ERROR, "unexpected response from page server with tag 0x%02x", resp->tag);
+			neon_log(ERROR, "unexpected response from page server with tag 0x%02x", resp->tag);
 	}
 	pfree(resp);
 	return exists;
@@ -1587,7 +1642,7 @@ neon_create(SMgrRelation reln, ForkNumber forkNum, bool isRedo)
 	switch (reln->smgr_relpersistence)
 	{
 		case 0:
-			elog(ERROR, "cannot call smgrcreate() on rel with unknown persistence");
+			neon_log(ERROR, "cannot call smgrcreate() on rel with unknown persistence");

 		case RELPERSISTENCE_PERMANENT:
 			break;
@@ -1598,10 +1653,10 @@ neon_create(SMgrRelation reln, ForkNumber forkNum, bool isRedo)
 			return;

 		default:
-			elog(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
+			neon_log(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
 	}

-	elog(SmgrTrace, "Create relation %u/%u/%u.%u",
+	neon_log(SmgrTrace, "Create relation %u/%u/%u.%u",
 		 RelFileInfoFmt(InfoFromSMgrRel(reln)),
 		 forkNum);

@@ -1696,7 +1751,7 @@ neon_extend(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno,
 	switch (reln->smgr_relpersistence)
 	{
 		case 0:
-			elog(ERROR, "cannot call smgrextend() on rel with unknown persistence");
+			neon_log(ERROR, "cannot call smgrextend() on rel with unknown persistence");

 		case RELPERSISTENCE_PERMANENT:
 			break;
@@ -1707,7 +1762,7 @@ neon_extend(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno,
 			return;

 		default:
-			elog(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
+			neon_log(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
 	}

 	/*
@@ -1726,7 +1781,7 @@ neon_extend(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno,
 		if (current_size >= ((uint64) max_cluster_size) * 1024 * 1024)
 			ereport(ERROR,
 					(errcode(ERRCODE_DISK_FULL),
-					 errmsg("could not extend file because project size limit (%d MB) has been exceeded",
+					 errmsg(NEON_TAG "could not extend file because project size limit (%d MB) has been exceeded",
 							max_cluster_size),
 					 errhint("This limit is defined externally by the project size limit, and internally by neon.max_cluster_size GUC")));
 	}
@@ -1745,7 +1800,7 @@ neon_extend(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno,
 	set_cached_relsize(InfoFromSMgrRel(reln), forkNum, blkno + 1);

 	lsn = PageGetLSN((Page) buffer);
-	elog(SmgrTrace, "smgrextend called for %u/%u/%u.%u blk %u, page LSN: %X/%08X",
+	neon_log(SmgrTrace, "smgrextend called for %u/%u/%u.%u blk %u, page LSN: %X/%08X",
 		 RelFileInfoFmt(InfoFromSMgrRel(reln)),
 		 forkNum, blkno,
 		 (uint32) (lsn >> 32), (uint32) lsn);
@@ -1785,7 +1840,7 @@ neon_zeroextend(SMgrRelation reln, ForkNumber forkNum, BlockNumber blocknum,
 	switch (reln->smgr_relpersistence)
 	{
 		case 0:
-			elog(ERROR, "cannot call smgrextend() on rel with unknown persistence");
+			neon_log(ERROR, "cannot call smgrextend() on rel with unknown persistence");

 		case RELPERSISTENCE_PERMANENT:
 			break;
@@ -1796,7 +1851,7 @@ neon_zeroextend(SMgrRelation reln, ForkNumber forkNum, BlockNumber blocknum,
 			return;

 		default:
-			elog(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
+			neon_log(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
 	}

 	if (max_cluster_size > 0 &&
@@ -1808,7 +1863,7 @@ neon_zeroextend(SMgrRelation reln, ForkNumber forkNum, BlockNumber blocknum,
 		if (current_size >= ((uint64) max_cluster_size) * 1024 * 1024)
 			ereport(ERROR,
 					(errcode(ERRCODE_DISK_FULL),
-					 errmsg("could not extend file because cluster size limit (%d MB) has been exceeded",
+					 errmsg(NEON_TAG "could not extend file because cluster size limit (%d MB) has been exceeded",
 							max_cluster_size),
 					 errhint("This limit is defined by neon.max_cluster_size GUC")));
 	}
@@ -1821,7 +1876,7 @@ neon_zeroextend(SMgrRelation reln, ForkNumber forkNum, BlockNumber blocknum,
 	if ((uint64) blocknum + nblocks >= (uint64) InvalidBlockNumber)
 		ereport(ERROR,
 				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
-				 errmsg("cannot extend file \"%s\" beyond %u blocks",
+				 errmsg(NEON_TAG "cannot extend file \"%s\" beyond %u blocks",
 						relpath(reln->smgr_rlocator, forkNum),
 						InvalidBlockNumber)));

@@ -1882,7 +1937,7 @@ neon_open(SMgrRelation reln)
 	mdopen(reln);

 	/* no work */
-	elog(SmgrTrace, "[NEON_SMGR] open noop");
+	neon_log(SmgrTrace, "open noop");
 }

 /*
@@ -1919,7 +1974,7 @@ neon_prefetch(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum)
 			return mdprefetch(reln, forknum, blocknum);

 		default:
-			elog(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
+			neon_log(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
 	}

 	if (lfc_cache_contains(InfoFromSMgrRel(reln), forknum, blocknum))
@@ -1964,11 +2019,11 @@ neon_writeback(SMgrRelation reln, ForkNumber forknum,
 			return;

 		default:
-			elog(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
+			neon_log(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
 	}

 	/* not implemented */
-	elog(SmgrTrace, "[NEON_SMGR] writeback noop");
+	neon_log(SmgrTrace, "writeback noop");

 #ifdef DEBUG_COMPARE_LOCAL
 	if (IS_LOCAL_REL(reln))
@@ -2098,8 +2153,8 @@ neon_read_at_lsn(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
 		case T_NeonErrorResponse:
 			ereport(ERROR,
 					(errcode(ERRCODE_IO_ERROR),
-					 errmsg("could not read block %u in rel %u/%u/%u.%u from page server at lsn %X/%08X",
-							blkno,
+					 errmsg(NEON_TAG "[shard %d] could not read block %u in rel %u/%u/%u.%u from page server at lsn %X/%08X",
+							slot->shard_no, blkno,
 							RelFileInfoFmt(rinfo),
 							forkNum,
 							(uint32) (request_lsn >> 32), (uint32) request_lsn),
@@ -2107,7 +2162,7 @@ neon_read_at_lsn(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
 							   ((NeonErrorResponse *) resp)->message)));
 			break;
 		default:
-			elog(ERROR, "unexpected response from page server with tag 0x%02x", resp->tag);
+			neon_log(ERROR, "unexpected response from page server with tag 0x%02x", resp->tag);
 	}

 	/* buffer was used, clean up for later reuse */
@@ -2131,7 +2186,7 @@ neon_read(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno, void *buffer
 	switch (reln->smgr_relpersistence)
 	{
 		case 0:
-			elog(ERROR, "cannot call smgrread() on rel with unknown persistence");
+			neon_log(ERROR, "cannot call smgrread() on rel with unknown persistence");

 		case RELPERSISTENCE_PERMANENT:
 			break;
@@ -2142,7 +2197,7 @@ neon_read(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno, void *buffer
 			return;

 		default:
-			elog(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
+			neon_log(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
 	}

 	/* Try to read from local file cache */
@@ -2170,7 +2225,7 @@ neon_read(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno, void *buffer
 		{
 			if (!PageIsNew((Page) pageserver_masked))
 			{
-				elog(PANIC, "page is new in MD but not in Page Server at blk %u in rel %u/%u/%u fork %u (request LSN %X/%08X):\n%s\n",
+				neon_log(PANIC, "page is new in MD but not in Page Server at blk %u in rel %u/%u/%u fork %u (request LSN %X/%08X):\n%s\n",
 					 blkno,
 					 RelFileInfoFmt(InfoFromSMgrRel(reln)),
 					 forkNum,
@@ -2180,7 +2235,7 @@ neon_read(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno, void *buffer
 		}
 		else if (PageIsNew((Page) buffer))
 		{
-			elog(PANIC, "page is new in Page Server but not in MD at blk %u in rel %u/%u/%u fork %u (request LSN %X/%08X):\n%s\n",
+			neon_log(PANIC, "page is new in Page Server but not in MD at blk %u in rel %u/%u/%u fork %u (request LSN %X/%08X):\n%s\n",
 				 blkno,
 				 RelFileInfoFmt(InfoFromSMgrRel(reln)),
 				 forkNum,
@@ -2195,7 +2250,7 @@ neon_read(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno, void *buffer

 			if (memcmp(mdbuf_masked, pageserver_masked, BLCKSZ) != 0)
 			{
-				elog(PANIC, "heap buffers differ at blk %u in rel %u/%u/%u fork %u (request LSN %X/%08X):\n------ MD ------\n%s\n------ Page Server ------\n%s\n",
+				neon_log(PANIC, "heap buffers differ at blk %u in rel %u/%u/%u fork %u (request LSN %X/%08X):\n------ MD ------\n%s\n------ Page Server ------\n%s\n",
 					 blkno,
 					 RelFileInfoFmt(InfoFromSMgrRel(reln)),
 					 forkNum,
@@ -2214,7 +2269,7 @@ neon_read(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno, void *buffer

 				if (memcmp(mdbuf_masked, pageserver_masked, BLCKSZ) != 0)
 				{
-					elog(PANIC, "btree buffers differ at blk %u in rel %u/%u/%u fork %u (request LSN %X/%08X):\n------ MD ------\n%s\n------ Page Server ------\n%s\n",
+					neon_log(PANIC, "btree buffers differ at blk %u in rel %u/%u/%u fork %u (request LSN %X/%08X):\n------ MD ------\n%s\n------ Page Server ------\n%s\n",
 						 blkno,
 						 RelFileInfoFmt(InfoFromSMgrRel(reln)),
 						 forkNum,
@@ -2294,13 +2349,13 @@ neon_write(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, const vo
 			return;

 		default:
-			elog(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
+			neon_log(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
 	}

 	neon_wallog_page(reln, forknum, blocknum, buffer, false);

 	lsn = PageGetLSN((Page) buffer);
-	elog(SmgrTrace, "smgrwrite called for %u/%u/%u.%u blk %u, page LSN: %X/%08X",
+	neon_log(SmgrTrace, "smgrwrite called for %u/%u/%u.%u blk %u, page LSN: %X/%08X",
 		 RelFileInfoFmt(InfoFromSMgrRel(reln)),
 		 forknum, blocknum,
 		 (uint32) (lsn >> 32), (uint32) lsn);
@@ -2327,7 +2382,7 @@ neon_nblocks(SMgrRelation reln, ForkNumber forknum)
 	switch (reln->smgr_relpersistence)
 	{
 		case 0:
-			elog(ERROR, "cannot call smgrnblocks() on rel with unknown persistence");
+			neon_log(ERROR, "cannot call smgrnblocks() on rel with unknown persistence");
 			break;

 		case RELPERSISTENCE_PERMANENT:
@@ -2338,12 +2393,12 @@ neon_nblocks(SMgrRelation reln, ForkNumber forknum)
 			return mdnblocks(reln, forknum);

 		default:
-			elog(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
+			neon_log(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
 	}

 	if (get_cached_relsize(InfoFromSMgrRel(reln), forknum, &n_blocks))
 	{
-		elog(SmgrTrace, "cached nblocks for %u/%u/%u.%u: %u blocks",
+		neon_log(SmgrTrace, "cached nblocks for %u/%u/%u.%u: %u blocks",
 			 RelFileInfoFmt(InfoFromSMgrRel(reln)),
 			 forknum, n_blocks);
 		return n_blocks;
@@ -2371,7 +2426,7 @@ neon_nblocks(SMgrRelation reln, ForkNumber forknum)
 		case T_NeonErrorResponse:
 			ereport(ERROR,
 					(errcode(ERRCODE_IO_ERROR),
-					 errmsg("could not read relation size of rel %u/%u/%u.%u from page server at lsn %X/%08X",
+					 errmsg(NEON_TAG "could not read relation size of rel %u/%u/%u.%u from page server at lsn %X/%08X",
 							RelFileInfoFmt(InfoFromSMgrRel(reln)),
 							forknum,
 							(uint32) (request_lsn >> 32), (uint32) request_lsn),
@@ -2380,11 +2435,11 @@ neon_nblocks(SMgrRelation reln, ForkNumber forknum)
 			break;

 		default:
-			elog(ERROR, "unexpected response from page server with tag 0x%02x", resp->tag);
+			neon_log(ERROR, "unexpected response from page server with tag 0x%02x", resp->tag);
 	}
 	update_cached_relsize(InfoFromSMgrRel(reln), forknum, n_blocks);

-	elog(SmgrTrace, "neon_nblocks: rel %u/%u/%u fork %u (request LSN %X/%08X): %u blocks",
+	neon_log(SmgrTrace, "neon_nblocks: rel %u/%u/%u fork %u (request LSN %X/%08X): %u blocks",
 		 RelFileInfoFmt(InfoFromSMgrRel(reln)),
 		 forknum,
 		 (uint32) (request_lsn >> 32), (uint32) request_lsn,
@@ -2427,7 +2482,7 @@ neon_dbsize(Oid dbNode)
 		case T_NeonErrorResponse:
 			ereport(ERROR,
 					(errcode(ERRCODE_IO_ERROR),
-					 errmsg("could not read db size of db %u from page server at lsn %X/%08X",
+					 errmsg(NEON_TAG "could not read db size of db %u from page server at lsn %X/%08X",
 							dbNode,
 							(uint32) (request_lsn >> 32), (uint32) request_lsn),
 					 errdetail("page server returned error: %s",
@@ -2435,10 +2490,10 @@ neon_dbsize(Oid dbNode)
 			break;

 		default:
-			elog(ERROR, "unexpected response from page server with tag 0x%02x", resp->tag);
+			neon_log(ERROR, "unexpected response from page server with tag 0x%02x", resp->tag);
 	}

-	elog(SmgrTrace, "neon_dbsize: db %u (request LSN %X/%08X): %ld bytes",
+	neon_log(SmgrTrace, "neon_dbsize: db %u (request LSN %X/%08X): %ld bytes",
 		 dbNode,
 		 (uint32) (request_lsn >> 32), (uint32) request_lsn,
 		 db_size);
@@ -2458,7 +2513,7 @@ neon_truncate(SMgrRelation reln, ForkNumber forknum, BlockNumber nblocks)
 	switch (reln->smgr_relpersistence)
 	{
 		case 0:
-			elog(ERROR, "cannot call smgrtruncate() on rel with unknown persistence");
+			neon_log(ERROR, "cannot call smgrtruncate() on rel with unknown persistence");
 			break;

 		case RELPERSISTENCE_PERMANENT:
@@ -2470,7 +2525,7 @@ neon_truncate(SMgrRelation reln, ForkNumber forknum, BlockNumber nblocks)
 			return;

 		default:
-			elog(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
+			neon_log(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
 	}

 	set_cached_relsize(InfoFromSMgrRel(reln), forknum, nblocks);
@@ -2526,7 +2581,7 @@ neon_immedsync(SMgrRelation reln, ForkNumber forknum)
 	switch (reln->smgr_relpersistence)
 	{
 		case 0:
-			elog(ERROR, "cannot call smgrimmedsync() on rel with unknown persistence");
+			neon_log(ERROR, "cannot call smgrimmedsync() on rel with unknown persistence");
 			break;

 		case RELPERSISTENCE_PERMANENT:
@@ -2538,10 +2593,10 @@ neon_immedsync(SMgrRelation reln, ForkNumber forknum)
 			return;

 		default:
-			elog(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
+			neon_log(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
 	}

-	elog(SmgrTrace, "[NEON_SMGR] immedsync noop");
+	neon_log(SmgrTrace, "[NEON_SMGR] immedsync noop");

 #ifdef DEBUG_COMPARE_LOCAL
 	if (IS_LOCAL_REL(reln))
@@ -2566,17 +2621,17 @@ neon_start_unlogged_build(SMgrRelation reln)
 	 * progress at a time. That's enough for the current usage.
 	 */
 	if (unlogged_build_phase != UNLOGGED_BUILD_NOT_IN_PROGRESS)
-		elog(ERROR, "unlogged relation build is already in progress");
+		neon_log(ERROR, "unlogged relation build is already in progress");
 	Assert(unlogged_build_rel == NULL);

 	ereport(SmgrTrace,
-			(errmsg("starting unlogged build of relation %u/%u/%u",
+			(errmsg(NEON_TAG "starting unlogged build of relation %u/%u/%u",
 					RelFileInfoFmt(InfoFromSMgrRel(reln)))));

 	switch (reln->smgr_relpersistence)
 	{
 		case 0:
-			elog(ERROR, "cannot call smgr_start_unlogged_build() on rel with unknown persistence");
+			neon_log(ERROR, "cannot call smgr_start_unlogged_build() on rel with unknown persistence");
 			break;

 		case RELPERSISTENCE_PERMANENT:
@@ -2589,11 +2644,11 @@ neon_start_unlogged_build(SMgrRelation reln)
 			return;

 		default:
-			elog(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
+			neon_log(ERROR, "unknown relpersistence '%c'", reln->smgr_relpersistence);
 	}

 	if (smgrnblocks(reln, MAIN_FORKNUM) != 0)
-		elog(ERROR, "cannot perform unlogged index build, index is not empty ");
+		neon_log(ERROR, "cannot perform unlogged index build, index is not empty ");

 	unlogged_build_rel = reln;
 	unlogged_build_phase = UNLOGGED_BUILD_PHASE_1;
@@ -2620,7 +2675,7 @@ neon_finish_unlogged_build_phase_1(SMgrRelation reln)
 	Assert(unlogged_build_rel == reln);

 	ereport(SmgrTrace,
-			(errmsg("finishing phase 1 of unlogged build of relation %u/%u/%u",
+			(errmsg(NEON_TAG "finishing phase 1 of unlogged build of relation %u/%u/%u",
 					RelFileInfoFmt(InfoFromSMgrRel(reln)))));

 	if (unlogged_build_phase == UNLOGGED_BUILD_NOT_PERMANENT)
@@ -2649,7 +2704,7 @@ neon_end_unlogged_build(SMgrRelation reln)
 	Assert(unlogged_build_rel == reln);

 	ereport(SmgrTrace,
-			(errmsg("ending unlogged build of relation %u/%u/%u",
+			(errmsg(NEON_TAG "ending unlogged build of relation %u/%u/%u",
 					RelFileInfoFmt(InfoFromNInfoB(rinfob)))));

 	if (unlogged_build_phase != UNLOGGED_BUILD_NOT_PERMANENT)
@@ -2664,7 +2719,7 @@ neon_end_unlogged_build(SMgrRelation reln)
 		rinfob = InfoBFromSMgrRel(reln);
 		for (int forknum = 0; forknum <= MAX_FORKNUM; forknum++)
 		{
-			elog(SmgrTrace, "forgetting cached relsize for %u/%u/%u.%u",
+			neon_log(SmgrTrace, "forgetting cached relsize for %u/%u/%u.%u",
 				 RelFileInfoFmt(InfoFromNInfoB(rinfob)),
 				 forknum);

@@ -2707,7 +2762,7 @@ AtEOXact_neon(XactEvent event, void *arg)
 				unlogged_build_phase = UNLOGGED_BUILD_NOT_IN_PROGRESS;
 				ereport(ERROR,
 						(errcode(ERRCODE_INTERNAL_ERROR),
-						 (errmsg("unlogged index build was not properly finished"))));
+						 (errmsg(NEON_TAG "unlogged index build was not properly finished"))));
 			}
 			break;
 	}
@@ -2806,14 +2861,14 @@ neon_extend_rel_size(NRelFileInfo rinfo, ForkNumber forknum, BlockNumber blkno,
 		set_cached_relsize(rinfo, forknum, relsize);
 		SetLastWrittenLSNForRelation(end_recptr, rinfo, forknum);

-		elog(SmgrTrace, "Set length to %d", relsize);
+		neon_log(SmgrTrace, "Set length to %d", relsize);
 	}
 }

 #define FSM_TREE_DEPTH	((SlotsPerFSMPage >= 1626) ? 3 : 4)

 /*
- * TODO: May be it is better to make correspondent fgunctio from freespace.c public?
+ * TODO: May be it is better to make correspondent function from freespace.c public?
 */
 static BlockNumber
 get_fsm_physical_block(BlockNumber heapblk)
@@ -2894,7 +2949,7 @@ neon_redo_read_buffer_filter(XLogReaderState *record, uint8 block_id)

 #if PG_VERSION_NUM < 150000
 	if (!XLogRecGetBlockTag(record, block_id, &rinfo, &forknum, &blkno))
-		elog(PANIC, "failed to locate backup block with ID %d", block_id);
+		neon_log(PANIC, "failed to locate backup block with ID %d", block_id);
 #else
 	XLogRecGetBlockTag(record, block_id, &rinfo, &forknum, &blkno);
 #endif
@@ -1482,21 +1482,6 @@ walprop_pg_wait_event_set(WalProposer *wp, long timeout, Safekeeper **sk, uint32
 #if PG_MAJORVERSION_NUM >= 16
 	if (WalSndCtl != NULL)
 		ConditionVariablePrepareToSleep(&WalSndCtl->wal_flush_cv);
-
-	/*
-	 * Now that we prepared the condvar, check flush ptr again -- it might have
-	 * changed before we subscribed to cv so we missed the wakeup.
-	 *
-	 * Do that only when we're interested in new WAL: without sync-safekeepers
-	 * and if election already passed.
-	 */
-	if (!wp->config->syncSafekeepers && wp->availableLsn != InvalidXLogRecPtr && GetFlushRecPtr(NULL) > wp->availableLsn)
-	{
-		ConditionVariableCancelSleep();
-		ResetLatch(MyLatch);
-		*events = WL_LATCH_SET;
-		return 1;
-	}
 #endif

 	/*
@@ -6,6 +6,7 @@
 use std::{net::SocketAddr, sync::Arc};

 use futures::future::Either;
+use itertools::Itertools;
 use proxy::config::TlsServerEndPoint;
 use proxy::proxy::run_until_cancelled;
 use tokio::net::TcpListener;
@@ -75,12 +76,10 @@ async fn main() -> anyhow::Result<()> {
            let key = {
                let key_bytes = std::fs::read(key_path).context("TLS key file")?;
                let mut keys = rustls_pemfile::pkcs8_private_keys(&mut &key_bytes[..])
-                    .collect::<Result<Vec<_>, _>>()
                    .context(format!("Failed to read TLS keys at '{key_path}'"))?;

                ensure!(keys.len() == 1, "keys.len() = {} (should be 1)", keys.len());
-                let bytes = keys.pop().unwrap().secret_pkcs8_der().to_owned();
-                rustls::pki_types::PrivateKeyDer::Pkcs1(bytes.into())
+                keys.pop().map(rustls::PrivateKey).unwrap()
            };

            let cert_chain_bytes = std::fs::read(cert_path)
@@ -88,23 +87,25 @@ async fn main() -> anyhow::Result<()> {

            let cert_chain = {
                rustls_pemfile::certs(&mut &cert_chain_bytes[..])
-                .collect::<Result<Vec<_>,_>>()
                    .context(format!(
                        "Failed to read TLS certificate chain from bytes from file at '{cert_path}'."
                    ))?
+                    .into_iter()
+                    .map(rustls::Certificate)
+                    .collect_vec()
            };

            // needed for channel bindings
            let first_cert = cert_chain.first().context("missing certificate")?;
            let tls_server_end_point = TlsServerEndPoint::new(first_cert)?;

-            let tls_config = rustls::ServerConfig::builder_with_protocol_versions(&[
-                &rustls::version::TLS13,
-                &rustls::version::TLS12,
-            ])
-            .with_no_client_auth()
-            .with_single_cert(cert_chain, key)?
-            .into();
+            let tls_config = rustls::ServerConfig::builder()
+                .with_safe_default_cipher_suites()
+                .with_safe_default_kx_groups()
+                .with_protocol_versions(&[&rustls::version::TLS13, &rustls::version::TLS12])?
+                .with_no_client_auth()
+                .with_single_cert(cert_chain, key)?
+                .into();

            (tls_config, tls_server_end_point)
        }
@@ -1,9 +1,6 @@
 use crate::{auth, rate_limiter::RateBucketInfo};
 use anyhow::{bail, ensure, Context, Ok};
-use rustls::{
-    crypto::ring::sign,
-    pki_types::{CertificateDer, PrivateKeyDer},
-};
+use rustls::{sign, Certificate, PrivateKey};
 use sha2::{Digest, Sha256};
 use std::{
    collections::{HashMap, HashSet},
@@ -88,14 +85,14 @@ pub fn configure_tls(

    let cert_resolver = Arc::new(cert_resolver);

-    // allow TLS 1.2 to be compatible with older client libraries
-    let config = rustls::ServerConfig::builder_with_protocol_versions(&[
-        &rustls::version::TLS13,
-        &rustls::version::TLS12,
-    ])
-    .with_no_client_auth()
-    .with_cert_resolver(cert_resolver.clone())
-    .into();
+    let config = rustls::ServerConfig::builder()
+        .with_safe_default_cipher_suites()
+        .with_safe_default_kx_groups()
+        // allow TLS 1.2 to be compatible with older client libraries
+        .with_protocol_versions(&[&rustls::version::TLS13, &rustls::version::TLS12])?
+        .with_no_client_auth()
+        .with_cert_resolver(cert_resolver.clone())
+        .into();

    Ok(TlsConfig {
        config,
@@ -133,14 +130,14 @@ pub enum TlsServerEndPoint {
 }

 impl TlsServerEndPoint {
-    pub fn new(cert: &CertificateDer) -> anyhow::Result<Self> {
+    pub fn new(cert: &Certificate) -> 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)
+        let pem = x509_parser::parse_x509_certificate(&cert.0)
            .context("Failed to parse PEM object from cerficiate")?
            .1;

@@ -150,7 +147,8 @@ impl TlsServerEndPoint {
        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();
+            let tls_server_end_point: [u8; 32] =
+                Sha256::new().chain_update(&cert.0).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 {
@@ -164,7 +162,7 @@ impl TlsServerEndPoint {
    }
 }

-#[derive(Default, Debug)]
+#[derive(Default)]
 pub struct CertResolver {
    certs: HashMap<String, (Arc<rustls::sign::CertifiedKey>, TlsServerEndPoint)>,
    default: Option<(Arc<rustls::sign::CertifiedKey>, TlsServerEndPoint)>,
@@ -184,12 +182,11 @@ impl CertResolver {
        let priv_key = {
            let key_bytes = std::fs::read(key_path)
                .context(format!("Failed to read TLS keys at '{key_path}'"))?;
-            let keys: Result<Vec<_>, _> =
-                rustls_pemfile::pkcs8_private_keys(&mut &key_bytes[..]).collect();
-            let mut keys = keys.context(format!("Failed to parse TLS keys at '{key_path}'"))?;
+            let mut keys = rustls_pemfile::pkcs8_private_keys(&mut &key_bytes[..])
+                .context(format!("Failed to parse TLS keys at '{key_path}'"))?;

            ensure!(keys.len() == 1, "keys.len() = {} (should be 1)", keys.len());
-            keys.pop().unwrap()
+            keys.pop().map(rustls::PrivateKey).unwrap()
        };

        let cert_chain_bytes = std::fs::read(cert_path)
@@ -197,28 +194,30 @@ impl CertResolver {

        let cert_chain = {
            rustls_pemfile::certs(&mut &cert_chain_bytes[..])
-                .collect::<Result<Vec<_>, _>>()
                .with_context(|| {
                    format!(
                    "Failed to read TLS certificate chain from bytes from file at '{cert_path}'."
                )
                })?
+                .into_iter()
+                .map(rustls::Certificate)
+                .collect()
        };

-        self.add_cert(PrivateKeyDer::Pkcs8(priv_key), cert_chain, is_default)
+        self.add_cert(priv_key, cert_chain, is_default)
    }

    pub fn add_cert(
        &mut self,
-        priv_key: PrivateKeyDer,
-        cert_chain: Vec<CertificateDer<'static>>,
+        priv_key: PrivateKey,
+        cert_chain: Vec<Certificate>,
        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)
+        let pem = x509_parser::parse_x509_certificate(&first_cert.0)
            .context("Failed to parse PEM object from cerficiate")?
            .1;

@@ -328,23 +328,19 @@ impl<T: AsyncRead> AsyncRead for WithClientIp<T> {

 impl AsyncAccept for ProxyProtocolAccept {
    type Connection = WithClientIp<AddrStream>;
-    type Address = std::net::SocketAddr;
+
    type Error = io::Error;

    fn poll_accept(
        mut self: Pin<&mut Self>,
        cx: &mut Context<'_>,
-    ) -> Poll<Result<(Self::Connection, Self::Address), Self::Error>> {
-        use hyper::server::accept::Accept;
+    ) -> Poll<Option<Result<Self::Connection, Self::Error>>> {
        let conn = ready!(Pin::new(&mut self.incoming).poll_accept(cx)?);
        let Some(conn) = conn else {
-            return Poll::Ready(Err(io::Error::new(
-                io::ErrorKind::NotConnected,
-                "no incoming connection?",
-            )));
+            return Poll::Ready(None);
        };
-        let addr = conn.remote_addr();
-        Poll::Ready(Ok((WithClientIp::new(conn), addr)))
+
+        Poll::Ready(Some(Ok(WithClientIp::new(conn))))
    }
 }

@@ -11,21 +11,16 @@ use crate::console::{CachedNodeInfo, NodeInfo};
 use crate::proxy::retry::{retry_after, NUM_RETRIES_CONNECT};
 use crate::{auth, http, sasl, scram};
 use async_trait::async_trait;
-use postgres_backend::{MakeRustlsConnect, RustlsStream};
 use rstest::rstest;
-use rustls::pki_types::{CertificateDer, PrivateKeyDer, PrivatePkcs1KeyDer};
 use tokio_postgres::config::SslMode;
 use tokio_postgres::tls::{MakeTlsConnect, NoTls};
+use tokio_postgres_rustls::{MakeRustlsConnect, RustlsStream};

 /// Generate a set of TLS certificates: CA + server.
 fn generate_certs(
    hostname: &str,
    common_name: &str,
-) -> anyhow::Result<(
-    CertificateDer<'static>,
-    CertificateDer<'static>,
-    PrivateKeyDer<'static>,
-)> {
+) -> anyhow::Result<(rustls::Certificate, rustls::Certificate, rustls::PrivateKey)> {
    let ca = rcgen::Certificate::from_params({
        let mut params = rcgen::CertificateParams::default();
        params.is_ca = rcgen::IsCa::Ca(rcgen::BasicConstraints::Unconstrained);
@@ -42,9 +37,9 @@ fn generate_certs(
    })?;

    Ok((
-        CertificateDer::from(ca.serialize_der()?),
-        CertificateDer::from(cert.serialize_der_with_signer(&ca)?),
-        PrivateKeyDer::Pkcs1(PrivatePkcs1KeyDer::from(cert.serialize_private_key_der())),
+        rustls::Certificate(ca.serialize_der()?),
+        rustls::Certificate(cert.serialize_der_with_signer(&ca)?),
+        rustls::PrivateKey(cert.serialize_private_key_der()),
    ))
 }

@@ -78,10 +73,10 @@ fn generate_tls_config<'a>(
    let (ca, cert, key) = generate_certs(hostname, common_name)?;

    let tls_config = {
-        let key_clone = rustls::pki_types::PrivateKeyDer::Pkcs1(key.secret_der().to_owned().into());
        let config = rustls::ServerConfig::builder()
+            .with_safe_defaults()
            .with_no_client_auth()
-            .with_single_cert(vec![cert.clone()], key_clone)?
+            .with_single_cert(vec![cert.clone()], key.clone())?
            .into();

        let mut cert_resolver = CertResolver::new();
@@ -98,9 +93,10 @@ fn generate_tls_config<'a>(

    let client_config = {
        let config = rustls::ClientConfig::builder()
+            .with_safe_defaults()
            .with_root_certificates({
                let mut store = rustls::RootCertStore::empty();
-                store.add(ca)?;
+                store.add(&ca)?;
                store
            })
            .with_no_client_auth();
@@ -77,19 +77,14 @@ pub async fn task_main(
    let ws_connections = tokio_util::task::task_tracker::TaskTracker::new();
    ws_connections.close(); // allows `ws_connections.wait to complete`

-    let tls_listener = TlsListener::new(tls_acceptor, addr_incoming)
-        .map(|x| match x {
-            Ok((conn, _)) => Ok(conn),
-            Err(e) => Err(e),
-        })
-        .filter(|conn| {
-            if let Err(err) = conn {
-                error!("failed to accept TLS connection for websockets: {err:?}");
-                ready(false)
-            } else {
-                ready(true)
-            }
-        });
+    let tls_listener = TlsListener::new(tls_acceptor, addr_incoming).filter(|conn| {
+        if let Err(err) = conn {
+            error!("failed to accept TLS connection for websockets: {err:?}");
+            ready(false)
+        } else {
+            ready(true)
+        }
+    });

    let make_svc = hyper::service::make_service_fn(
        |stream: &tokio_rustls::server::TlsStream<WithClientIp<AddrStream>>| {
@@ -510,21 +510,13 @@ class PageserverHttpClient(requests.Session):
        assert res_json is None

    def timeline_get_lsn_by_timestamp(
-        self,
-        tenant_id: TenantId,
-        timeline_id: TimelineId,
-        timestamp,
-        version: Optional[int] = None,
+        self, tenant_id: TenantId, timeline_id: TimelineId, timestamp, version: int
    ):
        log.info(
            f"Requesting lsn by timestamp {timestamp}, tenant {tenant_id}, timeline {timeline_id}"
        )
-        if version is None:
-            version_str = ""
-        else:
-            version_str = f"&version={version}"
        res = self.get(
-            f"http://localhost:{self.port}/v1/tenant/{tenant_id}/timeline/{timeline_id}/get_lsn_by_timestamp?timestamp={timestamp}{version_str}",
+            f"http://localhost:{self.port}/v1/tenant/{tenant_id}/timeline/{timeline_id}/get_lsn_by_timestamp?timestamp={timestamp}&version={version}",
        )
        self.verbose_error(res)
        res_json = res.json()
@@ -17,27 +17,6 @@ class LabelledQuery:
    query: str


-# This must run before all tests in this module
-# create extension pg_stat_statements if it does not exist
-# and TEST_OLAP_COLLECT_PG_STAT_STATEMENTS is set to true (default false)
-# Theoretically this could be in a module or session scope fixture,
-# however the code depends on other fixtures that have function scope
-@pytest.mark.skipif(
-    os.getenv("TEST_OLAP_COLLECT_PG_STAT_STATEMENTS", "false").lower() == "false",
-    reason="Skipping - Creating extension pg_stat_statements",
-)
-@pytest.mark.remote_cluster
-def test_clickbench_create_pg_stat_statements(remote_compare: RemoteCompare):
-    log.info("Creating extension pg_stat_statements")
-    query = LabelledQuery(
-        "Q_CREATE_EXTENSION", r"CREATE EXTENSION IF NOT EXISTS pg_stat_statements;"
-    )
-    run_psql(remote_compare, query, times=1, explain=False)
-    log.info("Reset pg_stat_statements")
-    query = LabelledQuery("Q_RESET", r"SELECT pg_stat_statements_reset();")
-    run_psql(remote_compare, query, times=1, explain=False)
-
-
 # A list of queries to run.
 # Please do not alter the label for the query, as it is used to identify it.
 # Labels for ClickBench queries match the labels in ClickBench reports
@@ -99,8 +78,6 @@ QUERIES: Tuple[LabelledQuery, ...] = (
    # fmt: on
 )

-EXPLAIN_STRING: str = "EXPLAIN (ANALYZE, VERBOSE, BUFFERS, COSTS, SETTINGS, FORMAT JSON)"
-

 def get_scale() -> List[str]:
    # We parametrize each tpc-h and clickbench test with scale
@@ -111,10 +88,7 @@ def get_scale() -> List[str]:
    return [scale]


-# run the query times times plus once with EXPLAIN VERBOSE if explain is requestd
-def run_psql(
-    env: RemoteCompare, labelled_query: LabelledQuery, times: int, explain: bool = False
-) -> None:
+def run_psql(env: RemoteCompare, labelled_query: LabelledQuery, times: int) -> None:
    # prepare connstr:
    # - cut out password from connstr to pass it via env
    # - add options to connstr
@@ -134,13 +108,6 @@ def run_psql(
        log.info(f"Run {run}/{times}")
        with env.zenbenchmark.record_duration(f"{label}/{run}"):
            env.pg_bin.run_capture(["psql", connstr, "-c", query], env=environ)
-    if explain:
-        log.info(f"Explaining query {label}")
-        run += 1
-        with env.zenbenchmark.record_duration(f"{label}/EXPLAIN"):
-            env.pg_bin.run_capture(
-                ["psql", connstr, "-c", f"{EXPLAIN_STRING} {query}"], env=environ
-            )


@pytest.mark.parametrize("scale", get_scale())
@@ -153,9 +120,8 @@ def test_clickbench(query: LabelledQuery, remote_compare: RemoteCompare, scale:
    Based on https://github.com/ClickHouse/ClickBench/tree/c00135ca5b6a0d86fedcdbf998fdaa8ed85c1c3b/aurora-postgresql
    The DB prepared manually in advance
    """
-    explain: bool = os.getenv("TEST_OLAP_COLLECT_EXPLAIN", "false").lower() == "true"

-    run_psql(remote_compare, query, times=3, explain=explain)
+    run_psql(remote_compare, query, times=3)


 def tpch_queuies() -> Tuple[ParameterSet, ...]:
@@ -229,16 +195,3 @@ def test_user_examples(remote_compare: RemoteCompare):
        """,
    )
    run_psql(remote_compare, query, times=3)
-
-
-# This must run after all tests in this module
-# Collect pg_stat_statements after running the tests if TEST_OLAP_COLLECT_PG_STAT_STATEMENTS is set to true (default false)
-@pytest.mark.skipif(
-    os.getenv("TEST_OLAP_COLLECT_PG_STAT_STATEMENTS", "false").lower() == "false",
-    reason="Skipping - Collecting pg_stat_statements",
-)
-@pytest.mark.remote_cluster
-def test_clickbench_collect_pg_stat_statements(remote_compare: RemoteCompare):
-    log.info("Collecting pg_stat_statements")
-    query = LabelledQuery("Q_COLLECT_PG_STAT_STATEMENTS", r"SELECT * from pg_stat_statements;")
-    run_psql(remote_compare, query, times=1, explain=False)
@@ -8,6 +8,71 @@ from fixtures.types import Lsn
 from fixtures.utils import query_scalar


+#
+# Test pageserver get_lsn_by_timestamp API
+#
+def test_lsn_mapping_old(neon_env_builder: NeonEnvBuilder):
+    env = neon_env_builder.init_start()
+
+    new_timeline_id = env.neon_cli.create_branch("test_lsn_mapping")
+    endpoint_main = env.endpoints.create_start("test_lsn_mapping")
+    log.info("postgres is running on 'test_lsn_mapping' branch")
+
+    cur = endpoint_main.connect().cursor()
+    # Create table, and insert rows, each in a separate transaction
+    # Disable synchronous_commit to make this initialization go faster.
+    #
+    # Each row contains current insert LSN and the current timestamp, when
+    # the row was inserted.
+    cur.execute("SET synchronous_commit=off")
+    cur.execute("CREATE TABLE foo (x integer)")
+    tbl = []
+    for i in range(1000):
+        cur.execute("INSERT INTO foo VALUES(%s)", (i,))
+        # Get the timestamp at UTC
+        after_timestamp = query_scalar(cur, "SELECT clock_timestamp()").replace(tzinfo=None)
+        tbl.append([i, after_timestamp])
+
+    # Execute one more transaction with synchronous_commit enabled, to flush
+    # all the previous transactions
+    cur.execute("SET synchronous_commit=on")
+    cur.execute("INSERT INTO foo VALUES (-1)")
+
+    # Wait until WAL is received by pageserver
+    wait_for_last_flush_lsn(env, endpoint_main, env.initial_tenant, new_timeline_id)
+
+    with env.pageserver.http_client() as client:
+        # Check edge cases: timestamp in the future
+        probe_timestamp = tbl[-1][1] + timedelta(hours=1)
+        result = client.timeline_get_lsn_by_timestamp(
+            env.initial_tenant, new_timeline_id, f"{probe_timestamp.isoformat()}Z", 1
+        )
+        assert result == "future"
+
+        # timestamp too the far history
+        probe_timestamp = tbl[0][1] - timedelta(hours=10)
+        result = client.timeline_get_lsn_by_timestamp(
+            env.initial_tenant, new_timeline_id, f"{probe_timestamp.isoformat()}Z", 1
+        )
+        assert result == "past"
+
+        # Probe a bunch of timestamps in the valid range
+        for i in range(1, len(tbl), 100):
+            probe_timestamp = tbl[i][1]
+            lsn = client.timeline_get_lsn_by_timestamp(
+                env.initial_tenant, new_timeline_id, f"{probe_timestamp.isoformat()}Z", 1
+            )
+            # Call get_lsn_by_timestamp to get the LSN
+            # Launch a new read-only node at that LSN, and check that only the rows
+            # that were supposed to be committed at that point in time are visible.
+            endpoint_here = env.endpoints.create_start(
+                branch_name="test_lsn_mapping", endpoint_id="ep-lsn_mapping_read", lsn=lsn
+            )
+            assert endpoint_here.safe_psql("SELECT max(x) FROM foo")[0][0] == i
+
+            endpoint_here.stop_and_destroy()
+
+
 #
 # Test pageserver get_lsn_by_timestamp API
 #
@@ -65,7 +130,7 @@ def test_lsn_mapping(neon_env_builder: NeonEnvBuilder):
        # Timestamp is in the future
        probe_timestamp = tbl[-1][1] + timedelta(hours=1)
        result = client.timeline_get_lsn_by_timestamp(
-            tenant_id, timeline_id, f"{probe_timestamp.isoformat()}Z"
+            tenant_id, timeline_id, f"{probe_timestamp.isoformat()}Z", 2
        )
        assert result["kind"] == "future"
        # make sure that we return a well advanced lsn here
@@ -74,7 +139,7 @@ def test_lsn_mapping(neon_env_builder: NeonEnvBuilder):
        # Timestamp is in the unreachable past
        probe_timestamp = tbl[0][1] - timedelta(hours=10)
        result = client.timeline_get_lsn_by_timestamp(
-            tenant_id, timeline_id, f"{probe_timestamp.isoformat()}Z"
+            tenant_id, timeline_id, f"{probe_timestamp.isoformat()}Z", 2
        )
        assert result["kind"] == "past"
        # make sure that we return the minimum lsn here at the start of the range
@@ -84,7 +149,7 @@ def test_lsn_mapping(neon_env_builder: NeonEnvBuilder):
        for i in range(1, len(tbl), 100):
            probe_timestamp = tbl[i][1]
            result = client.timeline_get_lsn_by_timestamp(
-                tenant_id, timeline_id, f"{probe_timestamp.isoformat()}Z"
+                tenant_id, timeline_id, f"{probe_timestamp.isoformat()}Z", 2
            )
            assert result["kind"] not in ["past", "nodata"]
            lsn = result["lsn"]
@@ -1,5 +1,5 @@
 {
-    "postgres-v16": "863b71572bc441581efb3bbee2ad18af037be1bb",
-    "postgres-v15": "24333abb81a9ecae4541019478f0bf7d0b289df7",
-    "postgres-v14": "0bb356aa0cd1582112926fbcf0b5370222c2db6d"
+    "postgres-v16": "de8242c400f7870084861ac5796e0b5088b1898d",
+    "postgres-v15": "a2dc225ddfc8cae1849aa2316f435c58f0333d8c",
+    "postgres-v14": "03358bb0b5e0d33c238710139e768db9e75cfcc8"
 }
@@ -33,8 +33,11 @@ dashmap = { version = "5", default-features = false, features = ["raw-api"] }
 either = { version = "1" }
 fail = { version = "0.5", default-features = false, features = ["failpoints"] }
 futures = { version = "0.3" }
+futures-channel = { version = "0.3", features = ["sink"] }
+futures-core = { version = "0.3" }
 futures-executor = { version = "0.3" }
 futures-io = { version = "0.3" }
+futures-sink = { version = "0.3" }
 futures-util = { version = "0.3", features = ["channel", "io", "sink"] }
 hex = { version = "0.4", features = ["serde"] }
 hmac = { version = "0.12", default-features = false, features = ["reset"] }
@@ -53,6 +56,7 @@ regex = { version = "1" }
 regex-automata = { version = "0.4", default-features = false, features = ["dfa-onepass", "hybrid", "meta", "nfa-backtrack", "perf-inline", "perf-literal", "unicode"] }
 regex-syntax = { version = "0.8" }
 reqwest = { version = "0.11", default-features = false, features = ["blocking", "default-tls", "json", "multipart", "rustls-tls", "stream"] }
+ring = { version = "0.16", features = ["std"] }
 rustls = { version = "0.21", features = ["dangerous_configuration"] }
 scopeguard = { version = "1" }
 serde = { version = "1", features = ["alloc", "derive"] }
@@ -71,8 +75,8 @@ tracing-core = { version = "0.1" }
 tungstenite = { version = "0.20" }
 url = { version = "2", features = ["serde"] }
 uuid = { version = "1", features = ["serde", "v4"] }
-zstd = { version = "0.13" }
-zstd-safe = { version = "7", default-features = false, features = ["arrays", "legacy", "std", "zdict_builder"] }
+zstd = { version = "0.12" }
+zstd-safe = { version = "6", default-features = false, features = ["arrays", "legacy", "std", "zdict_builder"] }
 zstd-sys = { version = "2", default-features = false, features = ["legacy", "std", "zdict_builder"] }

 [build-dependencies]
Author	SHA1	Message	Date
Konstantin Knizhnik	8de5b63ac4	Remove assert for non-zero number of shards from AssignPageserverConnstring	2023-12-20 09:16:54 +02:00
Konstantin Knizhnik	338bf7a446	Bump postgres version	2023-12-19 22:17:07 +02:00
Konstantin Knizhnik	767ce2b187	Bump Postgres version	2023-12-19 16:21:15 +02:00
Konstantin Knizhnik	80bf8d4761	Allow empty connection string	2023-12-19 14:45:25 +02:00
Konstantin Knizhnik	c2b396905f	Merge with main	2023-12-19 10:20:56 +02:00
Konstantin Knizhnik	b43ae6e26c	Address review comments	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	151870b666	Update pgxn/neon/libpagestore.c Co-authored-by: Sasha Krassovsky <sasha@neon.tech>	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	4cdfff75f2	Update pgxn/neon/libpagestore.c Co-authored-by: Sasha Krassovsky <sasha@neon.tech>	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	02abe3c82e	Fix problem with stats collector at pg14	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	5c8b3d997f	Add [NEON_SMGR] to all messages produced by Neon exrtension	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	b465185738	Add [NEON_SMGR] to all messages produced by Neon exrtension	2023-12-19 09:22:42 +02:00
John Spray	4e47d3a984	pgxn: amend key hashing	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	3853646c75	[see #6052 ] make connection logging shard-aware	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	86c76d4b59	Fix shard map reload synchronization	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	864d5f84e9	Fix shard map reload mechanism	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	38f64ea8ef	Load shard map only at postmaster	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	2855074f55	Fix comments	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	16b0348c1f	Do not deop PS connections of config reload if connection strings are not changed	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	93c75c97e1	Fix shard hash caclulation	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	d3221243f8	Minor refectoring	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	3cbd2df0b3	Add neon.stripe_size	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	8ddfa0515e	Undo occsional changed in control_place_connector.c	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	3019f8fe8f	Merge with main	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	e720b5f9d4	Load shardmap from postgresql.conf	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	039fa60446	Take in account stripe size when calculating shard hash number	2023-12-19 09:22:42 +02:00
Konstantin Knizhnik	7ca20dd12b	Add support for PS shardoing in compute	2023-12-19 09:22:42 +02:00