optimise array encoding

.github/workflows/build_and_test.yml

@@ -87,24 +87,6 @@ jobs:
     uses: ./.github/workflows/build-build-tools-image.yml
     secrets: inherit
 
-  lint-openapi-spec:
-    runs-on: ubuntu-22.04
-    needs: [ meta, check-permissions ]
-    # We do need to run this in `.*-rc-pr` because of hotfixes.
-    if: ${{ contains(fromJSON('["pr", "push-main", "storage-rc-pr", "proxy-rc-pr", "compute-rc-pr"]'), needs.meta.outputs.run-kind) }}
-    steps:
-      - name: Harden the runner (Audit all outbound calls)
-        uses: step-security/harden-runner@4d991eb9b905ef189e4c376166672c3f2f230481 # v2.11.0
-        with:
-          egress-policy: audit
-      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
-      - uses: docker/login-action@74a5d142397b4f367a81961eba4e8cd7edddf772 # v3.4.0
-        with:
-          registry: ghcr.io
-          username: ${{ github.actor }}
-          password: ${{ secrets.GITHUB_TOKEN }}
-      - run: make lint-openapi-spec
-
   check-codestyle-python:
     needs: [ meta, check-permissions, build-build-tools-image ]
     # No need to run on `main` because we this in the merge queue. We do need to run this in `.*-rc-pr` because of hotfixes.
@@ -1004,7 +986,6 @@ jobs:
       - name: Verify docker-compose example and test extensions
         timeout-minutes: 60
         env:
-          PARALLEL_COMPUTES: 3
           TAG: >-
             ${{
               needs.meta.outputs.run-kind == 'compute-rc-pr'

.gitignore

@@ -15,8 +15,6 @@ neon.iml
 /.neon
 /integration_tests/.neon
 compaction-suite-results.*
-pgxn/neon/communicator/communicator_bindings.h
-docker-compose/docker-compose-parallel.yml
 
 # Coverage
 *.profraw

Cargo.lock

@@ -253,17 +253,6 @@ version = "1.1.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "a8ab6b55fe97976e46f91ddbed8d147d966475dc29b2032757ba47e02376fbc3"
 
-[[package]]
-name = "atomic_enum"
-version = "0.3.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "99e1aca718ea7b89985790c94aad72d77533063fe00bc497bb79a7c2dae6a661"
-dependencies = [
- "proc-macro2",
- "quote",
- "syn 2.0.100",
-]
-
 [[package]]
 name = "autocfg"
 version = "1.1.0"
@@ -698,40 +687,13 @@ dependencies = [
  "tracing",
 ]
 
-[[package]]
-name = "axum"
-version = "0.7.9"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "edca88bc138befd0323b20752846e6587272d3b03b0343c8ea28a6f819e6e71f"
-dependencies = [
- "async-trait",
- "axum-core 0.4.5",
- "bytes",
- "futures-util",
- "http 1.1.0",
- "http-body 1.0.0",
- "http-body-util",
- "itoa",
- "matchit 0.7.3",
- "memchr",
- "mime",
- "percent-encoding",
- "pin-project-lite",
- "rustversion",
- "serde",
- "sync_wrapper 1.0.1",
- "tower 0.5.2",
- "tower-layer",
- "tower-service",
-]
-
 [[package]]
 name = "axum"
 version = "0.8.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "6d6fd624c75e18b3b4c6b9caf42b1afe24437daaee904069137d8bab077be8b8"
 dependencies = [
- "axum-core 0.5.0",
+ "axum-core",
  "base64 0.22.1",
  "bytes",
  "form_urlencoded",
@@ -739,10 +701,10 @@ dependencies = [
  "http 1.1.0",
  "http-body 1.0.0",
  "http-body-util",
- "hyper 1.6.0",
+ "hyper 1.4.1",
  "hyper-util",
  "itoa",
- "matchit 0.8.4",
+ "matchit",
  "memchr",
  "mime",
  "percent-encoding",
@@ -762,26 +724,6 @@ dependencies = [
  "tracing",
 ]
 
-[[package]]
-name = "axum-core"
-version = "0.4.5"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "09f2bd6146b97ae3359fa0cc6d6b376d9539582c7b4220f041a33ec24c226199"
-dependencies = [
- "async-trait",
- "bytes",
- "futures-util",
- "http 1.1.0",
- "http-body 1.0.0",
- "http-body-util",
- "mime",
- "pin-project-lite",
- "rustversion",
- "sync_wrapper 1.0.1",
- "tower-layer",
- "tower-service",
-]
-
 [[package]]
 name = "axum-core"
 version = "0.5.0"
@@ -808,8 +750,8 @@ version = "0.10.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "460fc6f625a1f7705c6cf62d0d070794e94668988b1c38111baeec177c715f7b"
 dependencies = [
- "axum 0.8.1",
- "axum-core 0.5.0",
+ "axum",
+ "axum-core",
  "bytes",
  "form_urlencoded",
  "futures-util",
@@ -1346,31 +1288,10 @@ dependencies = [
 
 [[package]]
 name = "communicator"
-version = "0.0.0"
+version = "0.1.0"
 dependencies = [
- "atomic_enum",
- "axum 0.8.1",
- "bytes",
  "cbindgen",
- "clashmap",
- "http 1.1.0",
- "libc",
- "metrics",
  "neon-shmem",
- "nix 0.30.1",
- "pageserver_api",
- "pageserver_client_grpc",
- "pageserver_page_api",
- "prometheus",
- "prost 0.13.5",
- "thiserror 1.0.69",
- "tokio",
- "tokio-pipe",
- "tonic 0.12.3",
- "tracing",
- "tracing-subscriber",
- "uring-common",
- "utils",
  "workspace_hack",
 ]
 
@@ -1400,7 +1321,7 @@ dependencies = [
  "aws-sdk-kms",
  "aws-sdk-s3",
  "aws-smithy-types",
- "axum 0.8.1",
+ "axum",
  "axum-extra",
  "base64 0.22.1",
  "bytes",
@@ -1427,7 +1348,6 @@ dependencies = [
  "p256 0.13.2",
  "pageserver_page_api",
  "postgres",
- "postgres-types",
  "postgres_initdb",
  "postgres_versioninfo",
  "regex",
@@ -1705,9 +1625,9 @@ dependencies = [
 
 [[package]]
 name = "crossbeam-utils"
-version = "0.8.21"
+version = "0.8.19"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "d0a5c400df2834b80a4c3327b3aad3a4c4cd4de0629063962b03235697506a28"
+checksum = "248e3bacc7dc6baa3b21e405ee045c3047101a49145e7e9eca583ab4c2ca5345"
 
 [[package]]
 name = "crossterm"
@@ -2161,7 +2081,7 @@ name = "endpoint_storage"
 version = "0.0.1"
 dependencies = [
  "anyhow",
- "axum 0.8.1",
+ "axum",
  "axum-extra",
  "camino",
  "camino-tempfile",
@@ -2422,12 +2342,6 @@ version = "1.0.7"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "3f9eec918d3f24069decb9af1554cad7c880e2da24a9afd88aca000531ab82c1"
 
-[[package]]
-name = "foldhash"
-version = "0.1.5"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "d9c4f5dac5e15c24eb999c26181a6ca40b39fe946cbe4c263c7209467bc83af2"
-
 [[package]]
 name = "form_urlencoded"
 version = "1.2.1"
@@ -2448,7 +2362,7 @@ dependencies = [
  "futures-core",
  "futures-sink",
  "http-body-util",
- "hyper 1.6.0",
+ "hyper 1.4.1",
  "hyper-util",
  "pin-project",
  "rand 0.8.5",
@@ -2618,18 +2532,6 @@ dependencies = [
  "wasm-bindgen",
 ]
 
-[[package]]
-name = "getrandom"
-version = "0.3.3"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "26145e563e54f2cadc477553f1ec5ee650b00862f0a58bcd12cbdc5f0ea2d2f4"
-dependencies = [
- "cfg-if",
- "libc",
- "r-efi",
- "wasi 0.14.2+wasi-0.2.4",
-]
-
 [[package]]
 name = "gettid"
 version = "0.1.3"
@@ -2795,16 +2697,6 @@ version = "0.15.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "bf151400ff0baff5465007dd2f3e717f3fe502074ca563069ce3a6629d07b289"
 
-[[package]]
-name = "hashbrown"
-version = "0.15.4"
-source = "git+https://github.com/quantumish/hashbrown.git?rev=6610e6d#6610e6d2b1f288ef7b0709a3efefbc846395dc5e"
-dependencies = [
- "allocator-api2",
- "equivalent",
- "foldhash",
-]
-
 [[package]]
 name = "hashlink"
 version = "0.9.1"
@@ -3029,9 +2921,9 @@ dependencies = [
 
 [[package]]
 name = "httparse"
-version = "1.10.1"
+version = "1.8.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6dbf3de79e51f3d586ab4cb9d5c3e2c14aa28ed23d180cf89b4df0454a69cc87"
+checksum = "d897f394bad6a705d5f4104762e116a75639e470d80901eed05a860a95cb1904"
 
 [[package]]
 name = "httpdate"
@@ -3081,9 +2973,9 @@ dependencies = [
 
 [[package]]
 name = "hyper"
-version = "1.6.0"
+version = "1.4.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "cc2b571658e38e0c01b1fdca3bbbe93c00d3d71693ff2770043f8c29bc7d6f80"
+checksum = "50dfd22e0e76d0f662d429a5f80fcaf3855009297eab6a0a9f8543834744ba05"
 dependencies = [
  "bytes",
  "futures-channel",
@@ -3123,7 +3015,7 @@ checksum = "a0bea761b46ae2b24eb4aef630d8d1c398157b6fc29e6350ecf090a0b70c952c"
 dependencies = [
  "futures-util",
  "http 1.1.0",
- "hyper 1.6.0",
+ "hyper 1.4.1",
  "hyper-util",
  "rustls 0.22.4",
  "rustls-pki-types",
@@ -3138,7 +3030,7 @@ version = "0.5.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "3203a961e5c83b6f5498933e78b6b263e208c197b63e9c6c53cc82ffd3f63793"
 dependencies = [
- "hyper 1.6.0",
+ "hyper 1.4.1",
  "hyper-util",
  "pin-project-lite",
  "tokio",
@@ -3147,21 +3039,20 @@ dependencies = [
 
 [[package]]
 name = "hyper-util"
-version = "0.1.14"
+version = "0.1.7"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "dc2fdfdbff08affe55bb779f33b053aa1fe5dd5b54c257343c17edfa55711bdb"
+checksum = "cde7055719c54e36e95e8719f95883f22072a48ede39db7fc17a4e1d5281e9b9"
 dependencies = [
  "bytes",
  "futures-channel",
- "futures-core",
  "futures-util",
  "http 1.1.0",
  "http-body 1.0.0",
- "hyper 1.6.0",
- "libc",
+ "hyper 1.4.1",
  "pin-project-lite",
  "socket2",
  "tokio",
+ "tower 0.4.13",
  "tower-service",
  "tracing",
 ]
@@ -3714,9 +3605,9 @@ checksum = "4ee93343901ab17bd981295f2cf0026d4ad018c7c31ba84549a4ddbb47a45104"
 
 [[package]]
 name = "lock_api"
-version = "0.4.13"
+version = "0.4.10"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "96936507f153605bddfcda068dd804796c84324ed2510809e5b2a624c81da765"
+checksum = "c1cc9717a20b1bb222f333e6a92fd32f7d8a18ddc5a3191a11af45dcbf4dcd16"
 dependencies = [
  "autocfg",
  "scopeguard",
@@ -3759,12 +3650,6 @@ dependencies = [
  "regex-automata 0.1.10",
 ]
 
-[[package]]
-name = "matchit"
-version = "0.7.3"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0e7465ac9959cc2b1404e8e2367b43684a6d13790fe23056cc8c6c5a6b7bcb94"
-
 [[package]]
 name = "matchit"
 version = "0.8.4"
@@ -3872,8 +3757,8 @@ dependencies = [
  "procfs",
  "prometheus",
  "rand 0.8.5",
- "rand_distr 0.4.3",
- "twox-hash 1.6.3",
+ "rand_distr",
+ "twox-hash",
 ]
 
 [[package]]
@@ -3960,33 +3845,10 @@ checksum = "e5ce46fe64a9d73be07dcbe690a38ce1b293be448fd8ce1e6c1b8062c9f72c6a"
 name = "neon-shmem"
 version = "0.1.0"
 dependencies = [
- "ahash",
- "criterion",
- "foldhash",
- "hashbrown 0.15.4",
- "libc",
- "lock_api",
  "nix 0.30.1",
- "rand 0.9.1",
- "rand_distr 0.5.1",
- "rustc-hash 2.1.1",
- "seahash",
  "tempfile",
  "thiserror 1.0.69",
- "twox-hash 2.1.1",
  "workspace_hack",
- "xxhash-rust",
-]
-
-[[package]]
-name = "neonart"
-version = "0.1.0"
-dependencies = [
- "crossbeam-utils",
- "rand 0.9.1",
- "rand_distr 0.5.1",
- "spin",
- "tracing",
 ]
 
 [[package]]
@@ -4422,19 +4284,15 @@ version = "0.1.0"
 dependencies = [
  "anyhow",
  "async-trait",
- "axum 0.8.1",
  "bytes",
  "camino",
  "clap",
  "futures",
  "hdrhistogram",
- "http 1.1.0",
  "humantime",
  "humantime-serde",
- "metrics",
  "pageserver_api",
  "pageserver_client",
- "pageserver_client_grpc",
  "pageserver_page_api",
  "rand 0.8.5",
  "reqwest",
@@ -4464,7 +4322,6 @@ dependencies = [
  "pageserver_api",
  "postgres_ffi",
  "remote_storage",
- "serde",
  "serde_json",
  "svg_fmt",
  "thiserror 1.0.69",
@@ -4482,7 +4339,6 @@ dependencies = [
  "arc-swap",
  "async-compression",
  "async-stream",
- "base64 0.22.1",
  "bincode",
  "bit_field",
  "byteorder",
@@ -4520,7 +4376,6 @@ dependencies = [
  "pageserver_client",
  "pageserver_compaction",
  "pageserver_page_api",
- "peekable",
  "pem",
  "pin-project-lite",
  "postgres-protocol",
@@ -4534,7 +4389,6 @@ dependencies = [
  "pprof",
  "pq_proto",
  "procfs",
- "prost 0.13.5",
  "rand 0.8.5",
  "range-set-blaze",
  "regex",
@@ -4571,7 +4425,7 @@ dependencies = [
  "tower 0.5.2",
  "tracing",
  "tracing-utils",
- "twox-hash 1.6.3",
+ "twox-hash",
  "url",
  "utils",
  "uuid",
@@ -4638,26 +4492,6 @@ dependencies = [
  "workspace_hack",
 ]
 
-[[package]]
-name = "pageserver_client_grpc"
-version = "0.1.0"
-dependencies = [
- "anyhow",
- "arc-swap",
- "bytes",
- "compute_api",
- "futures",
- "pageserver_api",
- "pageserver_page_api",
- "tokio",
- "tokio-stream",
- "tokio-util",
- "tonic 0.13.1",
- "tracing",
- "utils",
- "workspace_hack",
-]
-
 [[package]]
 name = "pageserver_compaction"
 version = "0.1.0"
@@ -4783,7 +4617,7 @@ dependencies = [
  "paste",
  "seq-macro",
  "thrift",
- "twox-hash 1.6.3",
+ "twox-hash",
  "zstd",
  "zstd-sys",
 ]
@@ -4829,15 +4663,6 @@ dependencies = [
  "sha2",
 ]
 
-[[package]]
-name = "peekable"
-version = "0.3.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "225f9651e475709164f871dc2f5724956be59cb9edb055372ffeeab01ec2d20b"
-dependencies = [
- "smallvec",
-]
-
 [[package]]
 name = "pem"
 version = "3.0.3"
@@ -5471,7 +5296,7 @@ dependencies = [
  "humantime",
  "humantime-serde",
  "hyper 0.14.30",
- "hyper 1.6.0",
+ "hyper 1.4.1",
  "hyper-util",
  "indexmap 2.9.0",
  "ipnet",
@@ -5495,7 +5320,7 @@ dependencies = [
  "postgres_backend",
  "pq_proto",
  "rand 0.8.5",
- "rand_distr 0.4.3",
+ "rand_distr",
  "rcgen",
  "redis",
  "regex",
@@ -5599,12 +5424,6 @@ dependencies = [
  "proc-macro2",
 ]
 
-[[package]]
-name = "r-efi"
-version = "5.3.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "69cdb34c158ceb288df11e18b4bd39de994f6657d83847bdffdbd7f346754b0f"
-
 [[package]]
 name = "rand"
 version = "0.7.3"
@@ -5629,16 +5448,6 @@ dependencies = [
  "rand_core 0.6.4",
 ]
 
-[[package]]
-name = "rand"
-version = "0.9.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9fbfd9d094a40bf3ae768db9361049ace4c0e04a4fd6b359518bd7b73a73dd97"
-dependencies = [
- "rand_chacha 0.9.0",
- "rand_core 0.9.3",
-]
-
 [[package]]
 name = "rand_chacha"
 version = "0.2.2"
@@ -5659,16 +5468,6 @@ dependencies = [
  "rand_core 0.6.4",
 ]
 
-[[package]]
-name = "rand_chacha"
-version = "0.9.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "d3022b5f1df60f26e1ffddd6c66e8aa15de382ae63b3a0c1bfc0e4d3e3f325cb"
-dependencies = [
- "ppv-lite86",
- "rand_core 0.9.3",
-]
-
 [[package]]
 name = "rand_core"
 version = "0.5.1"
@@ -5687,15 +5486,6 @@ dependencies = [
  "getrandom 0.2.11",
 ]
 
-[[package]]
-name = "rand_core"
-version = "0.9.3"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "99d9a13982dcf210057a8a78572b2217b667c3beacbf3a0d8b454f6f82837d38"
-dependencies = [
- "getrandom 0.3.3",
-]
-
 [[package]]
 name = "rand_distr"
 version = "0.4.3"
@@ -5706,16 +5496,6 @@ dependencies = [
  "rand 0.8.5",
 ]
 
-[[package]]
-name = "rand_distr"
-version = "0.5.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6a8615d50dcf34fa31f7ab52692afec947c4dd0ab803cc87cb3b0b4570ff7463"
-dependencies = [
- "num-traits",
- "rand 0.9.1",
-]
-
 [[package]]
 name = "rand_hc"
 version = "0.2.0"
@@ -5904,8 +5684,6 @@ dependencies = [
  "azure_identity",
  "azure_storage",
  "azure_storage_blobs",
- "base64 0.22.1",
- "byteorder",
  "bytes",
  "camino",
  "camino-tempfile",
@@ -5914,7 +5692,7 @@ dependencies = [
  "http-body-util",
  "http-types",
  "humantime-serde",
- "hyper 1.6.0",
+ "hyper 1.4.1",
  "itertools 0.10.5",
  "metrics",
  "once_cell",
@@ -5954,7 +5732,7 @@ dependencies = [
  "http 1.1.0",
  "http-body 1.0.0",
  "http-body-util",
- "hyper 1.6.0",
+ "hyper 1.4.1",
  "hyper-rustls 0.26.0",
  "hyper-util",
  "ipnet",
@@ -6011,7 +5789,7 @@ dependencies = [
  "futures",
  "getrandom 0.2.11",
  "http 1.1.0",
- "hyper 1.6.0",
+ "hyper 1.4.1",
  "parking_lot 0.11.2",
  "reqwest",
  "reqwest-middleware",
@@ -6032,7 +5810,7 @@ dependencies = [
  "async-trait",
  "getrandom 0.2.11",
  "http 1.1.0",
- "matchit 0.8.4",
+ "matchit",
  "opentelemetry",
  "reqwest",
  "reqwest-middleware",
@@ -6519,12 +6297,6 @@ version = "0.4.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "621e3680f3e07db4c9c2c3fb07c6223ab2fab2e54bd3c04c3ae037990f428c32"
 
-[[package]]
-name = "seahash"
-version = "4.1.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "1c107b6f4780854c8b126e228ea8869f4d7b71260f962fefb57b996b8959ba6b"
-
 [[package]]
 name = "sec1"
 version = "0.3.0"
@@ -6986,12 +6758,12 @@ dependencies = [
 
 [[package]]
 name = "socket2"
-version = "0.5.10"
+version = "0.5.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e22376abed350d73dd1cd119b57ffccad95b4e585a7cda43e286245ce23c0678"
+checksum = "7b5fac59a5cb5dd637972e5fca70daf0523c9067fcdc4842f053dae04a18f8e9"
 dependencies = [
  "libc",
- "windows-sys 0.52.0",
+ "windows-sys 0.48.0",
 ]
 
 [[package]]
@@ -6999,9 +6771,6 @@ name = "spin"
 version = "0.9.8"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "6980e8d7511241f8acf4aebddbb1ff938df5eebe98691418c4468d0b72a96a67"
-dependencies = [
- "lock_api",
-]
 
 [[package]]
 name = "spinning_top"
@@ -7060,7 +6829,7 @@ dependencies = [
  "http-body-util",
  "http-utils",
  "humantime",
- "hyper 1.6.0",
+ "hyper 1.4.1",
  "hyper-util",
  "metrics",
  "once_cell",
@@ -7669,16 +7438,6 @@ dependencies = [
  "syn 2.0.100",
 ]
 
-[[package]]
-name = "tokio-pipe"
-version = "0.2.12"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f213a84bffbd61b8fa0ba8a044b4bbe35d471d0b518867181e82bd5c15542784"
-dependencies = [
- "libc",
- "tokio",
-]
-
 [[package]]
 name = "tokio-postgres"
 version = "0.7.10"
@@ -7873,25 +7632,16 @@ version = "0.12.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "877c5b330756d856ffcc4553ab34a5684481ade925ecc54bcd1bf02b1d0d4d52"
 dependencies = [
- "async-stream",
  "async-trait",
- "axum 0.7.9",
  "base64 0.22.1",
  "bytes",
- "h2 0.4.4",
  "http 1.1.0",
  "http-body 1.0.0",
  "http-body-util",
- "hyper 1.6.0",
- "hyper-timeout",
- "hyper-util",
  "percent-encoding",
  "pin-project",
  "prost 0.13.5",
- "socket2",
- "tokio",
  "tokio-stream",
- "tower 0.4.13",
  "tower-layer",
  "tower-service",
  "tracing",
@@ -7904,7 +7654,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "7e581ba15a835f4d9ea06c55ab1bd4dce26fc53752c69a04aac00703bfb49ba9"
 dependencies = [
  "async-trait",
- "axum 0.8.1",
+ "axum",
  "base64 0.22.1",
  "bytes",
  "flate2",
@@ -7912,7 +7662,7 @@ dependencies = [
  "http 1.1.0",
  "http-body 1.0.0",
  "http-body-util",
- "hyper 1.6.0",
+ "hyper 1.4.1",
  "hyper-timeout",
  "hyper-util",
  "percent-encoding",
@@ -7965,16 +7715,11 @@ checksum = "b8fa9be0de6cf49e536ce1851f987bd21a43b771b09473c3549a6c853db37c1c"
 dependencies = [
  "futures-core",
  "futures-util",
- "indexmap 1.9.3",
  "pin-project",
  "pin-project-lite",
- "rand 0.8.5",
- "slab",
  "tokio",
- "tokio-util",
  "tower-layer",
  "tower-service",
- "tracing",
 ]
 
 [[package]]
@@ -8245,15 +7990,6 @@ dependencies = [
  "static_assertions",
 ]
 
-[[package]]
-name = "twox-hash"
-version = "2.1.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "8b907da542cbced5261bd3256de1b3a1bf340a3d37f93425a07362a1d687de56"
-dependencies = [
- "rand 0.9.1",
-]
-
 [[package]]
 name = "typed-json"
 version = "0.1.1"
@@ -8467,7 +8203,7 @@ name = "vm_monitor"
 version = "0.1.0"
 dependencies = [
  "anyhow",
- "axum 0.8.1",
+ "axum",
  "cgroups-rs",
  "clap",
  "futures",
@@ -8579,15 +8315,6 @@ version = "0.11.0+wasi-snapshot-preview1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9c8d87e72b64a3b4db28d11ce29237c246188f4f51057d65a7eab63b7987e423"
 
-[[package]]
-name = "wasi"
-version = "0.14.2+wasi-0.2.4"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9683f9a5a998d873c0d21fcbe3c083009670149a8fab228644b8bd36b2c48cb3"
-dependencies = [
- "wit-bindgen-rt",
-]
-
 [[package]]
 name = "wasite"
 version = "0.1.0"
@@ -8945,15 +8672,6 @@ dependencies = [
  "windows-sys 0.48.0",
 ]
 
-[[package]]
-name = "wit-bindgen-rt"
-version = "0.39.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6f42320e61fe2cfd34354ecb597f86f413484a798ba44a8ca1165c58d42da6c1"
-dependencies = [
- "bitflags 2.8.0",
-]
-
 [[package]]
 name = "workspace_hack"
 version = "0.1.0"
@@ -8961,8 +8679,8 @@ dependencies = [
  "ahash",
  "anstream",
  "anyhow",
- "axum 0.8.1",
- "axum-core 0.5.0",
+ "axum",
+ "axum-core",
  "base64 0.21.7",
  "base64ct",
  "bytes",
@@ -8996,7 +8714,7 @@ dependencies = [
  "hex",
  "hmac",
  "hyper 0.14.30",
- "hyper 1.6.0",
+ "hyper 1.4.1",
  "hyper-util",
  "indexmap 2.9.0",
  "itertools 0.12.1",
@@ -9121,12 +8839,6 @@ version = "0.13.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4d25c75bf9ea12c4040a97f829154768bbbce366287e2dc044af160cd79a13fd"
 
-[[package]]
-name = "xxhash-rust"
-version = "0.8.15"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "fdd20c5420375476fbd4394763288da7eb0cc0b8c11deed431a91562af7335d3"
-
 [[package]]
 name = "yasna"
 version = "0.5.2"

Cargo.toml

@@ -8,7 +8,6 @@ members = [
     "pageserver/compaction",
     "pageserver/ctl",
     "pageserver/client",
-    "pageserver/client_grpc",
     "pageserver/pagebench",
     "pageserver/page_api",
     "proxy",
@@ -35,7 +34,6 @@ members = [
     "libs/pq_proto",
     "libs/tenant_size_model",
     "libs/metrics",
-    "libs/neonart",
     "libs/postgres_connection",
     "libs/remote_storage",
     "libs/tracing-utils",
@@ -93,7 +91,6 @@ clap = { version = "4.0", features = ["derive", "env"] }
 clashmap = { version = "1.0", features = ["raw-api"] }
 comfy-table = "7.1"
 const_format = "0.2"
-crossbeam-utils = "0.8.21"
 crc32c = "0.6"
 diatomic-waker = { version = "0.2.3" }
 either = "1.8"
@@ -152,7 +149,6 @@ parquet = { version = "53", default-features = false, features = ["zstd"] }
 parquet_derive = "53"
 pbkdf2 = { version = "0.12.1", features = ["simple", "std"] }
 pem = "3.0.3"
-peekable = "0.3.0"
 pin-project-lite = "0.2"
 pprof = { version = "0.14", features = ["criterion", "flamegraph", "frame-pointer", "prost-codec"] }
 procfs = "0.16"
@@ -189,7 +185,6 @@ smallvec = "1.11"
 smol_str = { version = "0.2.0", features = ["serde"] }
 socket2 = "0.5"
 spki = "0.7.3"
-spin = "0.9.8"
 strum = "0.26"
 strum_macros = "0.26"
 "subtle"  = "2.5.0"
@@ -201,6 +196,7 @@ thiserror = "1.0"
 tikv-jemallocator = { version = "0.6", features = ["profiling", "stats", "unprefixed_malloc_on_supported_platforms"] }
 tikv-jemalloc-ctl = { version = "0.6", features = ["stats"] }
 tokio = { version = "1.43.1", features = ["macros"] }
+tokio-epoll-uring = { git = "https://github.com/neondatabase/tokio-epoll-uring.git" , branch = "main" }
 tokio-io-timeout = "1.2.0"
 tokio-postgres-rustls = "0.12.0"
 tokio-rustls = { version = "0.26.0", default-features = false, features = ["tls12", "ring"]}
@@ -242,9 +238,6 @@ x509-cert = { version = "0.2.5" }
 env_logger = "0.11"
 log = "0.4"
 
-tokio-epoll-uring = { git = "https://github.com/neondatabase/tokio-epoll-uring.git" , branch = "main" }
-uring-common = { git = "https://github.com/neondatabase/tokio-epoll-uring.git" , branch = "main" }
-
 ## Libraries from neondatabase/ git forks, ideally with changes to be upstreamed
 postgres = { git = "https://github.com/neondatabase/rust-postgres.git", branch = "neon" }
 postgres-protocol = { git = "https://github.com/neondatabase/rust-postgres.git", branch = "neon" }
@@ -268,7 +261,6 @@ neon-shmem = { version = "0.1", path = "./libs/neon-shmem/" }
 pageserver = { path = "./pageserver" }
 pageserver_api = { version = "0.1", path = "./libs/pageserver_api/" }
 pageserver_client = { path = "./pageserver/client" }
-pageserver_client_grpc = { path = "./pageserver/client_grpc" }
 pageserver_compaction = { version = "0.1", path = "./pageserver/compaction/" }
 pageserver_page_api = { path = "./pageserver/page_api" }
 postgres_backend = { version = "0.1", path = "./libs/postgres_backend/" }

Makefile

@@ -220,15 +220,6 @@ neon-pgindent: postgres-v17-pg-bsd-indent neon-pg-ext-v17
 setup-pre-commit-hook:
 	ln -s -f $(ROOT_PROJECT_DIR)/pre-commit.py .git/hooks/pre-commit
 
-.PHONY: lint-openapi-spec
-lint-openapi-spec:
-	# operation-2xx-response: pageserver timeline delete returns 404 on success
-	find . -iname "openapi_spec.y*ml" -exec\
-		docker run --rm -v ${PWD}:/spec ghcr.io/redocly/cli:1.34.4\
-			--skip-rule=operation-operationId --skip-rule=operation-summary --extends=minimal\
-			--skip-rule=no-server-example.com --skip-rule=operation-2xx-response\
-			lint {} \+
-
 # Targets for building PostgreSQL are defined in postgres.mk.
 #
 # But if the caller has indicated that PostgreSQL is already

clippy.toml

@@ -1,12 +1,9 @@
 disallowed-methods = [
     "tokio::task::block_in_place",
-
     # Allow this for now, to deny it later once we stop using Handle::block_on completely
     # "tokio::runtime::Handle::block_on",
-
-    # tokio-epoll-uring:
-    # - allow-invalid because the method doesn't exist on macOS
-    { path = "tokio_epoll_uring::thread_local_system", replacement = "tokio_epoll_uring_ext module inside pageserver crate", allow-invalid = true }
+    # use tokio_epoll_uring_ext instead
+    "tokio_epoll_uring::thread_local_system",
 ]
 
 disallowed-macros = [

compute/compute-node.Dockerfile

@@ -1915,10 +1915,10 @@ RUN cd /ext-src/pg_repack-src && patch -p1 </ext-src/pg_repack.patch && rm -f /e
 
 COPY --chmod=755 docker-compose/run-tests.sh /run-tests.sh
 RUN echo /usr/local/pgsql/lib > /etc/ld.so.conf.d/00-neon.conf && /sbin/ldconfig
-RUN apt-get update && apt-get install -y libtap-parser-sourcehandler-pgtap-perl jq parallel \
+RUN apt-get update && apt-get install -y libtap-parser-sourcehandler-pgtap-perl jq \
    && apt clean && rm -rf /ext-src/*.tar.gz /ext-src/*.patch /var/lib/apt/lists/*
 ENV PATH=/usr/local/pgsql/bin:$PATH
-ENV PGHOST=compute1
+ENV PGHOST=compute
 ENV PGPORT=55433
 ENV PGUSER=cloud_admin
 ENV PGDATABASE=postgres

compute_tools/Cargo.toml

@@ -66,7 +66,7 @@ url.workspace = true
 uuid.workspace = true
 walkdir.workspace = true
 x509-cert.workspace = true
-postgres-types.workspace = true
+
 postgres_versioninfo.workspace = true
 postgres_initdb.workspace = true
 compute_api.workspace = true

compute_tools/README.md

@@ -46,14 +46,11 @@ stateDiagram-v2
   Configuration --> Failed : Failed to configure the compute
   Configuration --> Running : Compute has been configured
   Empty --> Init : Compute spec is immediately available
-  Empty --> TerminationPendingFast : Requested termination
-  Empty --> TerminationPendingImmediate : Requested termination
+  Empty --> TerminationPending : Requested termination
   Init --> Failed : Failed to start Postgres
   Init --> Running : Started Postgres
-  Running --> TerminationPendingFast : Requested termination
-  Running --> TerminationPendingImmediate : Requested termination
-  TerminationPendingFast --> Terminated compute with 30s delay for cplane to inspect status
-  TerminationPendingImmediate --> Terminated : Terminated compute immediately
+  Running --> TerminationPending : Requested termination
+  TerminationPending --> Terminated : Terminated compute
   Failed --> [*] : Compute exited
   Terminated --> [*] : Compute exited
 ```

compute_tools/src/compute.rs

@@ -1,13 +1,12 @@
-use anyhow::{Context, Result, anyhow};
+use anyhow::{Context, Result};
 use chrono::{DateTime, Utc};
 use compute_api::privilege::Privilege;
 use compute_api::responses::{
     ComputeConfig, ComputeCtlConfig, ComputeMetrics, ComputeStatus, LfcOffloadState,
-    LfcPrewarmState, PromoteState, TlsConfig,
+    LfcPrewarmState, TlsConfig,
 };
 use compute_api::spec::{
-    ComputeAudit, ComputeFeature, ComputeMode, ComputeSpec, ExtVersion, PageserverConnectionInfo,
-    PageserverShardConnectionInfo, PgIdent,
+    ComputeAudit, ComputeFeature, ComputeMode, ComputeSpec, ExtVersion, PageserverProtocol, PgIdent,
 };
 use futures::StreamExt;
 use futures::future::join_all;
@@ -30,7 +29,8 @@ use std::sync::atomic::{AtomicU32, AtomicU64, Ordering};
 use std::sync::{Arc, Condvar, Mutex, RwLock};
 use std::time::{Duration, Instant};
 use std::{env, fs};
-use tokio::{spawn, sync::watch, task::JoinHandle, time};
+use tokio::task::JoinHandle;
+use tokio::{spawn, time};
 use tracing::{Instrument, debug, error, info, instrument, warn};
 use url::Url;
 use utils::id::{TenantId, TimelineId};
@@ -175,7 +175,6 @@ pub struct ComputeState {
     /// WAL flush LSN that is set after terminating Postgres and syncing safekeepers if
     /// mode == ComputeMode::Primary. None otherwise
     pub terminate_flush_lsn: Option<Lsn>,
-    pub promote_state: Option<watch::Receiver<PromoteState>>,
 
     pub metrics: ComputeMetrics,
 }
@@ -193,7 +192,6 @@ impl ComputeState {
             lfc_prewarm_state: LfcPrewarmState::default(),
             lfc_offload_state: LfcOffloadState::default(),
             terminate_flush_lsn: None,
-            promote_state: None,
         }
     }
 
@@ -226,7 +224,7 @@ pub struct ParsedSpec {
     pub spec: ComputeSpec,
     pub tenant_id: TenantId,
     pub timeline_id: TimelineId,
-    pub pageserver_conninfo: PageserverConnectionInfo,
+    pub pageserver_connstr: String,
     pub safekeeper_connstrings: Vec<String>,
     pub storage_auth_token: Option<String>,
     /// k8s dns name and port
@@ -273,27 +271,6 @@ impl ParsedSpec {
     }
 }
 
-fn extract_pageserver_conninfo_from_guc(
-    pageserver_connstring_guc: &str,
-) -> PageserverConnectionInfo {
-    PageserverConnectionInfo {
-        shards: pageserver_connstring_guc
-            .split(',')
-            .enumerate()
-            .map(|(i, connstr)| {
-                (
-                    i as u32,
-                    PageserverShardConnectionInfo {
-                        libpq_url: Some(connstr.to_string()),
-                        grpc_url: None,
-                    },
-                )
-            })
-            .collect(),
-        prefer_grpc: false,
-    }
-}
-
 impl TryFrom<ComputeSpec> for ParsedSpec {
     type Error = String;
     fn try_from(spec: ComputeSpec) -> Result<Self, String> {
@@ -303,17 +280,11 @@ impl TryFrom<ComputeSpec> for ParsedSpec {
         // For backwards-compatibility, the top-level fields in the spec file
         // may be empty. In that case, we need to dig them from the GUCs in the
         // cluster.settings field.
-        let pageserver_conninfo = match &spec.pageserver_connection_info {
-            Some(x) => x.clone(),
-            None => {
-                if let Some(guc) = spec.cluster.settings.find("neon.pageserver_connstring") {
-                    extract_pageserver_conninfo_from_guc(&guc)
-                } else {
-                    return Err("pageserver connstr should be provided".to_string());
-                }
-            }
-        };
-
+        let pageserver_connstr = spec
+            .pageserver_connstring
+            .clone()
+            .or_else(|| spec.cluster.settings.find("neon.pageserver_connstring"))
+            .ok_or("pageserver connstr should be provided")?;
         let safekeeper_connstrings = if spec.safekeeper_connstrings.is_empty() {
             if matches!(spec.mode, ComputeMode::Primary) {
                 spec.cluster
@@ -363,7 +334,7 @@ impl TryFrom<ComputeSpec> for ParsedSpec {
 
         let res = ParsedSpec {
             spec,
-            pageserver_conninfo,
+            pageserver_connstr,
             safekeeper_connstrings,
             storage_auth_token,
             tenant_id,
@@ -453,7 +424,7 @@ impl ComputeNode {
 
         let mut new_state = ComputeState::new();
         if let Some(spec) = config.spec {
-            let pspec = ParsedSpec::try_from(spec).map_err(|msg| anyhow!(msg))?;
+            let pspec = ParsedSpec::try_from(spec).map_err(|msg| anyhow::anyhow!(msg))?;
             new_state.pspec = Some(pspec);
         }
 
@@ -984,20 +955,14 @@ impl ComputeNode {
             None
         };
 
+        let mut delay_exit = false;
         let mut state = self.state.lock().unwrap();
         state.terminate_flush_lsn = lsn;
-
-        let delay_exit = state.status == ComputeStatus::TerminationPendingFast;
-        if state.status == ComputeStatus::TerminationPendingFast
-            || state.status == ComputeStatus::TerminationPendingImmediate
-        {
-            info!(
-                "Changing compute status from {} to {}",
-                state.status,
-                ComputeStatus::Terminated
-            );
+        if let ComputeStatus::TerminationPending { mode } = state.status {
             state.status = ComputeStatus::Terminated;
             self.state_changed.notify_all();
+            // we were asked to terminate gracefully, don't exit to avoid restart
+            delay_exit = mode == compute_api::responses::TerminateMode::Fast
         }
         drop(state);
 
@@ -1060,11 +1025,12 @@ impl ComputeNode {
     fn try_get_basebackup(&self, compute_state: &ComputeState, lsn: Lsn) -> Result<()> {
         let spec = compute_state.pspec.as_ref().expect("spec must be set");
 
+        let shard0_connstr = spec.pageserver_connstr.split(',').next().unwrap();
         let started = Instant::now();
-        let (connected, size) = if spec.pageserver_conninfo.prefer_grpc {
-            self.try_get_basebackup_grpc(spec, lsn)?
-        } else {
-            self.try_get_basebackup_libpq(spec, lsn)?
+
+        let (connected, size) = match PageserverProtocol::from_connstring(shard0_connstr)? {
+            PageserverProtocol::Libpq => self.try_get_basebackup_libpq(spec, lsn)?,
+            PageserverProtocol::Grpc => self.try_get_basebackup_grpc(spec, lsn)?,
         };
 
         let mut state = self.state.lock().unwrap();
@@ -1079,21 +1045,20 @@ impl ComputeNode {
     /// Fetches a basebackup via gRPC. The connstring must use grpc://. Returns the timestamp when
     /// the connection was established, and the (compressed) size of the basebackup.
     fn try_get_basebackup_grpc(&self, spec: &ParsedSpec, lsn: Lsn) -> Result<(Instant, usize)> {
-        let shard0 = spec
-            .pageserver_conninfo
-            .shards
-            .get(&0)
-            .expect("shard 0 connection info missing");
-        let shard0_url = shard0.grpc_url.clone().expect("no grpc_url for shard 0");
-
-        let shard_index = match spec.pageserver_conninfo.shards.len() as u8 {
+        let shard0_connstr = spec
+            .pageserver_connstr
+            .split(',')
+            .next()
+            .unwrap()
+            .to_string();
+        let shard_index = match spec.pageserver_connstr.split(',').count() as u8 {
             0 | 1 => ShardIndex::unsharded(),
             count => ShardIndex::new(ShardNumber(0), ShardCount(count)),
         };
 
         let (reader, connected) = tokio::runtime::Handle::current().block_on(async move {
-            let mut client = page_api::Client::connect(
-                shard0_url,
+            let mut client = page_api::Client::new(
+                shard0_connstr,
                 spec.tenant_id,
                 spec.timeline_id,
                 shard_index,
@@ -1128,13 +1093,8 @@ impl ComputeNode {
     /// Fetches a basebackup via libpq. The connstring must use postgresql://. Returns the timestamp
     /// when the connection was established, and the (compressed) size of the basebackup.
     fn try_get_basebackup_libpq(&self, spec: &ParsedSpec, lsn: Lsn) -> Result<(Instant, usize)> {
-        let shard0 = spec
-            .pageserver_conninfo
-            .shards
-            .get(&0)
-            .expect("shard 0 connection info missing");
-        let shard0_connstr = shard0.libpq_url.clone().expect("no libpq_url for shard 0");
-        let mut config = postgres::Config::from_str(&shard0_connstr)?;
+        let shard0_connstr = spec.pageserver_connstr.split(',').next().unwrap();
+        let mut config = postgres::Config::from_str(shard0_connstr)?;
 
         // Use the storage auth token from the config file, if given.
         // Note: this overrides any password set in the connection string.
@@ -1220,7 +1180,10 @@ impl ComputeNode {
                     return result;
                 }
                 Err(ref e) if attempts < max_attempts => {
-                    warn!("Failed to get basebackup: {e:?} (attempt {attempts}/{max_attempts})");
+                    warn!(
+                        "Failed to get basebackup: {} (attempt {}/{})",
+                        e, attempts, max_attempts
+                    );
                     std::thread::sleep(std::time::Duration::from_millis(retry_period_ms as u64));
                     retry_period_ms *= 1.5;
                 }
@@ -1429,8 +1392,16 @@ impl ComputeNode {
             }
         };
 
-        self.get_basebackup(compute_state, lsn)
-            .with_context(|| format!("failed to get basebackup@{lsn}"))?;
+        info!(
+            "getting basebackup@{} from pageserver {}",
+            lsn, &pspec.pageserver_connstr
+        );
+        self.get_basebackup(compute_state, lsn).with_context(|| {
+            format!(
+                "failed to get basebackup@{} from pageserver {}",
+                lsn, &pspec.pageserver_connstr
+            )
+        })?;
 
         // Update pg_hba.conf received with basebackup.
         update_pg_hba(pgdata_path)?;
@@ -1833,8 +1804,6 @@ impl ComputeNode {
             tls_config,
         )?;
 
-        self.pg_reload_conf()?;
-
         if !spec.skip_pg_catalog_updates {
             let max_concurrent_connections = spec.reconfigure_concurrency;
             // Temporarily reset max_cluster_size in config
@@ -1854,9 +1823,10 @@ impl ComputeNode {
 
                 Ok(())
             })?;
-            self.pg_reload_conf()?;
         }
 
+        self.pg_reload_conf()?;
+
         let unknown_op = "unknown".to_string();
         let op_id = spec.operation_uuid.as_ref().unwrap_or(&unknown_op);
         info!(
@@ -1929,8 +1899,7 @@ impl ComputeNode {
 
                             // exit loop
                             ComputeStatus::Failed
-                            | ComputeStatus::TerminationPendingFast
-                            | ComputeStatus::TerminationPendingImmediate
+                            | ComputeStatus::TerminationPending { .. }
                             | ComputeStatus::Terminated => break 'cert_update,
 
                             // wait
@@ -2096,7 +2065,7 @@ LIMIT 100",
             self.params
                 .remote_ext_base_url
                 .as_ref()
-                .ok_or(DownloadError::BadInput(anyhow!(
+                .ok_or(DownloadError::BadInput(anyhow::anyhow!(
                     "Remote extensions storage is not configured",
                 )))?;
 
@@ -2292,7 +2261,7 @@ LIMIT 100",
         let remote_extensions = spec
             .remote_extensions
             .as_ref()
-            .ok_or(anyhow!("Remote extensions are not configured"))?;
+            .ok_or(anyhow::anyhow!("Remote extensions are not configured"))?;
 
         info!("parse shared_preload_libraries from spec.cluster.settings");
         let mut libs_vec = Vec::new();
@@ -2371,22 +2340,22 @@ LIMIT 100",
     /// The operation will time out after a specified duration.
     pub fn wait_timeout_while_pageserver_connstr_unchanged(&self, duration: Duration) {
         let state = self.state.lock().unwrap();
-        let old_pageserver_conninfo = state
+        let old_pageserver_connstr = state
             .pspec
             .as_ref()
             .expect("spec must be set")
-            .pageserver_conninfo
+            .pageserver_connstr
             .clone();
         let mut unchanged = true;
         let _ = self
             .state_changed
             .wait_timeout_while(state, duration, |s| {
-                let pageserver_conninfo = &s
+                let pageserver_connstr = &s
                     .pspec
                     .as_ref()
                     .expect("spec must be set")
-                    .pageserver_conninfo;
-                unchanged = pageserver_conninfo == &old_pageserver_conninfo;
+                    .pageserver_connstr;
+                unchanged = pageserver_connstr == &old_pageserver_connstr;
                 unchanged
             })
             .unwrap();
@@ -2464,11 +2433,19 @@ LIMIT 100",
         // If the value is -1, we never suspend so set the value to default collection.
         // If the value is 0, it means default, we will just continue to use the default.
         if spec.suspend_timeout_seconds == -1 || spec.suspend_timeout_seconds == 0 {
+            info!(
+                "[NEON_EXT_INT_UPD] Spec Timeout: {}, New Timeout: {}",
+                spec.suspend_timeout_seconds, DEFAULT_INSTALLED_EXTENSIONS_COLLECTION_INTERVAL
+            );
             self.params.installed_extensions_collection_interval.store(
                 DEFAULT_INSTALLED_EXTENSIONS_COLLECTION_INTERVAL,
                 std::sync::atomic::Ordering::SeqCst,
             );
         } else {
+            info!(
+                "[NEON_EXT_INT_UPD] Spec Timeout: {}",
+                spec.suspend_timeout_seconds
+            );
             self.params.installed_extensions_collection_interval.store(
                 spec.suspend_timeout_seconds as u64,
                 std::sync::atomic::Ordering::SeqCst,

compute_tools/src/compute_prewarm.rs

@@ -70,7 +70,7 @@ impl ComputeNode {
             }
         };
         let row = match client
-            .query_one("select * from neon.get_prewarm_info()", &[])
+            .query_one("select * from get_prewarm_info()", &[])
             .await
         {
             Ok(row) => row,
@@ -105,8 +105,7 @@ impl ComputeNode {
                 cloned.state.lock().unwrap().lfc_prewarm_state = LfcPrewarmState::Completed;
                 return;
             };
-            crate::metrics::LFC_PREWARM_ERRORS.inc();
-            error!(%err, "prewarming lfc");
+            error!(%err);
             cloned.state.lock().unwrap().lfc_prewarm_state = LfcPrewarmState::Failed {
                 error: err.to_string(),
             };
@@ -146,7 +145,7 @@ impl ComputeNode {
         ComputeNode::get_maintenance_client(&self.tokio_conn_conf)
             .await
             .context("connecting to postgres")?
-            .query_one("select neon.prewarm_local_cache($1)", &[&uncompressed])
+            .query_one("select prewarm_local_cache($1)", &[&uncompressed])
             .await
             .context("loading LFC state into postgres")
             .map(|_| ())
@@ -181,8 +180,7 @@ impl ComputeNode {
             self.state.lock().unwrap().lfc_offload_state = LfcOffloadState::Completed;
             return;
         };
-        crate::metrics::LFC_OFFLOAD_ERRORS.inc();
-        error!(%err, "offloading lfc");
+        error!(%err);
         self.state.lock().unwrap().lfc_offload_state = LfcOffloadState::Failed {
             error: err.to_string(),
         };
@@ -196,7 +194,7 @@ impl ComputeNode {
         ComputeNode::get_maintenance_client(&self.tokio_conn_conf)
             .await
             .context("connecting to postgres")?
-            .query_one("select neon.get_local_cache_state()", &[])
+            .query_one("select get_local_cache_state()", &[])
             .await
             .context("querying LFC state")?
             .try_get::<usize, &[u8]>(0)

compute_tools/src/compute_promote.rs

@@ -1,132 +0,0 @@
-use crate::compute::ComputeNode;
-use anyhow::{Context, Result, bail};
-use compute_api::{
-    responses::{LfcPrewarmState, PromoteState, SafekeepersLsn},
-    spec::ComputeMode,
-};
-use std::{sync::Arc, time::Duration};
-use tokio::time::sleep;
-use utils::lsn::Lsn;
-
-impl ComputeNode {
-    /// Returns only when promote fails or succeeds. If a network error occurs
-    /// and http client disconnects, this does not stop promotion, and subsequent
-    /// calls block until promote finishes.
-    /// Called by control plane on secondary after primary endpoint is terminated
-    pub async fn promote(self: &Arc<Self>, safekeepers_lsn: SafekeepersLsn) -> PromoteState {
-        let cloned = self.clone();
-        let start_promotion = || {
-            let (tx, rx) = tokio::sync::watch::channel(PromoteState::NotPromoted);
-            tokio::spawn(async move {
-                tx.send(match cloned.promote_impl(safekeepers_lsn).await {
-                    Ok(_) => PromoteState::Completed,
-                    Err(err) => {
-                        tracing::error!(%err, "promoting");
-                        PromoteState::Failed {
-                            error: err.to_string(),
-                        }
-                    }
-                })
-            });
-            rx
-        };
-
-        let mut task;
-        // self.state is unlocked after block ends so we lock it in promote_impl
-        // and task.changed() is reached
-        {
-            task = self
-                .state
-                .lock()
-                .unwrap()
-                .promote_state
-                .get_or_insert_with(start_promotion)
-                .clone()
-        }
-        task.changed().await.expect("promote sender dropped");
-        task.borrow().clone()
-    }
-
-    // Why do we have to supply safekeepers?
-    // For secondary we use primary_connection_conninfo so safekeepers field is empty
-    async fn promote_impl(&self, safekeepers_lsn: SafekeepersLsn) -> Result<()> {
-        {
-            let state = self.state.lock().unwrap();
-            let mode = &state.pspec.as_ref().unwrap().spec.mode;
-            if *mode != ComputeMode::Replica {
-                bail!("{} is not replica", mode.to_type_str());
-            }
-
-            // we don't need to query Postgres so not self.lfc_prewarm_state()
-            match &state.lfc_prewarm_state {
-                LfcPrewarmState::NotPrewarmed | LfcPrewarmState::Prewarming => {
-                    bail!("prewarm not requested or pending")
-                }
-                LfcPrewarmState::Failed { error } => {
-                    tracing::warn!(%error, "replica prewarm failed")
-                }
-                _ => {}
-            }
-        }
-
-        let client = ComputeNode::get_maintenance_client(&self.tokio_conn_conf)
-            .await
-            .context("connecting to postgres")?;
-
-        let primary_lsn = safekeepers_lsn.wal_flush_lsn;
-        let mut last_wal_replay_lsn: Lsn = Lsn::INVALID;
-        const RETRIES: i32 = 20;
-        for i in 0..=RETRIES {
-            let row = client
-                .query_one("SELECT pg_last_wal_replay_lsn()", &[])
-                .await
-                .context("getting last replay lsn")?;
-            let lsn: u64 = row.get::<usize, postgres_types::PgLsn>(0).into();
-            last_wal_replay_lsn = lsn.into();
-            if last_wal_replay_lsn >= primary_lsn {
-                break;
-            }
-            tracing::info!("Try {i}, replica lsn {last_wal_replay_lsn}, primary lsn {primary_lsn}");
-            sleep(Duration::from_secs(1)).await;
-        }
-        if last_wal_replay_lsn < primary_lsn {
-            bail!("didn't catch up with primary in {RETRIES} retries");
-        }
-
-        // using $1 doesn't work with ALTER SYSTEM SET
-        let safekeepers_sql = format!(
-            "ALTER SYSTEM SET neon.safekeepers='{}'",
-            safekeepers_lsn.safekeepers
-        );
-        client
-            .query(&safekeepers_sql, &[])
-            .await
-            .context("setting safekeepers")?;
-        client
-            .query("SELECT pg_reload_conf()", &[])
-            .await
-            .context("reloading postgres config")?;
-        let row = client
-            .query_one("SELECT * FROM pg_promote()", &[])
-            .await
-            .context("pg_promote")?;
-        if !row.get::<usize, bool>(0) {
-            bail!("pg_promote() returned false");
-        }
-
-        let client = ComputeNode::get_maintenance_client(&self.tokio_conn_conf)
-            .await
-            .context("connecting to postgres")?;
-        let row = client
-            .query_one("SHOW transaction_read_only", &[])
-            .await
-            .context("getting transaction_read_only")?;
-        if row.get::<usize, &str>(0) == "on" {
-            bail!("replica in read only mode after promotion");
-        }
-
-        let mut state = self.state.lock().unwrap();
-        state.pspec.as_mut().unwrap().spec.mode = ComputeMode::Primary;
-        Ok(())
-    }
-}

compute_tools/src/config.rs

@@ -56,51 +56,9 @@ pub fn write_postgres_conf(
 
     // Add options for connecting to storage
     writeln!(file, "# Neon storage settings")?;
-
-    if let Some(conninfo) = &spec.pageserver_connection_info {
-        let mut libpq_urls: Option<Vec<String>> = Some(Vec::new());
-        let mut grpc_urls: Option<Vec<String>> = Some(Vec::new());
-
-        for shardno in 0..conninfo.shards.len() {
-            let info = conninfo.shards.get(&(shardno as u32)).ok_or_else(|| {
-                anyhow::anyhow!("shard {shardno} missing from pageserver_connection_info shard map")
-            })?;
-
-            if let Some(url) = &info.libpq_url {
-                if let Some(ref mut urls) = libpq_urls {
-                    urls.push(url.clone());
-                }
-            } else {
-                libpq_urls = None
-            }
-            if let Some(url) = &info.grpc_url {
-                if let Some(ref mut urls) = grpc_urls {
-                    urls.push(url.clone());
-                }
-            } else {
-                grpc_urls = None
-            }
-        }
-        if let Some(libpq_urls) = libpq_urls {
-            writeln!(
-                file,
-                "neon.pageserver_connstring={}",
-                escape_conf_value(&libpq_urls.join(","))
-            )?;
-        } else {
-            writeln!(file, "# no neon.pageserver_connstring")?;
-        }
-        if let Some(grpc_urls) = grpc_urls {
-            writeln!(
-                file,
-                "neon.pageserver_grpc_urls={}",
-                escape_conf_value(&grpc_urls.join(","))
-            )?;
-        } else {
-            writeln!(file, "# no neon.pageserver_grpc_urls")?;
-        }
+    if let Some(s) = &spec.pageserver_connstring {
+        writeln!(file, "neon.pageserver_connstring={}", escape_conf_value(s))?;
     }
-
     if let Some(stripe_size) = spec.shard_stripe_size {
         writeln!(file, "neon.stripe_size={stripe_size}")?;
     }

compute_tools/src/http/openapi_spec.yaml

@@ -83,87 +83,6 @@ paths:
               schema:
                 $ref: "#/components/schemas/DbsAndRoles"
 
-  /promote:
-    post:
-      tags:
-        - Promotion
-      summary: Promote secondary replica to primary
-      description: ""
-      operationId: promoteReplica
-      requestBody:
-        description: Promote requests data
-        required: true
-        content:
-          application/json:
-            schema:
-                $ref: "#/components/schemas/SafekeepersLsn"
-      responses:
-        200:
-          description: Promote succeeded or wasn't started
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/PromoteState"
-        500:
-          description: Promote failed
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/PromoteState"
-
-  /lfc/prewarm:
-    post:
-      summary: Request LFC Prewarm
-      parameters:
-        - name: from_endpoint
-          in: query
-          schema:
-            type: string
-      description: ""
-      operationId: lfcPrewarm
-      responses:
-        202:
-          description: LFC prewarm started
-        429:
-          description: LFC prewarm ongoing
-    get:
-      tags:
-        - Prewarm
-      summary: Get LFC prewarm state
-      description: ""
-      operationId: getLfcPrewarmState
-      responses:
-        200:
-          description: Prewarm state
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/LfcPrewarmState"
-
-  /lfc/offload:
-    post:
-      summary: Request LFC offload
-      description: ""
-      operationId: lfcOffload
-      responses:
-        202:
-          description: LFC offload started
-        429:
-          description: LFC offload ongoing
-    get:
-      tags:
-        - Prewarm
-      summary: Get LFC offloading state
-      description: ""
-      operationId: getLfcOffloadState
-      responses:
-        200:
-          description: Offload state
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/LfcOffloadState"
-
   /database_schema:
     get:
       tags:
@@ -371,28 +290,9 @@ paths:
       summary: Terminate Postgres and wait for it to exit
       description: ""
       operationId: terminate
-      parameters:
-        - name: mode
-          in: query
-          description: "Terminate mode: fast (wait 30s before returning) and immediate"
-          required: false
-          schema:
-            type: string
-            enum: ["fast", "immediate"]
-            default: fast
       responses:
         200:
           description: Result
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/TerminateResponse"
-        201:
-          description: Result if compute is already terminated
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/TerminateResponse"
         412:
           description: "wrong state"
           content:
@@ -435,6 +335,15 @@ components:
         total_startup_ms:
           type: integer
 
+    Info:
+      type: object
+      description: Information about VM/Pod.
+      required:
+        - num_cpus
+      properties:
+        num_cpus:
+          type: integer
+
     DbsAndRoles:
       type: object
       description: Databases and Roles
@@ -549,14 +458,11 @@ components:
       type: string
       enum:
         - empty
-        - configuration_pending
         - init
-        - running
-        - configuration
         - failed
-        - termination_pending_fast
-        - termination_pending_immediate
-        - terminated
+        - running
+        - configuration_pending
+        - configuration
       example: running
 
     ExtensionInstallRequest:
@@ -591,69 +497,25 @@ components:
           type: string
           example: "1.0.0"
 
-    SafekeepersLsn:
+    InstalledExtensions:
       type: object
-      required:
-        - safekeepers
-        - wal_flush_lsn
       properties:
-        safekeepers:
-          description: Primary replica safekeepers
-          type: string
-        wal_flush_lsn:
-          description: Primary last WAL flush LSN
-          type: string
-
-    LfcPrewarmState:
-      type: object
-      required:
-        - status
-        - total
-        - prewarmed
-        - skipped
-      properties:
-        status:
-          description: Lfc prewarm status
-          enum: [not_prewarmed, prewarming, completed, failed]
-          type: string
-        error:
-          description: Lfc prewarm error, if any
-          type: string
-        total:
-          description: Total pages processed
-          type: integer
-        prewarmed:
-          description: Total pages prewarmed
-          type: integer
-        skipped:
-          description: Pages processed but not prewarmed
-          type: integer
-
-    LfcOffloadState:
-      type: object
-      required:
-        - status
-      properties:
-        status:
-          description: Lfc offload status
-          enum: [not_offloaded, offloading, completed, failed]
-          type: string
-        error:
-          description: Lfc offload error, if any
-          type: string
-
-    PromoteState:
-      type: object
-      required:
-        - status
-      properties:
-        status:
-          description: Promote result
-          enum: [not_promoted, completed, failed]
-          type: string
-        error:
-          description: Promote error, if any
-          type: string
+        extensions:
+          description: Contains list of installed extensions.
+          type: array
+          items:
+            type: object
+            properties:
+              extname:
+                type: string
+              version:
+                type: string
+                items:
+                  type: string
+              n_databases:
+                type: integer
+              owned_by_superuser:
+                type: integer
 
     SetRoleGrantsRequest:
       type: object
@@ -682,17 +544,6 @@ components:
           description: Role name.
           example: "neon"
 
-    TerminateResponse:
-      type: object
-      required:
-        - lsn
-      properties:
-        lsn:
-          type: string
-          nullable: true
-          description: "last WAL flush LSN"
-          example: "0/028F10D8"
-
     SetRoleGrantsResponse:
       type: object
       required:

compute_tools/src/http/routes/mod.rs

@@ -14,7 +14,6 @@ pub(in crate::http) mod insights;
 pub(in crate::http) mod lfc;
 pub(in crate::http) mod metrics;
 pub(in crate::http) mod metrics_json;
-pub(in crate::http) mod promote;
 pub(in crate::http) mod status;
 pub(in crate::http) mod terminate;
 

compute_tools/src/http/routes/promote.rs

@@ -1,14 +0,0 @@
-use crate::http::JsonResponse;
-use axum::Form;
-use http::StatusCode;
-
-pub(in crate::http) async fn promote(
-    compute: axum::extract::State<std::sync::Arc<crate::compute::ComputeNode>>,
-    Form(safekeepers_lsn): Form<compute_api::responses::SafekeepersLsn>,
-) -> axum::response::Response {
-    let state = compute.promote(safekeepers_lsn).await;
-    if let compute_api::responses::PromoteState::Failed { error } = state {
-        return JsonResponse::error(StatusCode::INTERNAL_SERVER_ERROR, error);
-    }
-    JsonResponse::success(StatusCode::OK, state)
-}

compute_tools/src/http/routes/terminate.rs

@@ -3,7 +3,7 @@ use crate::http::JsonResponse;
 use axum::extract::State;
 use axum::response::Response;
 use axum_extra::extract::OptionalQuery;
-use compute_api::responses::{ComputeStatus, TerminateMode, TerminateResponse};
+use compute_api::responses::{ComputeStatus, TerminateResponse};
 use http::StatusCode;
 use serde::Deserialize;
 use std::sync::Arc;
@@ -12,7 +12,7 @@ use tracing::info;
 
 #[derive(Deserialize, Default)]
 pub struct TerminateQuery {
-    mode: TerminateMode,
+    mode: compute_api::responses::TerminateMode,
 }
 
 /// Terminate the compute.
@@ -24,16 +24,16 @@ pub(in crate::http) async fn terminate(
     {
         let mut state = compute.state.lock().unwrap();
         if state.status == ComputeStatus::Terminated {
-            let response = TerminateResponse {
-                lsn: state.terminate_flush_lsn,
-            };
-            return JsonResponse::success(StatusCode::CREATED, response);
+            return JsonResponse::success(StatusCode::CREATED, state.terminate_flush_lsn);
         }
 
         if !matches!(state.status, ComputeStatus::Empty | ComputeStatus::Running) {
             return JsonResponse::invalid_status(state.status);
         }
-        state.set_status(mode.into(), &compute.state_changed);
+        state.set_status(
+            ComputeStatus::TerminationPending { mode },
+            &compute.state_changed,
+        );
     }
 
     forward_termination_signal(false);

compute_tools/src/http/server.rs

@@ -23,7 +23,7 @@ use super::{
     middleware::authorize::Authorize,
     routes::{
         check_writability, configure, database_schema, dbs_and_roles, extension_server, extensions,
-        grants, insights, lfc, metrics, metrics_json, promote, status, terminate,
+        grants, insights, lfc, metrics, metrics_json, status, terminate,
     },
 };
 use crate::compute::ComputeNode;
@@ -87,7 +87,6 @@ impl From<&Server> for Router<Arc<ComputeNode>> {
                 let authenticated_router = Router::<Arc<ComputeNode>>::new()
                     .route("/lfc/prewarm", get(lfc::prewarm_state).post(lfc::prewarm))
                     .route("/lfc/offload", get(lfc::offload_state).post(lfc::offload))
-                    .route("/promote", post(promote::promote))
                     .route("/check_writability", post(check_writability::is_writable))
                     .route("/configure", post(configure::configure))
                     .route("/database_schema", get(database_schema::get_schema_dump))

compute_tools/src/lib.rs

@@ -12,7 +12,6 @@ pub mod logger;
 pub mod catalog;
 pub mod compute;
 pub mod compute_prewarm;
-pub mod compute_promote;
 pub mod disk_quota;
 pub mod extension_server;
 pub mod installed_extensions;

compute_tools/src/lsn_lease.rs

@@ -4,7 +4,8 @@ use std::thread;
 use std::time::{Duration, SystemTime};
 
 use anyhow::{Result, bail};
-use compute_api::spec::{ComputeMode, PageserverConnectionInfo};
+use compute_api::spec::{ComputeMode, PageserverProtocol};
+use itertools::Itertools as _;
 use pageserver_page_api as page_api;
 use postgres::{NoTls, SimpleQueryMessage};
 use tracing::{info, warn};
@@ -77,16 +78,17 @@ fn acquire_lsn_lease_with_retry(
 
     loop {
         // Note: List of pageservers is dynamic, need to re-read configs before each attempt.
-        let (conninfo, auth) = {
+        let (connstrings, auth) = {
             let state = compute.state.lock().unwrap();
             let spec = state.pspec.as_ref().expect("spec must be set");
             (
-                spec.pageserver_conninfo.clone(),
+                spec.pageserver_connstr.clone(),
                 spec.storage_auth_token.clone(),
             )
         };
 
-        let result = try_acquire_lsn_lease(conninfo, auth.as_deref(), tenant_id, timeline_id, lsn);
+        let result =
+            try_acquire_lsn_lease(&connstrings, auth.as_deref(), tenant_id, timeline_id, lsn);
         match result {
             Ok(Some(res)) => {
                 return Ok(res);
@@ -110,16 +112,17 @@ fn acquire_lsn_lease_with_retry(
 
 /// Tries to acquire LSN leases on all Pageserver shards.
 fn try_acquire_lsn_lease(
-    conninfo: PageserverConnectionInfo,
+    connstrings: &str,
     auth: Option<&str>,
     tenant_id: TenantId,
     timeline_id: TimelineId,
     lsn: Lsn,
 ) -> Result<Option<SystemTime>> {
-    let shard_count = conninfo.shards.len();
+    let connstrings = connstrings.split(',').collect_vec();
+    let shard_count = connstrings.len();
     let mut leases = Vec::new();
 
-    for (shard_number, shard) in conninfo.shards.into_iter() {
+    for (shard_number, &connstring) in connstrings.iter().enumerate() {
         let tenant_shard_id = match shard_count {
             0 | 1 => TenantShardId::unsharded(tenant_id),
             shard_count => TenantShardId {
@@ -129,22 +132,13 @@ fn try_acquire_lsn_lease(
             },
         };
 
-        let lease = if conninfo.prefer_grpc {
-            acquire_lsn_lease_grpc(
-                &shard.grpc_url.unwrap(),
-                auth,
-                tenant_shard_id,
-                timeline_id,
-                lsn,
-            )?
-        } else {
-            acquire_lsn_lease_libpq(
-                &shard.libpq_url.unwrap(),
-                auth,
-                tenant_shard_id,
-                timeline_id,
-                lsn,
-            )?
+        let lease = match PageserverProtocol::from_connstring(connstring)? {
+            PageserverProtocol::Libpq => {
+                acquire_lsn_lease_libpq(connstring, auth, tenant_shard_id, timeline_id, lsn)?
+            }
+            PageserverProtocol::Grpc => {
+                acquire_lsn_lease_grpc(connstring, auth, tenant_shard_id, timeline_id, lsn)?
+            }
         };
         leases.push(lease);
     }
@@ -198,7 +192,7 @@ fn acquire_lsn_lease_grpc(
     lsn: Lsn,
 ) -> Result<Option<SystemTime>> {
     tokio::runtime::Handle::current().block_on(async move {
-        let mut client = page_api::Client::connect(
+        let mut client = page_api::Client::new(
             connstring.to_string(),
             tenant_shard_id.tenant_id,
             timeline_id,

compute_tools/src/metrics.rs

@@ -105,14 +105,6 @@ pub(crate) static LFC_PREWARMS: Lazy<IntCounter> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
-pub(crate) static LFC_PREWARM_ERRORS: Lazy<IntCounter> = Lazy::new(|| {
-    register_int_counter!(
-        "compute_ctl_lfc_prewarm_errors_total",
-        "Total number of LFC prewarm errors",
-    )
-    .expect("failed to define a metric")
-});
-
 pub(crate) static LFC_OFFLOADS: Lazy<IntCounter> = Lazy::new(|| {
     register_int_counter!(
         "compute_ctl_lfc_offloads_total",
@@ -121,14 +113,6 @@ pub(crate) static LFC_OFFLOADS: Lazy<IntCounter> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
-pub(crate) static LFC_OFFLOAD_ERRORS: Lazy<IntCounter> = Lazy::new(|| {
-    register_int_counter!(
-        "compute_ctl_lfc_offload_errors_total",
-        "Total number of LFC offload errors",
-    )
-    .expect("failed to define a metric")
-});
-
 pub fn collect() -> Vec<MetricFamily> {
     let mut metrics = COMPUTE_CTL_UP.collect();
     metrics.extend(INSTALLED_EXTENSIONS.collect());
@@ -139,8 +123,6 @@ pub fn collect() -> Vec<MetricFamily> {
     metrics.extend(PG_CURR_DOWNTIME_MS.collect());
     metrics.extend(PG_TOTAL_DOWNTIME_MS.collect());
     metrics.extend(LFC_PREWARMS.collect());
-    metrics.extend(LFC_PREWARM_ERRORS.collect());
     metrics.extend(LFC_OFFLOADS.collect());
-    metrics.extend(LFC_OFFLOAD_ERRORS.collect());
     metrics
 }

compute_tools/src/migrations/0002-alter_roles.sql

@@ -1,16 +1,3 @@
--- On December 8th, 2023, an engineering escalation (INC-110) was opened after
--- it was found that BYPASSRLS was being applied to all roles.
---
--- PR that introduced the issue: https://github.com/neondatabase/neon/pull/5657
--- Subsequent commit on main: https://github.com/neondatabase/neon/commit/ad99fa5f0393e2679e5323df653c508ffa0ac072
---
--- NOBYPASSRLS and INHERIT are the defaults for a Postgres role, but because it
--- isn't easy to know if a Postgres cluster is affected by the issue, we need to
--- keep the migration around for a long time, if not indefinitely, so any
--- cluster can be fixed.
---
--- Branching is the gift that keeps on giving...
-
 DO $$
 DECLARE
     role_name text;

compute_tools/src/migrations/0012-grant_pg_signal_backend_to_neon_superuser.sql

@@ -1 +0,0 @@
-GRANT pg_signal_backend TO neon_superuser WITH ADMIN OPTION;

compute_tools/src/migrations/tests/0004-grant_pg_monitor_to_neon_superuser.sql

@@ -7,17 +7,13 @@ BEGIN
         INTO monitor
         FROM pg_auth_members
         WHERE roleid = 'pg_monitor'::regrole
-            AND member = 'neon_superuser'::regrole;
+            AND member = 'pg_monitor'::regrole;
 
-    IF monitor IS NULL THEN
-        RAISE EXCEPTION 'no entry in pg_auth_members for neon_superuser and pg_monitor';
-    END IF;
-
-    IF monitor.admin IS NULL OR NOT monitor.member THEN
+    IF NOT monitor.member THEN
         RAISE EXCEPTION 'neon_superuser is not a member of pg_monitor';
     END IF;
 
-    IF monitor.admin IS NULL OR NOT monitor.admin THEN
+    IF NOT monitor.admin THEN
         RAISE EXCEPTION 'neon_superuser cannot grant pg_monitor';
     END IF;
 END $$;

compute_tools/src/migrations/tests/0012-grant_pg_signal_backend_to_neon_superuser.sql

@@ -1,23 +0,0 @@
-DO $$
-DECLARE
-    signal_backend record;
-BEGIN
-    SELECT pg_has_role('neon_superuser', 'pg_signal_backend', 'member') AS member,
-            admin_option AS admin
-        INTO signal_backend
-        FROM pg_auth_members
-        WHERE roleid = 'pg_signal_backend'::regrole
-            AND member = 'neon_superuser'::regrole;
-
-    IF signal_backend IS NULL THEN
-        RAISE EXCEPTION 'no entry in pg_auth_members for neon_superuser and pg_signal_backend';
-    END IF;
-
-    IF signal_backend.member IS NULL OR NOT signal_backend.member THEN
-        RAISE EXCEPTION 'neon_superuser is not a member of pg_signal_backend';
-    END IF;
-
-    IF signal_backend.admin IS NULL OR NOT signal_backend.admin THEN
-        RAISE EXCEPTION 'neon_superuser cannot grant pg_signal_backend';
-    END IF;
-END $$;

compute_tools/src/monitor.rs

@@ -84,8 +84,7 @@ impl ComputeMonitor {
         if matches!(
             compute_status,
             ComputeStatus::Terminated
-                | ComputeStatus::TerminationPendingFast
-                | ComputeStatus::TerminationPendingImmediate
+                | ComputeStatus::TerminationPending { .. }
                 | ComputeStatus::Failed
         ) {
             info!(

compute_tools/src/spec.rs

@@ -197,7 +197,6 @@ pub async fn handle_migrations(client: &mut Client) -> Result<()> {
         include_str!(
             "./migrations/0011-grant_pg_show_replication_origin_status_to_neon_superuser.sql"
         ),
-        include_str!("./migrations/0012-grant_pg_signal_backend_to_neon_superuser.sql"),
     ];
 
     MigrationRunner::new(client, &migrations)

control_plane/src/bin/neon_local.rs

@@ -16,7 +16,7 @@ use std::time::Duration;
 use anyhow::{Context, Result, anyhow, bail};
 use clap::Parser;
 use compute_api::requests::ComputeClaimsScope;
-use compute_api::spec::{ComputeMode, PageserverConnectionInfo, PageserverShardConnectionInfo};
+use compute_api::spec::{ComputeMode, PageserverProtocol};
 use control_plane::broker::StorageBroker;
 use control_plane::endpoint::{ComputeControlPlane, EndpointTerminateMode};
 use control_plane::endpoint_storage::{ENDPOINT_STORAGE_DEFAULT_ADDR, EndpointStorage};
@@ -1516,35 +1516,29 @@ async fn handle_endpoint(subcmd: &EndpointCmd, env: &local_env::LocalEnv) -> Res
                 )?;
             }
 
-            let (shards, stripe_size) = if let Some(ps_id) = pageserver_id {
-                let conf = env.get_pageserver_conf(ps_id).unwrap();
-                let libpq_url = Some({
-                    let (host, port) = parse_host_port(&conf.listen_pg_addr)?;
-                    let port = port.unwrap_or(5432);
-                    format!("postgres://no_user@{host}:{port}")
-                });
-                let grpc_url = if let Some(grpc_addr) = &conf.listen_grpc_addr {
+            let (pageservers, stripe_size) = if let Some(pageserver_id) = pageserver_id {
+                let conf = env.get_pageserver_conf(pageserver_id).unwrap();
+                // Use gRPC if requested.
+                let pageserver = if endpoint.grpc {
+                    let grpc_addr = conf.listen_grpc_addr.as_ref().expect("bad config");
                     let (host, port) = parse_host_port(grpc_addr)?;
                     let port = port.unwrap_or(DEFAULT_PAGESERVER_GRPC_PORT);
-                    Some(format!("grpc://no_user@{host}:{port}"))
+                    (PageserverProtocol::Grpc, host, port)
                 } else {
-                    None
+                    let (host, port) = parse_host_port(&conf.listen_pg_addr)?;
+                    let port = port.unwrap_or(5432);
+                    (PageserverProtocol::Libpq, host, port)
                 };
-                let pageserver = PageserverShardConnectionInfo {
-                    libpq_url,
-                    grpc_url,
-                };
-
                 // If caller is telling us what pageserver to use, this is not a tenant which is
                 // fully managed by storage controller, therefore not sharded.
-                (vec![(0, pageserver)], DEFAULT_STRIPE_SIZE)
+                (vec![pageserver], DEFAULT_STRIPE_SIZE)
             } else {
                 // Look up the currently attached location of the tenant, and its striping metadata,
                 // to pass these on to postgres.
                 let storage_controller = StorageController::from_env(env);
                 let locate_result = storage_controller.tenant_locate(endpoint.tenant_id).await?;
-                let shards = futures::future::try_join_all(locate_result.shards.into_iter().map(
-                    |shard| async move {
+                let pageservers = futures::future::try_join_all(
+                    locate_result.shards.into_iter().map(|shard| async move {
                         if let ComputeMode::Static(lsn) = endpoint.mode {
                             // Initialize LSN leases for static computes.
                             let conf = env.get_pageserver_conf(shard.node_id).unwrap();
@@ -1556,34 +1550,28 @@ async fn handle_endpoint(subcmd: &EndpointCmd, env: &local_env::LocalEnv) -> Res
                                 .await?;
                         }
 
-                        let libpq_host = Host::parse(&shard.listen_pg_addr)?;
-                        let libpq_port = shard.listen_pg_port;
-                        let libpq_url =
-                            Some(format!("postgres://no_user@{libpq_host}:{libpq_port}"));
-
-                        let grpc_url = if let Some(grpc_host) = shard.listen_grpc_addr {
-                            let grpc_port = shard.listen_grpc_port.expect("no gRPC port");
-                            Some(format!("grpc://no_user@{grpc_host}:{grpc_port}"))
+                        let pageserver = if endpoint.grpc {
+                            (
+                                PageserverProtocol::Grpc,
+                                Host::parse(&shard.listen_grpc_addr.expect("no gRPC address"))?,
+                                shard.listen_grpc_port.expect("no gRPC port"),
+                            )
                         } else {
-                            None
+                            (
+                                PageserverProtocol::Libpq,
+                                Host::parse(&shard.listen_pg_addr)?,
+                                shard.listen_pg_port,
+                            )
                         };
-                        let pageserver = PageserverShardConnectionInfo {
-                            libpq_url,
-                            grpc_url,
-                        };
-                        anyhow::Ok((shard.shard_id.shard_number.0 as u32, pageserver))
-                    },
-                ))
+                        anyhow::Ok(pageserver)
+                    }),
+                )
                 .await?;
                 let stripe_size = locate_result.shard_params.stripe_size;
 
-                (shards, stripe_size)
-            };
-            assert!(!shards.is_empty());
-            let pageserver_conninfo = PageserverConnectionInfo {
-                shards: shards.into_iter().collect(),
-                prefer_grpc: endpoint.grpc,
+                (pageservers, stripe_size)
             };
+            assert!(!pageservers.is_empty());
 
             let ps_conf = env.get_pageserver_conf(DEFAULT_PAGESERVER_ID)?;
             let auth_token = if matches!(ps_conf.pg_auth_type, AuthType::NeonJWT) {
@@ -1613,7 +1601,7 @@ async fn handle_endpoint(subcmd: &EndpointCmd, env: &local_env::LocalEnv) -> Res
                 endpoint_storage_addr,
                 safekeepers_generation,
                 safekeepers,
-                pageserver_conninfo,
+                pageservers,
                 remote_ext_base_url: remote_ext_base_url.clone(),
                 shard_stripe_size: stripe_size.0 as usize,
                 create_test_user: args.create_test_user,
@@ -1632,27 +1620,20 @@ async fn handle_endpoint(subcmd: &EndpointCmd, env: &local_env::LocalEnv) -> Res
                 .endpoints
                 .get(endpoint_id.as_str())
                 .with_context(|| format!("postgres endpoint {endpoint_id} is not found"))?;
-            let shards = if let Some(ps_id) = args.endpoint_pageserver_id {
+            let pageservers = if let Some(ps_id) = args.endpoint_pageserver_id {
                 let conf = env.get_pageserver_conf(ps_id)?;
-                let libpq_url = Some({
-                    let (host, port) = parse_host_port(&conf.listen_pg_addr)?;
-                    let port = port.unwrap_or(5432);
-                    format!("postgres://no_user@{host}:{port}")
-                });
-                let grpc_url = if let Some(grpc_addr) = &conf.listen_grpc_addr {
+                // Use gRPC if requested.
+                let pageserver = if endpoint.grpc {
+                    let grpc_addr = conf.listen_grpc_addr.as_ref().expect("bad config");
                     let (host, port) = parse_host_port(grpc_addr)?;
                     let port = port.unwrap_or(DEFAULT_PAGESERVER_GRPC_PORT);
-                    Some(format!("grpc://no_user@{host}:{port}"))
+                    (PageserverProtocol::Grpc, host, port)
                 } else {
-                    None
+                    let (host, port) = parse_host_port(&conf.listen_pg_addr)?;
+                    let port = port.unwrap_or(5432);
+                    (PageserverProtocol::Libpq, host, port)
                 };
-                let pageserver = PageserverShardConnectionInfo {
-                    libpq_url,
-                    grpc_url,
-                };
-                // If caller is telling us what pageserver to use, this is not a tenant which is
-                // fully managed by storage controller, therefore not sharded.
-                vec![(0, pageserver)]
+                vec![pageserver]
             } else {
                 let storage_controller = StorageController::from_env(env);
                 storage_controller
@@ -1662,36 +1643,28 @@ async fn handle_endpoint(subcmd: &EndpointCmd, env: &local_env::LocalEnv) -> Res
                     .into_iter()
                     .map(|shard| {
                         // Use gRPC if requested.
-                        let libpq_host = Host::parse(&shard.listen_pg_addr).expect("bad hostname");
-                        let libpq_port = shard.listen_pg_port;
-                        let libpq_url =
-                            Some(format!("postgres://no_user@{libpq_host}:{libpq_port}"));
-
-                        let grpc_url = if let Some(grpc_host) = shard.listen_grpc_addr {
-                            let grpc_port = shard.listen_grpc_port.expect("no gRPC port");
-                            Some(format!("grpc://no_user@{grpc_host}:{grpc_port}"))
+                        if endpoint.grpc {
+                            (
+                                PageserverProtocol::Grpc,
+                                Host::parse(&shard.listen_grpc_addr.expect("no gRPC address"))
+                                    .expect("bad hostname"),
+                                shard.listen_grpc_port.expect("no gRPC port"),
+                            )
                         } else {
-                            None
-                        };
-                        (
-                            shard.shard_id.shard_number.0 as u32,
-                            PageserverShardConnectionInfo {
-                                libpq_url,
-                                grpc_url,
-                            },
-                        )
+                            (
+                                PageserverProtocol::Libpq,
+                                Host::parse(&shard.listen_pg_addr).expect("bad hostname"),
+                                shard.listen_pg_port,
+                            )
+                        }
                     })
                     .collect::<Vec<_>>()
             };
-            let pageserver_conninfo = PageserverConnectionInfo {
-                shards: shards.into_iter().collect(),
-                prefer_grpc: endpoint.grpc,
-            };
             // If --safekeepers argument is given, use only the listed
             // safekeeper nodes; otherwise all from the env.
             let safekeepers = parse_safekeepers(&args.safekeepers)?;
             endpoint
-                .reconfigure(Some(pageserver_conninfo), None, safekeepers, None)
+                .reconfigure(Some(pageservers), None, safekeepers, None)
                 .await?;
         }
         EndpointCmd::Stop(args) => {

control_plane/src/endpoint.rs

@@ -56,13 +56,9 @@ use compute_api::responses::{
     TlsConfig,
 };
 use compute_api::spec::{
-    Cluster, ComputeAudit, ComputeFeature, ComputeMode, ComputeSpec, Database, PgIdent,
-    RemoteExtSpec, Role,
+    Cluster, ComputeAudit, ComputeFeature, ComputeMode, ComputeSpec, Database, PageserverProtocol,
+    PgIdent, RemoteExtSpec, Role,
 };
-
-// re-export these, because they're used in the reconfigure() function
-pub use compute_api::spec::{PageserverConnectionInfo, PageserverShardConnectionInfo};
-
 use jsonwebtoken::jwk::{
     AlgorithmParameters, CommonParameters, EllipticCurve, Jwk, JwkSet, KeyAlgorithm, KeyOperations,
     OctetKeyPairParameters, OctetKeyPairType, PublicKeyUse,
@@ -78,6 +74,7 @@ use sha2::{Digest, Sha256};
 use spki::der::Decode;
 use spki::{SubjectPublicKeyInfo, SubjectPublicKeyInfoRef};
 use tracing::debug;
+use url::Host;
 use utils::id::{NodeId, TenantId, TimelineId};
 
 use crate::local_env::LocalEnv;
@@ -382,7 +379,7 @@ pub struct EndpointStartArgs {
     pub endpoint_storage_addr: String,
     pub safekeepers_generation: Option<SafekeeperGeneration>,
     pub safekeepers: Vec<NodeId>,
-    pub pageserver_conninfo: PageserverConnectionInfo,
+    pub pageservers: Vec<(PageserverProtocol, Host, u16)>,
     pub remote_ext_base_url: Option<String>,
     pub shard_stripe_size: usize,
     pub create_test_user: bool,
@@ -656,6 +653,14 @@ impl Endpoint {
         }
     }
 
+    fn build_pageserver_connstr(pageservers: &[(PageserverProtocol, Host, u16)]) -> String {
+        pageservers
+            .iter()
+            .map(|(scheme, host, port)| format!("{scheme}://no_user@{host}:{port}"))
+            .collect::<Vec<_>>()
+            .join(",")
+    }
+
     /// Map safekeepers ids to the actual connection strings.
     fn build_safekeepers_connstrs(&self, sk_ids: Vec<NodeId>) -> Result<Vec<String>> {
         let mut safekeeper_connstrings = Vec::new();
@@ -701,6 +706,9 @@ impl Endpoint {
             std::fs::remove_dir_all(self.pgdata())?;
         }
 
+        let pageserver_connstring = Self::build_pageserver_connstr(&args.pageservers);
+        assert!(!pageserver_connstring.is_empty());
+
         let safekeeper_connstrings = self.build_safekeepers_connstrs(args.safekeepers)?;
 
         // check for file remote_extensions_spec.json
@@ -759,7 +767,7 @@ impl Endpoint {
                 branch_id: None,
                 endpoint_id: Some(self.endpoint_id.clone()),
                 mode: self.mode,
-                pageserver_connection_info: Some(args.pageserver_conninfo),
+                pageserver_connstring: Some(pageserver_connstring),
                 safekeepers_generation: args.safekeepers_generation.map(|g| g.into_inner()),
                 safekeeper_connstrings,
                 storage_auth_token: args.auth_token.clone(),
@@ -914,8 +922,7 @@ impl Endpoint {
                         ComputeStatus::Empty
                         | ComputeStatus::ConfigurationPending
                         | ComputeStatus::Configuration
-                        | ComputeStatus::TerminationPendingFast
-                        | ComputeStatus::TerminationPendingImmediate
+                        | ComputeStatus::TerminationPending { .. }
                         | ComputeStatus::Terminated => {
                             bail!("unexpected compute status: {:?}", state.status)
                         }
@@ -973,7 +980,7 @@ impl Endpoint {
 
     pub async fn reconfigure(
         &self,
-        pageserver_conninfo: Option<PageserverConnectionInfo>,
+        pageservers: Option<Vec<(PageserverProtocol, Host, u16)>>,
         stripe_size: Option<ShardStripeSize>,
         safekeepers: Option<Vec<NodeId>>,
         safekeeper_generation: Option<SafekeeperGeneration>,
@@ -989,17 +996,15 @@ impl Endpoint {
         let postgresql_conf = self.read_postgresql_conf()?;
         spec.cluster.postgresql_conf = Some(postgresql_conf);
 
-        if let Some(pageserver_conninfo) = pageserver_conninfo {
-            // If pageservers are provided, we need to ensure that they are not empty.
-            // This is a requirement for the compute_ctl configuration.
-            anyhow::ensure!(
-                !pageserver_conninfo.shards.is_empty(),
-                "no pageservers provided"
-            );
-            spec.pageserver_connection_info = Some(pageserver_conninfo);
-        }
-        if stripe_size.is_some() {
-            spec.shard_stripe_size = stripe_size.map(|s| s.0 as usize);
+        // If pageservers are not specified, don't change them.
+        if let Some(pageservers) = pageservers {
+            anyhow::ensure!(!pageservers.is_empty(), "no pageservers provided");
+
+            let pageserver_connstr = Self::build_pageserver_connstr(&pageservers);
+            spec.pageserver_connstring = Some(pageserver_connstr);
+            if stripe_size.is_some() {
+                spec.shard_stripe_size = stripe_size.map(|s| s.0 as usize);
+            }
         }
 
         // If safekeepers are not specified, don't change them.
@@ -1048,7 +1053,7 @@ impl Endpoint {
 
     pub async fn reconfigure_pageservers(
         &self,
-        pageservers: PageserverConnectionInfo,
+        pageservers: Vec<(PageserverProtocol, Host, u16)>,
         stripe_size: Option<ShardStripeSize>,
     ) -> Result<()> {
         self.reconfigure(Some(pageservers), stripe_size, None, None)

control_plane/src/pageserver.rs

@@ -452,12 +452,6 @@ impl PageServerNode {
                 .map(|x| x.parse::<usize>())
                 .transpose()
                 .context("Failed to parse 'image_creation_threshold' as non zero integer")?,
-            // HADRON
-            image_layer_force_creation_period: settings
-                .remove("image_layer_force_creation_period")
-                .map(humantime::parse_duration)
-                .transpose()
-                .context("Failed to parse 'image_layer_force_creation_period' as duration")?,
             image_layer_creation_check_threshold: settings
                 .remove("image_layer_creation_check_threshold")
                 .map(|x| x.parse::<u8>())

docker-compose/compute_wrapper/shell/compute.sh

@@ -54,16 +54,14 @@ else
     printf '%s\n' "${result}" | jq .
   fi
 
-  if [[ "${RUN_PARALLEL:-false}" != "true" ]]; then
-    echo "Check if a timeline present"
-    PARAMS=(
-         -X GET
-         -H "Content-Type: application/json"
-        "http://pageserver:9898/v1/tenant/${tenant_id}/timeline"
-    )
-    timeline_id=$(curl "${PARAMS[@]}" | jq -r .[0].timeline_id)
-  fi
-  if [[ -z "${timeline_id:-}" || "${timeline_id:-}" = null ]]; then
+  echo "Check if a timeline present"
+  PARAMS=(
+       -X GET
+       -H "Content-Type: application/json"
+       "http://pageserver:9898/v1/tenant/${tenant_id}/timeline"
+  )
+  timeline_id=$(curl "${PARAMS[@]}" | jq -r .[0].timeline_id)
+  if [[ -z "${timeline_id}" || "${timeline_id}" = null ]]; then
     generate_id timeline_id
     PARAMS=(
         -sbf

docker-compose/docker-compose.yml

@@ -142,7 +142,7 @@ services:
       - "storage_broker"
       - "--listen-addr=0.0.0.0:50051"
 
-  compute1:
+  compute:
     restart: always
     build:
       context: ./compute_wrapper/
@@ -152,7 +152,6 @@ services:
         - TAG=${COMPUTE_TAG:-${TAG:-latest}}
         - http_proxy=${http_proxy:-}
         - https_proxy=${https_proxy:-}
-    image: built-compute
     environment:
       - PG_VERSION=${PG_VERSION:-16}
       - TENANT_ID=${TENANT_ID:-}
@@ -167,11 +166,6 @@ services:
       - 3080:3080 # http endpoints
     entrypoint:
       - "/shell/compute.sh"
-    # Ad an alias for compute1 for compatibility
-    networks:
-      default:
-        aliases:
-            - compute
     depends_on:
       - safekeeper1
       - safekeeper2
@@ -180,20 +174,15 @@ services:
 
   compute_is_ready:
     image: postgres:latest
-    environment:
-      - PARALLEL_COMPUTES=1
     entrypoint:
-      - "/bin/sh"
+      - "/bin/bash"
       - "-c"
     command:
-      - "for i in $(seq 1 $${PARALLEL_COMPUTES}); do
-           until pg_isready -h compute$$i -p 55433 -U cloud_admin ; do
-             sleep 1;
-           done;
-         done;
-         echo All computes are started"
+      - "until pg_isready -h compute -p 55433 -U cloud_admin ; do
+            echo 'Waiting to start compute...' && sleep 1;
+         done"
     depends_on:
-      - compute1
+      - compute
 
   neon-test-extensions:
     profiles: ["test-extensions"]
@@ -207,4 +196,4 @@ services:
     command:
       - sleep 3600
     depends_on:
-      - compute1
+      - compute

docker-compose/docker_compose_test.sh

@@ -1,4 +1,4 @@
-#!/usr/bin/env bash
+#!/bin/bash
 
 # A basic test to ensure Docker images are built correctly.
 # Build a wrapper around the compute, start all services and runs a simple SQL query.
@@ -13,36 +13,9 @@
 #
 set -eux -o pipefail
 
-cd "$(dirname "${0}")"
 export COMPOSE_FILE='docker-compose.yml'
 export COMPOSE_PROFILES=test-extensions
-export PARALLEL_COMPUTES=${PARALLEL_COMPUTES:-1}
-READY_MESSAGE="All computes are started"
-COMPUTES=()
-for i in $(seq 1 "${PARALLEL_COMPUTES}"); do
-  COMPUTES+=("compute${i}")
-done
-CURRENT_TMPDIR=$(mktemp -d)
-trap 'rm -rf ${CURRENT_TMPDIR} docker-compose-parallel.yml' EXIT
-if [[ ${PARALLEL_COMPUTES} -gt 1 ]]; then
-  export COMPOSE_FILE=docker-compose-parallel.yml
-  cp docker-compose.yml docker-compose-parallel.yml
-  # Replace the environment variable PARALLEL_COMPUTES with the actual value
-  yq eval -i ".services.compute_is_ready.environment |=  map(select(. | test(\"^PARALLEL_COMPUTES=\") | not)) + [\"PARALLEL_COMPUTES=${PARALLEL_COMPUTES}\"]" ${COMPOSE_FILE}
-  for i in $(seq 2 "${PARALLEL_COMPUTES}"); do
-    # Duplicate compute1 as compute${i} for parallel execution
-    yq eval -i ".services.compute${i} = .services.compute1" ${COMPOSE_FILE}
-    # We don't need these sections, so delete them
-    yq eval -i "(del .services.compute${i}.build) | (del .services.compute${i}.ports) | (del .services.compute${i}.networks)" ${COMPOSE_FILE}
-    # Let the compute 1 be the only dependence
-    yq eval -i ".services.compute${i}.depends_on = [\"compute1\"]" ${COMPOSE_FILE}
-    # Set RUN_PARALLEL=true for compute2. They will generate tenant_id and timeline_id to avoid using the same as other computes
-    yq eval -i ".services.compute${i}.environment += [\"RUN_PARALLEL=true\"]" ${COMPOSE_FILE}
-    # Remove TENANT_ID and TIMELINE_ID from the environment variables of the generated computes
-    # They will create new TENANT_ID and TIMELINE_ID anyway.
-    yq eval -i ".services.compute${i}.environment |= map(select(. | (test(\"^TENANT_ID=\") or test(\"^TIMELINE_ID=\")) | not))" ${COMPOSE_FILE}
-  done
-fi
+cd "$(dirname "${0}")"
 PSQL_OPTION="-h localhost -U cloud_admin -p 55433 -d postgres"
 
 function cleanup() {
@@ -54,11 +27,11 @@ function cleanup() {
 
 for pg_version in ${TEST_VERSION_ONLY-14 15 16 17}; do
     pg_version=${pg_version/v/}
-    echo "clean up containers if exist"
+    echo "clean up containers if exists"
     cleanup
     PG_TEST_VERSION=$((pg_version < 16 ? 16 : pg_version))
-    PG_VERSION=${pg_version} PG_TEST_VERSION=${PG_TEST_VERSION} docker compose build compute1
-    PG_VERSION=${pg_version} PG_TEST_VERSION=${PG_TEST_VERSION} docker compose up --quiet-pull -d
+    PG_VERSION=${pg_version} PG_TEST_VERSION=${PG_TEST_VERSION} docker compose up --quiet-pull --build -d
+
     echo "wait until the compute is ready. timeout after 60s. "
     cnt=0
     while sleep 3; do
@@ -68,50 +41,45 @@ for pg_version in ${TEST_VERSION_ONLY-14 15 16 17}; do
             echo "timeout before the compute is ready."
             exit 1
         fi
-        if docker compose logs compute_is_ready | grep -q "${READY_MESSAGE}"; then
+        if docker compose logs "compute_is_ready" | grep -q "accepting connections"; then
             echo "OK. The compute is ready to connect."
             echo "execute simple queries."
-            for compute in "${COMPUTES[@]}"; do
-              docker compose exec "${compute}" /bin/bash -c "psql ${PSQL_OPTION} -c 'SELECT 1'"
-            done
+            docker compose exec compute /bin/bash -c "psql ${PSQL_OPTION} -c 'SELECT 1'"
             break
         fi
     done
 
     if [[ ${pg_version} -ge 16 ]]; then
-        mkdir "${CURRENT_TMPDIR}"/{pg_hint_plan-src,file_fdw,postgis-src}
-        docker compose cp neon-test-extensions:/ext-src/postgis-src/raster/test "${CURRENT_TMPDIR}/postgis-src/test"
-        docker compose cp neon-test-extensions:/ext-src/postgis-src/regress/00-regress-install "${CURRENT_TMPDIR}/postgis-src/00-regress-install"
-        docker compose cp neon-test-extensions:/ext-src/pg_hint_plan-src/data "${CURRENT_TMPDIR}/pg_hint_plan-src/data"
-        docker compose cp neon-test-extensions:/postgres/contrib/file_fdw/data "${CURRENT_TMPDIR}/file_fdw/data"
-
-        for compute in "${COMPUTES[@]}"; do
-          # This is required for the pg_hint_plan test, to prevent flaky log message causing the test to fail
-          # It cannot be moved to Dockerfile now because the database directory is created after the start of the container
-          echo Adding dummy config on "${compute}"
-          docker compose exec "${compute}" touch /var/db/postgres/compute/compute_ctl_temp_override.conf
-          # Prepare for the PostGIS test
-          docker compose exec "${compute}" mkdir -p /tmp/pgis_reg/pgis_reg_tmp /ext-src/postgis-src/raster /ext-src/postgis-src/regress /ext-src/postgis-src/regress/00-regress-install
-          docker compose cp "${CURRENT_TMPDIR}/postgis-src/test" "${compute}":/ext-src/postgis-src/raster/test
-          docker compose cp "${CURRENT_TMPDIR}/postgis-src/00-regress-install" "${compute}":/ext-src/postgis-src/regress
-          # The following block copies the files for the pg_hintplan test to the compute node for the extension test in an isolated docker-compose environment
-          docker compose cp "${CURRENT_TMPDIR}/pg_hint_plan-src/data" "${compute}":/ext-src/pg_hint_plan-src/
-          # The following block does the same for the contrib/file_fdw test
-          docker compose cp "${CURRENT_TMPDIR}/file_fdw/data" "${compute}":/postgres/contrib/file_fdw/data
-        done
+        # This is required for the pg_hint_plan test, to prevent flaky log message causing the test to fail
+        # It cannot be moved to Dockerfile now because the database directory is created after the start of the container
+        echo Adding dummy config
+        docker compose exec compute touch /var/db/postgres/compute/compute_ctl_temp_override.conf
+        # Prepare for the PostGIS test
+        docker compose exec compute mkdir -p /tmp/pgis_reg/pgis_reg_tmp
+        TMPDIR=$(mktemp -d)
+        docker compose cp neon-test-extensions:/ext-src/postgis-src/raster/test "${TMPDIR}"
+        docker compose cp neon-test-extensions:/ext-src/postgis-src/regress/00-regress-install "${TMPDIR}"
+        docker compose exec compute mkdir -p /ext-src/postgis-src/raster /ext-src/postgis-src/regress /ext-src/postgis-src/regress/00-regress-install
+        docker compose cp "${TMPDIR}/test" compute:/ext-src/postgis-src/raster/test
+        docker compose cp "${TMPDIR}/00-regress-install" compute:/ext-src/postgis-src/regress
+        rm -rf "${TMPDIR}"
+        # The following block copies the files for the pg_hintplan test to the compute node for the extension test in an isolated docker-compose environment
+        TMPDIR=$(mktemp -d)
+        docker compose cp neon-test-extensions:/ext-src/pg_hint_plan-src/data "${TMPDIR}/data"
+        docker compose cp "${TMPDIR}/data" compute:/ext-src/pg_hint_plan-src/
+        rm -rf "${TMPDIR}"
+        # The following block does the same for the contrib/file_fdw test
+        TMPDIR=$(mktemp -d)
+        docker compose cp neon-test-extensions:/postgres/contrib/file_fdw/data "${TMPDIR}/data"
+        docker compose cp "${TMPDIR}/data" compute:/postgres/contrib/file_fdw/data
+        rm -rf "${TMPDIR}"
         # Apply patches
         docker compose exec -T neon-test-extensions bash -c "(cd /postgres && patch -p1)" <"../compute/patches/contrib_pg${pg_version}.patch"
         # We are running tests now
         rm -f testout.txt testout_contrib.txt
-        # We want to run the longest tests first to better utilize parallelization and reduce overall test time.
-        # Tests listed in the RUN_FIRST variable will be run before others.
-        # If parallelization is not used, this environment variable will be ignored.
-
         docker compose exec -e USE_PGXS=1 -e SKIP=timescaledb-src,rdkit-src,pg_jsonschema-src,kq_imcx-src,wal2json_2_5-src,rag_jina_reranker_v1_tiny_en-src,rag_bge_small_en_v15-src \
-        -e RUN_FIRST=hll-src,postgis-src,pgtap-src -e PARALLEL_COMPUTES="${PARALLEL_COMPUTES}" \
         neon-test-extensions /run-tests.sh /ext-src | tee testout.txt && EXT_SUCCESS=1 || EXT_SUCCESS=0
         docker compose exec -e SKIP=start-scripts,postgres_fdw,ltree_plpython,jsonb_plpython,jsonb_plperl,hstore_plpython,hstore_plperl,dblink,bool_plperl \
-        -e PARALLEL_COMPUTES="${PARALLEL_COMPUTES}" \
         neon-test-extensions /run-tests.sh /postgres/contrib | tee testout_contrib.txt && CONTRIB_SUCCESS=1 || CONTRIB_SUCCESS=0
         if [[ ${EXT_SUCCESS} -eq 0 || ${CONTRIB_SUCCESS} -eq 0 ]]; then
             CONTRIB_FAILED=

docker-compose/run-tests.sh

@@ -1,4 +1,4 @@
-#!/usr/bin/env bash
+#!/bin/bash
 set -x
 
 if [[ -v BENCHMARK_CONNSTR ]]; then
@@ -26,9 +26,8 @@ if [[ -v BENCHMARK_CONNSTR ]]; then
   fi
 fi
 REGULAR_USER=false
-PARALLEL_COMPUTES=${PARALLEL_COMPUTES:-1}
-while getopts pr arg; do
-  case ${arg} in
+while getopts r arg; do
+  case $arg in
   r)
     REGULAR_USER=true
     shift $((OPTIND-1))
@@ -42,49 +41,26 @@ extdir=${1}
 
 cd "${extdir}" || exit 2
 FAILED=
-export FAILED_FILE=/tmp/failed
-rm -f ${FAILED_FILE}
-mapfile -t LIST < <( (echo -e "${SKIP//","/"\n"}"; ls) | sort | uniq -u)
-if [[ ${PARALLEL_COMPUTES} -gt 1 ]]; then
-  # Avoid errors if RUN_FIRST is not defined
-  RUN_FIRST=${RUN_FIRST:-}
-  # Move entries listed in the RUN_FIRST variable to the beginning
-  ORDERED_LIST=$(printf "%s\n" "${LIST[@]}" | grep -x -Ff <(echo -e "${RUN_FIRST//,/$'\n'}"); printf "%s\n" "${LIST[@]}" | grep -vx -Ff <(echo -e "${RUN_FIRST//,/$'\n'}"))
-  parallel -j"${PARALLEL_COMPUTES}" "[[ -d {} ]] || exit 0
-                export PGHOST=compute{%}
-                if ! psql -c 'select 1'>/dev/null; then
-                  exit 1
-                fi
-                echo Running on \${PGHOST}
-                if [[ -f ${extdir}/{}/neon-test.sh ]]; then
-                  echo Running from script
-                  ${extdir}/{}/neon-test.sh || echo {} >> ${FAILED_FILE};
-                else
-                  echo Running using make;
-                  USE_PGXS=1 make -C {} installcheck || echo {} >> ${FAILED_FILE};
-                fi" ::: ${ORDERED_LIST}
-  [[ ! -f ${FAILED_FILE} ]] && exit 0
-else
-  for d in "${LIST[@]}"; do
-      [ -d "${d}" ] || continue
-      if ! psql -w -c "select 1" >/dev/null; then
-        FAILED="${d} ${FAILED}"
-        break
-      fi
-      if [[ ${REGULAR_USER} = true ]] && [ -f "${d}"/regular-test.sh ]; then
-        "${d}/regular-test.sh" || FAILED="${d} ${FAILED}"
-        continue
-      fi
+LIST=$( (echo -e "${SKIP//","/"\n"}"; ls) | sort | uniq -u)
+for d in ${LIST}; do
+    [ -d "${d}" ] || continue
+    if ! psql -w -c "select 1" >/dev/null; then
+      FAILED="${d} ${FAILED}"
+      break
+    fi
+    if [[ ${REGULAR_USER} = true ]] && [ -f "${d}"/regular-test.sh ]; then
+       "${d}/regular-test.sh" || FAILED="${d} ${FAILED}"
+       continue
+    fi
 
-      if [ -f "${d}/neon-test.sh" ]; then
-        "${d}/neon-test.sh" || FAILED="${d} ${FAILED}"
-      else
-        USE_PGXS=1 make -C "${d}" installcheck || FAILED="${d} ${FAILED}"
-      fi
-  done
-  [[ -z ${FAILED} ]]  && exit 0
-fi
-for d in ${FAILED} $([[ ! -f ${FAILED_FILE} ]] || cat ${FAILED_FILE}); do
+    if [ -f "${d}/neon-test.sh" ]; then
+       "${d}/neon-test.sh" || FAILED="${d} ${FAILED}"
+    else
+       USE_PGXS=1 make -C "${d}" installcheck || FAILED="${d} ${FAILED}"
+    fi
+done
+[ -z "${FAILED}" ] && exit 0
+for d in ${FAILED}; do
   cat "$(find $d -name regression.diffs)"
 done
 for postgis_diff in /tmp/pgis_reg/*_diff; do
@@ -92,5 +68,4 @@ for postgis_diff in /tmp/pgis_reg/*_diff; do
   cat "${postgis_diff}"
 done
 echo "${FAILED}"
-cat ${FAILED_FILE}
 exit 1

docker-compose/test_extensions_upgrade.sh

@@ -1,4 +1,4 @@
-#!/usr/bin/env bash
+#!/bin/bash
 set -eux -o pipefail
 cd "$(dirname "${0}")"
 # Takes a variable name as argument. The result is stored in that variable.
@@ -60,8 +60,8 @@ function check_timeline() {
 # Restarts the compute node with the required compute tag and timeline.
 # Accepts the tag for the compute node and the timeline as parameters.
 function restart_compute() {
-  docker compose down compute1 compute_is_ready
-  COMPUTE_TAG=${1} TENANT_ID=${tenant_id} TIMELINE_ID=${2} docker compose up --quiet-pull -d --build compute1 compute_is_ready
+  docker compose down compute compute_is_ready
+  COMPUTE_TAG=${1} TENANT_ID=${tenant_id} TIMELINE_ID=${2} docker compose up --quiet-pull -d --build compute compute_is_ready
   wait_for_ready
   check_timeline ${2}
 }

endpoint_storage/src/app.rs

@@ -13,8 +13,6 @@ use utils::backoff::retry;
 pub fn app(state: Arc<Storage>) -> Router<()> {
     use axum::routing::{delete as _delete, get as _get};
     let delete_prefix = _delete(delete_prefix);
-    // NB: On any changes do not forget to update the OpenAPI spec
-    // in /endpoint_storage/src/openapi_spec.yml.
     Router::new()
         .route(
             "/{tenant_id}/{timeline_id}/{endpoint_id}/{*path}",

endpoint_storage/src/openapi_spec.yml

@@ -1,146 +0,0 @@
-openapi: "3.0.2"
-info:
-  title: Endpoint Storage API
-  description: Endpoint Storage API
-  version: "1.0"
-  license:
-    name: "Apache"
-    url: https://github.com/neondatabase/neon/blob/main/LICENSE
-servers:
-  - url: ""
-paths:
-  /status:
-    description: Healthcheck endpoint
-    get:
-      description: Healthcheck
-      security: []
-      responses:
-        "200":
-          description: OK
-
-  /{tenant_id}/{timeline_id}/{endpoint_id}/{key}:
-    parameters:
-      - name: tenant_id
-        in: path
-        required: true
-        schema:
-          type: string
-      - name: timeline_id
-        in: path
-        required: true
-        schema:
-          type: string
-      - name: endpoint_id
-        in: path
-        required: true
-        schema:
-          type: string
-      - name: key
-        in: path
-        required: true
-        schema:
-          type: string
-    get:
-      description: Get file from blob storage
-      responses:
-        "200":
-          description: "File stream from blob storage"
-          content:
-            application/octet-stream:
-              schema:
-                type: string
-                format: binary
-        "400":
-          description: File was not found
-        "403":
-          description: JWT does not authorize request to this route
-    put:
-      description: Insert file into blob storage. If file exists, override it
-      requestBody:
-        content:
-          application/octet-stream:
-            schema:
-              type: string
-              format: binary
-      responses:
-        "200":
-          description: File was inserted successfully
-        "403":
-          description: JWT does not authorize request to this route
-    delete:
-      description: Delete file from blob storage
-      responses:
-        "200":
-          description: File was successfully deleted or not found
-        "403":
-          description: JWT does not authorize request to this route
-
-  /{tenant_id}/{timeline_id}/{endpoint_id}:
-    parameters:
-      - name: tenant_id
-        in: path
-        required: true
-        schema:
-          type: string
-      - name: timeline_id
-        in: path
-        required: true
-        schema:
-          type: string
-      - name: endpoint_id
-        in: path
-        required: true
-        schema:
-          type: string
-    delete:
-      description: Delete endpoint data from blob storage
-      responses:
-        "200":
-          description: Endpoint data was deleted
-        "403":
-          description: JWT does not authorize request to this route
-
-  /{tenant_id}/{timeline_id}:
-    parameters:
-      - name: tenant_id
-        in: path
-        required: true
-        schema:
-          type: string
-      - name: timeline_id
-        in: path
-        required: true
-        schema:
-          type: string
-    delete:
-      description: Delete timeline data from blob storage
-      responses:
-        "200":
-          description: Timeline data was deleted
-        "403":
-          description: JWT does not authorize request to this route
-
-  /{tenant_id}:
-    parameters:
-      - name: tenant_id
-        in: path
-        required: true
-        schema:
-          type: string
-    delete:
-      description: Delete tenant data from blob storage
-      responses:
-        "200":
-          description: Tenant data was deleted
-        "403":
-          description: JWT does not authorize request to this route
-
-components:
-  securitySchemes:
-    JWT:
-      type: http
-      scheme: bearer
-      bearerFormat: JWT
-
-security:
-  - JWT: []

libs/compute_api/src/responses.rs

@@ -46,7 +46,7 @@ pub struct ExtensionInstallResponse {
     pub version: ExtVersion,
 }
 
-#[derive(Serialize, Default, Debug, Clone, PartialEq)]
+#[derive(Serialize, Default, Debug, Clone)]
 #[serde(tag = "status", rename_all = "snake_case")]
 pub enum LfcPrewarmState {
     #[default]
@@ -58,17 +58,6 @@ pub enum LfcPrewarmState {
     },
 }
 
-impl Display for LfcPrewarmState {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        match self {
-            LfcPrewarmState::NotPrewarmed => f.write_str("NotPrewarmed"),
-            LfcPrewarmState::Prewarming => f.write_str("Prewarming"),
-            LfcPrewarmState::Completed => f.write_str("Completed"),
-            LfcPrewarmState::Failed { error } => write!(f, "Error({error})"),
-        }
-    }
-}
-
 #[derive(Serialize, Default, Debug, Clone, PartialEq)]
 #[serde(tag = "status", rename_all = "snake_case")]
 pub enum LfcOffloadState {
@@ -81,23 +70,6 @@ pub enum LfcOffloadState {
     },
 }
 
-#[derive(Serialize, Debug, Clone, PartialEq)]
-#[serde(tag = "status", rename_all = "snake_case")]
-/// Response of /promote
-pub enum PromoteState {
-    NotPromoted,
-    Completed,
-    Failed { error: String },
-}
-
-#[derive(Deserialize, Serialize, Default, Debug, Clone)]
-#[serde(rename_all = "snake_case")]
-/// Result of /safekeepers_lsn
-pub struct SafekeepersLsn {
-    pub safekeepers: String,
-    pub wal_flush_lsn: utils::lsn::Lsn,
-}
-
 /// Response of the /status API
 #[derive(Serialize, Debug, Deserialize)]
 #[serde(rename_all = "snake_case")]
@@ -121,15 +93,6 @@ pub enum TerminateMode {
     Immediate,
 }
 
-impl From<TerminateMode> for ComputeStatus {
-    fn from(mode: TerminateMode) -> Self {
-        match mode {
-            TerminateMode::Fast => ComputeStatus::TerminationPendingFast,
-            TerminateMode::Immediate => ComputeStatus::TerminationPendingImmediate,
-        }
-    }
-}
-
 #[derive(Serialize, Clone, Copy, Debug, Deserialize, PartialEq, Eq)]
 #[serde(rename_all = "snake_case")]
 pub enum ComputeStatus {
@@ -150,9 +113,7 @@ pub enum ComputeStatus {
     // control-plane to terminate it.
     Failed,
     // Termination requested
-    TerminationPendingFast,
-    // Termination requested, without waiting 30s before returning from /terminate
-    TerminationPendingImmediate,
+    TerminationPending { mode: TerminateMode },
     // Terminated Postgres
     Terminated,
 }
@@ -171,10 +132,7 @@ impl Display for ComputeStatus {
             ComputeStatus::Running => f.write_str("running"),
             ComputeStatus::Configuration => f.write_str("configuration"),
             ComputeStatus::Failed => f.write_str("failed"),
-            ComputeStatus::TerminationPendingFast => f.write_str("termination-pending-fast"),
-            ComputeStatus::TerminationPendingImmediate => {
-                f.write_str("termination-pending-immediate")
-            }
+            ComputeStatus::TerminationPending { .. } => f.write_str("termination-pending"),
             ComputeStatus::Terminated => f.write_str("terminated"),
         }
     }

libs/compute_api/src/spec.rs

@@ -105,11 +105,7 @@ pub struct ComputeSpec {
     // updated to fill these fields, we can make these non optional.
     pub tenant_id: Option<TenantId>,
     pub timeline_id: Option<TimelineId>,
-
-    // Pageserver information can be passed in two different ways:
-    // 1. Here
-    // 2. in cluster.settings. This is legacy, we are switching to method 1.
-    pub pageserver_connection_info: Option<PageserverConnectionInfo>,
+    pub pageserver_connstring: Option<String>,
 
     // More neon ids that we expose to the compute_ctl
     // and to postgres as neon extension GUCs.
@@ -218,20 +214,6 @@ pub enum ComputeFeature {
     UnknownFeature,
 }
 
-/// Feature flag to signal `compute_ctl` to enable certain experimental functionality.
-#[derive(Clone, Debug, Default, Deserialize, Serialize, Eq, PartialEq)]
-pub struct PageserverConnectionInfo {
-    pub shards: HashMap<u32, PageserverShardConnectionInfo>,
-
-    pub prefer_grpc: bool,
-}
-
-#[derive(Clone, Debug, Default, Deserialize, Serialize, Eq, PartialEq)]
-pub struct PageserverShardConnectionInfo {
-    pub libpq_url: Option<String>,
-    pub grpc_url: Option<String>,
-}
-
 #[derive(Clone, Debug, Default, Deserialize, Serialize)]
 pub struct RemoteExtSpec {
     pub public_extensions: Option<Vec<String>>,
@@ -349,12 +331,6 @@ impl ComputeMode {
     }
 }
 
-impl Display for ComputeMode {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        f.write_str(self.to_type_str())
-    }
-}
-
 /// Log level for audit logging
 #[derive(Clone, Debug, Default, Eq, PartialEq, Deserialize, Serialize)]
 pub enum ComputeAudit {
@@ -466,7 +442,7 @@ pub struct JwksSettings {
 }
 
 /// Protocol used to connect to a Pageserver. Parsed from the connstring scheme.
-#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)]
+#[derive(Clone, Copy, Debug, Default)]
 pub enum PageserverProtocol {
     /// The original protocol based on libpq and COPY. Uses postgresql:// or postgres:// scheme.
     #[default]

libs/http-utils/src/endpoint.rs

@@ -20,7 +20,6 @@ use tokio_stream::wrappers::ReceiverStream;
 use tokio_util::io::ReaderStream;
 use tracing::{Instrument, debug, info, info_span, warn};
 use utils::auth::{AuthError, Claims, SwappableJwtAuth};
-use utils::metrics_collector::{METRICS_COLLECTOR, METRICS_STALE_MILLIS};
 
 use crate::error::{ApiError, api_error_handler, route_error_handler};
 use crate::request::{get_query_param, parse_query_param};
@@ -251,28 +250,9 @@ impl std::io::Write for ChannelWriter {
     }
 }
 
-pub async fn prometheus_metrics_handler(
-    req: Request<Body>,
-    force_metric_collection_on_scrape: bool,
-) -> Result<Response<Body>, ApiError> {
+pub async fn prometheus_metrics_handler(_req: Request<Body>) -> Result<Response<Body>, ApiError> {
     SERVE_METRICS_COUNT.inc();
 
-    // HADRON
-    let requested_use_latest = parse_query_param(&req, "use_latest")?;
-
-    let use_latest = match requested_use_latest {
-        None => force_metric_collection_on_scrape,
-        Some(true) => true,
-        Some(false) => {
-            if force_metric_collection_on_scrape {
-                // We don't cache in this case
-                true
-            } else {
-                false
-            }
-        }
-    };
-
     let started_at = std::time::Instant::now();
 
     let (tx, rx) = mpsc::channel(1);
@@ -297,18 +277,12 @@ pub async fn prometheus_metrics_handler(
 
         let _span = span.entered();
 
-        // HADRON
-        let collected = if use_latest {
-            // Skip caching the results if we always force metric collection on scrape.
-            METRICS_COLLECTOR.run_once(!force_metric_collection_on_scrape)
-        } else {
-            METRICS_COLLECTOR.last_collected()
-        };
+        let metrics = metrics::gather();
 
         let gathered_at = std::time::Instant::now();
 
         let res = encoder
-            .encode(&collected.metrics, &mut writer)
+            .encode(&metrics, &mut writer)
             .and_then(|_| writer.flush().map_err(|e| e.into()));
 
         // this instant is not when we finally got the full response sent, sending is done by hyper
@@ -321,10 +295,6 @@ pub async fn prometheus_metrics_handler(
         let encoded_in = encoded_at - gathered_at - writer.wait_time();
         let total = encoded_at - started_at;
 
-        // HADRON
-        let staleness_ms = (encoded_at - collected.collected_at).as_millis();
-        METRICS_STALE_MILLIS.set(staleness_ms as i64);
-
         match res {
             Ok(()) => {
                 tracing::info!(
@@ -333,7 +303,6 @@ pub async fn prometheus_metrics_handler(
                     spawning_ms = spawned_in.as_millis(),
                     collection_ms = collected_in.as_millis(),
                     encoding_ms = encoded_in.as_millis(),
-                    stalenss_ms = staleness_ms,
                     "responded /metrics"
                 );
             }

libs/http-utils/src/request.rs

@@ -41,35 +41,17 @@ pub fn get_query_param<'a>(
         Some(q) => q,
         None => return Ok(None),
     };
-    let values = url::form_urlencoded::parse(query.as_bytes())
+    let mut values = url::form_urlencoded::parse(query.as_bytes())
         .filter_map(|(k, v)| if k == param_name { Some(v) } else { None })
         // we call .next() twice below. If it's None the first time, .fuse() ensures it's None afterwards
         .fuse();
 
-    // Work around an issue with Alloy's pyroscope scrape where the "seconds"
-    // parameter is added several times. https://github.com/grafana/alloy/issues/3026
-    // TODO: revert after Alloy is fixed.
-    let value1 = values
-        .map(Ok)
-        .reduce(|acc, i| {
-            match acc {
-                Err(_) => acc,
-
-                // It's okay to have duplicates as along as they have the same value.
-                Ok(ref a) if a == &i.unwrap() => acc,
-
-                _ => Err(ApiError::BadRequest(anyhow!(
-                    "param {param_name} specified more than once"
-                ))),
-            }
-        })
-        .transpose()?;
-    // if values.next().is_some() {
-    //     return Err(ApiError::BadRequest(anyhow!(
-    //         "param {param_name} specified more than once"
-    //     )));
-    // }
-
+    let value1 = values.next();
+    if values.next().is_some() {
+        return Err(ApiError::BadRequest(anyhow!(
+            "param {param_name} specified more than once"
+        )));
+    }
     Ok(value1)
 }
 
@@ -110,39 +92,3 @@ pub async fn ensure_no_body(request: &mut Request<Body>) -> Result<(), ApiError>
         None => Ok(()),
     }
 }
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_get_query_param_duplicate() {
-        let req = Request::builder()
-            .uri("http://localhost:12345/testuri?testparam=1")
-            .body(hyper::Body::empty())
-            .unwrap();
-        let value = get_query_param(&req, "testparam").unwrap();
-        assert_eq!(value.unwrap(), "1");
-
-        let req = Request::builder()
-            .uri("http://localhost:12345/testuri?testparam=1&testparam=1")
-            .body(hyper::Body::empty())
-            .unwrap();
-        let value = get_query_param(&req, "testparam").unwrap();
-        assert_eq!(value.unwrap(), "1");
-
-        let req = Request::builder()
-            .uri("http://localhost:12345/testuri")
-            .body(hyper::Body::empty())
-            .unwrap();
-        let value = get_query_param(&req, "testparam").unwrap();
-        assert!(value.is_none());
-
-        let req = Request::builder()
-            .uri("http://localhost:12345/testuri?testparam=1&testparam=2&testparam=3")
-            .body(hyper::Body::empty())
-            .unwrap();
-        let value = get_query_param(&req, "testparam");
-        assert!(value.is_err());
-    }
-}

libs/neon-shmem/Cargo.toml

@@ -6,27 +6,8 @@ license.workspace = true
 
 [dependencies]
 thiserror.workspace = true
-nix.workspace = true
+nix.workspace=true
 workspace_hack = { version = "0.1", path = "../../workspace_hack" }
-rustc-hash = { version = "2.1.1" }
-rand = "0.9.1"
-libc.workspace = true
-lock_api = "0.4.13"
-
-[dev-dependencies]
-criterion = { workspace = true, features = ["html_reports"] }
-rand_distr = "0.5.1"
-xxhash-rust = { version = "0.8.15", features = ["xxh3"] }
-ahash.workspace = true
-twox-hash = { version = "2.1.1" }
-seahash = "4.1.0"
-hashbrown = { git = "https://github.com/quantumish/hashbrown.git", rev = "6610e6d" }
-foldhash = "0.1.5"
-
 
 [target.'cfg(target_os = "macos")'.dependencies]
 tempfile = "3.14.0"
-
-[[bench]]
-name = "hmap_resize"
-harness = false

libs/neon-shmem/benches/hmap_resize.rs

@@ -1,330 +0,0 @@
-use criterion::{BatchSize, BenchmarkId, Criterion, criterion_group, criterion_main};
-use neon_shmem::hash::HashMapAccess;
-use neon_shmem::hash::HashMapInit;
-use neon_shmem::hash::entry::Entry;
-use rand::distr::{Distribution, StandardUniform};
-use rand::prelude::*;
-use std::default::Default;
-use std::hash::BuildHasher;
-
-// Taken from bindings to C code
-
-#[derive(Clone, Debug, Hash, Eq, PartialEq)]
-#[repr(C)]
-pub struct FileCacheKey {
-    pub _spc_id: u32,
-    pub _db_id: u32,
-    pub _rel_number: u32,
-    pub _fork_num: u32,
-    pub _block_num: u32,
-}
-
-impl Distribution<FileCacheKey> for StandardUniform {
-    // questionable, but doesn't need to be good randomness
-    fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> FileCacheKey {
-        FileCacheKey {
-            _spc_id: rng.random(),
-            _db_id: rng.random(),
-            _rel_number: rng.random(),
-            _fork_num: rng.random(),
-            _block_num: rng.random(),
-        }
-    }
-}
-
-#[derive(Clone, Debug)]
-#[repr(C)]
-pub struct FileCacheEntry {
-    pub _offset: u32,
-    pub _access_count: u32,
-    pub _prev: *mut FileCacheEntry,
-    pub _next: *mut FileCacheEntry,
-    pub _state: [u32; 8],
-}
-
-impl FileCacheEntry {
-    fn dummy() -> Self {
-        Self {
-            _offset: 0,
-            _access_count: 0,
-            _prev: std::ptr::null_mut(),
-            _next: std::ptr::null_mut(),
-            _state: [0; 8],
-        }
-    }
-}
-
-// Utilities for applying operations.
-
-#[derive(Clone, Debug)]
-struct TestOp<K, V>(K, Option<V>);
-
-fn apply_op<K: Clone + std::hash::Hash + Eq, V, S: std::hash::BuildHasher>(
-    op: TestOp<K, V>,
-    map: &mut HashMapAccess<K, V, S>,
-) {
-    let entry = map.entry(op.0);
-
-    match op.1 {
-        Some(new) => match entry {
-            Entry::Occupied(mut e) => Some(e.insert(new)),
-            Entry::Vacant(e) => {
-                _ = e.insert(new).unwrap();
-                None
-            }
-        },
-        None => match entry {
-            Entry::Occupied(e) => Some(e.remove()),
-            Entry::Vacant(_) => None,
-        },
-    };
-}
-
-// Hash utilities
-
-struct SeaRandomState {
-    k1: u64,
-    k2: u64,
-    k3: u64,
-    k4: u64,
-}
-
-impl std::hash::BuildHasher for SeaRandomState {
-    type Hasher = seahash::SeaHasher;
-
-    fn build_hasher(&self) -> Self::Hasher {
-        seahash::SeaHasher::with_seeds(self.k1, self.k2, self.k3, self.k4)
-    }
-}
-
-impl SeaRandomState {
-    fn new() -> Self {
-        let mut rng = rand::rng();
-        Self {
-            k1: rng.random(),
-            k2: rng.random(),
-            k3: rng.random(),
-            k4: rng.random(),
-        }
-    }
-}
-
-fn small_benchs(c: &mut Criterion) {
-    let mut group = c.benchmark_group("Small maps");
-    group.sample_size(10);
-
-    group.bench_function("small_rehash", |b| {
-        let ideal_filled = 4_000_000;
-        let size = 5_000_000;
-        let mut writer = HashMapInit::new_resizeable(size, size * 2).attach_writer();
-        let mut rng = rand::rng();
-        while writer.get_num_buckets_in_use() < ideal_filled as usize {
-            let key: FileCacheKey = rng.random();
-            let val = FileCacheEntry::dummy();
-            apply_op(TestOp(key, Some(val)), &mut writer);
-        }
-        b.iter(|| writer.shuffle());
-    });
-
-    group.bench_function("small_rehash_xxhash", |b| {
-        let ideal_filled = 4_000_000;
-        let size = 5_000_000;
-        let mut writer = HashMapInit::new_resizeable(size, size * 2)
-            .with_hasher(twox_hash::xxhash64::RandomState::default())
-            .attach_writer();
-        let mut rng = rand::rng();
-        while writer.get_num_buckets_in_use() < ideal_filled as usize {
-            let key: FileCacheKey = rng.random();
-            let val = FileCacheEntry::dummy();
-            apply_op(TestOp(key, Some(val)), &mut writer);
-        }
-        b.iter(|| writer.shuffle());
-    });
-
-    group.bench_function("small_rehash_ahash", |b| {
-        let ideal_filled = 4_000_000;
-        let size = 5_000_000;
-        let mut writer = HashMapInit::new_resizeable(size, size * 2)
-            .with_hasher(ahash::RandomState::default())
-            .attach_writer();
-        let mut rng = rand::rng();
-        while writer.get_num_buckets_in_use() < ideal_filled as usize {
-            let key: FileCacheKey = rng.random();
-            let val = FileCacheEntry::dummy();
-            apply_op(TestOp(key, Some(val)), &mut writer);
-        }
-        b.iter(|| writer.shuffle());
-    });
-
-    group.bench_function("small_rehash_seahash", |b| {
-        let ideal_filled = 4_000_000;
-        let size = 5_000_000;
-        let mut writer = HashMapInit::new_resizeable(size, size * 2)
-            .with_hasher(SeaRandomState::new())
-            .attach_writer();
-        let mut rng = rand::rng();
-        while writer.get_num_buckets_in_use() < ideal_filled as usize {
-            let key: FileCacheKey = rng.random();
-            let val = FileCacheEntry::dummy();
-            apply_op(TestOp(key, Some(val)), &mut writer);
-        }
-        b.iter(|| writer.shuffle());
-    });
-
-    group.finish();
-}
-
-fn real_benchs(c: &mut Criterion) {
-    let mut group = c.benchmark_group("Realistic workloads");
-    group.sample_size(10);
-    group.bench_function("real_bulk_insert", |b| {
-        let size = 125_000_000;
-        let ideal_filled = 100_000_000;
-        let mut rng = rand::rng();
-        b.iter_batched(
-            || HashMapInit::new_resizeable(size, size * 2).attach_writer(),
-            |writer| {
-                for _ in 0..ideal_filled {
-                    let key: FileCacheKey = rng.random();
-                    let val = FileCacheEntry::dummy();
-                    let entry = writer.entry(key);
-                    std::hint::black_box(match entry {
-                        Entry::Occupied(mut e) => {
-                            e.insert(val);
-                        }
-                        Entry::Vacant(e) => {
-                            _ = e.insert(val).unwrap();
-                        }
-                    })
-                }
-            },
-            BatchSize::SmallInput,
-        )
-    });
-
-    group.bench_function("real_rehash", |b| {
-        let size = 125_000_000;
-        let ideal_filled = 100_000_000;
-        let mut writer = HashMapInit::new_resizeable(size, size).attach_writer();
-        let mut rng = rand::rng();
-        while writer.get_num_buckets_in_use() < ideal_filled {
-            let key: FileCacheKey = rng.random();
-            let val = FileCacheEntry::dummy();
-            apply_op(TestOp(key, Some(val)), &mut writer);
-        }
-        b.iter(|| writer.shuffle());
-    });
-
-    group.bench_function("real_rehash_hashbrown", |b| {
-        let size = 125_000_000;
-        let ideal_filled = 100_000_000;
-        let mut writer = hashbrown::raw::RawTable::new();
-        let mut rng = rand::rng();
-        let hasher = rustc_hash::FxBuildHasher::default();
-        unsafe {
-            writer
-                .resize(
-                    size,
-                    |(k, _)| hasher.hash_one(&k),
-                    hashbrown::raw::Fallibility::Infallible,
-                )
-                .unwrap();
-        }
-        while writer.len() < ideal_filled as usize {
-            let key: FileCacheKey = rng.random();
-            let val = FileCacheEntry::dummy();
-            writer.insert(hasher.hash_one(&key), (key, val), |(k, _)| {
-                hasher.hash_one(&k)
-            });
-        }
-        b.iter(|| unsafe {
-            writer.table.rehash_in_place(
-                &|table, index| {
-                    hasher.hash_one(
-                        &table
-                            .bucket::<(FileCacheKey, FileCacheEntry)>(index)
-                            .as_ref()
-                            .0,
-                    )
-                },
-                std::mem::size_of::<(FileCacheKey, FileCacheEntry)>(),
-                if std::mem::needs_drop::<(FileCacheKey, FileCacheEntry)>() {
-                    Some(|ptr| std::ptr::drop_in_place(ptr as *mut (FileCacheKey, FileCacheEntry)))
-                } else {
-                    None
-                },
-            )
-        });
-    });
-
-    for elems in [2, 4, 8, 16, 32, 64, 96, 112] {
-        group.bench_with_input(
-            BenchmarkId::new("real_rehash_varied", elems),
-            &elems,
-            |b, &size| {
-                let ideal_filled = size * 1_000_000;
-                let size = 125_000_000;
-                let mut writer = HashMapInit::new_resizeable(size, size).attach_writer();
-                let mut rng = rand::rng();
-                while writer.get_num_buckets_in_use() < ideal_filled as usize {
-                    let key: FileCacheKey = rng.random();
-                    let val = FileCacheEntry::dummy();
-                    apply_op(TestOp(key, Some(val)), &mut writer);
-                }
-                b.iter(|| writer.shuffle());
-            },
-        );
-        group.bench_with_input(
-            BenchmarkId::new("real_rehash_varied_hashbrown", elems),
-            &elems,
-            |b, &size| {
-                let ideal_filled = size * 1_000_000;
-                let size = 125_000_000;
-                let mut writer = hashbrown::raw::RawTable::new();
-                let mut rng = rand::rng();
-                let hasher = rustc_hash::FxBuildHasher::default();
-                unsafe {
-                    writer
-                        .resize(
-                            size,
-                            |(k, _)| hasher.hash_one(&k),
-                            hashbrown::raw::Fallibility::Infallible,
-                        )
-                        .unwrap();
-                }
-                while writer.len() < ideal_filled as usize {
-                    let key: FileCacheKey = rng.random();
-                    let val = FileCacheEntry::dummy();
-                    writer.insert(hasher.hash_one(&key), (key, val), |(k, _)| {
-                        hasher.hash_one(&k)
-                    });
-                }
-                b.iter(|| unsafe {
-                    writer.table.rehash_in_place(
-                        &|table, index| {
-                            hasher.hash_one(
-                                &table
-                                    .bucket::<(FileCacheKey, FileCacheEntry)>(index)
-                                    .as_ref()
-                                    .0,
-                            )
-                        },
-                        std::mem::size_of::<(FileCacheKey, FileCacheEntry)>(),
-                        if std::mem::needs_drop::<(FileCacheKey, FileCacheEntry)>() {
-                            Some(|ptr| {
-                                std::ptr::drop_in_place(ptr as *mut (FileCacheKey, FileCacheEntry))
-                            })
-                        } else {
-                            None
-                        },
-                    )
-                });
-            },
-        );
-    }
-
-    group.finish();
-}
-
-criterion_group!(benches, small_benchs, real_benchs);
-criterion_main!(benches);

libs/neon-shmem/src/hash.rs

@@ -1,598 +0,0 @@
-//! Resizable hash table implementation on top of byte-level storage (either a [`ShmemHandle`] or a fixed byte array).
-//!
-//! This hash table has two major components: the bucket array and the dictionary. Each bucket within the
-//! bucket array contains a `Option<(K, V)>` and an index of another bucket. In this way there is both an
-//! implicit freelist within the bucket array (`None` buckets point to other `None` entries) and various hash
-//! chains within the bucket array (a Some bucket will point to other Some buckets that had the same hash).
-//!
-//! Buckets are never moved unless they are within a region that is being shrunk, and so the actual hash-
-//! dependent component is done with the dictionary. When a new key is inserted into the map, a position
-//! within the dictionary is decided based on its hash, the data is inserted into an empty bucket based
-//! off of the freelist, and then the index of said bucket is placed in the dictionary.
-//!
-//! This map is resizable (if initialized on top of a [`ShmemHandle`]). Both growing and shrinking happen
-//! in-place and are at a high level achieved by expanding/reducing the bucket array and rebuilding the
-//! dictionary by rehashing all keys.
-
-use std::fmt::Debug;
-use std::hash::{BuildHasher, Hash};
-use std::mem::MaybeUninit;
-
-use crate::shmem::ShmemHandle;
-use crate::{shmem, sync::*};
-
-mod core;
-pub mod entry;
-
-#[cfg(test)]
-mod tests;
-
-use core::{Bucket, CoreHashMap, INVALID_POS};
-use entry::{Entry, OccupiedEntry, PrevPos, VacantEntry};
-
-/// This represents a hash table that (possibly) lives in shared memory.
-/// If a new process is launched with fork(), the child process inherits
-/// this struct.
-#[must_use]
-pub struct HashMapInit<'a, K, V, S = rustc_hash::FxBuildHasher> {
-    shmem_handle: Option<ShmemHandle>,
-    shared_ptr: *mut HashMapShared<'a, K, V>,
-    shared_size: usize,
-    hasher: S,
-    num_buckets: u32,
-}
-
-impl<'a, K, V, S> Debug for HashMapInit<'a, K, V, S>
-where
-    K: Debug,
-    V: Debug,
-{
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        f.debug_struct("HashMapInit")
-            .field("shmem_handle", &self.shmem_handle)
-            .field("shared_ptr", &self.shared_ptr)
-            .field("shared_size", &self.shared_size)
-            // .field("hasher", &self.hasher)
-            .field("num_buckets", &self.num_buckets)
-            .finish()
-    }
-}
-
-/// This is a per-process handle to a hash table that (possibly) lives in shared memory.
-/// If a child process is launched with fork(), the child process should
-/// get its own HashMapAccess by calling HashMapInit::attach_writer/reader().
-///
-/// XXX: We're not making use of it at the moment, but this struct could
-/// hold process-local information in the future.
-pub struct HashMapAccess<'a, K, V, S = rustc_hash::FxBuildHasher> {
-    shmem_handle: Option<ShmemHandle>,
-    shared_ptr: *mut HashMapShared<'a, K, V>,
-    hasher: S,
-}
-
-unsafe impl<K: Sync, V: Sync, S> Sync for HashMapAccess<'_, K, V, S> {}
-unsafe impl<K: Send, V: Send, S> Send for HashMapAccess<'_, K, V, S> {}
-
-impl<'a, K, V, S> Debug for HashMapAccess<'a, K, V, S>
-where
-    K: Debug,
-    V: Debug,
-{
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        f.debug_struct("HashMapAccess")
-            .field("shmem_handle", &self.shmem_handle)
-            .field("shared_ptr", &self.shared_ptr)
-            // .field("hasher", &self.hasher)
-            .finish()
-    }
-}
-
-impl<'a, K: Clone + Hash + Eq, V, S> HashMapInit<'a, K, V, S> {
-    /// Change the 'hasher' used by the hash table.
-    ///
-    /// NOTE: This must be called right after creating the hash table,
-    /// before inserting any entries and before calling attach_writer/reader.
-    /// Otherwise different accessors could be using different hash function,
-    /// with confusing results.
-    pub fn with_hasher<T: BuildHasher>(self, hasher: T) -> HashMapInit<'a, K, V, T> {
-        HashMapInit {
-            hasher,
-            shmem_handle: self.shmem_handle,
-            shared_ptr: self.shared_ptr,
-            shared_size: self.shared_size,
-            num_buckets: self.num_buckets,
-        }
-    }
-
-    /// Loosely (over)estimate the size needed to store a hash table with `num_buckets` buckets.
-    pub fn estimate_size(num_buckets: u32) -> usize {
-        // add some margin to cover alignment etc.
-        CoreHashMap::<K, V>::estimate_size(num_buckets) + size_of::<HashMapShared<K, V>>() + 1000
-    }
-
-    fn new(
-        num_buckets: u32,
-        shmem_handle: Option<ShmemHandle>,
-        area_ptr: *mut u8,
-        area_size: usize,
-        hasher: S,
-    ) -> Self {
-        let mut ptr: *mut u8 = area_ptr;
-        let end_ptr: *mut u8 = unsafe { ptr.add(area_size) };
-
-        // carve out area for the One Big Lock (TM) and the HashMapShared.
-        ptr = unsafe { ptr.add(ptr.align_offset(align_of::<libc::pthread_rwlock_t>())) };
-        let raw_lock_ptr = ptr;
-        ptr = unsafe { ptr.add(size_of::<libc::pthread_rwlock_t>()) };
-        ptr = unsafe { ptr.add(ptr.align_offset(align_of::<HashMapShared<K, V>>())) };
-        let shared_ptr: *mut HashMapShared<K, V> = ptr.cast();
-        ptr = unsafe { ptr.add(size_of::<HashMapShared<K, V>>()) };
-
-        // carve out the buckets
-        ptr = unsafe { ptr.byte_add(ptr.align_offset(align_of::<core::Bucket<K, V>>())) };
-        let buckets_ptr = ptr;
-        ptr = unsafe { ptr.add(size_of::<core::Bucket<K, V>>() * num_buckets as usize) };
-
-        // use remaining space for the dictionary
-        ptr = unsafe { ptr.byte_add(ptr.align_offset(align_of::<u32>())) };
-        assert!(ptr.addr() < end_ptr.addr());
-        let dictionary_ptr = ptr;
-        let dictionary_size = unsafe { end_ptr.byte_offset_from(ptr) / size_of::<u32>() as isize };
-        assert!(dictionary_size > 0);
-
-        let buckets =
-            unsafe { std::slice::from_raw_parts_mut(buckets_ptr.cast(), num_buckets as usize) };
-        let dictionary = unsafe {
-            std::slice::from_raw_parts_mut(dictionary_ptr.cast(), dictionary_size as usize)
-        };
-
-        let hashmap = CoreHashMap::new(buckets, dictionary);
-        let lock = RwLock::from_raw(PthreadRwLock::new(raw_lock_ptr.cast()), hashmap);
-        unsafe {
-            std::ptr::write(shared_ptr, lock);
-        }
-
-        Self {
-            num_buckets,
-            shmem_handle,
-            shared_ptr,
-            shared_size: area_size,
-            hasher,
-        }
-    }
-
-    /// Attach to a hash table for writing.
-    pub fn attach_writer(self) -> HashMapAccess<'a, K, V, S> {
-        HashMapAccess {
-            shmem_handle: self.shmem_handle,
-            shared_ptr: self.shared_ptr,
-            hasher: self.hasher,
-        }
-    }
-
-    /// Initialize a table for reading. Currently identical to [`HashMapInit::attach_writer`].
-    pub fn attach_reader(self) -> HashMapAccess<'a, K, V, S> {
-        self.attach_writer()
-    }
-}
-
-/// Hash table data that is actually stored in the shared memory area.
-///
-/// NOTE: We carve out the parts from a contiguous chunk. Growing and shrinking the hash table
-/// relies on the memory layout! The data structures are laid out in the contiguous shared memory
-/// area as follows:
-///
-/// [`libc::pthread_rwlock_t`]
-/// [`HashMapShared`]
-/// [buckets]
-/// [dictionary]
-///
-/// In between the above parts, there can be padding bytes to align the parts correctly.
-type HashMapShared<'a, K, V> = RwLock<CoreHashMap<'a, K, V>>;
-
-impl<'a, K, V> HashMapInit<'a, K, V, rustc_hash::FxBuildHasher>
-where
-    K: Clone + Hash + Eq,
-{
-    /// Place the hash table within a user-supplied fixed memory area.
-    pub fn with_fixed(num_buckets: u32, area: &'a mut [MaybeUninit<u8>]) -> Self {
-        Self::new(
-            num_buckets,
-            None,
-            area.as_mut_ptr().cast(),
-            area.len(),
-            rustc_hash::FxBuildHasher,
-        )
-    }
-
-    /// Place a new hash map in the given shared memory area
-    ///
-    /// # Panics
-    /// Will panic on failure to resize area to expected map size.
-    pub fn with_shmem(num_buckets: u32, shmem: ShmemHandle) -> Self {
-        let size = Self::estimate_size(num_buckets);
-        shmem
-            .set_size(size)
-            .expect("could not resize shared memory area");
-        let ptr = shmem.data_ptr.as_ptr().cast();
-        Self::new(
-            num_buckets,
-            Some(shmem),
-            ptr,
-            size,
-            rustc_hash::FxBuildHasher,
-        )
-    }
-
-    /// Make a resizable hash map within a new shared memory area with the given name.
-    pub fn new_resizeable_named(num_buckets: u32, max_buckets: u32, name: &str) -> Self {
-        let size = Self::estimate_size(num_buckets);
-        let max_size = Self::estimate_size(max_buckets);
-        let shmem =
-            ShmemHandle::new(name, size, max_size).expect("failed to make shared memory area");
-        let ptr = shmem.data_ptr.as_ptr().cast();
-
-        Self::new(
-            num_buckets,
-            Some(shmem),
-            ptr,
-            size,
-            rustc_hash::FxBuildHasher,
-        )
-    }
-
-    /// Make a resizable hash map within a new anonymous shared memory area.
-    pub fn new_resizeable(num_buckets: u32, max_buckets: u32) -> Self {
-        use std::sync::atomic::{AtomicUsize, Ordering};
-        static COUNTER: AtomicUsize = AtomicUsize::new(0);
-        let val = COUNTER.fetch_add(1, Ordering::Relaxed);
-        let name = format!("neon_shmem_hmap{val}");
-        Self::new_resizeable_named(num_buckets, max_buckets, &name)
-    }
-}
-
-impl<'a, K, V, S: BuildHasher> HashMapAccess<'a, K, V, S>
-where
-    K: Clone + Hash + Eq,
-{
-    /// Hash a key using the map's hasher.
-    #[inline]
-    fn get_hash_value(&self, key: &K) -> u64 {
-        self.hasher.hash_one(key)
-    }
-
-    fn entry_with_hash(&self, key: K, hash: u64) -> Entry<'a, '_, K, V> {
-        let mut map = unsafe { self.shared_ptr.as_ref() }.unwrap().write();
-        let dict_pos = hash as usize % map.dictionary.len();
-        let first = map.dictionary[dict_pos];
-        if first == INVALID_POS {
-            // no existing entry
-            return Entry::Vacant(VacantEntry {
-                map,
-                key,
-                dict_pos: dict_pos as u32,
-            });
-        }
-
-        let mut prev_pos = PrevPos::First(dict_pos as u32);
-        let mut next = first;
-        loop {
-            let bucket = &mut map.buckets[next as usize];
-            let (bucket_key, _bucket_value) = bucket.inner.as_mut().expect("entry is in use");
-            if *bucket_key == key {
-                // found existing entry
-                return Entry::Occupied(OccupiedEntry {
-                    map,
-                    _key: key,
-                    prev_pos,
-                    bucket_pos: next,
-                });
-            }
-
-            if bucket.next == INVALID_POS {
-                // No existing entry
-                return Entry::Vacant(VacantEntry {
-                    map,
-                    key,
-                    dict_pos: dict_pos as u32,
-                });
-            }
-            prev_pos = PrevPos::Chained(next);
-            next = bucket.next;
-        }
-    }
-
-    /// Get a reference to the corresponding value for a key.
-    pub fn get<'e>(&'e self, key: &K) -> Option<ValueReadGuard<'e, V>> {
-        let hash = self.get_hash_value(key);
-        let map = unsafe { self.shared_ptr.as_ref() }.unwrap().read();
-        RwLockReadGuard::try_map(map, |m| m.get_with_hash(key, hash)).ok()
-    }
-
-    /// Get a reference to the entry containing a key.
-    pub fn entry(&self, key: K) -> Entry<'a, '_, K, V> {
-        let hash = self.get_hash_value(&key);
-        self.entry_with_hash(key, hash)
-    }
-
-    /// Remove a key given its hash. Returns the associated value if it existed.
-    pub fn remove(&self, key: &K) -> Option<V> {
-        let hash = self.get_hash_value(&key);
-        match self.entry_with_hash(key.clone(), hash) {
-            Entry::Occupied(e) => Some(e.remove()),
-            Entry::Vacant(_) => None,
-        }
-    }
-
-    /// Insert/update a key. Returns the previous associated value if it existed.
-    ///
-    /// # Errors
-    /// Will return [`core::FullError`] if there is no more space left in the map.
-    pub fn insert(&self, key: K, value: V) -> Result<Option<V>, core::FullError> {
-        let hash = self.get_hash_value(&key);
-        match self.entry_with_hash(key.clone(), hash) {
-            Entry::Occupied(mut e) => Ok(Some(e.insert(value))),
-            Entry::Vacant(e) => {
-                _ = e.insert(value)?;
-                Ok(None)
-            }
-        }
-    }
-
-    /// Optionally return the entry for a bucket at a given index if it exists.
-    ///
-    /// Has more overhead than one would intuitively expect: performs both a clone of the key
-    /// due to the [`OccupiedEntry`] type owning the key and also a hash of the key in order
-    /// to enable repairing the hash chain if the entry is removed.
-    pub fn entry_at_bucket(&self, pos: usize) -> Option<OccupiedEntry<'a, '_, K, V>> {
-        let map = unsafe { self.shared_ptr.as_mut() }.unwrap().write();
-        if pos >= map.buckets.len() {
-            return None;
-        }
-
-        let entry = map.buckets[pos].inner.as_ref();
-        match entry {
-            Some((key, _)) => Some(OccupiedEntry {
-                _key: key.clone(),
-                bucket_pos: pos as u32,
-                prev_pos: entry::PrevPos::Unknown(self.get_hash_value(&key)),
-                map,
-            }),
-            _ => None,
-        }
-    }
-
-    /// Returns the number of buckets in the table.
-    pub fn get_num_buckets(&self) -> usize {
-        let map = unsafe { self.shared_ptr.as_ref() }.unwrap().read();
-        map.get_num_buckets()
-    }
-
-    /// Return the key and value stored in bucket with given index. This can be used to
-    /// iterate through the hash map.
-    // TODO: An Iterator might be nicer. The communicator's clock algorithm needs to
-    // _slowly_ iterate through all buckets with its clock hand,  without holding a lock.
-    // If we switch to an Iterator, it must not hold the lock.
-    pub fn get_at_bucket(&self, pos: usize) -> Option<ValueReadGuard<(K, V)>> {
-        let map = unsafe { self.shared_ptr.as_ref() }.unwrap().read();
-        if pos >= map.buckets.len() {
-            return None;
-        }
-        RwLockReadGuard::try_map(map, |m| m.buckets[pos].inner.as_ref()).ok()
-    }
-
-    /// Returns the index of the bucket a given value corresponds to.
-    pub fn get_bucket_for_value(&self, val_ptr: *const V) -> usize {
-        let map = unsafe { self.shared_ptr.as_ref() }.unwrap().read();
-
-        let origin = map.buckets.as_ptr();
-        let idx = (val_ptr as usize - origin as usize) / size_of::<Bucket<K, V>>();
-        assert!(idx < map.buckets.len());
-
-        idx
-    }
-
-    /// Returns the number of occupied buckets in the table.
-    pub fn get_num_buckets_in_use(&self) -> usize {
-        let map = unsafe { self.shared_ptr.as_ref() }.unwrap().read();
-        map.buckets_in_use as usize
-    }
-
-    /// Clears all entries in a table. Does not reset any shrinking operations.
-    pub fn clear(&self) {
-        let mut map = unsafe { self.shared_ptr.as_mut() }.unwrap().write();
-        map.clear();
-    }
-
-    /// Perform an in-place rehash of some region (0..`rehash_buckets`) of the table and reset
-    /// the `buckets` and `dictionary` slices to be as long as `num_buckets`. Resets the freelist
-    /// in the process.
-    fn rehash_dict(
-        &self,
-        inner: &mut CoreHashMap<'a, K, V>,
-        buckets_ptr: *mut core::Bucket<K, V>,
-        end_ptr: *mut u8,
-        num_buckets: u32,
-        rehash_buckets: u32,
-    ) {
-        inner.free_head = INVALID_POS;
-
-        let buckets;
-        let dictionary;
-        unsafe {
-            let buckets_end_ptr = buckets_ptr.add(num_buckets as usize);
-            let dictionary_ptr: *mut u32 = buckets_end_ptr
-                .byte_add(buckets_end_ptr.align_offset(align_of::<u32>()))
-                .cast();
-            let dictionary_size: usize =
-                end_ptr.byte_offset_from(buckets_end_ptr) as usize / size_of::<u32>();
-
-            buckets = std::slice::from_raw_parts_mut(buckets_ptr, num_buckets as usize);
-            dictionary = std::slice::from_raw_parts_mut(dictionary_ptr, dictionary_size);
-        }
-        for e in dictionary.iter_mut() {
-            *e = INVALID_POS;
-        }
-
-        for (i, bucket) in buckets.iter_mut().enumerate().take(rehash_buckets as usize) {
-            if bucket.inner.is_none() {
-                bucket.next = inner.free_head;
-                inner.free_head = i as u32;
-                continue;
-            }
-
-            let hash = self.hasher.hash_one(&bucket.inner.as_ref().unwrap().0);
-            let pos: usize = (hash % dictionary.len() as u64) as usize;
-            bucket.next = dictionary[pos];
-            dictionary[pos] = i as u32;
-        }
-
-        inner.dictionary = dictionary;
-        inner.buckets = buckets;
-    }
-
-    /// Rehash the map without growing or shrinking.
-    pub fn shuffle(&self) {
-        let mut map = unsafe { self.shared_ptr.as_mut() }.unwrap().write();
-        let num_buckets = map.get_num_buckets() as u32;
-        let size_bytes = HashMapInit::<K, V, S>::estimate_size(num_buckets);
-        let end_ptr: *mut u8 = unsafe { self.shared_ptr.byte_add(size_bytes).cast() };
-        let buckets_ptr = map.buckets.as_mut_ptr();
-        self.rehash_dict(&mut map, buckets_ptr, end_ptr, num_buckets, num_buckets);
-    }
-
-    /// Grow the number of buckets within the table.
-    ///
-    /// 1. Grows the underlying shared memory area
-    /// 2. Initializes new buckets and overwrites the current dictionary
-    /// 3. Rehashes the dictionary
-    ///
-    /// # Panics
-    /// Panics if called on a map initialized with [`HashMapInit::with_fixed`].
-    ///
-    /// # Errors
-    /// Returns an [`shmem::Error`] if any errors occur resizing the memory region.
-    pub fn grow(&self, num_buckets: u32) -> Result<(), shmem::Error> {
-        let mut map = unsafe { self.shared_ptr.as_mut() }.unwrap().write();
-        let old_num_buckets = map.buckets.len() as u32;
-
-        assert!(
-            num_buckets >= old_num_buckets,
-            "grow called with a smaller number of buckets"
-        );
-        if num_buckets == old_num_buckets {
-            return Ok(());
-        }
-        let shmem_handle = self
-            .shmem_handle
-            .as_ref()
-            .expect("grow called on a fixed-size hash table");
-
-        let size_bytes = HashMapInit::<K, V, S>::estimate_size(num_buckets);
-        shmem_handle.set_size(size_bytes)?;
-        let end_ptr: *mut u8 = unsafe { shmem_handle.data_ptr.as_ptr().add(size_bytes) };
-
-        // Initialize new buckets. The new buckets are linked to the free list.
-        // NB: This overwrites the dictionary!
-        let buckets_ptr = map.buckets.as_mut_ptr();
-        unsafe {
-            for i in old_num_buckets..num_buckets {
-                let bucket = buckets_ptr.add(i as usize);
-                bucket.write(core::Bucket {
-                    next: if i < num_buckets - 1 {
-                        i + 1
-                    } else {
-                        map.free_head
-                    },
-                    inner: None,
-                });
-            }
-        }
-
-        self.rehash_dict(&mut map, buckets_ptr, end_ptr, num_buckets, old_num_buckets);
-        map.free_head = old_num_buckets;
-
-        Ok(())
-    }
-
-    /// Begin a shrink, limiting all new allocations to be in buckets with index below `num_buckets`.
-    ///
-    /// # Panics
-    /// Panics if called on a map initialized with [`HashMapInit::with_fixed`] or if `num_buckets` is
-    /// greater than the number of buckets in the map.
-    pub fn begin_shrink(&mut self, num_buckets: u32) {
-        let mut map = unsafe { self.shared_ptr.as_mut() }.unwrap().write();
-        assert!(
-            num_buckets <= map.get_num_buckets() as u32,
-            "shrink called with a larger number of buckets"
-        );
-        _ = self
-            .shmem_handle
-            .as_ref()
-            .expect("shrink called on a fixed-size hash table");
-        map.alloc_limit = num_buckets;
-    }
-
-    /// If a shrink operation is underway, returns the target size of the map. Otherwise, returns None.
-    pub fn shrink_goal(&self) -> Option<usize> {
-        let map = unsafe { self.shared_ptr.as_mut() }.unwrap().read();
-        let goal = map.alloc_limit;
-        if goal == INVALID_POS {
-            None
-        } else {
-            Some(goal as usize)
-        }
-    }
-
-    /// Complete a shrink after caller has evicted entries, removing the unused buckets and rehashing.
-    ///
-    /// # Panics
-    /// The following cases result in a panic:
-    /// - Calling this function on a map initialized with [`HashMapInit::with_fixed`].
-    /// - Calling this function on a map when no shrink operation is in progress.
-    /// - Calling this function on a map with `shrink_mode` set to [`HashMapShrinkMode::Remap`] and
-    ///   there are more buckets in use than the value returned by [`HashMapAccess::shrink_goal`].
-    ///
-    /// # Errors
-    /// Returns an [`shmem::Error`] if any errors occur resizing the memory region.
-    pub fn finish_shrink(&self) -> Result<(), shmem::Error> {
-        let mut map = unsafe { self.shared_ptr.as_mut() }.unwrap().write();
-        assert!(
-            map.alloc_limit != INVALID_POS,
-            "called finish_shrink when no shrink is in progress"
-        );
-
-        let num_buckets = map.alloc_limit;
-
-        if map.get_num_buckets() == num_buckets as usize {
-            return Ok(());
-        }
-
-        assert!(
-            map.buckets_in_use <= num_buckets,
-            "called finish_shrink before enough entries were removed"
-        );
-
-        for i in (num_buckets as usize)..map.buckets.len() {
-            if let Some((k, v)) = map.buckets[i].inner.take() {
-                // alloc_bucket increases count, so need to decrease since we're just moving
-                map.buckets_in_use -= 1;
-                map.alloc_bucket(k, v).unwrap();
-            }
-        }
-
-        let shmem_handle = self
-            .shmem_handle
-            .as_ref()
-            .expect("shrink called on a fixed-size hash table");
-
-        let size_bytes = HashMapInit::<K, V, S>::estimate_size(num_buckets);
-        shmem_handle.set_size(size_bytes)?;
-        let end_ptr: *mut u8 = unsafe { shmem_handle.data_ptr.as_ptr().add(size_bytes) };
-        let buckets_ptr = map.buckets.as_mut_ptr();
-        self.rehash_dict(&mut map, buckets_ptr, end_ptr, num_buckets, num_buckets);
-        map.alloc_limit = INVALID_POS;
-
-        Ok(())
-    }
-}

libs/neon-shmem/src/hash/core.rs

@@ -1,208 +0,0 @@
-//! Simple hash table with chaining.
-
-use std::fmt::Debug;
-use std::hash::Hash;
-use std::mem::MaybeUninit;
-
-use crate::hash::entry::*;
-
-/// Invalid position within the map (either within the dictionary or bucket array).
-pub(crate) const INVALID_POS: u32 = u32::MAX;
-
-/// Fundamental storage unit within the hash table. Either empty or contains a key-value pair.
-/// Always part of a chain of some kind (either a freelist if empty or a hash chain if full).
-pub(crate) struct Bucket<K, V> {
-    /// Index of next bucket in the chain.
-    pub(crate) next: u32,
-    /// Key-value pair contained within bucket.
-    pub(crate) inner: Option<(K, V)>,
-}
-
-impl<K, V> Debug for Bucket<K, V>
-where
-    K: Debug,
-    V: Debug,
-{
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        f.debug_struct("Bucket")
-            .field("next", &self.next)
-            .field("inner", &self.inner)
-            .finish()
-    }
-}
-
-/// Core hash table implementation.
-pub(crate) struct CoreHashMap<'a, K, V> {
-    /// Dictionary used to map hashes to bucket indices.
-    pub(crate) dictionary: &'a mut [u32],
-    /// Buckets containing key-value pairs.
-    pub(crate) buckets: &'a mut [Bucket<K, V>],
-    /// Head of the freelist.
-    pub(crate) free_head: u32,
-    /// Maximum index of a bucket allowed to be allocated. [`INVALID_POS`] if no limit.
-    pub(crate) alloc_limit: u32,
-    /// The number of currently occupied buckets.
-    pub(crate) buckets_in_use: u32,
-    // pub(crate) lock: libc::pthread_mutex_t,
-    // Unclear what the purpose of this is.
-    pub(crate) _user_list_head: u32,
-}
-
-impl<'a, K, V> Debug for CoreHashMap<'a, K, V>
-where
-    K: Debug,
-    V: Debug,
-{
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        f.debug_struct("CoreHashMap")
-            .field("dictionary", &self.dictionary)
-            .field("buckets", &self.buckets)
-            .field("free_head", &self.free_head)
-            .field("alloc_limit", &self.alloc_limit)
-            .field("buckets_in_use", &self.buckets_in_use)
-            .finish()
-    }
-}
-
-/// Error for when there are no empty buckets left but one is needed.
-#[derive(Debug, PartialEq)]
-pub struct FullError();
-
-impl<'a, K: Clone + Hash + Eq, V> CoreHashMap<'a, K, V> {
-    const FILL_FACTOR: f32 = 0.60;
-
-    /// Estimate the size of data contained within the the hash map.
-    pub fn estimate_size(num_buckets: u32) -> usize {
-        let mut size = 0;
-
-        // buckets
-        size += size_of::<Bucket<K, V>>() * num_buckets as usize;
-
-        // dictionary
-        size += (f32::ceil((size_of::<u32>() * num_buckets as usize) as f32 / Self::FILL_FACTOR))
-            as usize;
-
-        size
-    }
-
-    pub fn new(
-        buckets: &'a mut [MaybeUninit<Bucket<K, V>>],
-        dictionary: &'a mut [MaybeUninit<u32>],
-    ) -> Self {
-        // Initialize the buckets
-        for i in 0..buckets.len() {
-            buckets[i].write(Bucket {
-                next: if i < buckets.len() - 1 {
-                    i as u32 + 1
-                } else {
-                    INVALID_POS
-                },
-                inner: None,
-            });
-        }
-
-        // Initialize the dictionary
-        for e in dictionary.iter_mut() {
-            e.write(INVALID_POS);
-        }
-
-        // TODO: use std::slice::assume_init_mut() once it stabilizes
-        let buckets =
-            unsafe { std::slice::from_raw_parts_mut(buckets.as_mut_ptr().cast(), buckets.len()) };
-        let dictionary = unsafe {
-            std::slice::from_raw_parts_mut(dictionary.as_mut_ptr().cast(), dictionary.len())
-        };
-
-        Self {
-            dictionary,
-            buckets,
-            free_head: 0,
-            buckets_in_use: 0,
-            _user_list_head: INVALID_POS,
-            alloc_limit: INVALID_POS,
-        }
-    }
-
-    /// Get the value associated with a key (if it exists) given its hash.
-    pub fn get_with_hash(&self, key: &K, hash: u64) -> Option<&V> {
-        let mut next = self.dictionary[hash as usize % self.dictionary.len()];
-        loop {
-            if next == INVALID_POS {
-                return None;
-            }
-
-            let bucket = &self.buckets[next as usize];
-            let (bucket_key, bucket_value) = bucket.inner.as_ref().expect("entry is in use");
-            if bucket_key == key {
-                return Some(bucket_value);
-            }
-            next = bucket.next;
-        }
-    }
-
-    /// Get number of buckets in map.
-    pub fn get_num_buckets(&self) -> usize {
-        self.buckets.len()
-    }
-
-    /// Clears all entries from the hashmap.
-    ///
-    /// Does not reset any allocation limits, but does clear any entries beyond them.
-    pub fn clear(&mut self) {
-        for i in 0..self.buckets.len() {
-            self.buckets[i] = Bucket {
-                next: if i < self.buckets.len() - 1 {
-                    i as u32 + 1
-                } else {
-                    INVALID_POS
-                },
-                inner: None,
-            }
-        }
-        for i in 0..self.dictionary.len() {
-            self.dictionary[i] = INVALID_POS;
-        }
-
-        self.free_head = 0;
-        self.buckets_in_use = 0;
-    }
-
-    /// Find the position of an unused bucket via the freelist and initialize it.
-    pub(crate) fn alloc_bucket(&mut self, key: K, value: V) -> Result<u32, FullError> {
-        let mut pos = self.free_head;
-
-        // Find the first bucket we're *allowed* to use.
-        let mut prev = PrevPos::First(self.free_head);
-        while pos != INVALID_POS && pos >= self.alloc_limit {
-            let bucket = &mut self.buckets[pos as usize];
-            prev = PrevPos::Chained(pos);
-            pos = bucket.next;
-        }
-        if pos == INVALID_POS {
-            return Err(FullError());
-        }
-
-        // Repair the freelist.
-        match prev {
-            PrevPos::First(_) => {
-                let next_pos = self.buckets[pos as usize].next;
-                self.free_head = next_pos;
-            }
-            PrevPos::Chained(p) => {
-                if p != INVALID_POS {
-                    let next_pos = self.buckets[pos as usize].next;
-                    self.buckets[p as usize].next = next_pos;
-                }
-            }
-            _ => unreachable!(),
-        }
-
-        // Initialize the bucket.
-        let bucket = &mut self.buckets[pos as usize];
-        self.buckets_in_use += 1;
-        bucket.next = INVALID_POS;
-        bucket.inner = Some((key, value));
-
-        Ok(pos)
-    }
-}

libs/neon-shmem/src/hash/entry.rs

@@ -1,138 +0,0 @@
-//! Equivalent of [`std::collections::hash_map::Entry`] for this hashmap.
-
-use crate::hash::core::{CoreHashMap, FullError, INVALID_POS};
-use crate::sync::{RwLockWriteGuard, ValueWriteGuard};
-
-use std::hash::Hash;
-use std::mem;
-
-pub enum Entry<'a, 'b, K, V> {
-    Occupied(OccupiedEntry<'a, 'b, K, V>),
-    Vacant(VacantEntry<'a, 'b, K, V>),
-}
-
-/// Enum representing the previous position within a chain.
-#[derive(Clone, Copy)]
-pub(crate) enum PrevPos {
-    /// Starting index within the dictionary.  
-    First(u32),
-    /// Regular index within the buckets.
-    Chained(u32),
-    /// Unknown - e.g. the associated entry was retrieved by index instead of chain.
-    Unknown(u64),
-}
-
-pub struct OccupiedEntry<'a, 'b, K, V> {
-    /// Mutable reference to the map containing this entry.
-    pub(crate) map: RwLockWriteGuard<'b, CoreHashMap<'a, K, V>>,
-    /// The key of the occupied entry
-    pub(crate) _key: K,
-    /// The index of the previous entry in the chain.
-    pub(crate) prev_pos: PrevPos,
-    /// The position of the bucket in the [`CoreHashMap`] bucket array.
-    pub(crate) bucket_pos: u32,
-}
-
-impl<K, V> OccupiedEntry<'_, '_, K, V> {
-    pub fn get(&self) -> &V {
-        &self.map.buckets[self.bucket_pos as usize]
-            .inner
-            .as_ref()
-            .unwrap()
-            .1
-    }
-
-    pub fn get_mut(&mut self) -> &mut V {
-        &mut self.map.buckets[self.bucket_pos as usize]
-            .inner
-            .as_mut()
-            .unwrap()
-            .1
-    }
-
-    /// Inserts a value into the entry, replacing (and returning) the existing value.
-    pub fn insert(&mut self, value: V) -> V {
-        let bucket = &mut self.map.buckets[self.bucket_pos as usize];
-        // This assumes inner is Some, which it must be for an OccupiedEntry
-        mem::replace(&mut bucket.inner.as_mut().unwrap().1, value)
-    }
-
-    /// Removes the entry from the hash map, returning the value originally stored within it.
-    ///
-    /// This may result in multiple bucket accesses if the entry was obtained by index as the
-    /// previous chain entry needs to be discovered in this case.
-    ///
-    /// # Panics
-    /// Panics if the `prev_pos` field is equal to [`PrevPos::Unknown`]. In practice, this means
-    /// the entry was obtained via calling something like [`CoreHashMap::entry_at_bucket`].
-    pub fn remove(mut self) -> V {
-        // If this bucket was queried by index, go ahead and follow its chain from the start.
-        let prev = if let PrevPos::Unknown(hash) = self.prev_pos {
-            let dict_idx = hash as usize % self.map.dictionary.len();
-            let mut prev = PrevPos::First(dict_idx as u32);
-            let mut curr = self.map.dictionary[dict_idx];
-            while curr != self.bucket_pos {
-                assert!(curr != INVALID_POS);
-                prev = PrevPos::Chained(curr);
-                curr = self.map.buckets[curr as usize].next;
-            }
-            prev
-        } else {
-            self.prev_pos
-        };
-
-        // CoreHashMap::remove returns Option<(K, V)>. We know it's Some for an OccupiedEntry.
-        let bucket = &mut self.map.buckets[self.bucket_pos as usize];
-
-        // unlink it from the chain
-        match prev {
-            PrevPos::First(dict_pos) => {
-                self.map.dictionary[dict_pos as usize] = bucket.next;
-            }
-            PrevPos::Chained(bucket_pos) => {
-                // println!("we think prev of {} is {bucket_pos}", self.bucket_pos);
-                self.map.buckets[bucket_pos as usize].next = bucket.next;
-            }
-            _ => unreachable!(),
-        }
-
-        // and add it to the freelist
-        let free = self.map.free_head;
-        let bucket = &mut self.map.buckets[self.bucket_pos as usize];
-        let old_value = bucket.inner.take();
-        bucket.next = free;
-        self.map.free_head = self.bucket_pos;
-        self.map.buckets_in_use -= 1;
-
-        old_value.unwrap().1
-    }
-}
-
-/// An abstract view into a vacant entry within the map.
-pub struct VacantEntry<'a, 'b, K, V> {
-    /// Mutable reference to the map containing this entry.
-    pub(crate) map: RwLockWriteGuard<'b, CoreHashMap<'a, K, V>>,
-    /// The key to be inserted into this entry.
-    pub(crate) key: K,
-    /// The position within the dictionary corresponding to the key's hash.
-    pub(crate) dict_pos: u32,
-}
-
-impl<'b, K: Clone + Hash + Eq, V> VacantEntry<'_, 'b, K, V> {
-    /// Insert a value into the vacant entry, finding and populating an empty bucket in the process.
-    ///
-    /// # Errors
-    /// Will return [`FullError`] if there are no unoccupied buckets in the map.
-    pub fn insert(mut self, value: V) -> Result<ValueWriteGuard<'b, V>, FullError> {
-        let pos = self.map.alloc_bucket(self.key, value)?;
-        if pos == INVALID_POS {
-            return Err(FullError());
-        }
-        self.map.buckets[pos as usize].next = self.map.dictionary[self.dict_pos as usize];
-        self.map.dictionary[self.dict_pos as usize] = pos;
-
-        Ok(RwLockWriteGuard::map(self.map, |m| {
-            &mut m.buckets[pos as usize].inner.as_mut().unwrap().1
-        }))
-    }
-}

libs/neon-shmem/src/hash/tests.rs

@@ -1,429 +0,0 @@
-use std::collections::BTreeMap;
-use std::collections::HashSet;
-use std::fmt::Debug;
-use std::mem::MaybeUninit;
-
-use crate::hash::Entry;
-use crate::hash::HashMapAccess;
-use crate::hash::HashMapInit;
-use crate::hash::core::FullError;
-
-use rand::seq::SliceRandom;
-use rand::{Rng, RngCore};
-use rand_distr::Zipf;
-
-const TEST_KEY_LEN: usize = 16;
-
-#[derive(Clone, Copy, Debug, Hash, PartialEq, Eq, PartialOrd, Ord)]
-struct TestKey([u8; TEST_KEY_LEN]);
-
-impl From<&TestKey> for u128 {
-    fn from(val: &TestKey) -> u128 {
-        u128::from_be_bytes(val.0)
-    }
-}
-
-impl From<u128> for TestKey {
-    fn from(val: u128) -> TestKey {
-        TestKey(val.to_be_bytes())
-    }
-}
-
-impl<'a> From<&'a [u8]> for TestKey {
-    fn from(bytes: &'a [u8]) -> TestKey {
-        TestKey(bytes.try_into().unwrap())
-    }
-}
-
-fn test_inserts<K: Into<TestKey> + Copy>(keys: &[K]) {
-    let w = HashMapInit::<TestKey, usize>::new_resizeable_named(100000, 120000, "test_inserts")
-        .attach_writer();
-
-    for (idx, k) in keys.iter().enumerate() {
-        let res = w.entry((*k).into());
-        match res {
-            Entry::Occupied(mut e) => {
-                e.insert(idx);
-            }
-            Entry::Vacant(e) => {
-                let res = e.insert(idx);
-                assert!(res.is_ok());
-            }
-        };
-    }
-
-    for (idx, k) in keys.iter().enumerate() {
-        let x = w.get(&(*k).into());
-        let value = x.as_deref().copied();
-        assert_eq!(value, Some(idx));
-    }
-}
-
-#[test]
-fn dense() {
-    // This exercises splitting a node with prefix
-    let keys: &[u128] = &[0, 1, 2, 3, 256];
-    test_inserts(keys);
-
-    // Dense keys
-    let mut keys: Vec<u128> = (0..10000).collect();
-    test_inserts(&keys);
-
-    // Do the same in random orders
-    for _ in 1..10 {
-        keys.shuffle(&mut rand::rng());
-        test_inserts(&keys);
-    }
-}
-
-#[test]
-fn sparse() {
-    // sparse keys
-    let mut keys: Vec<TestKey> = Vec::new();
-    let mut used_keys = HashSet::new();
-    for _ in 0..10000 {
-        loop {
-            let key = rand::random::<u128>();
-            if used_keys.contains(&key) {
-                continue;
-            }
-            used_keys.insert(key);
-            keys.push(key.into());
-            break;
-        }
-    }
-    test_inserts(&keys);
-}
-
-#[derive(Clone, Debug)]
-struct TestOp(TestKey, Option<usize>);
-
-fn apply_op(
-    op: &TestOp,
-    map: &mut HashMapAccess<TestKey, usize>,
-    shadow: &mut BTreeMap<TestKey, usize>,
-) {
-    // apply the change to the shadow tree first
-    let shadow_existing = if let Some(v) = op.1 {
-        shadow.insert(op.0, v)
-    } else {
-        shadow.remove(&op.0)
-    };
-
-    let entry = map.entry(op.0);
-    let hash_existing = match op.1 {
-        Some(new) => match entry {
-            Entry::Occupied(mut e) => Some(e.insert(new)),
-            Entry::Vacant(e) => {
-                _ = e.insert(new).unwrap();
-                None
-            }
-        },
-        None => match entry {
-            Entry::Occupied(e) => Some(e.remove()),
-            Entry::Vacant(_) => None,
-        },
-    };
-
-    assert_eq!(shadow_existing, hash_existing);
-}
-
-fn do_random_ops(
-    num_ops: usize,
-    size: u32,
-    del_prob: f64,
-    writer: &mut HashMapAccess<TestKey, usize>,
-    shadow: &mut BTreeMap<TestKey, usize>,
-    rng: &mut rand::rngs::ThreadRng,
-) {
-    for i in 0..num_ops {
-        let key: TestKey = ((rng.next_u32() % size) as u128).into();
-        let op = TestOp(
-            key,
-            if rng.random_bool(del_prob) {
-                Some(i)
-            } else {
-                None
-            },
-        );
-        apply_op(&op, writer, shadow);
-    }
-}
-
-fn do_deletes(
-    num_ops: usize,
-    writer: &mut HashMapAccess<TestKey, usize>,
-    shadow: &mut BTreeMap<TestKey, usize>,
-) {
-    for _ in 0..num_ops {
-        let (k, _) = shadow.pop_first().unwrap();
-        writer.remove(&k);
-    }
-}
-
-fn do_shrink(
-    writer: &mut HashMapAccess<TestKey, usize>,
-    shadow: &mut BTreeMap<TestKey, usize>,
-    to: u32,
-) {
-    assert!(writer.shrink_goal().is_none());
-    writer.begin_shrink(to);
-    assert_eq!(writer.shrink_goal(), Some(to as usize));
-    while writer.get_num_buckets_in_use() > to as usize {
-        let (k, _) = shadow.pop_first().unwrap();
-        let entry = writer.entry(k);
-        if let Entry::Occupied(e) = entry {
-            e.remove();
-        }
-    }
-    let old_usage = writer.get_num_buckets_in_use();
-    writer.finish_shrink().unwrap();
-    assert!(writer.shrink_goal().is_none());
-    assert_eq!(writer.get_num_buckets_in_use(), old_usage);
-}
-
-#[test]
-fn random_ops() {
-    let mut writer =
-        HashMapInit::<TestKey, usize>::new_resizeable_named(100000, 120000, "test_random")
-            .attach_writer();
-    let mut shadow: std::collections::BTreeMap<TestKey, usize> = BTreeMap::new();
-
-    let distribution = Zipf::new(u128::MAX as f64, 1.1).unwrap();
-    let mut rng = rand::rng();
-    for i in 0..100000 {
-        let key: TestKey = (rng.sample(distribution) as u128).into();
-
-        let op = TestOp(key, if rng.random_bool(0.75) { Some(i) } else { None });
-
-        apply_op(&op, &mut writer, &mut shadow);
-    }
-}
-
-#[test]
-fn test_shuffle() {
-    let mut writer = HashMapInit::<TestKey, usize>::new_resizeable_named(1000, 1200, "test_shuf")
-        .attach_writer();
-    let mut shadow: std::collections::BTreeMap<TestKey, usize> = BTreeMap::new();
-    let mut rng = rand::rng();
-
-    do_random_ops(10000, 1000, 0.75, &mut writer, &mut shadow, &mut rng);
-    writer.shuffle();
-    do_random_ops(10000, 1000, 0.75, &mut writer, &mut shadow, &mut rng);
-}
-
-#[test]
-fn test_grow() {
-    let mut writer = HashMapInit::<TestKey, usize>::new_resizeable_named(1000, 2000, "test_grow")
-        .attach_writer();
-    let mut shadow: std::collections::BTreeMap<TestKey, usize> = BTreeMap::new();
-    let mut rng = rand::rng();
-
-    do_random_ops(10000, 1000, 0.75, &mut writer, &mut shadow, &mut rng);
-    let old_usage = writer.get_num_buckets_in_use();
-    writer.grow(1500).unwrap();
-    assert_eq!(writer.get_num_buckets_in_use(), old_usage);
-    assert_eq!(writer.get_num_buckets(), 1500);
-    do_random_ops(10000, 1500, 0.75, &mut writer, &mut shadow, &mut rng);
-}
-
-#[test]
-fn test_clear() {
-    let mut writer = HashMapInit::<TestKey, usize>::new_resizeable_named(1500, 2000, "test_clear")
-        .attach_writer();
-    let mut shadow: std::collections::BTreeMap<TestKey, usize> = BTreeMap::new();
-    let mut rng = rand::rng();
-    do_random_ops(2000, 1500, 0.75, &mut writer, &mut shadow, &mut rng);
-    writer.clear();
-    assert_eq!(writer.get_num_buckets_in_use(), 0);
-    assert_eq!(writer.get_num_buckets(), 1500);
-    while let Some((key, _)) = shadow.pop_first() {
-        assert!(writer.get(&key).is_none());
-    }
-    do_random_ops(2000, 1500, 0.75, &mut writer, &mut shadow, &mut rng);
-    for i in 0..(1500 - writer.get_num_buckets_in_use()) {
-        writer.insert((1500 + i as u128).into(), 0).unwrap();
-    }
-    assert_eq!(writer.insert(5000.into(), 0), Err(FullError {}));
-    writer.clear();
-    assert!(writer.insert(5000.into(), 0).is_ok());
-}
-
-#[test]
-fn test_idx_remove() {
-    let mut writer = HashMapInit::<TestKey, usize>::new_resizeable_named(1500, 2000, "test_clear")
-        .attach_writer();
-    let mut shadow: std::collections::BTreeMap<TestKey, usize> = BTreeMap::new();
-    let mut rng = rand::rng();
-    do_random_ops(2000, 1500, 0.25, &mut writer, &mut shadow, &mut rng);
-    for _ in 0..100 {
-        let idx = (rng.next_u32() % 1500) as usize;
-        if let Some(e) = writer.entry_at_bucket(idx) {
-            shadow.remove(&e._key);
-            e.remove();
-        }
-    }
-    while let Some((key, val)) = shadow.pop_first() {
-        assert_eq!(*writer.get(&key).unwrap(), val);
-    }
-}
-
-#[test]
-fn test_idx_get() {
-    let mut writer = HashMapInit::<TestKey, usize>::new_resizeable_named(1500, 2000, "test_clear")
-        .attach_writer();
-    let mut shadow: std::collections::BTreeMap<TestKey, usize> = BTreeMap::new();
-    let mut rng = rand::rng();
-    do_random_ops(2000, 1500, 0.25, &mut writer, &mut shadow, &mut rng);
-    for _ in 0..100 {
-        let idx = (rng.next_u32() % 1500) as usize;
-        if let Some(pair) = writer.get_at_bucket(idx) {
-            {
-                let v: *const usize = &pair.1;
-                assert_eq!(writer.get_bucket_for_value(v), idx);
-            }
-            {
-                let v: *const usize = &pair.1;
-                assert_eq!(writer.get_bucket_for_value(v), idx);
-            }
-        }
-    }
-}
-
-#[test]
-fn test_shrink() {
-    let mut writer = HashMapInit::<TestKey, usize>::new_resizeable_named(1500, 2000, "test_shrink")
-        .attach_writer();
-    let mut shadow: std::collections::BTreeMap<TestKey, usize> = BTreeMap::new();
-    let mut rng = rand::rng();
-
-    do_random_ops(10000, 1500, 0.75, &mut writer, &mut shadow, &mut rng);
-    do_shrink(&mut writer, &mut shadow, 1000);
-    assert_eq!(writer.get_num_buckets(), 1000);
-    do_deletes(500, &mut writer, &mut shadow);
-    do_random_ops(10000, 500, 0.75, &mut writer, &mut shadow, &mut rng);
-    assert!(writer.get_num_buckets_in_use() <= 1000);
-}
-
-#[test]
-fn test_shrink_grow_seq() {
-    let mut writer =
-        HashMapInit::<TestKey, usize>::new_resizeable_named(1000, 20000, "test_grow_seq")
-            .attach_writer();
-    let mut shadow: std::collections::BTreeMap<TestKey, usize> = BTreeMap::new();
-    let mut rng = rand::rng();
-
-    do_random_ops(500, 1000, 0.1, &mut writer, &mut shadow, &mut rng);
-    eprintln!("Shrinking to 750");
-    do_shrink(&mut writer, &mut shadow, 750);
-    do_random_ops(200, 1000, 0.5, &mut writer, &mut shadow, &mut rng);
-    eprintln!("Growing to 1500");
-    writer.grow(1500).unwrap();
-    do_random_ops(600, 1500, 0.1, &mut writer, &mut shadow, &mut rng);
-    eprintln!("Shrinking to 200");
-    while shadow.len() > 100 {
-        do_deletes(1, &mut writer, &mut shadow);
-    }
-    do_shrink(&mut writer, &mut shadow, 200);
-    do_random_ops(50, 1500, 0.25, &mut writer, &mut shadow, &mut rng);
-    eprintln!("Growing to 10k");
-    writer.grow(10000).unwrap();
-    do_random_ops(10000, 5000, 0.25, &mut writer, &mut shadow, &mut rng);
-}
-
-#[test]
-fn test_bucket_ops() {
-    let writer = HashMapInit::<TestKey, usize>::new_resizeable_named(1000, 1200, "test_bucket_ops")
-        .attach_writer();
-    match writer.entry(1.into()) {
-        Entry::Occupied(mut e) => {
-            e.insert(2);
-        }
-        Entry::Vacant(e) => {
-            _ = e.insert(2).unwrap();
-        }
-    }
-    assert_eq!(writer.get_num_buckets_in_use(), 1);
-    assert_eq!(writer.get_num_buckets(), 1000);
-    assert_eq!(*writer.get(&1.into()).unwrap(), 2);
-    let pos = match writer.entry(1.into()) {
-        Entry::Occupied(e) => {
-            assert_eq!(e._key, 1.into());
-            let pos = e.bucket_pos as usize;
-            pos
-        }
-        Entry::Vacant(_) => {
-            panic!("Insert didn't affect entry");
-        }
-    };
-    assert_eq!(writer.entry_at_bucket(pos).unwrap()._key, 1.into());
-    assert_eq!(*writer.get_at_bucket(pos).unwrap(), (1.into(), 2));
-    {
-        let ptr: *const usize = &*writer.get(&1.into()).unwrap();
-        assert_eq!(writer.get_bucket_for_value(ptr), pos);
-    }
-    writer.remove(&1.into());
-    assert!(writer.get(&1.into()).is_none());
-}
-
-#[test]
-fn test_shrink_zero() {
-    let mut writer =
-        HashMapInit::<TestKey, usize>::new_resizeable_named(1500, 2000, "test_shrink_zero")
-            .attach_writer();
-    writer.begin_shrink(0);
-    for i in 0..1500 {
-        writer.entry_at_bucket(i).map(|x| x.remove());
-    }
-    writer.finish_shrink().unwrap();
-    assert_eq!(writer.get_num_buckets_in_use(), 0);
-    let entry = writer.entry(1.into());
-    if let Entry::Vacant(v) = entry {
-        assert!(v.insert(2).is_err());
-    } else {
-        panic!("Somehow got non-vacant entry in empty map.")
-    }
-    writer.grow(50).unwrap();
-    let entry = writer.entry(1.into());
-    if let Entry::Vacant(v) = entry {
-        assert!(v.insert(2).is_ok());
-    } else {
-        panic!("Somehow got non-vacant entry in empty map.")
-    }
-    assert_eq!(writer.get_num_buckets_in_use(), 1);
-}
-
-#[test]
-#[should_panic]
-fn test_grow_oom() {
-    let writer = HashMapInit::<TestKey, usize>::new_resizeable_named(1500, 2000, "test_grow_oom")
-        .attach_writer();
-    writer.grow(20000).unwrap();
-}
-
-#[test]
-#[should_panic]
-fn test_shrink_bigger() {
-    let mut writer =
-        HashMapInit::<TestKey, usize>::new_resizeable_named(1500, 2500, "test_shrink_bigger")
-            .attach_writer();
-    writer.begin_shrink(2000);
-}
-
-#[test]
-#[should_panic]
-fn test_shrink_early_finish() {
-    let writer =
-        HashMapInit::<TestKey, usize>::new_resizeable_named(1500, 2500, "test_shrink_early_finish")
-            .attach_writer();
-    writer.finish_shrink().unwrap();
-}
-
-#[test]
-#[should_panic]
-fn test_shrink_fixed_size() {
-    let mut area = [MaybeUninit::uninit(); 10000];
-    let init_struct = HashMapInit::<TestKey, usize>::with_fixed(3, &mut area);
-    let mut writer = init_struct.attach_writer();
-    writer.begin_shrink(1);
-}

libs/neon-shmem/src/lib.rs

@@ -1,5 +1,418 @@
 //! Shared memory utilities for neon communicator
 
-pub mod hash;
-pub mod shmem;
-pub mod sync;
+use std::num::NonZeroUsize;
+use std::os::fd::{AsFd, BorrowedFd, OwnedFd};
+use std::ptr::NonNull;
+use std::sync::atomic::{AtomicUsize, Ordering};
+
+use nix::errno::Errno;
+use nix::sys::mman::MapFlags;
+use nix::sys::mman::ProtFlags;
+use nix::sys::mman::mmap as nix_mmap;
+use nix::sys::mman::munmap as nix_munmap;
+use nix::unistd::ftruncate as nix_ftruncate;
+
+/// ShmemHandle represents a shared memory area that can be shared by processes over fork().
+/// Unlike shared memory allocated by Postgres, this area is resizable, up to 'max_size' that's
+/// specified at creation.
+///
+/// The area is backed by an anonymous file created with memfd_create(). The full address space for
+/// 'max_size' is reserved up-front with mmap(), but whenever you call [`ShmemHandle::set_size`],
+/// the underlying file is resized. Do not access the area beyond the current size. Currently, that
+/// will cause the file to be expanded, but we might use mprotect() etc. to enforce that in the
+/// future.
+pub struct ShmemHandle {
+    /// memfd file descriptor
+    fd: OwnedFd,
+
+    max_size: usize,
+
+    // Pointer to the beginning of the shared memory area. The header is stored there.
+    shared_ptr: NonNull<SharedStruct>,
+
+    // Pointer to the beginning of the user data
+    pub data_ptr: NonNull<u8>,
+}
+
+/// This is stored at the beginning in the shared memory area.
+struct SharedStruct {
+    max_size: usize,
+
+    /// Current size of the backing file. The high-order bit is used for the RESIZE_IN_PROGRESS flag
+    current_size: AtomicUsize,
+}
+
+const RESIZE_IN_PROGRESS: usize = 1 << 63;
+
+const HEADER_SIZE: usize = std::mem::size_of::<SharedStruct>();
+
+/// Error type returned by the ShmemHandle functions.
+#[derive(thiserror::Error, Debug)]
+#[error("{msg}: {errno}")]
+pub struct Error {
+    pub msg: String,
+    pub errno: Errno,
+}
+
+impl Error {
+    fn new(msg: &str, errno: Errno) -> Error {
+        Error {
+            msg: msg.to_string(),
+            errno,
+        }
+    }
+}
+
+impl ShmemHandle {
+    /// Create a new shared memory area. To communicate between processes, the processes need to be
+    /// fork()'d after calling this, so that the ShmemHandle is inherited by all processes.
+    ///
+    /// If the ShmemHandle is dropped, the memory is unmapped from the current process. Other
+    /// processes can continue using it, however.
+    pub fn new(name: &str, initial_size: usize, max_size: usize) -> Result<ShmemHandle, Error> {
+        // create the backing anonymous file.
+        let fd = create_backing_file(name)?;
+
+        Self::new_with_fd(fd, initial_size, max_size)
+    }
+
+    fn new_with_fd(
+        fd: OwnedFd,
+        initial_size: usize,
+        max_size: usize,
+    ) -> Result<ShmemHandle, Error> {
+        // We reserve the high-order bit for the RESIZE_IN_PROGRESS flag, and the actual size
+        // is a little larger than this because of the SharedStruct header. Make the upper limit
+        // somewhat smaller than that, because with anything close to that, you'll run out of
+        // memory anyway.
+        if max_size >= 1 << 48 {
+            panic!("max size {max_size} too large");
+        }
+        if initial_size > max_size {
+            panic!("initial size {initial_size} larger than max size {max_size}");
+        }
+
+        // The actual initial / max size is the one given by the caller, plus the size of
+        // 'SharedStruct'.
+        let initial_size = HEADER_SIZE + initial_size;
+        let max_size = NonZeroUsize::new(HEADER_SIZE + max_size).unwrap();
+
+        // Reserve address space for it with mmap
+        //
+        // TODO: Use MAP_HUGETLB if possible
+        let start_ptr = unsafe {
+            nix_mmap(
+                None,
+                max_size,
+                ProtFlags::PROT_READ | ProtFlags::PROT_WRITE,
+                MapFlags::MAP_SHARED,
+                &fd,
+                0,
+            )
+        }
+        .map_err(|e| Error::new("mmap failed: {e}", e))?;
+
+        // Reserve space for the initial size
+        enlarge_file(fd.as_fd(), initial_size as u64)?;
+
+        // Initialize the header
+        let shared: NonNull<SharedStruct> = start_ptr.cast();
+        unsafe {
+            shared.write(SharedStruct {
+                max_size: max_size.into(),
+                current_size: AtomicUsize::new(initial_size),
+            })
+        };
+
+        // The user data begins after the header
+        let data_ptr = unsafe { start_ptr.cast().add(HEADER_SIZE) };
+
+        Ok(ShmemHandle {
+            fd,
+            max_size: max_size.into(),
+            shared_ptr: shared,
+            data_ptr,
+        })
+    }
+
+    // return reference to the header
+    fn shared(&self) -> &SharedStruct {
+        unsafe { self.shared_ptr.as_ref() }
+    }
+
+    /// Resize the shared memory area. 'new_size' must not be larger than the 'max_size' specified
+    /// when creating the area.
+    ///
+    /// This may only be called from one process/thread concurrently. We detect that case
+    /// and return an Error.
+    pub fn set_size(&self, new_size: usize) -> Result<(), Error> {
+        let new_size = new_size + HEADER_SIZE;
+        let shared = self.shared();
+
+        if new_size > self.max_size {
+            panic!(
+                "new size ({} is greater than max size ({})",
+                new_size, self.max_size
+            );
+        }
+        assert_eq!(self.max_size, shared.max_size);
+
+        // Lock the area by setting the bit in 'current_size'
+        //
+        // Ordering::Relaxed would probably be sufficient here, as we don't access any other memory
+        // and the posix_fallocate/ftruncate call is surely a synchronization point anyway. But
+        // since this is not performance-critical, better safe than sorry .
+        let mut old_size = shared.current_size.load(Ordering::Acquire);
+        loop {
+            if (old_size & RESIZE_IN_PROGRESS) != 0 {
+                return Err(Error::new(
+                    "concurrent resize detected",
+                    Errno::UnknownErrno,
+                ));
+            }
+            match shared.current_size.compare_exchange(
+                old_size,
+                new_size,
+                Ordering::Acquire,
+                Ordering::Relaxed,
+            ) {
+                Ok(_) => break,
+                Err(x) => old_size = x,
+            }
+        }
+
+        // Ok, we got the lock.
+        //
+        // NB: If anything goes wrong, we *must* clear the bit!
+        let result = {
+            use std::cmp::Ordering::{Equal, Greater, Less};
+            match new_size.cmp(&old_size) {
+                Less => nix_ftruncate(&self.fd, new_size as i64).map_err(|e| {
+                    Error::new("could not shrink shmem segment, ftruncate failed: {e}", e)
+                }),
+                Equal => Ok(()),
+                Greater => enlarge_file(self.fd.as_fd(), new_size as u64),
+            }
+        };
+
+        // Unlock
+        shared.current_size.store(
+            if result.is_ok() { new_size } else { old_size },
+            Ordering::Release,
+        );
+
+        result
+    }
+
+    /// Returns the current user-visible size of the shared memory segment.
+    ///
+    /// NOTE: a concurrent set_size() call can change the size at any time. It is the caller's
+    /// responsibility not to access the area beyond the current size.
+    pub fn current_size(&self) -> usize {
+        let total_current_size =
+            self.shared().current_size.load(Ordering::Relaxed) & !RESIZE_IN_PROGRESS;
+        total_current_size - HEADER_SIZE
+    }
+}
+
+impl Drop for ShmemHandle {
+    fn drop(&mut self) {
+        // SAFETY: The pointer was obtained from mmap() with the given size.
+        // We unmap the entire region.
+        let _ = unsafe { nix_munmap(self.shared_ptr.cast(), self.max_size) };
+        // The fd is dropped automatically by OwnedFd.
+    }
+}
+
+/// Create a "backing file" for the shared memory area. On Linux, use memfd_create(), to create an
+/// anonymous in-memory file. One macos, fall back to a regular file. That's good enough for
+/// development and testing, but in production we want the file to stay in memory.
+///
+/// disable 'unused_variables' warnings, because in the macos path, 'name' is unused.
+#[allow(unused_variables)]
+fn create_backing_file(name: &str) -> Result<OwnedFd, Error> {
+    #[cfg(not(target_os = "macos"))]
+    {
+        nix::sys::memfd::memfd_create(name, nix::sys::memfd::MFdFlags::empty())
+            .map_err(|e| Error::new("memfd_create failed: {e}", e))
+    }
+    #[cfg(target_os = "macos")]
+    {
+        let file = tempfile::tempfile().map_err(|e| {
+            Error::new(
+                "could not create temporary file to back shmem area: {e}",
+                nix::errno::Errno::from_raw(e.raw_os_error().unwrap_or(0)),
+            )
+        })?;
+        Ok(OwnedFd::from(file))
+    }
+}
+
+fn enlarge_file(fd: BorrowedFd, size: u64) -> Result<(), Error> {
+    // Use posix_fallocate() to enlarge the file. It reserves the space correctly, so that
+    // we don't get a segfault later when trying to actually use it.
+    #[cfg(not(target_os = "macos"))]
+    {
+        nix::fcntl::posix_fallocate(fd, 0, size as i64).map_err(|e| {
+            Error::new(
+                "could not grow shmem segment, posix_fallocate failed: {e}",
+                e,
+            )
+        })
+    }
+    // As a fallback on macos, which doesn't have posix_fallocate, use plain 'fallocate'
+    #[cfg(target_os = "macos")]
+    {
+        nix::unistd::ftruncate(fd, size as i64)
+            .map_err(|e| Error::new("could not grow shmem segment, ftruncate failed: {e}", e))
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    use nix::unistd::ForkResult;
+    use std::ops::Range;
+
+    /// check that all bytes in given range have the expected value.
+    fn assert_range(ptr: *const u8, expected: u8, range: Range<usize>) {
+        for i in range {
+            let b = unsafe { *(ptr.add(i)) };
+            assert_eq!(expected, b, "unexpected byte at offset {i}");
+        }
+    }
+
+    /// Write 'b' to all bytes in the given range
+    fn write_range(ptr: *mut u8, b: u8, range: Range<usize>) {
+        unsafe { std::ptr::write_bytes(ptr.add(range.start), b, range.end - range.start) };
+    }
+
+    // simple single-process test of growing and shrinking
+    #[test]
+    fn test_shmem_resize() -> Result<(), Error> {
+        let max_size = 1024 * 1024;
+        let init_struct = ShmemHandle::new("test_shmem_resize", 0, max_size)?;
+
+        assert_eq!(init_struct.current_size(), 0);
+
+        // Initial grow
+        let size1 = 10000;
+        init_struct.set_size(size1).unwrap();
+        assert_eq!(init_struct.current_size(), size1);
+
+        // Write some data
+        let data_ptr = init_struct.data_ptr.as_ptr();
+        write_range(data_ptr, 0xAA, 0..size1);
+        assert_range(data_ptr, 0xAA, 0..size1);
+
+        // Shrink
+        let size2 = 5000;
+        init_struct.set_size(size2).unwrap();
+        assert_eq!(init_struct.current_size(), size2);
+
+        // Grow again
+        let size3 = 20000;
+        init_struct.set_size(size3).unwrap();
+        assert_eq!(init_struct.current_size(), size3);
+
+        // Try to read it. The area that was shrunk and grown again should read as all zeros now
+        assert_range(data_ptr, 0xAA, 0..5000);
+        assert_range(data_ptr, 0, 5000..size1);
+
+        // Try to grow beyond max_size
+        //let size4 = max_size + 1;
+        //assert!(init_struct.set_size(size4).is_err());
+
+        // Dropping init_struct should unmap the memory
+        drop(init_struct);
+
+        Ok(())
+    }
+
+    /// This is used in tests to coordinate between test processes. It's like std::sync::Barrier,
+    /// but is stored in the shared memory area and works across processes. It's implemented by
+    /// polling, because e.g. standard rust mutexes are not guaranteed to work across processes.
+    struct SimpleBarrier {
+        num_procs: usize,
+        count: AtomicUsize,
+    }
+
+    impl SimpleBarrier {
+        unsafe fn init(ptr: *mut SimpleBarrier, num_procs: usize) {
+            unsafe {
+                *ptr = SimpleBarrier {
+                    num_procs,
+                    count: AtomicUsize::new(0),
+                }
+            }
+        }
+
+        pub fn wait(&self) {
+            let old = self.count.fetch_add(1, Ordering::Relaxed);
+
+            let generation = old / self.num_procs;
+
+            let mut current = old + 1;
+            while current < (generation + 1) * self.num_procs {
+                std::thread::sleep(std::time::Duration::from_millis(10));
+                current = self.count.load(Ordering::Relaxed);
+            }
+        }
+    }
+
+    #[test]
+    fn test_multi_process() {
+        // Initialize
+        let max_size = 1_000_000_000_000;
+        let init_struct = ShmemHandle::new("test_multi_process", 0, max_size).unwrap();
+        let ptr = init_struct.data_ptr.as_ptr();
+
+        // Store the SimpleBarrier in the first 1k of the area.
+        init_struct.set_size(10000).unwrap();
+        let barrier_ptr: *mut SimpleBarrier = unsafe {
+            ptr.add(ptr.align_offset(std::mem::align_of::<SimpleBarrier>()))
+                .cast()
+        };
+        unsafe { SimpleBarrier::init(barrier_ptr, 2) };
+        let barrier = unsafe { barrier_ptr.as_ref().unwrap() };
+
+        // Fork another test process. The code after this runs in both processes concurrently.
+        let fork_result = unsafe { nix::unistd::fork().unwrap() };
+
+        // In the parent, fill bytes between 1000..2000. In the child, between 2000..3000
+        if fork_result.is_parent() {
+            write_range(ptr, 0xAA, 1000..2000);
+        } else {
+            write_range(ptr, 0xBB, 2000..3000);
+        }
+        barrier.wait();
+        // Verify the contents. (in both processes)
+        assert_range(ptr, 0xAA, 1000..2000);
+        assert_range(ptr, 0xBB, 2000..3000);
+
+        // Grow, from the child this time
+        let size = 10_000_000;
+        if !fork_result.is_parent() {
+            init_struct.set_size(size).unwrap();
+        }
+        barrier.wait();
+
+        // make some writes at the end
+        if fork_result.is_parent() {
+            write_range(ptr, 0xAA, (size - 10)..size);
+        } else {
+            write_range(ptr, 0xBB, (size - 20)..(size - 10));
+        }
+        barrier.wait();
+
+        // Verify the contents. (This runs in both processes)
+        assert_range(ptr, 0, (size - 1000)..(size - 20));
+        assert_range(ptr, 0xBB, (size - 20)..(size - 10));
+        assert_range(ptr, 0xAA, (size - 10)..size);
+
+        if let ForkResult::Parent { child } = fork_result {
+            nix::sys::wait::waitpid(child, None).unwrap();
+        }
+    }
+}

libs/neon-shmem/src/shmem.rs

@@ -1,411 +0,0 @@
-//! Dynamically resizable contiguous chunk of shared memory
-
-use std::num::NonZeroUsize;
-use std::os::fd::{AsFd, BorrowedFd, OwnedFd};
-use std::ptr::NonNull;
-use std::sync::atomic::{AtomicUsize, Ordering};
-
-use nix::errno::Errno;
-use nix::sys::mman::MapFlags;
-use nix::sys::mman::ProtFlags;
-use nix::sys::mman::mmap as nix_mmap;
-use nix::sys::mman::munmap as nix_munmap;
-use nix::unistd::ftruncate as nix_ftruncate;
-
-/// `ShmemHandle` represents a shared memory area that can be shared by processes over `fork()`.
-/// Unlike shared memory allocated by Postgres, this area is resizable, up to `max_size` that's
-/// specified at creation.
-///
-/// The area is backed by an anonymous file created with `memfd_create()`. The full address space for
-/// `max_size` is reserved up-front with `mmap()`, but whenever you call [`ShmemHandle::set_size`],
-/// the underlying file is resized. Do not access the area beyond the current size. Currently, that
-/// will cause the file to be expanded, but we might use `mprotect()` etc. to enforce that in the
-/// future.
-#[derive(Debug)]
-pub struct ShmemHandle {
-    /// memfd file descriptor
-    fd: OwnedFd,
-
-    max_size: usize,
-
-    // Pointer to the beginning of the shared memory area. The header is stored there.
-    shared_ptr: NonNull<SharedStruct>,
-
-    // Pointer to the beginning of the user data
-    pub data_ptr: NonNull<u8>,
-}
-
-/// This is stored at the beginning in the shared memory area.
-#[derive(Debug)]
-struct SharedStruct {
-    max_size: usize,
-
-    /// Current size of the backing file. The high-order bit is used for the [`RESIZE_IN_PROGRESS`] flag.
-    current_size: AtomicUsize,
-}
-
-const RESIZE_IN_PROGRESS: usize = 1 << 63;
-
-const HEADER_SIZE: usize = std::mem::size_of::<SharedStruct>();
-
-/// Error type returned by the [`ShmemHandle`] functions.
-#[derive(thiserror::Error, Debug)]
-#[error("{msg}: {errno}")]
-pub struct Error {
-    pub msg: String,
-    pub errno: Errno,
-}
-
-impl Error {
-    fn new(msg: &str, errno: Errno) -> Self {
-        Self {
-            msg: msg.to_string(),
-            errno,
-        }
-    }
-}
-
-impl ShmemHandle {
-    /// Create a new shared memory area. To communicate between processes, the processes need to be
-    /// `fork()`'d after calling this, so that the `ShmemHandle` is inherited by all processes.
-    ///
-    /// If the `ShmemHandle` is dropped, the memory is unmapped from the current process. Other
-    /// processes can continue using it, however.
-    pub fn new(name: &str, initial_size: usize, max_size: usize) -> Result<Self, Error> {
-        // create the backing anonymous file.
-        let fd = create_backing_file(name)?;
-
-        Self::new_with_fd(fd, initial_size, max_size)
-    }
-
-    fn new_with_fd(fd: OwnedFd, initial_size: usize, max_size: usize) -> Result<Self, Error> {
-        // We reserve the high-order bit for the `RESIZE_IN_PROGRESS` flag, and the actual size
-        // is a little larger than this because of the SharedStruct header. Make the upper limit
-        // somewhat smaller than that, because with anything close to that, you'll run out of
-        // memory anyway.
-        assert!(max_size < 1 << 48, "max size {max_size} too large");
-
-        assert!(
-            initial_size <= max_size,
-            "initial size {initial_size} larger than max size {max_size}"
-        );
-
-        // The actual initial / max size is the one given by the caller, plus the size of
-        // 'SharedStruct'.
-        let initial_size = HEADER_SIZE + initial_size;
-        let max_size = NonZeroUsize::new(HEADER_SIZE + max_size).unwrap();
-
-        // Reserve address space for it with mmap
-        //
-        // TODO: Use MAP_HUGETLB if possible
-        let start_ptr = unsafe {
-            nix_mmap(
-                None,
-                max_size,
-                ProtFlags::PROT_READ | ProtFlags::PROT_WRITE,
-                MapFlags::MAP_SHARED,
-                &fd,
-                0,
-            )
-        }
-        .map_err(|e| Error::new("mmap failed", e))?;
-
-        // Reserve space for the initial size
-        enlarge_file(fd.as_fd(), initial_size as u64)?;
-
-        // Initialize the header
-        let shared: NonNull<SharedStruct> = start_ptr.cast();
-        unsafe {
-            shared.write(SharedStruct {
-                max_size: max_size.into(),
-                current_size: AtomicUsize::new(initial_size),
-            });
-        }
-
-        // The user data begins after the header
-        let data_ptr = unsafe { start_ptr.cast().add(HEADER_SIZE) };
-
-        Ok(Self {
-            fd,
-            max_size: max_size.into(),
-            shared_ptr: shared,
-            data_ptr,
-        })
-    }
-
-    // return reference to the header
-    fn shared(&self) -> &SharedStruct {
-        unsafe { self.shared_ptr.as_ref() }
-    }
-
-    /// Resize the shared memory area. `new_size` must not be larger than the `max_size` specified
-    /// when creating the area.
-    ///
-    /// This may only be called from one process/thread concurrently. We detect that case
-    /// and return an [`shmem::Error`](Error).
-    pub fn set_size(&self, new_size: usize) -> Result<(), Error> {
-        let new_size = new_size + HEADER_SIZE;
-        let shared = self.shared();
-
-        assert!(
-            new_size <= self.max_size,
-            "new size ({new_size}) is greater than max size ({})",
-            self.max_size
-        );
-
-        assert_eq!(self.max_size, shared.max_size);
-
-        // Lock the area by setting the bit in `current_size`
-        //
-        // Ordering::Relaxed would probably be sufficient here, as we don't access any other memory
-        // and the `posix_fallocate`/`ftruncate` call is surely a synchronization point anyway. But
-        // since this is not performance-critical, better safe than sorry.
-        let mut old_size = shared.current_size.load(Ordering::Acquire);
-        loop {
-            if (old_size & RESIZE_IN_PROGRESS) != 0 {
-                return Err(Error::new(
-                    "concurrent resize detected",
-                    Errno::UnknownErrno,
-                ));
-            }
-            match shared.current_size.compare_exchange(
-                old_size,
-                new_size,
-                Ordering::Acquire,
-                Ordering::Relaxed,
-            ) {
-                Ok(_) => break,
-                Err(x) => old_size = x,
-            }
-        }
-
-        // Ok, we got the lock.
-        //
-        // NB: If anything goes wrong, we *must* clear the bit!
-        let result = {
-            use std::cmp::Ordering::{Equal, Greater, Less};
-            match new_size.cmp(&old_size) {
-                Less => nix_ftruncate(&self.fd, new_size as i64)
-                    .map_err(|e| Error::new("could not shrink shmem segment, ftruncate failed", e)),
-                Equal => Ok(()),
-                Greater => enlarge_file(self.fd.as_fd(), new_size as u64),
-            }
-        };
-
-        // Unlock
-        shared.current_size.store(
-            if result.is_ok() { new_size } else { old_size },
-            Ordering::Release,
-        );
-
-        result
-    }
-
-    /// Returns the current user-visible size of the shared memory segment.
-    ///
-    /// NOTE: a concurrent [`ShmemHandle::set_size()`] call can change the size at any time.
-    /// It is the caller's responsibility not to access the area beyond the current size.
-    pub fn current_size(&self) -> usize {
-        let total_current_size =
-            self.shared().current_size.load(Ordering::Relaxed) & !RESIZE_IN_PROGRESS;
-        total_current_size - HEADER_SIZE
-    }
-}
-
-impl Drop for ShmemHandle {
-    fn drop(&mut self) {
-        // SAFETY: The pointer was obtained from mmap() with the given size.
-        // We unmap the entire region.
-        let _ = unsafe { nix_munmap(self.shared_ptr.cast(), self.max_size) };
-        // The fd is dropped automatically by OwnedFd.
-    }
-}
-
-/// Create a "backing file" for the shared memory area. On Linux, use `memfd_create()`, to create an
-/// anonymous in-memory file. One macos, fall back to a regular file. That's good enough for
-/// development and testing, but in production we want the file to stay in memory.
-///
-/// Disable unused variables warnings because `name` is unused in the macos path.
-#[allow(unused_variables)]
-fn create_backing_file(name: &str) -> Result<OwnedFd, Error> {
-    #[cfg(not(target_os = "macos"))]
-    {
-        nix::sys::memfd::memfd_create(name, nix::sys::memfd::MFdFlags::empty())
-            .map_err(|e| Error::new("memfd_create failed", e))
-    }
-    #[cfg(target_os = "macos")]
-    {
-        let file = tempfile::tempfile().map_err(|e| {
-            Error::new(
-                "could not create temporary file to back shmem area",
-                nix::errno::Errno::from_raw(e.raw_os_error().unwrap_or(0)),
-            )
-        })?;
-        Ok(OwnedFd::from(file))
-    }
-}
-
-fn enlarge_file(fd: BorrowedFd, size: u64) -> Result<(), Error> {
-    // Use posix_fallocate() to enlarge the file. It reserves the space correctly, so that
-    // we don't get a segfault later when trying to actually use it.
-    #[cfg(not(target_os = "macos"))]
-    {
-        nix::fcntl::posix_fallocate(fd, 0, size as i64)
-            .map_err(|e| Error::new("could not grow shmem segment, posix_fallocate failed", e))
-    }
-    // As a fallback on macos, which doesn't have posix_fallocate, use plain 'fallocate'
-    #[cfg(target_os = "macos")]
-    {
-        nix::unistd::ftruncate(fd, size as i64)
-            .map_err(|e| Error::new("could not grow shmem segment, ftruncate failed", e))
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    use nix::unistd::ForkResult;
-    use std::ops::Range;
-
-    /// check that all bytes in given range have the expected value.
-    fn assert_range(ptr: *const u8, expected: u8, range: Range<usize>) {
-        for i in range {
-            let b = unsafe { *(ptr.add(i)) };
-            assert_eq!(expected, b, "unexpected byte at offset {i}");
-        }
-    }
-
-    /// Write 'b' to all bytes in the given range
-    fn write_range(ptr: *mut u8, b: u8, range: Range<usize>) {
-        unsafe { std::ptr::write_bytes(ptr.add(range.start), b, range.end - range.start) };
-    }
-
-    // simple single-process test of growing and shrinking
-    #[test]
-    fn test_shmem_resize() -> Result<(), Error> {
-        let max_size = 1024 * 1024;
-        let init_struct = ShmemHandle::new("test_shmem_resize", 0, max_size)?;
-
-        assert_eq!(init_struct.current_size(), 0);
-
-        // Initial grow
-        let size1 = 10000;
-        init_struct.set_size(size1).unwrap();
-        assert_eq!(init_struct.current_size(), size1);
-
-        // Write some data
-        let data_ptr = init_struct.data_ptr.as_ptr();
-        write_range(data_ptr, 0xAA, 0..size1);
-        assert_range(data_ptr, 0xAA, 0..size1);
-
-        // Shrink
-        let size2 = 5000;
-        init_struct.set_size(size2).unwrap();
-        assert_eq!(init_struct.current_size(), size2);
-
-        // Grow again
-        let size3 = 20000;
-        init_struct.set_size(size3).unwrap();
-        assert_eq!(init_struct.current_size(), size3);
-
-        // Try to read it. The area that was shrunk and grown again should read as all zeros now
-        assert_range(data_ptr, 0xAA, 0..5000);
-        assert_range(data_ptr, 0, 5000..size1);
-
-        // Try to grow beyond max_size
-        //let size4 = max_size + 1;
-        //assert!(init_struct.set_size(size4).is_err());
-
-        // Dropping init_struct should unmap the memory
-        drop(init_struct);
-
-        Ok(())
-    }
-
-    /// This is used in tests to coordinate between test processes. It's like `std::sync::Barrier`,
-    /// but is stored in the shared memory area and works across processes. It's implemented by
-    /// polling, because e.g. standard rust mutexes are not guaranteed to work across processes.
-    struct SimpleBarrier {
-        num_procs: usize,
-        count: AtomicUsize,
-    }
-
-    impl SimpleBarrier {
-        unsafe fn init(ptr: *mut SimpleBarrier, num_procs: usize) {
-            unsafe {
-                *ptr = SimpleBarrier {
-                    num_procs,
-                    count: AtomicUsize::new(0),
-                }
-            }
-        }
-
-        pub fn wait(&self) {
-            let old = self.count.fetch_add(1, Ordering::Relaxed);
-
-            let generation = old / self.num_procs;
-
-            let mut current = old + 1;
-            while current < (generation + 1) * self.num_procs {
-                std::thread::sleep(std::time::Duration::from_millis(10));
-                current = self.count.load(Ordering::Relaxed);
-            }
-        }
-    }
-
-    #[test]
-    fn test_multi_process() {
-        // Initialize
-        let max_size = 1_000_000_000_000;
-        let init_struct = ShmemHandle::new("test_multi_process", 0, max_size).unwrap();
-        let ptr = init_struct.data_ptr.as_ptr();
-
-        // Store the SimpleBarrier in the first 1k of the area.
-        init_struct.set_size(10000).unwrap();
-        let barrier_ptr: *mut SimpleBarrier = unsafe {
-            ptr.add(ptr.align_offset(std::mem::align_of::<SimpleBarrier>()))
-                .cast()
-        };
-        unsafe { SimpleBarrier::init(barrier_ptr, 2) };
-        let barrier = unsafe { barrier_ptr.as_ref().unwrap() };
-
-        // Fork another test process. The code after this runs in both processes concurrently.
-        let fork_result = unsafe { nix::unistd::fork().unwrap() };
-
-        // In the parent, fill bytes between 1000..2000. In the child, between 2000..3000
-        if fork_result.is_parent() {
-            write_range(ptr, 0xAA, 1000..2000);
-        } else {
-            write_range(ptr, 0xBB, 2000..3000);
-        }
-        barrier.wait();
-        // Verify the contents. (in both processes)
-        assert_range(ptr, 0xAA, 1000..2000);
-        assert_range(ptr, 0xBB, 2000..3000);
-
-        // Grow, from the child this time
-        let size = 10_000_000;
-        if !fork_result.is_parent() {
-            init_struct.set_size(size).unwrap();
-        }
-        barrier.wait();
-
-        // make some writes at the end
-        if fork_result.is_parent() {
-            write_range(ptr, 0xAA, (size - 10)..size);
-        } else {
-            write_range(ptr, 0xBB, (size - 20)..(size - 10));
-        }
-        barrier.wait();
-
-        // Verify the contents. (This runs in both processes)
-        assert_range(ptr, 0, (size - 1000)..(size - 20));
-        assert_range(ptr, 0xBB, (size - 20)..(size - 10));
-        assert_range(ptr, 0xAA, (size - 10)..size);
-
-        if let ForkResult::Parent { child } = fork_result {
-            nix::sys::wait::waitpid(child, None).unwrap();
-        }
-    }
-}

libs/neon-shmem/src/sync.rs

@@ -1,104 +0,0 @@
-//! Simple utilities akin to what's in [`std::sync`] but designed to work with shared memory.
-
-use std::mem::MaybeUninit;
-use std::ptr::NonNull;
-
-use nix::errno::Errno;
-
-pub type RwLock<T> = lock_api::RwLock<PthreadRwLock, T>;
-pub(crate) type RwLockReadGuard<'a, T> = lock_api::RwLockReadGuard<'a, PthreadRwLock, T>;
-pub type RwLockWriteGuard<'a, T> = lock_api::RwLockWriteGuard<'a, PthreadRwLock, T>;
-pub type ValueReadGuard<'a, T> = lock_api::MappedRwLockReadGuard<'a, PthreadRwLock, T>;
-pub type ValueWriteGuard<'a, T> = lock_api::MappedRwLockWriteGuard<'a, PthreadRwLock, T>;
-
-/// Shared memory read-write lock.
-pub struct PthreadRwLock(Option<NonNull<libc::pthread_rwlock_t>>);
-
-impl PthreadRwLock {
-    pub fn new(lock: *mut libc::pthread_rwlock_t) -> Self {
-        unsafe {
-            let mut attrs = MaybeUninit::uninit();
-            // Ignoring return value here - only possible error is OOM.
-            libc::pthread_rwlockattr_init(attrs.as_mut_ptr());
-            libc::pthread_rwlockattr_setpshared(attrs.as_mut_ptr(), libc::PTHREAD_PROCESS_SHARED);
-            // TODO(quantumish): worth making this function return Result?
-            libc::pthread_rwlock_init(lock, attrs.as_mut_ptr());
-            // Safety: POSIX specifies that "any function affecting the attributes
-            // object (including destruction) shall not affect any previously
-            // initialized read-write locks".
-            libc::pthread_rwlockattr_destroy(attrs.as_mut_ptr());
-            Self(Some(NonNull::new_unchecked(lock)))
-        }
-    }
-
-    fn inner(&self) -> NonNull<libc::pthread_rwlock_t> {
-        match self.0 {
-            None => {
-                panic!("PthreadRwLock constructed badly - something likely used RawMutex::INIT")
-            }
-            Some(x) => x,
-        }
-    }
-}
-
-unsafe impl lock_api::RawRwLock for PthreadRwLock {
-    type GuardMarker = lock_api::GuardSend;
-    const INIT: Self = Self(None);
-
-    fn lock_shared(&self) {
-        unsafe {
-            let res = libc::pthread_rwlock_rdlock(self.inner().as_ptr());
-            if res != 0 {
-                panic!("rdlock failed with {}", Errno::from_raw(res));
-            }
-        }
-    }
-
-    fn try_lock_shared(&self) -> bool {
-        unsafe {
-            let res = libc::pthread_rwlock_tryrdlock(self.inner().as_ptr());
-            match res {
-                0 => true,
-                libc::EAGAIN => false,
-                _ => panic!("try_rdlock failed with {}", Errno::from_raw(res)),
-            }
-        }
-    }
-
-    fn lock_exclusive(&self) {
-        unsafe {
-            let res = libc::pthread_rwlock_wrlock(self.inner().as_ptr());
-            if res != 0 {
-                panic!("wrlock failed with {}", Errno::from_raw(res));
-            }
-        }
-    }
-
-    fn try_lock_exclusive(&self) -> bool {
-        unsafe {
-            let res = libc::pthread_rwlock_trywrlock(self.inner().as_ptr());
-            match res {
-                0 => true,
-                libc::EAGAIN => false,
-                _ => panic!("try_wrlock failed with {}", Errno::from_raw(res)),
-            }
-        }
-    }
-
-    unsafe fn unlock_exclusive(&self) {
-        unsafe {
-            let res = libc::pthread_rwlock_unlock(self.inner().as_ptr());
-            if res != 0 {
-                panic!("unlock failed with {}", Errno::from_raw(res));
-            }
-        }
-    }
-    unsafe fn unlock_shared(&self) {
-        unsafe {
-            let res = libc::pthread_rwlock_unlock(self.inner().as_ptr());
-            if res != 0 {
-                panic!("unlock failed with {}", Errno::from_raw(res));
-            }
-        }
-    }
-}

libs/neonart/Cargo.toml

@@ -1,14 +0,0 @@
-[package]
-name = "neonart"
-version = "0.1.0"
-edition.workspace = true
-license.workspace = true
-
-[dependencies]
-crossbeam-utils.workspace = true
-spin.workspace = true
-tracing.workspace = true
-
-[dev-dependencies]
-rand = "0.9.1"
-rand_distr = "0.5.1"

libs/neonart/src/algorithm.rs

@@ -1,599 +0,0 @@
-mod lock_and_version;
-pub(crate) mod node_ptr;
-mod node_ref;
-
-use std::vec::Vec;
-
-use crate::algorithm::lock_and_version::ConcurrentUpdateError;
-use crate::algorithm::node_ptr::MAX_PREFIX_LEN;
-use crate::algorithm::node_ref::{NewNodeRef, NodeRef, ReadLockedNodeRef, WriteLockedNodeRef};
-use crate::allocator::OutOfMemoryError;
-
-use crate::TreeWriteGuard;
-use crate::UpdateAction;
-use crate::allocator::ArtAllocator;
-use crate::epoch::EpochPin;
-use crate::{Key, Value};
-
-pub(crate) type RootPtr<V> = node_ptr::NodePtr<V>;
-
-#[derive(Debug)]
-pub enum ArtError {
-    ConcurrentUpdate, // need to retry
-    OutOfMemory,
-}
-
-impl From<ConcurrentUpdateError> for ArtError {
-    fn from(_: ConcurrentUpdateError) -> ArtError {
-        ArtError::ConcurrentUpdate
-    }
-}
-
-impl From<OutOfMemoryError> for ArtError {
-    fn from(_: OutOfMemoryError) -> ArtError {
-        ArtError::OutOfMemory
-    }
-}
-
-pub fn new_root<V: Value>(
-    allocator: &impl ArtAllocator<V>,
-) -> Result<RootPtr<V>, OutOfMemoryError> {
-    node_ptr::new_root(allocator)
-}
-
-pub(crate) fn search<'e, K: Key, V: Value>(
-    key: &K,
-    root: RootPtr<V>,
-    epoch_pin: &'e EpochPin,
-) -> Option<&'e V> {
-    loop {
-        let root_ref = NodeRef::from_root_ptr(root);
-        if let Ok(result) = lookup_recurse(key.as_bytes(), root_ref, None, epoch_pin) {
-            break result;
-        }
-        // retry
-    }
-}
-
-pub(crate) fn iter_next<'e, V: Value>(
-    key: &[u8],
-    root: RootPtr<V>,
-    epoch_pin: &'e EpochPin,
-) -> Option<(Vec<u8>, &'e V)> {
-    loop {
-        let mut path = Vec::new();
-        let root_ref = NodeRef::from_root_ptr(root);
-
-        match next_recurse(key, &mut path, root_ref, epoch_pin) {
-            Ok(Some(v)) => {
-                assert_eq!(path.len(), key.len());
-                break Some((path, v));
-            }
-            Ok(None) => break None,
-            Err(ConcurrentUpdateError()) => {
-                // retry
-                continue;
-            }
-        }
-    }
-}
-
-pub(crate) fn update_fn<'e, 'g, K: Key, V: Value, A: ArtAllocator<V>, F>(
-    key: &K,
-    value_fn: F,
-    root: RootPtr<V>,
-    guard: &'g mut TreeWriteGuard<'e, K, V, A>,
-) -> Result<(), OutOfMemoryError>
-where
-    F: FnOnce(Option<&V>) -> UpdateAction<V>,
-{
-    let value_fn_cell = std::cell::Cell::new(Some(value_fn));
-    loop {
-        let root_ref = NodeRef::from_root_ptr(root);
-        let this_value_fn = |arg: Option<&V>| value_fn_cell.take().unwrap()(arg);
-        let key_bytes = key.as_bytes();
-
-        match update_recurse(
-            key_bytes,
-            this_value_fn,
-            root_ref,
-            None,
-            None,
-            guard,
-            0,
-            key_bytes,
-        ) {
-            Ok(()) => break Ok(()),
-            Err(ArtError::ConcurrentUpdate) => {
-                continue; // retry
-            }
-            Err(ArtError::OutOfMemory) => break Err(OutOfMemoryError()),
-        }
-    }
-}
-
-// Error means you must retry.
-//
-// This corresponds to the 'lookupOpt' function in the paper
-#[allow(clippy::only_used_in_recursion)]
-fn lookup_recurse<'e, V: Value>(
-    key: &[u8],
-    node: NodeRef<'e, V>,
-    parent: Option<ReadLockedNodeRef<V>>,
-    epoch_pin: &'e EpochPin,
-) -> Result<Option<&'e V>, ConcurrentUpdateError> {
-    let rnode = node.read_lock_or_restart()?;
-    if let Some(parent) = parent {
-        parent.read_unlock_or_restart()?;
-    }
-
-    // check if the prefix matches, may increment level
-    let prefix_len = if let Some(prefix_len) = rnode.prefix_matches(key) {
-        prefix_len
-    } else {
-        rnode.read_unlock_or_restart()?;
-        return Ok(None);
-    };
-
-    if rnode.is_leaf() {
-        assert_eq!(key.len(), prefix_len);
-        let vptr = rnode.get_leaf_value_ptr()?;
-        // safety: It's OK to return a ref of the pointer because we checked the version
-        // and the lifetime of 'epoch_pin' enforces that the reference is only accessible
-        // as long as the epoch is pinned.
-        let v = unsafe { vptr.as_ref().unwrap() };
-        return Ok(Some(v));
-    }
-
-    let key = &key[prefix_len..];
-
-    // find child (or leaf value)
-    let next_node = rnode.find_child_or_restart(key[0])?;
-
-    match next_node {
-        None => Ok(None), // key not found
-        Some(child) => lookup_recurse(&key[1..], child, Some(rnode), epoch_pin),
-    }
-}
-
-#[allow(clippy::only_used_in_recursion)]
-fn next_recurse<'e, V: Value>(
-    min_key: &[u8],
-    path: &mut Vec<u8>,
-    node: NodeRef<'e, V>,
-    epoch_pin: &'e EpochPin,
-) -> Result<Option<&'e V>, ConcurrentUpdateError> {
-    let rnode = node.read_lock_or_restart()?;
-    let prefix = rnode.get_prefix();
-    if !prefix.is_empty() {
-        path.extend_from_slice(prefix);
-    }
-
-    use std::cmp::Ordering;
-    let comparison = path.as_slice().cmp(&min_key[0..path.len()]);
-    if comparison == Ordering::Less {
-        rnode.read_unlock_or_restart()?;
-        return Ok(None);
-    }
-
-    if rnode.is_leaf() {
-        assert_eq!(path.len(), min_key.len());
-        let vptr = rnode.get_leaf_value_ptr()?;
-        // safety: It's OK to return a ref of the pointer because we checked the version
-        // and the lifetime of 'epoch_pin' enforces that the reference is only accessible
-        // as long as the epoch is pinned.
-        let v = unsafe { vptr.as_ref().unwrap() };
-        return Ok(Some(v));
-    }
-
-    let mut min_key_byte = match comparison {
-        Ordering::Less => unreachable!(), // checked this above already
-        Ordering::Equal => min_key[path.len()],
-        Ordering::Greater => 0,
-    };
-
-    loop {
-        match rnode.find_next_child_or_restart(min_key_byte)? {
-            None => {
-                return Ok(None);
-            }
-            Some((key_byte, child_ref)) => {
-                let path_len = path.len();
-                path.push(key_byte);
-                let result = next_recurse(min_key, path, child_ref, epoch_pin)?;
-                if result.is_some() {
-                    return Ok(result);
-                }
-                if key_byte == u8::MAX {
-                    return Ok(None);
-                }
-                path.truncate(path_len);
-                min_key_byte = key_byte + 1;
-            }
-        }
-    }
-}
-
-// This corresponds to the 'insertOpt' function in the paper
-#[allow(clippy::only_used_in_recursion)]
-#[allow(clippy::too_many_arguments)]
-pub(crate) fn update_recurse<'e, K: Key, V: Value, A: ArtAllocator<V>, F>(
-    key: &[u8],
-    value_fn: F,
-    node: NodeRef<'e, V>,
-    rparent: Option<(ReadLockedNodeRef<V>, u8)>,
-    rgrandparent: Option<(ReadLockedNodeRef<V>, u8)>,
-    guard: &'_ mut TreeWriteGuard<'e, K, V, A>,
-    level: usize,
-    orig_key: &[u8],
-) -> Result<(), ArtError>
-where
-    F: FnOnce(Option<&V>) -> UpdateAction<V>,
-{
-    let rnode = node.read_lock_or_restart()?;
-
-    let prefix_match_len = rnode.prefix_matches(key);
-    if prefix_match_len.is_none() {
-        let (rparent, parent_key) = rparent.expect("direct children of the root have no prefix");
-        let mut wparent = rparent.upgrade_to_write_lock_or_restart()?;
-        let mut wnode = rnode.upgrade_to_write_lock_or_restart()?;
-
-        match value_fn(None) {
-            UpdateAction::Nothing => {}
-            UpdateAction::Insert(new_value) => {
-                insert_split_prefix(key, new_value, &mut wnode, &mut wparent, parent_key, guard)?;
-            }
-            UpdateAction::Remove => {
-                panic!("unexpected Remove action on insertion");
-            }
-        }
-        wnode.write_unlock();
-        wparent.write_unlock();
-        return Ok(());
-    }
-    let prefix_match_len = prefix_match_len.unwrap();
-    let key = &key[prefix_match_len..];
-    let level = level + prefix_match_len;
-
-    if rnode.is_leaf() {
-        assert_eq!(key.len(), 0);
-        let (rparent, parent_key) = rparent.expect("root cannot be leaf");
-        let mut wparent = rparent.upgrade_to_write_lock_or_restart()?;
-        let mut wnode = rnode.upgrade_to_write_lock_or_restart()?;
-
-        // safety: Now that we have acquired the write lock, we have exclusive access to the
-        // value. XXX: There might be concurrent reads though?
-        let value_mut = wnode.get_leaf_value_mut();
-
-        match value_fn(Some(value_mut)) {
-            UpdateAction::Nothing => {
-                wparent.write_unlock();
-                wnode.write_unlock();
-            }
-            UpdateAction::Insert(_) => panic!("cannot insert over existing value"),
-            UpdateAction::Remove => {
-                guard.remember_obsolete_node(wnode.as_ptr());
-                wparent.delete_child(parent_key);
-                wnode.write_unlock_obsolete();
-
-                if let Some(rgrandparent) = rgrandparent {
-                    // FIXME: Ignore concurrency error. It doesn't lead to
-                    // corruption, but it means we might leak something. Until
-                    // another update cleans it up.
-                    let _ = cleanup_parent(wparent, rgrandparent, guard);
-                }
-            }
-        }
-
-        return Ok(());
-    }
-
-    let next_node = rnode.find_child_or_restart(key[0])?;
-
-    if next_node.is_none() {
-        if rnode.is_full() {
-            let (rparent, parent_key) = rparent.expect("root node cannot become full");
-            let mut wparent = rparent.upgrade_to_write_lock_or_restart()?;
-            let wnode = rnode.upgrade_to_write_lock_or_restart()?;
-
-            match value_fn(None) {
-                UpdateAction::Nothing => {
-                    wnode.write_unlock();
-                    wparent.write_unlock();
-                }
-                UpdateAction::Insert(new_value) => {
-                    insert_and_grow(key, new_value, wnode, &mut wparent, parent_key, guard)?;
-                    wparent.write_unlock();
-                }
-                UpdateAction::Remove => {
-                    panic!("unexpected Remove action on insertion");
-                }
-            };
-        } else {
-            let mut wnode = rnode.upgrade_to_write_lock_or_restart()?;
-            if let Some((rparent, _)) = rparent {
-                rparent.read_unlock_or_restart()?;
-            }
-            match value_fn(None) {
-                UpdateAction::Nothing => {}
-                UpdateAction::Insert(new_value) => {
-                    insert_to_node(&mut wnode, key, new_value, guard)?;
-                }
-                UpdateAction::Remove => {
-                    panic!("unexpected Remove action on insertion");
-                }
-            };
-            wnode.write_unlock();
-        }
-        Ok(())
-    } else {
-        let next_child = next_node.unwrap(); // checked above it's not None
-        if let Some((ref rparent, _)) = rparent {
-            rparent.check_or_restart()?;
-        }
-
-        // recurse to next level
-        update_recurse(
-            &key[1..],
-            value_fn,
-            next_child,
-            Some((rnode, key[0])),
-            rparent,
-            guard,
-            level + 1,
-            orig_key,
-        )
-    }
-}
-
-#[derive(Clone)]
-enum PathElement {
-    Prefix(Vec<u8>),
-    KeyByte(u8),
-}
-
-impl std::fmt::Debug for PathElement {
-    fn fmt(&self, fmt: &mut std::fmt::Formatter<'_>) -> Result<(), std::fmt::Error> {
-        match self {
-            PathElement::Prefix(prefix) => write!(fmt, "{prefix:?}"),
-            PathElement::KeyByte(key_byte) => write!(fmt, "{key_byte}"),
-        }
-    }
-}
-
-pub(crate) fn dump_tree<V: Value + std::fmt::Debug>(
-    root: RootPtr<V>,
-    epoch_pin: &'_ EpochPin,
-    dst: &mut dyn std::io::Write,
-) {
-    let root_ref = NodeRef::from_root_ptr(root);
-
-    let _ = dump_recurse(&[], root_ref, epoch_pin, 0, dst);
-}
-
-// TODO: return an Err if writeln!() returns error, instead of unwrapping
-#[allow(clippy::only_used_in_recursion)]
-fn dump_recurse<'e, V: Value + std::fmt::Debug>(
-    path: &[PathElement],
-    node: NodeRef<'e, V>,
-    epoch_pin: &'e EpochPin,
-    level: usize,
-    dst: &mut dyn std::io::Write,
-) -> Result<(), ConcurrentUpdateError> {
-    let indent = str::repeat(" ", level);
-
-    let rnode = node.read_lock_or_restart()?;
-    let mut path = Vec::from(path);
-    let prefix = rnode.get_prefix();
-    if !prefix.is_empty() {
-        path.push(PathElement::Prefix(Vec::from(prefix)));
-    }
-
-    if rnode.is_leaf() {
-        let vptr = rnode.get_leaf_value_ptr()?;
-        // safety: It's OK to return a ref of the pointer because we checked the version
-        // and the lifetime of 'epoch_pin' enforces that the reference is only accessible
-        // as long as the epoch is pinned.
-        let val = unsafe { vptr.as_ref().unwrap() };
-        writeln!(dst, "{indent} {path:?}: {val:?}").unwrap();
-        return Ok(());
-    }
-
-    for key_byte in 0..=u8::MAX {
-        match rnode.find_child_or_restart(key_byte)? {
-            None => continue,
-            Some(child_ref) => {
-                let rchild = child_ref.read_lock_or_restart()?;
-                writeln!(
-                    dst,
-                    "{} {:?}, {}: prefix {:?}",
-                    indent,
-                    &path,
-                    key_byte,
-                    rchild.get_prefix()
-                )
-                .unwrap();
-
-                let mut child_path = path.clone();
-                child_path.push(PathElement::KeyByte(key_byte));
-
-                dump_recurse(&child_path, child_ref, epoch_pin, level + 1, dst)?;
-            }
-        }
-    }
-
-    Ok(())
-}
-
-///```text
-///        [fooba]r -> value
-///
-/// [foo]b -> [a]r  -> value
-///      e -> [ls]e -> value
-///```
-fn insert_split_prefix<K: Key, V: Value, A: ArtAllocator<V>>(
-    key: &[u8],
-    value: V,
-    node: &mut WriteLockedNodeRef<V>,
-    parent: &mut WriteLockedNodeRef<V>,
-    parent_key: u8,
-    guard: &'_ TreeWriteGuard<K, V, A>,
-) -> Result<(), OutOfMemoryError> {
-    let old_node = node;
-    let old_prefix = old_node.get_prefix();
-    let common_prefix_len = common_prefix(key, old_prefix);
-
-    // Allocate a node for the new value.
-    let new_value_node = allocate_node_for_value(
-        &key[common_prefix_len + 1..],
-        value,
-        guard.tree_writer.allocator,
-    )?;
-
-    // Allocate a new internal node with the common prefix
-    // FIXME: deallocate 'new_value_node' on OOM
-    let mut prefix_node =
-        node_ref::new_internal(&key[..common_prefix_len], guard.tree_writer.allocator)?;
-
-    // Add the old node and the new nodes to the new internal node
-    prefix_node.insert_old_child(old_prefix[common_prefix_len], old_node);
-    prefix_node.insert_new_child(key[common_prefix_len], new_value_node);
-
-    // Modify the prefix of the old child in place
-    old_node.truncate_prefix(old_prefix.len() - common_prefix_len - 1);
-
-    // replace the pointer in the parent
-    parent.replace_child(parent_key, prefix_node.into_ptr());
-
-    Ok(())
-}
-
-fn insert_to_node<K: Key, V: Value, A: ArtAllocator<V>>(
-    wnode: &mut WriteLockedNodeRef<V>,
-    key: &[u8],
-    value: V,
-    guard: &'_ TreeWriteGuard<K, V, A>,
-) -> Result<(), OutOfMemoryError> {
-    let value_child = allocate_node_for_value(&key[1..], value, guard.tree_writer.allocator)?;
-    wnode.insert_child(key[0], value_child.into_ptr());
-    Ok(())
-}
-
-// On entry: 'parent' and 'node' are locked
-fn insert_and_grow<'e, 'g, K: Key, V: Value, A: ArtAllocator<V>>(
-    key: &[u8],
-    value: V,
-    wnode: WriteLockedNodeRef<V>,
-    parent: &mut WriteLockedNodeRef<V>,
-    parent_key_byte: u8,
-    guard: &'g mut TreeWriteGuard<'e, K, V, A>,
-) -> Result<(), ArtError> {
-    let mut bigger_node = wnode.grow(guard.tree_writer.allocator)?;
-
-    // FIXME: deallocate 'bigger_node' on OOM
-    let value_child = allocate_node_for_value(&key[1..], value, guard.tree_writer.allocator)?;
-    bigger_node.insert_new_child(key[0], value_child);
-
-    // Replace the pointer in the parent
-    parent.replace_child(parent_key_byte, bigger_node.into_ptr());
-
-    guard.remember_obsolete_node(wnode.as_ptr());
-    wnode.write_unlock_obsolete();
-
-    Ok(())
-}
-
-fn cleanup_parent<'e, 'g, K: Key, V: Value, A: ArtAllocator<V>>(
-    wparent: WriteLockedNodeRef<V>,
-    rgrandparent: (ReadLockedNodeRef<V>, u8),
-    guard: &'g mut TreeWriteGuard<'e, K, V, A>,
-) -> Result<(), ArtError> {
-    let (rgrandparent, grandparent_key_byte) = rgrandparent;
-
-    // If the parent becomes completely empty after the deletion, remove the parent from the
-    // grandparent. (This case is possible because we reserve only 8 bytes for the prefix.)
-    // TODO: not implemented.
-
-    // If the parent has only one child, replace the parent with the remaining child. (This is not
-    // possible if the child's prefix field cannot absorb the parent's)
-    if wparent.num_children() == 1 {
-        // Try to lock the remaining child. This can fail if the child is updated
-        // concurrently.
-        let (key_byte, remaining_child) = wparent.find_remaining_child();
-
-        let mut wremaining_child = remaining_child.write_lock_or_restart()?;
-
-        if 1 + wremaining_child.get_prefix().len() + wparent.get_prefix().len() <= MAX_PREFIX_LEN {
-            let mut wgrandparent = rgrandparent.upgrade_to_write_lock_or_restart()?;
-
-            // Ok, we have locked the leaf, the parent, the grandparent, and the parent's only
-            // remaining leaf. Proceed with the updates.
-
-            // Update the prefix on the remaining leaf
-            wremaining_child.prepend_prefix(wparent.get_prefix(), key_byte);
-
-            // Replace the pointer in the grandparent to point directly to the remaining leaf
-            wgrandparent.replace_child(grandparent_key_byte, wremaining_child.as_ptr());
-
-            // Mark the parent as deleted.
-            guard.remember_obsolete_node(wparent.as_ptr());
-            wparent.write_unlock_obsolete();
-            return Ok(());
-        }
-    }
-
-    // If the parent's children would fit on a smaller node type after the deletion, replace it with
-    // a smaller node.
-    if wparent.can_shrink() {
-        let mut wgrandparent = rgrandparent.upgrade_to_write_lock_or_restart()?;
-        let smaller_node = wparent.shrink(guard.tree_writer.allocator)?;
-
-        // Replace the pointer in the grandparent
-        wgrandparent.replace_child(grandparent_key_byte, smaller_node.into_ptr());
-
-        guard.remember_obsolete_node(wparent.as_ptr());
-        wparent.write_unlock_obsolete();
-        return Ok(());
-    }
-
-    // nothing to do
-    wparent.write_unlock();
-    Ok(())
-}
-
-// Allocate a new leaf node to hold 'value'. If the key is long, we
-// may need to allocate new internal nodes to hold it too
-fn allocate_node_for_value<'a, V: Value, A: ArtAllocator<V>>(
-    key: &[u8],
-    value: V,
-    allocator: &'a A,
-) -> Result<NewNodeRef<'a, V, A>, OutOfMemoryError> {
-    let mut prefix_off = key.len().saturating_sub(MAX_PREFIX_LEN);
-
-    let leaf_node = node_ref::new_leaf(&key[prefix_off..key.len()], value, allocator)?;
-
-    let mut node = leaf_node;
-    while prefix_off > 0 {
-        // Need another internal node
-        let remain_prefix = &key[0..prefix_off];
-
-        prefix_off = remain_prefix.len().saturating_sub(MAX_PREFIX_LEN + 1);
-        let mut internal_node = node_ref::new_internal(
-            &remain_prefix[prefix_off..remain_prefix.len() - 1],
-            allocator,
-        )?;
-        internal_node.insert_new_child(*remain_prefix.last().unwrap(), node);
-        node = internal_node;
-    }
-
-    Ok(node)
-}
-
-fn common_prefix(a: &[u8], b: &[u8]) -> usize {
-    for i in 0..MAX_PREFIX_LEN {
-        if a[i] != b[i] {
-            return i;
-        }
-    }
-    panic!("prefixes are equal");
-}

libs/neonart/src/algorithm/lock_and_version.rs

@@ -1,117 +0,0 @@
-//! Each node in the tree has contains one atomic word that stores three things:
-//!
-//! Bit 0: set if the node is "obsolete". An obsolete node has been removed from the tree,
-//!        but might still be accessed by concurrent readers until the epoch expires.
-//! Bit 1: set if the node is currently write-locked. Used as a spinlock.
-//! Bits 2-63: Version number, incremented every time the node is modified.
-//!
-//! AtomicLockAndVersion represents that.
-
-use std::sync::atomic::{AtomicU64, Ordering};
-
-pub(crate) struct ConcurrentUpdateError();
-
-pub(crate) struct AtomicLockAndVersion {
-    inner: AtomicU64,
-}
-
-impl AtomicLockAndVersion {
-    pub(crate) fn new() -> AtomicLockAndVersion {
-        AtomicLockAndVersion {
-            inner: AtomicU64::new(0),
-        }
-    }
-}
-
-impl AtomicLockAndVersion {
-    pub(crate) fn read_lock_or_restart(&self) -> Result<u64, ConcurrentUpdateError> {
-        let version = self.await_node_unlocked();
-        if is_obsolete(version) {
-            return Err(ConcurrentUpdateError());
-        }
-        Ok(version)
-    }
-
-    pub(crate) fn check_or_restart(&self, version: u64) -> Result<(), ConcurrentUpdateError> {
-        self.read_unlock_or_restart(version)
-    }
-
-    pub(crate) fn read_unlock_or_restart(&self, version: u64) -> Result<(), ConcurrentUpdateError> {
-        if self.inner.load(Ordering::Acquire) != version {
-            return Err(ConcurrentUpdateError());
-        }
-        Ok(())
-    }
-
-    pub(crate) fn upgrade_to_write_lock_or_restart(
-        &self,
-        version: u64,
-    ) -> Result<(), ConcurrentUpdateError> {
-        if self
-            .inner
-            .compare_exchange(
-                version,
-                set_locked_bit(version),
-                Ordering::Acquire,
-                Ordering::Relaxed,
-            )
-            .is_err()
-        {
-            return Err(ConcurrentUpdateError());
-        }
-        Ok(())
-    }
-
-    pub(crate) fn write_lock_or_restart(&self) -> Result<(), ConcurrentUpdateError> {
-        let old = self.inner.load(Ordering::Relaxed);
-        if is_obsolete(old) || is_locked(old) {
-            return Err(ConcurrentUpdateError());
-        }
-        if self
-            .inner
-            .compare_exchange(
-                old,
-                set_locked_bit(old),
-                Ordering::Acquire,
-                Ordering::Relaxed,
-            )
-            .is_err()
-        {
-            return Err(ConcurrentUpdateError());
-        }
-        Ok(())
-    }
-
-    pub(crate) fn write_unlock(&self) {
-        // reset locked bit and overflow into version
-        self.inner.fetch_add(2, Ordering::Release);
-    }
-
-    pub(crate) fn write_unlock_obsolete(&self) {
-        // set obsolete, reset locked, overflow into version
-        self.inner.fetch_add(3, Ordering::Release);
-    }
-
-    // Helper functions
-    fn await_node_unlocked(&self) -> u64 {
-        let mut version = self.inner.load(Ordering::Acquire);
-        while is_locked(version) {
-            // spinlock
-            std::thread::yield_now();
-            version = self.inner.load(Ordering::Acquire)
-        }
-        version
-    }
-}
-
-fn set_locked_bit(version: u64) -> u64 {
-    version + 2
-}
-
-fn is_obsolete(version: u64) -> bool {
-    (version & 1) == 1
-}
-
-fn is_locked(version: u64) -> bool {
-    (version & 2) == 2
-}

libs/neonart/src/algorithm/node_ref.rs

@@ -1,349 +0,0 @@
-use std::fmt::Debug;
-use std::marker::PhantomData;
-
-use super::node_ptr;
-use super::node_ptr::NodePtr;
-use crate::EpochPin;
-use crate::Value;
-use crate::algorithm::lock_and_version::AtomicLockAndVersion;
-use crate::algorithm::lock_and_version::ConcurrentUpdateError;
-use crate::allocator::ArtAllocator;
-use crate::allocator::OutOfMemoryError;
-
-pub struct NodeRef<'e, V> {
-    ptr: NodePtr<V>,
-
-    phantom: PhantomData<&'e EpochPin<'e>>,
-}
-
-impl<'e, V> Debug for NodeRef<'e, V> {
-    fn fmt(&self, fmt: &mut std::fmt::Formatter<'_>) -> Result<(), std::fmt::Error> {
-        write!(fmt, "{:?}", self.ptr)
-    }
-}
-
-impl<'e, V: Value> NodeRef<'e, V> {
-    pub(crate) fn from_root_ptr(root_ptr: NodePtr<V>) -> NodeRef<'e, V> {
-        NodeRef {
-            ptr: root_ptr,
-            phantom: PhantomData,
-        }
-    }
-
-    pub(crate) fn read_lock_or_restart(
-        &self,
-    ) -> Result<ReadLockedNodeRef<'e, V>, ConcurrentUpdateError> {
-        let version = self.lockword().read_lock_or_restart()?;
-        Ok(ReadLockedNodeRef {
-            ptr: self.ptr,
-            version,
-            phantom: self.phantom,
-        })
-    }
-
-    pub(crate) fn write_lock_or_restart(
-        &self,
-    ) -> Result<WriteLockedNodeRef<'e, V>, ConcurrentUpdateError> {
-        self.lockword().write_lock_or_restart()?;
-        Ok(WriteLockedNodeRef {
-            ptr: self.ptr,
-            phantom: self.phantom,
-        })
-    }
-
-    fn lockword(&self) -> &AtomicLockAndVersion {
-        self.ptr.lockword()
-    }
-}
-
-/// A reference to a node that has been optimistically read-locked. The functions re-check
-/// the version after each read.
-pub struct ReadLockedNodeRef<'e, V> {
-    ptr: NodePtr<V>,
-    version: u64,
-
-    phantom: PhantomData<&'e EpochPin<'e>>,
-}
-
-impl<'e, V: Value> ReadLockedNodeRef<'e, V> {
-    pub(crate) fn is_leaf(&self) -> bool {
-        self.ptr.is_leaf()
-    }
-
-    pub(crate) fn is_full(&self) -> bool {
-        self.ptr.is_full()
-    }
-
-    pub(crate) fn get_prefix(&self) -> &[u8] {
-        self.ptr.get_prefix()
-    }
-
-    /// Note: because we're only holding a read lock, the prefix can change concurrently.
-    /// You must be prepared to restart, if read_unlock() returns error later.
-    ///
-    /// Returns the length of the prefix, or None if it's not a match
-    pub(crate) fn prefix_matches(&self, key: &[u8]) -> Option<usize> {
-        self.ptr.prefix_matches(key)
-    }
-
-    pub(crate) fn find_child_or_restart(
-        &self,
-        key_byte: u8,
-    ) -> Result<Option<NodeRef<'e, V>>, ConcurrentUpdateError> {
-        let child_or_value = self.ptr.find_child(key_byte);
-        self.ptr.lockword().check_or_restart(self.version)?;
-
-        match child_or_value {
-            None => Ok(None),
-            Some(child_ptr) => Ok(Some(NodeRef {
-                ptr: child_ptr,
-                phantom: self.phantom,
-            })),
-        }
-    }
-
-    pub(crate) fn find_next_child_or_restart(
-        &self,
-        min_key_byte: u8,
-    ) -> Result<Option<(u8, NodeRef<'e, V>)>, ConcurrentUpdateError> {
-        let child_or_value = self.ptr.find_next_child(min_key_byte);
-        self.ptr.lockword().check_or_restart(self.version)?;
-
-        match child_or_value {
-            None => Ok(None),
-            Some((k, child_ptr)) => Ok(Some((
-                k,
-                NodeRef {
-                    ptr: child_ptr,
-                    phantom: self.phantom,
-                },
-            ))),
-        }
-    }
-
-    pub(crate) fn get_leaf_value_ptr(&self) -> Result<*const V, ConcurrentUpdateError> {
-        let result = self.ptr.get_leaf_value();
-        self.ptr.lockword().check_or_restart(self.version)?;
-
-        // Extend the lifetime.
-        let result = std::ptr::from_ref(result);
-
-        Ok(result)
-    }
-
-    pub(crate) fn upgrade_to_write_lock_or_restart(
-        self,
-    ) -> Result<WriteLockedNodeRef<'e, V>, ConcurrentUpdateError> {
-        self.ptr
-            .lockword()
-            .upgrade_to_write_lock_or_restart(self.version)?;
-
-        Ok(WriteLockedNodeRef {
-            ptr: self.ptr,
-            phantom: self.phantom,
-        })
-    }
-
-    pub(crate) fn read_unlock_or_restart(self) -> Result<(), ConcurrentUpdateError> {
-        self.ptr.lockword().check_or_restart(self.version)?;
-        Ok(())
-    }
-
-    pub(crate) fn check_or_restart(&self) -> Result<(), ConcurrentUpdateError> {
-        self.ptr.lockword().check_or_restart(self.version)?;
-        Ok(())
-    }
-}
-
-/// A reference to a node that has been optimistically read-locked. The functions re-check
-/// the version after each read.
-pub struct WriteLockedNodeRef<'e, V> {
-    ptr: NodePtr<V>,
-    phantom: PhantomData<&'e EpochPin<'e>>,
-}
-
-impl<'e, V: Value> WriteLockedNodeRef<'e, V> {
-    pub(crate) fn can_shrink(&self) -> bool {
-        self.ptr.can_shrink()
-    }
-
-    pub(crate) fn num_children(&self) -> usize {
-        self.ptr.num_children()
-    }
-
-    pub(crate) fn write_unlock(mut self) {
-        self.ptr.lockword().write_unlock();
-        self.ptr = NodePtr::null();
-    }
-
-    pub(crate) fn write_unlock_obsolete(mut self) {
-        self.ptr.lockword().write_unlock_obsolete();
-        self.ptr = NodePtr::null();
-    }
-
-    pub(crate) fn get_prefix(&self) -> &[u8] {
-        self.ptr.get_prefix()
-    }
-
-    pub(crate) fn truncate_prefix(&mut self, new_prefix_len: usize) {
-        self.ptr.truncate_prefix(new_prefix_len)
-    }
-
-    pub(crate) fn prepend_prefix(&mut self, prefix: &[u8], prefix_byte: u8) {
-        self.ptr.prepend_prefix(prefix, prefix_byte)
-    }
-
-    pub(crate) fn insert_child(&mut self, key_byte: u8, child: NodePtr<V>) {
-        self.ptr.insert_child(key_byte, child)
-    }
-
-    pub(crate) fn get_leaf_value_mut(&mut self) -> &mut V {
-        self.ptr.get_leaf_value_mut()
-    }
-
-    pub(crate) fn grow<'a, A>(
-        &self,
-        allocator: &'a A,
-    ) -> Result<NewNodeRef<'a, V, A>, OutOfMemoryError>
-    where
-        A: ArtAllocator<V>,
-    {
-        let new_node = self.ptr.grow(allocator)?;
-        Ok(NewNodeRef {
-            ptr: new_node,
-            allocator,
-            extra_nodes: Vec::new(),
-        })
-    }
-
-    pub(crate) fn shrink<'a, A>(
-        &self,
-        allocator: &'a A,
-    ) -> Result<NewNodeRef<'a, V, A>, OutOfMemoryError>
-    where
-        A: ArtAllocator<V>,
-    {
-        let new_node = self.ptr.shrink(allocator)?;
-        Ok(NewNodeRef {
-            ptr: new_node,
-            allocator,
-            extra_nodes: Vec::new(),
-        })
-    }
-
-    pub(crate) fn as_ptr(&self) -> NodePtr<V> {
-        self.ptr
-    }
-
-    pub(crate) fn replace_child(&mut self, key_byte: u8, replacement: NodePtr<V>) {
-        self.ptr.replace_child(key_byte, replacement);
-    }
-
-    pub(crate) fn delete_child(&mut self, key_byte: u8) {
-        self.ptr.delete_child(key_byte);
-    }
-
-    pub(crate) fn find_remaining_child(&self) -> (u8, NodeRef<'e, V>) {
-        assert_eq!(self.num_children(), 1);
-        let child_or_value = self.ptr.find_next_child(0);
-
-        match child_or_value {
-            None => panic!("could not find only child in node"),
-            Some((k, child_ptr)) => (
-                k,
-                NodeRef {
-                    ptr: child_ptr,
-                    phantom: self.phantom,
-                },
-            ),
-        }
-    }
-}
-
-impl<'e, V> Drop for WriteLockedNodeRef<'e, V> {
-    fn drop(&mut self) {
-        if !self.ptr.is_null() {
-            self.ptr.lockword().write_unlock();
-        }
-    }
-}
-
-pub(crate) struct NewNodeRef<'a, V, A>
-where
-    V: Value,
-    A: ArtAllocator<V>,
-{
-    ptr: NodePtr<V>,
-    allocator: &'a A,
-
-    extra_nodes: Vec<NodePtr<V>>,
-}
-
-impl<'a, V, A> NewNodeRef<'a, V, A>
-where
-    V: Value,
-    A: ArtAllocator<V>,
-{
-    pub(crate) fn insert_old_child(&mut self, key_byte: u8, child: &WriteLockedNodeRef<V>) {
-        self.ptr.insert_child(key_byte, child.as_ptr())
-    }
-
-    pub(crate) fn into_ptr(mut self) -> NodePtr<V> {
-        let ptr = self.ptr;
-        self.ptr = NodePtr::null();
-        ptr
-    }
-
-    pub(crate) fn insert_new_child(&mut self, key_byte: u8, child: NewNodeRef<'a, V, A>) {
-        let child_ptr = child.into_ptr();
-        self.ptr.insert_child(key_byte, child_ptr);
-        self.extra_nodes.push(child_ptr);
-    }
-}
-
-impl<'a, V, A> Drop for NewNodeRef<'a, V, A>
-where
-    V: Value,
-    A: ArtAllocator<V>,
-{
-    /// This drop implementation deallocates the newly allocated node, if into_ptr() was not called.
-    fn drop(&mut self) {
-        if !self.ptr.is_null() {
-            self.ptr.deallocate(self.allocator);
-            for p in self.extra_nodes.iter() {
-                p.deallocate(self.allocator);
-            }
-        }
-    }
-}
-
-pub(crate) fn new_internal<'a, V, A>(
-    prefix: &[u8],
-    allocator: &'a A,
-) -> Result<NewNodeRef<'a, V, A>, OutOfMemoryError>
-where
-    V: Value,
-    A: ArtAllocator<V>,
-{
-    Ok(NewNodeRef {
-        ptr: node_ptr::new_internal(prefix, allocator)?,
-        allocator,
-        extra_nodes: Vec::new(),
-    })
-}
-
-pub(crate) fn new_leaf<'a, V, A>(
-    prefix: &[u8],
-    value: V,
-    allocator: &'a A,
-) -> Result<NewNodeRef<'a, V, A>, OutOfMemoryError>
-where
-    V: Value,
-    A: ArtAllocator<V>,
-{
-    Ok(NewNodeRef {
-        ptr: node_ptr::new_leaf(prefix, value, allocator)?,
-        allocator,
-        extra_nodes: Vec::new(),
-    })
-}

libs/neonart/src/allocator.rs

@@ -1,156 +0,0 @@
-pub mod block;
-mod multislab;
-mod slab;
-pub mod r#static;
-
-use std::alloc::Layout;
-use std::marker::PhantomData;
-use std::mem::MaybeUninit;
-use std::sync::atomic::Ordering;
-
-use crate::allocator::multislab::MultiSlabAllocator;
-use crate::allocator::r#static::alloc_from_slice;
-
-use spin;
-
-use crate::Tree;
-pub use crate::algorithm::node_ptr::{
-    NodeInternal4, NodeInternal16, NodeInternal48, NodeInternal256, NodeLeaf,
-};
-
-#[derive(Debug)]
-pub struct OutOfMemoryError();
-
-pub trait ArtAllocator<V: crate::Value> {
-    fn alloc_tree(&self) -> *mut Tree<V>;
-
-    fn alloc_node_internal4(&self) -> *mut NodeInternal4<V>;
-    fn alloc_node_internal16(&self) -> *mut NodeInternal16<V>;
-    fn alloc_node_internal48(&self) -> *mut NodeInternal48<V>;
-    fn alloc_node_internal256(&self) -> *mut NodeInternal256<V>;
-    fn alloc_node_leaf(&self) -> *mut NodeLeaf<V>;
-
-    fn dealloc_node_internal4(&self, ptr: *mut NodeInternal4<V>);
-    fn dealloc_node_internal16(&self, ptr: *mut NodeInternal16<V>);
-    fn dealloc_node_internal48(&self, ptr: *mut NodeInternal48<V>);
-    fn dealloc_node_internal256(&self, ptr: *mut NodeInternal256<V>);
-    fn dealloc_node_leaf(&self, ptr: *mut NodeLeaf<V>);
-}
-
-pub struct ArtMultiSlabAllocator<'t, V>
-where
-    V: crate::Value,
-{
-    tree_area: spin::Mutex<Option<&'t mut MaybeUninit<Tree<V>>>>,
-
-    pub(crate) inner: MultiSlabAllocator<'t, 5>,
-
-    phantom_val: PhantomData<V>,
-}
-
-impl<'t, V: crate::Value> ArtMultiSlabAllocator<'t, V> {
-    const LAYOUTS: [Layout; 5] = [
-        Layout::new::<NodeInternal4<V>>(),
-        Layout::new::<NodeInternal16<V>>(),
-        Layout::new::<NodeInternal48<V>>(),
-        Layout::new::<NodeInternal256<V>>(),
-        Layout::new::<NodeLeaf<V>>(),
-    ];
-
-    pub fn new(area: &'t mut [MaybeUninit<u8>]) -> &'t mut ArtMultiSlabAllocator<'t, V> {
-        let (allocator_area, remain) = alloc_from_slice::<ArtMultiSlabAllocator<V>>(area);
-        let (tree_area, remain) = alloc_from_slice::<Tree<V>>(remain);
-
-        allocator_area.write(ArtMultiSlabAllocator {
-            tree_area: spin::Mutex::new(Some(tree_area)),
-            inner: MultiSlabAllocator::new(remain, &Self::LAYOUTS),
-            phantom_val: PhantomData,
-        })
-    }
-}
-
-impl<'t, V: crate::Value> ArtAllocator<V> for ArtMultiSlabAllocator<'t, V> {
-    fn alloc_tree(&self) -> *mut Tree<V> {
-        let mut t = self.tree_area.lock();
-        if let Some(tree_area) = t.take() {
-            return tree_area.as_mut_ptr().cast();
-        }
-        panic!("cannot allocate more than one tree");
-    }
-
-    fn alloc_node_internal4(&self) -> *mut NodeInternal4<V> {
-        self.inner.alloc_slab(0).cast()
-    }
-    fn alloc_node_internal16(&self) -> *mut NodeInternal16<V> {
-        self.inner.alloc_slab(1).cast()
-    }
-    fn alloc_node_internal48(&self) -> *mut NodeInternal48<V> {
-        self.inner.alloc_slab(2).cast()
-    }
-    fn alloc_node_internal256(&self) -> *mut NodeInternal256<V> {
-        self.inner.alloc_slab(3).cast()
-    }
-    fn alloc_node_leaf(&self) -> *mut NodeLeaf<V> {
-        self.inner.alloc_slab(4).cast()
-    }
-
-    fn dealloc_node_internal4(&self, ptr: *mut NodeInternal4<V>) {
-        self.inner.dealloc_slab(0, ptr.cast())
-    }
-
-    fn dealloc_node_internal16(&self, ptr: *mut NodeInternal16<V>) {
-        self.inner.dealloc_slab(1, ptr.cast())
-    }
-    fn dealloc_node_internal48(&self, ptr: *mut NodeInternal48<V>) {
-        self.inner.dealloc_slab(2, ptr.cast())
-    }
-    fn dealloc_node_internal256(&self, ptr: *mut NodeInternal256<V>) {
-        self.inner.dealloc_slab(3, ptr.cast())
-    }
-    fn dealloc_node_leaf(&self, ptr: *mut NodeLeaf<V>) {
-        self.inner.dealloc_slab(4, ptr.cast())
-    }
-}
-
-impl<'t, V: crate::Value> ArtMultiSlabAllocator<'t, V> {
-    pub(crate) fn get_statistics(&self) -> ArtMultiSlabStats {
-        ArtMultiSlabStats {
-            num_internal4: self.inner.slab_descs[0]
-                .num_allocated
-                .load(Ordering::Relaxed),
-            num_internal16: self.inner.slab_descs[1]
-                .num_allocated
-                .load(Ordering::Relaxed),
-            num_internal48: self.inner.slab_descs[2]
-                .num_allocated
-                .load(Ordering::Relaxed),
-            num_internal256: self.inner.slab_descs[3]
-                .num_allocated
-                .load(Ordering::Relaxed),
-            num_leaf: self.inner.slab_descs[4]
-                .num_allocated
-                .load(Ordering::Relaxed),
-
-            num_blocks_internal4: self.inner.slab_descs[0].num_blocks.load(Ordering::Relaxed),
-            num_blocks_internal16: self.inner.slab_descs[1].num_blocks.load(Ordering::Relaxed),
-            num_blocks_internal48: self.inner.slab_descs[2].num_blocks.load(Ordering::Relaxed),
-            num_blocks_internal256: self.inner.slab_descs[3].num_blocks.load(Ordering::Relaxed),
-            num_blocks_leaf: self.inner.slab_descs[4].num_blocks.load(Ordering::Relaxed),
-        }
-    }
-}
-
-#[derive(Clone, Debug)]
-pub struct ArtMultiSlabStats {
-    pub num_internal4: u64,
-    pub num_internal16: u64,
-    pub num_internal48: u64,
-    pub num_internal256: u64,
-    pub num_leaf: u64,
-
-    pub num_blocks_internal4: u64,
-    pub num_blocks_internal16: u64,
-    pub num_blocks_internal48: u64,
-    pub num_blocks_internal256: u64,
-    pub num_blocks_leaf: u64,
-}

libs/neonart/src/allocator/block.rs

@@ -1,191 +0,0 @@
-//! Simple allocator of fixed-size blocks
-
-use std::mem::MaybeUninit;
-use std::sync::atomic::{AtomicU64, Ordering};
-
-use spin;
-
-pub const BLOCK_SIZE: usize = 16 * 1024;
-
-const INVALID_BLOCK: u64 = u64::MAX;
-
-pub(crate) struct BlockAllocator<'t> {
-    blocks_ptr: &'t [MaybeUninit<u8>],
-    num_blocks: u64,
-    num_initialized: AtomicU64,
-
-    freelist_head: spin::Mutex<u64>,
-}
-
-struct FreeListBlock {
-    inner: spin::Mutex<FreeListBlockInner>,
-}
-
-struct FreeListBlockInner {
-    next: u64,
-
-    num_free_blocks: u64,
-    free_blocks: [u64; 100], // FIXME: fill the rest of the block
-}
-
-impl<'t> BlockAllocator<'t> {
-    pub(crate) fn new(area: &'t mut [MaybeUninit<u8>]) -> Self {
-        // Use all the space for the blocks
-        let padding = area.as_ptr().align_offset(BLOCK_SIZE);
-        let remain = &mut area[padding..];
-
-        let num_blocks = (remain.len() / BLOCK_SIZE) as u64;
-
-        BlockAllocator {
-            blocks_ptr: remain,
-            num_blocks,
-            num_initialized: AtomicU64::new(0),
-            freelist_head: spin::Mutex::new(INVALID_BLOCK),
-        }
-    }
-
-    /// safety: you must hold a lock on the pointer to this block, otherwise it might get
-    /// reused for another kind of block
-    fn read_freelist_block(&self, blkno: u64) -> &FreeListBlock {
-        let ptr: *const FreeListBlock = self.get_block_ptr(blkno).cast();
-        unsafe { ptr.as_ref().unwrap() }
-    }
-
-    fn get_block_ptr(&self, blkno: u64) -> *mut u8 {
-        assert!(blkno < self.num_blocks);
-        unsafe {
-            self.blocks_ptr
-                .as_ptr()
-                .byte_offset(blkno as isize * BLOCK_SIZE as isize)
-        }
-        .cast_mut()
-        .cast()
-    }
-
-    #[allow(clippy::mut_from_ref)]
-    pub(crate) fn alloc_block(&self) -> &mut [MaybeUninit<u8>] {
-        // FIXME: handle OOM
-        let blkno = self.alloc_block_internal();
-        if blkno == INVALID_BLOCK {
-            panic!("out of memory");
-        }
-
-        let ptr: *mut MaybeUninit<u8> = self.get_block_ptr(blkno).cast();
-        unsafe { std::slice::from_raw_parts_mut(ptr, BLOCK_SIZE) }
-    }
-
-    fn alloc_block_internal(&self) -> u64 {
-        //  check the free list.
-        {
-            let mut freelist_head = self.freelist_head.lock();
-            if *freelist_head != INVALID_BLOCK {
-                let freelist_block = self.read_freelist_block(*freelist_head);
-
-                // acquire lock on the freelist block before releasing the lock on the parent (i.e. lock coupling)
-                let mut g = freelist_block.inner.lock();
-
-                if g.num_free_blocks > 0 {
-                    g.num_free_blocks -= 1;
-                    let result = g.free_blocks[g.num_free_blocks as usize];
-                    return result;
-                } else {
-                    // consume the freelist block itself
-                    let result = *freelist_head;
-                    *freelist_head = g.next;
-                    // This freelist block is now unlinked and can be repurposed
-                    drop(g);
-                    return result;
-                }
-            }
-        }
-
-        // If there are some blocks left that we've never used, pick next such block
-        let mut next_uninitialized = self.num_initialized.load(Ordering::Relaxed);
-        while next_uninitialized < self.num_blocks {
-            match self.num_initialized.compare_exchange(
-                next_uninitialized,
-                next_uninitialized + 1,
-                Ordering::Relaxed,
-                Ordering::Relaxed,
-            ) {
-                Ok(_) => {
-                    return next_uninitialized;
-                }
-                Err(old) => {
-                    next_uninitialized = old;
-                    continue;
-                }
-            }
-        }
-
-        // out of blocks
-        INVALID_BLOCK
-    }
-
-    // TODO: this is currently unused. The slab allocator never releases blocks
-    #[allow(dead_code)]
-    pub(crate) fn release_block(&self, block_ptr: *mut u8) {
-        let blockno = unsafe { block_ptr.byte_offset_from(self.blocks_ptr) / BLOCK_SIZE as isize };
-        self.release_block_internal(blockno as u64);
-    }
-
-    fn release_block_internal(&self, blockno: u64) {
-        let mut freelist_head = self.freelist_head.lock();
-        if *freelist_head != INVALID_BLOCK {
-            let freelist_block = self.read_freelist_block(*freelist_head);
-
-            // acquire lock on the freelist block before releasing the lock on the parent (i.e. lock coupling)
-            let mut g = freelist_block.inner.lock();
-
-            let num_free_blocks = g.num_free_blocks;
-            if num_free_blocks < g.free_blocks.len() as u64 {
-                g.free_blocks[num_free_blocks as usize] = blockno;
-                g.num_free_blocks += 1;
-                return;
-            }
-        }
-
-        // Convert the block into a new freelist block
-        let block_ptr: *mut FreeListBlock = self.get_block_ptr(blockno).cast();
-        let init = FreeListBlock {
-            inner: spin::Mutex::new(FreeListBlockInner {
-                next: *freelist_head,
-                num_free_blocks: 0,
-                free_blocks: [INVALID_BLOCK; 100],
-            }),
-        };
-        unsafe { (*block_ptr) = init };
-        *freelist_head = blockno;
-    }
-
-    // for debugging
-    pub(crate) fn get_statistics(&self) -> BlockAllocatorStats {
-        let mut num_free_blocks = 0;
-
-        let mut _prev_lock = None;
-        let head_lock = self.freelist_head.lock();
-        let mut next_blk = *head_lock;
-        let mut _head_lock = Some(head_lock);
-        while next_blk != INVALID_BLOCK {
-            let freelist_block = self.read_freelist_block(next_blk);
-            let lock = freelist_block.inner.lock();
-            num_free_blocks += lock.num_free_blocks;
-            next_blk = lock.next;
-            _prev_lock = Some(lock); // hold the lock until we've read the next block
-            _head_lock = None;
-        }
-
-        BlockAllocatorStats {
-            num_blocks: self.num_blocks,
-            num_initialized: self.num_initialized.load(Ordering::Relaxed),
-            num_free_blocks,
-        }
-    }
-}
-
-#[derive(Clone, Debug)]
-pub struct BlockAllocatorStats {
-    pub num_blocks: u64,
-    pub num_initialized: u64,
-    pub num_free_blocks: u64,
-}

libs/neonart/src/allocator/multislab.rs

@@ -1,33 +0,0 @@
-use std::alloc::Layout;
-use std::mem::MaybeUninit;
-
-use crate::allocator::block::BlockAllocator;
-use crate::allocator::slab::SlabDesc;
-
-pub struct MultiSlabAllocator<'t, const N: usize> {
-    pub(crate) block_allocator: BlockAllocator<'t>,
-
-    pub(crate) slab_descs: [SlabDesc; N],
-}
-
-impl<'t, const N: usize> MultiSlabAllocator<'t, N> {
-    pub(crate) fn new(
-        area: &'t mut [MaybeUninit<u8>],
-        layouts: &[Layout; N],
-    ) -> MultiSlabAllocator<'t, N> {
-        let block_allocator = BlockAllocator::new(area);
-        MultiSlabAllocator {
-            block_allocator,
-
-            slab_descs: std::array::from_fn(|i| SlabDesc::new(&layouts[i])),
-        }
-    }
-
-    pub(crate) fn alloc_slab(&self, slab_idx: usize) -> *mut u8 {
-        self.slab_descs[slab_idx].alloc_chunk(&self.block_allocator)
-    }
-
-    pub(crate) fn dealloc_slab(&self, slab_idx: usize, ptr: *mut u8) {
-        self.slab_descs[slab_idx].dealloc_chunk(ptr, &self.block_allocator)
-    }
-}

libs/neonart/src/allocator/slab.rs

@@ -1,433 +0,0 @@
-//! A slab allocator that carves out fixed-size chunks from larger blocks.
-//!
-//!
-
-use std::alloc::Layout;
-use std::mem::MaybeUninit;
-use std::ops::Deref;
-use std::sync::atomic::{AtomicU32, AtomicU64, Ordering};
-
-use spin;
-
-use super::alloc_from_slice;
-use super::block::BlockAllocator;
-
-use crate::allocator::block::BLOCK_SIZE;
-
-pub(crate) struct SlabDesc {
-    pub(crate) layout: Layout,
-
-    block_lists: spin::RwLock<BlockLists>,
-
-    pub(crate) num_blocks: AtomicU64,
-    pub(crate) num_allocated: AtomicU64,
-}
-
-// FIXME: Not sure if SlabDesc is really Sync or Send. It probably is when it's empty, but
-// 'block_lists' contains pointers when it's not empty. In the current use as part of the
-// the art tree, SlabDescs are only moved during initialization.
-unsafe impl Sync for SlabDesc {}
-unsafe impl Send for SlabDesc {}
-
-#[derive(Default, Debug)]
-struct BlockLists {
-    full_blocks: BlockList,
-    nonfull_blocks: BlockList,
-}
-
-impl BlockLists {
-    // Unlink a node. It must be in either one of the two lists.
-    unsafe fn unlink(&mut self, elem: *mut SlabBlockHeader) {
-        let list = unsafe {
-            if (*elem).next.is_null() {
-                if self.full_blocks.tail == elem {
-                    Some(&mut self.full_blocks)
-                } else {
-                    Some(&mut self.nonfull_blocks)
-                }
-            } else if (*elem).prev.is_null() {
-                if self.full_blocks.head == elem {
-                    Some(&mut self.full_blocks)
-                } else {
-                    Some(&mut self.nonfull_blocks)
-                }
-            } else {
-                None
-            }
-        };
-        unsafe { unlink_slab_block(list, elem) };
-    }
-}
-
-unsafe fn unlink_slab_block(mut list: Option<&mut BlockList>, elem: *mut SlabBlockHeader) {
-    unsafe {
-        if (*elem).next.is_null() {
-            assert_eq!(list.as_ref().unwrap().tail, elem);
-            list.as_mut().unwrap().tail = (*elem).prev;
-        } else {
-            assert_eq!((*(*elem).next).prev, elem);
-            (*(*elem).next).prev = (*elem).prev;
-        }
-        if (*elem).prev.is_null() {
-            assert_eq!(list.as_ref().unwrap().head, elem);
-            list.as_mut().unwrap().head = (*elem).next;
-        } else {
-            assert_eq!((*(*elem).prev).next, elem);
-            (*(*elem).prev).next = (*elem).next;
-        }
-    }
-}
-
-#[derive(Debug)]
-struct BlockList {
-    head: *mut SlabBlockHeader,
-    tail: *mut SlabBlockHeader,
-}
-
-impl Default for BlockList {
-    fn default() -> Self {
-        BlockList {
-            head: std::ptr::null_mut(),
-            tail: std::ptr::null_mut(),
-        }
-    }
-}
-
-impl BlockList {
-    unsafe fn push_head(&mut self, elem: *mut SlabBlockHeader) {
-        unsafe {
-            if self.is_empty() {
-                self.tail = elem;
-                (*elem).next = std::ptr::null_mut();
-            } else {
-                (*elem).next = self.head;
-                (*self.head).prev = elem;
-            }
-            (*elem).prev = std::ptr::null_mut();
-            self.head = elem;
-        }
-    }
-
-    fn is_empty(&self) -> bool {
-        self.head.is_null()
-    }
-
-    unsafe fn unlink(&mut self, elem: *mut SlabBlockHeader) {
-        unsafe { unlink_slab_block(Some(self), elem) }
-    }
-
-    #[cfg(test)]
-    fn dump(&self) {
-        let mut next = self.head;
-
-        while !next.is_null() {
-            let n = unsafe { next.as_ref() }.unwrap();
-            eprintln!(
-                "  blk {:?} (free {}/{})",
-                next,
-                n.num_free_chunks.load(Ordering::Relaxed),
-                n.num_chunks
-            );
-            next = n.next;
-        }
-    }
-}
-
-impl SlabDesc {
-    pub(crate) fn new(layout: &Layout) -> SlabDesc {
-        SlabDesc {
-            layout: *layout,
-            block_lists: spin::RwLock::new(BlockLists::default()),
-            num_allocated: AtomicU64::new(0),
-            num_blocks: AtomicU64::new(0),
-        }
-    }
-}
-
-#[derive(Debug)]
-struct SlabBlockHeader {
-    free_chunks_head: spin::Mutex<*mut FreeChunk>,
-    num_free_chunks: AtomicU32,
-    num_chunks: u32, // this is really a constant for a given Layout
-
-    // these fields are protected by the lock on the BlockLists
-    prev: *mut SlabBlockHeader,
-    next: *mut SlabBlockHeader,
-}
-
-struct FreeChunk {
-    next: *mut FreeChunk,
-}
-
-enum ReadOrWriteGuard<'a, T> {
-    Read(spin::RwLockReadGuard<'a, T>),
-    Write(spin::RwLockWriteGuard<'a, T>),
-}
-
-impl<'a, T> Deref for ReadOrWriteGuard<'a, T> {
-    type Target = T;
-
-    fn deref(&self) -> &<Self as Deref>::Target {
-        match self {
-            ReadOrWriteGuard::Read(g) => g.deref(),
-            ReadOrWriteGuard::Write(g) => g.deref(),
-        }
-    }
-}
-
-impl SlabDesc {
-    pub fn alloc_chunk(&self, block_allocator: &BlockAllocator) -> *mut u8 {
-        // Are there any free chunks?
-        let mut acquire_write = false;
-        'outer: loop {
-            let mut block_lists_guard = if acquire_write {
-                ReadOrWriteGuard::Write(self.block_lists.write())
-            } else {
-                ReadOrWriteGuard::Read(self.block_lists.read())
-            };
-            'inner: loop {
-                let block_ptr = block_lists_guard.nonfull_blocks.head;
-                if block_ptr.is_null() {
-                    break 'outer;
-                }
-                unsafe {
-                    let mut free_chunks_head = (*block_ptr).free_chunks_head.lock();
-                    if !(*free_chunks_head).is_null() {
-                        let result = *free_chunks_head;
-                        (*free_chunks_head) = (*result).next;
-                        let _old = (*block_ptr).num_free_chunks.fetch_sub(1, Ordering::Relaxed);
-
-                        self.num_allocated.fetch_add(1, Ordering::Relaxed);
-                        return result.cast();
-                    }
-                }
-
-                // The block at the head of the list was full. Grab write lock and retry
-                match block_lists_guard {
-                    ReadOrWriteGuard::Read(_) => {
-                        acquire_write = true;
-                        continue 'outer;
-                    }
-                    ReadOrWriteGuard::Write(ref mut g) => {
-                        // move the node to the list of full blocks
-                        unsafe {
-                            g.nonfull_blocks.unlink(block_ptr);
-                            g.full_blocks.push_head(block_ptr);
-                        };
-                        continue 'inner;
-                    }
-                }
-            }
-        }
-
-        // no free chunks. Allocate a new block (and the chunk from that)
-        let (new_block, new_chunk) = self.alloc_block_and_chunk(block_allocator);
-        self.num_blocks.fetch_add(1, Ordering::Relaxed);
-
-        // Add the block to the list in the SlabDesc
-        unsafe {
-            let mut block_lists_guard = self.block_lists.write();
-            block_lists_guard.nonfull_blocks.push_head(new_block);
-        }
-        self.num_allocated.fetch_add(1, Ordering::Relaxed);
-        new_chunk
-    }
-
-    pub fn dealloc_chunk(&self, chunk_ptr: *mut u8, _block_allocator: &BlockAllocator) {
-        // Find the block it belongs to. You can find the block from the address. (And knowing the
-        // layout, you could calculate the chunk number too.)
-        let block_ptr: *mut SlabBlockHeader = {
-            let block_addr = (chunk_ptr.addr() / BLOCK_SIZE) * BLOCK_SIZE;
-            chunk_ptr.with_addr(block_addr).cast()
-        };
-        let chunk_ptr: *mut FreeChunk = chunk_ptr.cast();
-
-        // Mark the chunk as free in 'freechunks' list
-        let num_chunks;
-        let num_free_chunks;
-        unsafe {
-            let mut free_chunks_head = (*block_ptr).free_chunks_head.lock();
-            (*chunk_ptr).next = *free_chunks_head;
-            *free_chunks_head = chunk_ptr;
-
-            num_free_chunks = (*block_ptr).num_free_chunks.fetch_add(1, Ordering::Relaxed) + 1;
-            num_chunks = (*block_ptr).num_chunks;
-        }
-
-        if num_free_chunks == 1 {
-            // If the block was full previously, add it to the nonfull blocks list. Note that
-            // we're not holding the lock anymore, so it can immediately become full again.
-            // That's harmless, it will be moved back to the full list again when a call
-            // to alloc_chunk() sees it.
-            let mut block_lists = self.block_lists.write();
-            unsafe {
-                block_lists.unlink(block_ptr);
-                block_lists.nonfull_blocks.push_head(block_ptr);
-            };
-        } else if num_free_chunks == num_chunks {
-            // If the block became completely empty, move it to the free list
-            // TODO
-            // FIXME: we're still holding the spinlock. It's not exactly safe to return it to
-            // the free blocks list, is it? Defer it as garbage to wait out concurrent updates?
-            //block_allocator.release_block()
-        }
-
-        // update stats
-        self.num_allocated.fetch_sub(1, Ordering::Relaxed);
-    }
-
-    fn alloc_block_and_chunk(
-        &self,
-        block_allocator: &BlockAllocator,
-    ) -> (*mut SlabBlockHeader, *mut u8) {
-        // fixme: handle OOM
-        let block_slice: &mut [MaybeUninit<u8>] = block_allocator.alloc_block();
-        let (block_header, remain) = alloc_from_slice::<SlabBlockHeader>(block_slice);
-
-        let padding = remain.as_ptr().align_offset(self.layout.align());
-
-        let num_chunks = (remain.len() - padding) / self.layout.size();
-
-        let first_chunk_ptr: *mut FreeChunk = remain[padding..].as_mut_ptr().cast();
-
-        unsafe {
-            let mut chunk_ptr = first_chunk_ptr;
-            for _ in 0..num_chunks - 1 {
-                let next_chunk_ptr = chunk_ptr.byte_add(self.layout.size());
-                (*chunk_ptr).next = next_chunk_ptr;
-                chunk_ptr = next_chunk_ptr;
-            }
-            (*chunk_ptr).next = std::ptr::null_mut();
-
-            let result_chunk = first_chunk_ptr;
-
-            let block_header = block_header.write(SlabBlockHeader {
-                free_chunks_head: spin::Mutex::new((*first_chunk_ptr).next),
-                prev: std::ptr::null_mut(),
-                next: std::ptr::null_mut(),
-                num_chunks: num_chunks as u32,
-                num_free_chunks: AtomicU32::new(num_chunks as u32 - 1),
-            });
-
-            (block_header, result_chunk.cast())
-        }
-    }
-
-    #[cfg(test)]
-    fn dump(&self) {
-        eprintln!(
-            "slab dump ({} blocks, {} allocated chunks)",
-            self.num_blocks.load(Ordering::Relaxed),
-            self.num_allocated.load(Ordering::Relaxed)
-        );
-        let lists = self.block_lists.read();
-
-        eprintln!("nonfull blocks:");
-        lists.nonfull_blocks.dump();
-        eprintln!("full blocks:");
-        lists.full_blocks.dump();
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    use rand::Rng;
-    use rand_distr::Zipf;
-
-    struct TestObject {
-        val: usize,
-        _dummy: [u8; BLOCK_SIZE / 4],
-    }
-
-    struct TestObjectSlab<'a>(SlabDesc, BlockAllocator<'a>);
-    impl<'a> TestObjectSlab<'a> {
-        fn new(block_allocator: BlockAllocator) -> TestObjectSlab {
-            TestObjectSlab(SlabDesc::new(&Layout::new::<TestObject>()), block_allocator)
-        }
-
-        fn alloc(&self, val: usize) -> *mut TestObject {
-            let obj: *mut TestObject = self.0.alloc_chunk(&self.1).cast();
-            unsafe { (*obj).val = val };
-            obj
-        }
-
-        fn dealloc(&self, obj: *mut TestObject) {
-            self.0.dealloc_chunk(obj.cast(), &self.1)
-        }
-    }
-
-    #[test]
-    fn test_slab_alloc() {
-        const MEM_SIZE: usize = 100000000;
-        let mut area = Box::new_uninit_slice(MEM_SIZE);
-        let block_allocator = BlockAllocator::new(&mut area);
-
-        let slab = TestObjectSlab::new(block_allocator);
-
-        let mut all: Vec<*mut TestObject> = Vec::new();
-        for i in 0..11 {
-            all.push(slab.alloc(i));
-        }
-        #[allow(clippy::needless_range_loop)]
-        for i in 0..11 {
-            assert!(unsafe { (*all[i]).val == i });
-        }
-
-        let distribution = Zipf::new(10.0, 1.1).unwrap();
-        let mut rng = rand::rng();
-        for _ in 0..100000 {
-            slab.0.dump();
-            let idx = rng.sample(distribution) as usize;
-            let ptr: *mut TestObject = all[idx];
-            if !ptr.is_null() {
-                assert_eq!(unsafe { (*ptr).val }, idx);
-                slab.dealloc(ptr);
-                all[idx] = std::ptr::null_mut();
-            } else {
-                all[idx] = slab.alloc(idx);
-            }
-        }
-    }
-
-    fn new_test_blk(i: u32) -> *mut SlabBlockHeader {
-        Box::into_raw(Box::new(SlabBlockHeader {
-            free_chunks_head: spin::Mutex::new(std::ptr::null_mut()),
-            num_free_chunks: AtomicU32::new(0),
-            num_chunks: i,
-            prev: std::ptr::null_mut(),
-            next: std::ptr::null_mut(),
-        }))
-    }
-
-    #[test]
-    fn test_block_linked_list() {
-        // note: these are leaked, but that's OK for tests
-        let a = new_test_blk(0);
-        let b = new_test_blk(1);
-
-        let mut list = BlockList::default();
-        assert!(list.is_empty());
-
-        unsafe {
-            list.push_head(a);
-            assert!(!list.is_empty());
-            list.unlink(a);
-        }
-        assert!(list.is_empty());
-
-        unsafe {
-            list.push_head(b);
-            list.push_head(a);
-            assert_eq!(list.head, a);
-            assert_eq!((*a).next, b);
-            assert_eq!((*b).prev, a);
-            assert_eq!(list.tail, b);
-
-            list.unlink(a);
-            list.unlink(b);
-            assert!(list.is_empty());
-        }
-    }
-}

libs/neonart/src/allocator/static.rs

@@ -1,44 +0,0 @@
-use std::mem::MaybeUninit;
-
-pub fn alloc_from_slice<T>(
-    area: &mut [MaybeUninit<u8>],
-) -> (&mut MaybeUninit<T>, &mut [MaybeUninit<u8>]) {
-    let layout = std::alloc::Layout::new::<T>();
-
-    let area_start = area.as_mut_ptr();
-
-    // pad to satisfy alignment requirements
-    let padding = area_start.align_offset(layout.align());
-    if padding + layout.size() > area.len() {
-        panic!("out of memory");
-    }
-    let area = &mut area[padding..];
-    let (result_area, remain) = area.split_at_mut(layout.size());
-
-    let result_ptr: *mut MaybeUninit<T> = result_area.as_mut_ptr().cast();
-    let result = unsafe { result_ptr.as_mut().unwrap() };
-
-    (result, remain)
-}
-
-pub fn alloc_array_from_slice<T>(
-    area: &mut [MaybeUninit<u8>],
-    len: usize,
-) -> (&mut [MaybeUninit<T>], &mut [MaybeUninit<u8>]) {
-    let layout = std::alloc::Layout::new::<T>();
-
-    let area_start = area.as_mut_ptr();
-
-    // pad to satisfy alignment requirements
-    let padding = area_start.align_offset(layout.align());
-    if padding + layout.size() * len > area.len() {
-        panic!("out of memory");
-    }
-    let area = &mut area[padding..];
-    let (result_area, remain) = area.split_at_mut(layout.size() * len);
-
-    let result_ptr: *mut MaybeUninit<T> = result_area.as_mut_ptr().cast();
-    let result = unsafe { std::slice::from_raw_parts_mut(result_ptr.as_mut().unwrap(), len) };
-
-    (result, remain)
-}

libs/neonart/src/epoch.rs

@@ -1,142 +0,0 @@
-//! This is similar to crossbeam_epoch crate, but works in shared memory
-
-use std::sync::atomic::{AtomicU64, AtomicUsize, Ordering};
-
-use crossbeam_utils::CachePadded;
-
-const NUM_SLOTS: usize = 1000;
-
-/// This is the struct that is stored in shmem
-///
-/// bit 0: is it pinned or not?
-/// rest of the bits are the epoch counter.
-pub struct EpochShared {
-    global_epoch: AtomicU64,
-    participants: [CachePadded<AtomicU64>; NUM_SLOTS],
-
-    broadcast_lock: spin::Mutex<()>,
-}
-
-impl EpochShared {
-    pub fn new() -> EpochShared {
-        EpochShared {
-            global_epoch: AtomicU64::new(2),
-            participants: [const { CachePadded::new(AtomicU64::new(2)) }; NUM_SLOTS],
-            broadcast_lock: spin::Mutex::new(()),
-        }
-    }
-
-    pub fn register(&self) -> LocalHandle {
-        LocalHandle {
-            global: self,
-            last_slot: AtomicUsize::new(0), // todo: choose more intelligently
-        }
-    }
-
-    fn release_pin(&self, slot: usize, _epoch: u64) {
-        let global_epoch = self.global_epoch.load(Ordering::Relaxed);
-        self.participants[slot].store(global_epoch, Ordering::Relaxed);
-    }
-
-    fn pin_internal(&self, slot_hint: usize) -> (usize, u64) {
-        // pick a slot
-        let mut slot = slot_hint;
-        let epoch = loop {
-            let old = self.participants[slot].fetch_or(1, Ordering::Relaxed);
-            if old & 1 == 0 {
-                // Got this slot
-                break old;
-            }
-
-            // the slot was busy by another thread / process. try a different slot
-            slot += 1;
-            if slot == NUM_SLOTS {
-                slot = 0;
-            }
-            continue;
-        };
-        (slot, epoch)
-    }
-
-    pub(crate) fn advance(&self) -> u64 {
-        // Advance the global epoch
-        let old_epoch = self.global_epoch.fetch_add(2, Ordering::Relaxed);
-        // Anyone that release their pin after this will update their slot.
-        old_epoch + 2
-    }
-
-    pub(crate) fn broadcast(&self) {
-        let Some(_guard) = self.broadcast_lock.try_lock() else {
-            return;
-        };
-
-        let epoch = self.global_epoch.load(Ordering::Relaxed);
-        let old_epoch = epoch.wrapping_sub(2);
-
-        // Update all free slots.
-        for i in 0..NUM_SLOTS {
-            // TODO: check result, as a sanity check. It should either be the old epoch, or pinned
-            let _ = self.participants[i].compare_exchange(
-                old_epoch,
-                epoch,
-                Ordering::Relaxed,
-                Ordering::Relaxed,
-            );
-        }
-
-        // FIXME: memory fence here, since we used Relaxed?
-    }
-
-    pub(crate) fn get_oldest(&self) -> u64 {
-        // Read all slots.
-        let now = self.global_epoch.load(Ordering::Relaxed);
-        let mut oldest = now;
-        for i in 0..NUM_SLOTS {
-            let this_epoch = self.participants[i].load(Ordering::Relaxed);
-            let delta = now.wrapping_sub(this_epoch);
-            if delta > u64::MAX / 2 {
-                // this is very recent
-            } else if delta > now.wrapping_sub(oldest) {
-                oldest = this_epoch;
-            }
-        }
-        oldest
-    }
-
-    pub(crate) fn get_current(&self) -> u64 {
-        self.global_epoch.load(Ordering::Relaxed)
-    }
-}
-
-pub(crate) struct EpochPin<'e> {
-    slot: usize,
-    pub(crate) epoch: u64,
-
-    handle: &'e LocalHandle<'e>,
-}
-
-impl<'e> Drop for EpochPin<'e> {
-    fn drop(&mut self) {
-        self.handle.global.release_pin(self.slot, self.epoch);
-    }
-}
-
-pub struct LocalHandle<'g> {
-    global: &'g EpochShared,
-
-    last_slot: AtomicUsize,
-}
-
-impl<'g> LocalHandle<'g> {
-    pub fn pin(&self) -> EpochPin {
-        let (slot, epoch) = self
-            .global
-            .pin_internal(self.last_slot.load(Ordering::Relaxed));
-        self.last_slot.store(slot, Ordering::Relaxed);
-        EpochPin {
-            handle: self,
-            epoch,
-            slot,
-        }
-    }
-}

libs/neonart/src/lib.rs

@@ -1,583 +0,0 @@
-//! Adaptive Radix Tree (ART) implementation, with Optimistic Lock Coupling.
-//!
-//! The data structure is described in these two papers:
-//!
-//! [1] Leis, V. & Kemper, Alfons & Neumann, Thomas. (2013).
-//!     The adaptive radix tree: ARTful indexing for main-memory databases.
-//!     Proceedings - International Conference on Data Engineering. 38-49. 10.1109/ICDE.2013.6544812.
-//!     https://db.in.tum.de/~leis/papers/ART.pdf
-//!
-//! [2] Leis, Viktor & Scheibner, Florian & Kemper, Alfons & Neumann, Thomas. (2016).
-//!     The ART of practical synchronization.
-//!     1-8. 10.1145/2933349.2933352.
-//!     https://db.in.tum.de/~leis/papers/artsync.pdf
-//!
-//! [1] describes the base data structure, and [2] describes the Optimistic Lock Coupling that we
-//! use.
-//!
-//! The papers mention a few different variants. We have made the following choices in this
-//! implementation:
-//!
-//! - All keys have the same length
-//!
-//! - Single-value leaves.
-//!
-//! - For collapsing inner nodes, we use the Pessimistic approach, where each inner node stores a
-//!   variable length "prefix", which stores the keys of all the one-way nodes which have been
-//!   removed. However, similar to the "hybrid" approach described in the paper, each node only has
-//!   space for a constant-size prefix of 8 bytes. If a node would have a longer prefix, then we
-//!   create create one-way nodes to store them. (There was no particular reason for this choice,
-//!   the "hybrid" approach described in the paper might be better.)
-//!
-//! - For concurrency, we use Optimistic Lock Coupling. The paper [2] also describes another method,
-//!   ROWEX, which generally performs better when there is contention, but that is not important
-//!   for use and Optimisic Lock Coupling is simpler to implement.
-//!
-//! ## Requirements
-//!
-//! This data structure is currently used for the integrated LFC, relsize and last-written LSN cache
-//! in the compute communicator, part of the 'neon' Postgres extension. We have some unique
-//! requirements, which is why we had to write our own. Namely:
-//!
-//! - The data structure has to live in fixed-sized shared memory segment. That rules out any
-//!   built-in Rust collections and most crates. (Except possibly with the 'allocator_api' rust
-//!   feature, which still nightly-only experimental as of this writing).
-//!
-//! - The data structure is accessed from multiple processes. Only one process updates the data
-//!   structure, but other processes perform reads. That rules out using built-in Rust locking
-//!   primitives like Mutex and RwLock, and most crates too.
-//!
-//! - Within the one process with write-access, multiple threads can perform updates concurrently.
-//!   That rules out using PostgreSQL LWLocks for the locking.
-//!
-//! The implementation is generic, and doesn't depend on any PostgreSQL specifics, but it has been
-//! written with that usage and the above constraints in mind. Some noteworthy assumptions:
-//!
-//! - Contention is assumed to be rare. In the integrated cache in PostgreSQL, there's higher level
-//!   locking in the PostgreSQL buffer manager, which ensures that two backends should not try to
-//!   read / write the same page at the same time. (Prefetching can conflict with actual reads,
-//!   however.)
-//!
-//!  - The keys in the integrated cache are 17 bytes long.
-//!
-//! ## Usage
-//!
-//! Because this is designed to be used as a Postgres shared memory data structure, initialization
-//! happens in three stages:
-//!
-//! 0. A fixed area of shared memory is allocated at postmaster startup.
-//!
-//! 1. TreeInitStruct::new() is called to initialize it, still in Postmaster process, before any
-//!    other process or thread is running. It returns a TreeInitStruct, which is inherited by all
-//!    the processes through fork().
-//!
-//! 2. One process may have write-access to the struct, by calling
-//!    [TreeInitStruct::attach_writer]. (That process is the communicator process.)
-//!
-//! 3. Other processes get read-access to the struct, by calling [TreeInitStruct::attach_reader]
-//!
-//! "Write access" means that you can insert / update / delete values in the tree.
-//!
-//! NOTE: The Values stored in the tree are sometimes moved, when a leaf node fills up and a new
-//! larger node needs to be allocated. The versioning and epoch-based allocator ensure that the data
-//! structure stays consistent, but if the Value has interior mutability, like atomic fields,
-//! updates to such fields might be lost if the leaf node is concurrently moved! If that becomes a
-//! problem, the version check could be passed up to the caller, so that the caller could detect the
-//! lost updates and retry the operation.
-//!
-//! ## Implementation
-//!
-//! node_ptr: Provides low-level implementations of the four different node types (eight actually,
-//! since there is an Internal and Leaf variant of each)
-//!
-//! lock_and_version.rs: Provides an abstraction for the combined lock and version counter on each
-//! node.
-//!
-//! node_ref.rs: The code in node_ptr.rs deals with raw pointers. node_ref.rs provides more type-safe
-//!   abstractions on top.
-//!
-//! algorithm.rs: Contains the functions to implement lookups and updates in the tree
-//!
-//! allocator.rs: Provides a facility to allocate memory for the tree nodes. (We must provide our
-//!   own abstraction for that because we need the data structure to live in a pre-allocated shared
-//!   memory segment).
-//!
-//! epoch.rs: The data structure requires that when a node is removed from the tree, it is not
-//!   immediately deallocated, but stays around for as long as concurrent readers might still have
-//!   pointers to them. This is enforced by an epoch system. This is similar to
-//!   e.g. crossbeam_epoch, but we couldn't use that either because it has to work across processes
-//!   communicating over the shared memory segment.
-//!
-//! ## See also
-//!
-//! There are some existing Rust ART implementations out there, but none of them filled all
-//! the requirements:
-//!
-//! - https://github.com/XiangpengHao/congee
-//! - https://github.com/declanvk/blart
-//!
-//! ## TODO
-//!
-//! - Removing values has not been implemented
-
-mod algorithm;
-pub mod allocator;
-mod epoch;
-
-use algorithm::RootPtr;
-use algorithm::node_ptr::NodePtr;
-
-use std::collections::VecDeque;
-use std::fmt::Debug;
-use std::marker::PhantomData;
-use std::ptr::NonNull;
-use std::sync::atomic::{AtomicBool, Ordering};
-
-use crate::epoch::EpochPin;
-
-#[cfg(test)]
-mod tests;
-
-use allocator::ArtAllocator;
-pub use allocator::ArtMultiSlabAllocator;
-pub use allocator::OutOfMemoryError;
-
-/// Fixed-length key type.
-///
-pub trait Key: Debug {
-    const KEY_LEN: usize;
-
-    fn as_bytes(&self) -> &[u8];
-}
-
-/// Values stored in the tree
-///
-/// Values need to be Cloneable, because when a node "grows", the value is copied to a new node and
-/// the old sticks around until all readers that might see the old value are gone.
-// fixme obsolete, no longer needs Clone
-pub trait Value {}
-
-const MAX_GARBAGE: usize = 1024;
-
-/// The root of the tree, plus other tree-wide data. This is stored in the shared memory.
-pub struct Tree<V: Value> {
-    /// For simplicity, so that we never need to grow or shrink the root, the root node is always an
-    /// Internal256 node. Also, it never has a prefix (that's actually a bit wasteful, incurring one
-    /// indirection to every lookup)
-    root: RootPtr<V>,
-
-    writer_attached: AtomicBool,
-
-    epoch: epoch::EpochShared,
-}
-
-unsafe impl<V: Value + Sync> Sync for Tree<V> {}
-unsafe impl<V: Value + Send> Send for Tree<V> {}
-
-struct GarbageQueue<V>(VecDeque<(NodePtr<V>, u64)>);
-
-unsafe impl<V: Value + Sync> Sync for GarbageQueue<V> {}
-unsafe impl<V: Value + Send> Send for GarbageQueue<V> {}
-
-impl<V> GarbageQueue<V> {
-    fn new() -> GarbageQueue<V> {
-        GarbageQueue(VecDeque::with_capacity(MAX_GARBAGE))
-    }
-
-    fn remember_obsolete_node(&mut self, ptr: NodePtr<V>, epoch: u64) {
-        self.0.push_front((ptr, epoch));
-    }
-
-    fn next_obsolete(&mut self, cutoff_epoch: u64) -> Option<NodePtr<V>> {
-        if let Some(back) = self.0.back() {
-            if back.1 < cutoff_epoch {
-                return Some(self.0.pop_back().unwrap().0);
-            }
-        }
-        None
-    }
-}
-
-/// Struct created at postmaster startup
-pub struct TreeInitStruct<'t, K: Key, V: Value, A: ArtAllocator<V>> {
-    tree: &'t Tree<V>,
-
-    allocator: &'t A,
-
-    phantom_key: PhantomData<K>,
-}
-
-/// The worker process has a reference to this. The write operations are only safe
-/// from the worker process
-pub struct TreeWriteAccess<'t, K: Key, V: Value, A: ArtAllocator<V>>
-where
-    K: Key,
-    V: Value,
-{
-    tree: &'t Tree<V>,
-
-    pub allocator: &'t A,
-
-    epoch_handle: epoch::LocalHandle<'t>,
-
-    phantom_key: PhantomData<K>,
-
-    /// Obsolete nodes that cannot be recycled until their epoch expires.
-    garbage: spin::Mutex<GarbageQueue<V>>,
-}
-
-/// The backends have a reference to this. It cannot be used to modify the tree
-pub struct TreeReadAccess<'t, K: Key, V: Value>
-where
-    K: Key,
-    V: Value,
-{
-    tree: &'t Tree<V>,
-
-    epoch_handle: epoch::LocalHandle<'t>,
-
-    phantom_key: PhantomData<K>,
-}
-
-impl<'t, K: Key, V: Value, A: ArtAllocator<V>> TreeInitStruct<'t, K, V, A> {
-    pub fn new(allocator: &'t A) -> TreeInitStruct<'t, K, V, A> {
-        let tree_ptr = allocator.alloc_tree();
-        let tree_ptr = NonNull::new(tree_ptr).expect("out of memory");
-        let init = Tree {
-            root: algorithm::new_root(allocator).expect("out of memory"),
-            writer_attached: AtomicBool::new(false),
-            epoch: epoch::EpochShared::new(),
-        };
-        unsafe { tree_ptr.write(init) };
-
-        TreeInitStruct {
-            tree: unsafe { tree_ptr.as_ref() },
-            allocator,
-            phantom_key: PhantomData,
-        }
-    }
-
-    pub fn attach_writer(self) -> TreeWriteAccess<'t, K, V, A> {
-        let previously_attached = self.tree.writer_attached.swap(true, Ordering::Relaxed);
-        if previously_attached {
-            panic!("writer already attached");
-        }
-        TreeWriteAccess {
-            tree: self.tree,
-            allocator: self.allocator,
-            phantom_key: PhantomData,
-            epoch_handle: self.tree.epoch.register(),
-            garbage: spin::Mutex::new(GarbageQueue::new()),
-        }
-    }
-
-    pub fn attach_reader(self) -> TreeReadAccess<'t, K, V> {
-        TreeReadAccess {
-            tree: self.tree,
-            phantom_key: PhantomData,
-            epoch_handle: self.tree.epoch.register(),
-        }
-    }
-}
-
-impl<'t, K: Key, V: Value, A: ArtAllocator<V>> TreeWriteAccess<'t, K, V, A> {
-    pub fn start_write<'g>(&'t self) -> TreeWriteGuard<'g, K, V, A>
-    where
-        't: 'g,
-    {
-        TreeWriteGuard {
-            tree_writer: self,
-            epoch_pin: self.epoch_handle.pin(),
-            phantom_key: PhantomData,
-            created_garbage: false,
-        }
-    }
-
-    pub fn start_read(&'t self) -> TreeReadGuard<'t, K, V> {
-        TreeReadGuard {
-            tree: self.tree,
-            epoch_pin: self.epoch_handle.pin(),
-            phantom_key: PhantomData,
-        }
-    }
-}
-
-impl<'t, K: Key, V: Value> TreeReadAccess<'t, K, V> {
-    pub fn start_read(&'t self) -> TreeReadGuard<'t, K, V> {
-        TreeReadGuard {
-            tree: self.tree,
-            epoch_pin: self.epoch_handle.pin(),
-            phantom_key: PhantomData,
-        }
-    }
-}
-
-pub struct TreeReadGuard<'e, K, V>
-where
-    K: Key,
-    V: Value,
-{
-    tree: &'e Tree<V>,
-
-    epoch_pin: EpochPin<'e>,
-    phantom_key: PhantomData<K>,
-}
-
-impl<'e, K: Key, V: Value> TreeReadGuard<'e, K, V> {
-    pub fn get(&'e self, key: &K) -> Option<&'e V> {
-        algorithm::search(key, self.tree.root, &self.epoch_pin)
-    }
-}
-
-pub struct TreeWriteGuard<'e, K, V, A>
-where
-    K: Key,
-    V: Value,
-    A: ArtAllocator<V>,
-{
-    tree_writer: &'e TreeWriteAccess<'e, K, V, A>,
-
-    epoch_pin: EpochPin<'e>,
-    phantom_key: PhantomData<K>,
-
-    created_garbage: bool,
-}
-
-pub enum UpdateAction<V> {
-    Nothing,
-    Insert(V),
-    Remove,
-}
-
-impl<'e, K: Key, V: Value, A: ArtAllocator<V>> TreeWriteGuard<'e, K, V, A> {
-    /// Get a value
-    pub fn get(&'e mut self, key: &K) -> Option<&'e V> {
-        algorithm::search(key, self.tree_writer.tree.root, &self.epoch_pin)
-    }
-
-    /// Insert a value
-    pub fn insert(self, key: &K, value: V) -> Result<bool, OutOfMemoryError> {
-        let mut success = None;
-
-        self.update_with_fn(key, |existing| {
-            if existing.is_some() {
-                success = Some(false);
-                UpdateAction::Nothing
-            } else {
-                success = Some(true);
-                UpdateAction::Insert(value)
-            }
-        })?;
-        Ok(success.expect("value_fn not called"))
-    }
-
-    /// Remove value. Returns true if it existed
-    pub fn remove(self, key: &K) -> bool {
-        let mut result = false;
-        // FIXME: It's not clear if OOM is expected while removing. It seems
-        // not nice, but shrinking a node can OOM. Then again, we could opt
-        // to not shrink a node if we cannot allocate, to live a little longer.
-        self.update_with_fn(key, |existing| match existing {
-            Some(_) => {
-                result = true;
-                UpdateAction::Remove
-            }
-            None => UpdateAction::Nothing,
-        })
-        .expect("out of memory while removing");
-        result
-    }
-
-    /// Try to remove value and return the old value.
-    pub fn remove_and_return(self, key: &K) -> Option<V>
-    where
-        V: Clone,
-    {
-        let mut old = None;
-        self.update_with_fn(key, |existing| {
-            old = existing.cloned();
-            UpdateAction::Remove
-        })
-        .expect("out of memory while removing");
-        old
-    }
-
-    /// Update key using the given function. All the other modifying operations are based on this.
-    ///
-    /// The function is passed a reference to the existing value, if any. If the function
-    /// returns None, the value is removed from the tree (or if there was no existing value,
-    /// does nothing). If the function returns Some, the existing value is replaced, of if there
-    /// was no existing value, it is inserted. FIXME: update comment
-    pub fn update_with_fn<F>(mut self, key: &K, value_fn: F) -> Result<(), OutOfMemoryError>
-    where
-        F: FnOnce(Option<&V>) -> UpdateAction<V>,
-    {
-        algorithm::update_fn(key, value_fn, self.tree_writer.tree.root, &mut self)?;
-
-        if self.created_garbage {
-            let _ = self.collect_garbage();
-        }
-        Ok(())
-    }
-
-    fn remember_obsolete_node(&mut self, ptr: NodePtr<V>) {
-        self.tree_writer
-            .garbage
-            .lock()
-            .remember_obsolete_node(ptr, self.epoch_pin.epoch);
-        self.created_garbage = true;
-    }
-
-    // returns number of nodes recycled
-    fn collect_garbage(&self) -> usize {
-        self.tree_writer.tree.epoch.advance();
-        self.tree_writer.tree.epoch.broadcast();
-
-        let cutoff_epoch = self.tree_writer.tree.epoch.get_oldest();
-
-        let mut result = 0;
-        let mut garbage_queue = self.tree_writer.garbage.lock();
-        while let Some(ptr) = garbage_queue.next_obsolete(cutoff_epoch) {
-            ptr.deallocate(self.tree_writer.allocator);
-            result += 1;
-        }
-        result
-    }
-}
-
-pub struct TreeIterator<K>
-where
-    K: Key + for<'a> From<&'a [u8]>,
-{
-    done: bool,
-    pub next_key: Vec<u8>,
-    max_key: Option<Vec<u8>>,
-
-    phantom_key: PhantomData<K>,
-}
-
-impl<K> TreeIterator<K>
-where
-    K: Key + for<'a> From<&'a [u8]>,
-{
-    pub fn new_wrapping() -> TreeIterator<K> {
-        TreeIterator {
-            done: false,
-            next_key: vec![0; K::KEY_LEN],
-            max_key: None,
-            phantom_key: PhantomData,
-        }
-    }
-
-    pub fn new(range: &std::ops::Range<K>) -> TreeIterator<K> {
-        let result = TreeIterator {
-            done: false,
-            next_key: Vec::from(range.start.as_bytes()),
-            max_key: Some(Vec::from(range.end.as_bytes())),
-            phantom_key: PhantomData,
-        };
-        assert_eq!(result.next_key.len(), K::KEY_LEN);
-        assert_eq!(result.max_key.as_ref().unwrap().len(), K::KEY_LEN);
-
-        result
-    }
-
-    pub fn next<'g, V>(&mut self, read_guard: &'g TreeReadGuard<'g, K, V>) -> Option<(K, &'g V)>
-    where
-        V: Value,
-    {
-        if self.done {
-            return None;
-        }
-
-        let mut wrapped_around = false;
-        loop {
-            assert_eq!(self.next_key.len(), K::KEY_LEN);
-            if let Some((k, v)) =
-                algorithm::iter_next(&self.next_key, read_guard.tree.root, &read_guard.epoch_pin)
-            {
-                assert_eq!(k.len(), K::KEY_LEN);
-                assert_eq!(self.next_key.len(), K::KEY_LEN);
-
-                // Check if we reached the end of the range
-                if let Some(max_key) = &self.max_key {
-                    if k.as_slice() >= max_key.as_slice() {
-                        self.done = true;
-                        break None;
-                    }
-                }
-
-                // increment the key
-                self.next_key = k.clone();
-                increment_key(self.next_key.as_mut_slice());
-                let k = k.as_slice().into();
-
-                break Some((k, v));
-            } else {
-                if self.max_key.is_some() {
-                    self.done = true;
-                } else {
-                    // Start from beginning
-                    if !wrapped_around {
-                        for i in 0..K::KEY_LEN {
-                            self.next_key[i] = 0;
-                        }
-                        wrapped_around = true;
-                        continue;
-                    } else {
-                        // The tree is completely empty
-                        // FIXME: perhaps we should remember the starting point instead.
-                        // Currently this will scan some ranges twice.
-                        break None;
-                    }
-                }
-                break None;
-            }
-        }
-    }
-}
-
-fn increment_key(key: &mut [u8]) -> bool {
-    for i in (0..key.len()).rev() {
-        let (byte, overflow) = key[i].overflowing_add(1);
-        key[i] = byte;
-        if !overflow {
-            return false;
-        }
-    }
-    true
-}
-
-// Debugging functions
-impl<'e, K: Key, V: Value + Debug, A: ArtAllocator<V>> TreeWriteGuard<'e, K, V, A> {
-    pub fn dump(&mut self, dst: &mut dyn std::io::Write) {
-        algorithm::dump_tree(self.tree_writer.tree.root, &self.epoch_pin, dst)
-    }
-}
-impl<'e, K: Key, V: Value + Debug> TreeReadGuard<'e, K, V> {
-    pub fn dump(&mut self, dst: &mut dyn std::io::Write) {
-        algorithm::dump_tree(self.tree.root, &self.epoch_pin, dst)
-    }
-}
-impl<'e, K: Key, V: Value> TreeWriteAccess<'e, K, V, ArtMultiSlabAllocator<'e, V>> {
-    pub fn get_statistics(&self) -> ArtTreeStatistics {
-        self.allocator.get_statistics();
-        ArtTreeStatistics {
-            blocks: self.allocator.inner.block_allocator.get_statistics(),
-            slabs: self.allocator.get_statistics(),
-            epoch: self.tree.epoch.get_current(),
-            oldest_epoch: self.tree.epoch.get_oldest(),
-            num_garbage: self.garbage.lock().0.len() as u64,
-        }
-    }
-}
-
-#[derive(Clone, Debug)]
-pub struct ArtTreeStatistics {
-    pub blocks: allocator::block::BlockAllocatorStats,
-    pub slabs: allocator::ArtMultiSlabStats,
-
-    pub epoch: u64,
-    pub oldest_epoch: u64,
-    pub num_garbage: u64,
-}

libs/neonart/src/tests.rs

@@ -1,236 +0,0 @@
-use std::collections::BTreeMap;
-use std::collections::HashSet;
-use std::fmt::{Debug, Formatter};
-use std::sync::atomic::{AtomicUsize, Ordering};
-
-use crate::ArtAllocator;
-use crate::ArtMultiSlabAllocator;
-use crate::TreeInitStruct;
-use crate::TreeIterator;
-use crate::TreeWriteAccess;
-use crate::UpdateAction;
-
-use crate::{Key, Value};
-
-use rand::Rng;
-use rand::seq::SliceRandom;
-use rand_distr::Zipf;
-
-const TEST_KEY_LEN: usize = 16;
-
-#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord)]
-struct TestKey([u8; TEST_KEY_LEN]);
-
-impl TestKey {
-    const MIN: TestKey = TestKey([0; TEST_KEY_LEN]);
-    const MAX: TestKey = TestKey([u8::MAX; TEST_KEY_LEN]);
-}
-
-impl Key for TestKey {
-    const KEY_LEN: usize = TEST_KEY_LEN;
-    fn as_bytes(&self) -> &[u8] {
-        &self.0
-    }
-}
-
-impl From<&TestKey> for u128 {
-    fn from(val: &TestKey) -> u128 {
-        u128::from_be_bytes(val.0)
-    }
-}
-
-impl From<u128> for TestKey {
-    fn from(val: u128) -> TestKey {
-        TestKey(val.to_be_bytes())
-    }
-}
-
-impl<'a> From<&'a [u8]> for TestKey {
-    fn from(bytes: &'a [u8]) -> TestKey {
-        TestKey(bytes.try_into().unwrap())
-    }
-}
-
-impl Value for usize {}
-
-fn test_inserts<K: Into<TestKey> + Copy>(keys: &[K]) {
-    const MEM_SIZE: usize = 10000000;
-    let mut area = Box::new_uninit_slice(MEM_SIZE);
-
-    let allocator = ArtMultiSlabAllocator::new(&mut area);
-
-    let init_struct = TreeInitStruct::<TestKey, usize, _>::new(allocator);
-    let tree_writer = init_struct.attach_writer();
-
-    for (idx, k) in keys.iter().enumerate() {
-        let w = tree_writer.start_write();
-        let res = w.insert(&(*k).into(), idx);
-        assert!(res.is_ok());
-    }
-
-    for (idx, k) in keys.iter().enumerate() {
-        let r = tree_writer.start_read();
-        let value = r.get(&(*k).into());
-        assert_eq!(value, Some(idx).as_ref());
-    }
-
-    eprintln!("stats: {:?}", tree_writer.get_statistics());
-}
-
-#[test]
-fn dense() {
-    // This exercises splitting a node with prefix
-    let keys: &[u128] = &[0, 1, 2, 3, 256];
-    test_inserts(keys);
-
-    // Dense keys
-    let mut keys: Vec<u128> = (0..10000).collect();
-    test_inserts(&keys);
-
-    // Do the same in random orders
-    for _ in 1..10 {
-        keys.shuffle(&mut rand::rng());
-        test_inserts(&keys);
-    }
-}
-
-#[test]
-fn sparse() {
-    // sparse keys
-    let mut keys: Vec<TestKey> = Vec::new();
-    let mut used_keys = HashSet::new();
-    for _ in 0..10000 {
-        loop {
-            let key = rand::random::<u128>();
-            if used_keys.contains(&key) {
-                continue;
-            }
-            used_keys.insert(key);
-            keys.push(key.into());
-            break;
-        }
-    }
-    test_inserts(&keys);
-}
-
-struct TestValue(AtomicUsize);
-
-impl TestValue {
-    fn new(val: usize) -> TestValue {
-        TestValue(AtomicUsize::new(val))
-    }
-
-    fn load(&self) -> usize {
-        self.0.load(Ordering::Relaxed)
-    }
-}
-
-impl Value for TestValue {}
-
-impl Clone for TestValue {
-    fn clone(&self) -> TestValue {
-        TestValue::new(self.load())
-    }
-}
-
-impl Debug for TestValue {
-    fn fmt(&self, fmt: &mut Formatter<'_>) -> Result<(), std::fmt::Error> {
-        write!(fmt, "{:?}", self.load())
-    }
-}
-
-#[derive(Clone, Debug)]
-struct TestOp(TestKey, Option<usize>);
-
-fn apply_op<A: ArtAllocator<TestValue>>(
-    op: &TestOp,
-    tree: &TreeWriteAccess<TestKey, TestValue, A>,
-    shadow: &mut BTreeMap<TestKey, usize>,
-) {
-    eprintln!("applying op: {op:?}");
-
-    // apply the change to the shadow tree first
-    let shadow_existing = if let Some(v) = op.1 {
-        shadow.insert(op.0, v)
-    } else {
-        shadow.remove(&op.0)
-    };
-
-    // apply to Art tree
-    let w = tree.start_write();
-    w.update_with_fn(&op.0, |existing| {
-        assert_eq!(existing.map(TestValue::load), shadow_existing);
-
-        match (existing, op.1) {
-            (None, None) => UpdateAction::Nothing,
-            (None, Some(new_val)) => UpdateAction::Insert(TestValue::new(new_val)),
-            (Some(_old_val), None) => UpdateAction::Remove,
-            (Some(old_val), Some(new_val)) => {
-                old_val.0.store(new_val, Ordering::Relaxed);
-                UpdateAction::Nothing
-            }
-        }
-    })
-    .expect("out of memory");
-}
-
-fn test_iter<A: ArtAllocator<TestValue>>(
-    tree: &TreeWriteAccess<TestKey, TestValue, A>,
-    shadow: &BTreeMap<TestKey, usize>,
-) {
-    let mut shadow_iter = shadow.iter();
-    let mut iter = TreeIterator::new(&(TestKey::MIN..TestKey::MAX));
-
-    loop {
-        let shadow_item = shadow_iter.next().map(|(k, v)| (*k, *v));
-        let r = tree.start_read();
-        let item = iter.next(&r);
-
-        if shadow_item != item.map(|(k, v)| (k, v.load())) {
-            eprintln!("FAIL: iterator returned {item:?}, expected {shadow_item:?}");
-            tree.start_read().dump(&mut std::io::stderr());
-
-            eprintln!("SHADOW:");
-            for si in shadow {
-                eprintln!("key: {:?}, val: {}", si.0, si.1);
-            }
-            panic!("FAIL: iterator returned {item:?}, expected {shadow_item:?}");
-        }
-        if item.is_none() {
-            break;
-        }
-    }
-}
-
-#[test]
-fn random_ops() {
-    const MEM_SIZE: usize = 10000000;
-    let mut area = Box::new_uninit_slice(MEM_SIZE);
-
-    let allocator = ArtMultiSlabAllocator::new(&mut area);
-
-    let init_struct = TreeInitStruct::<TestKey, TestValue, _>::new(allocator);
-    let tree_writer = init_struct.attach_writer();
-
-    let mut shadow: std::collections::BTreeMap<TestKey, usize> = BTreeMap::new();
-
-    let distribution = Zipf::new(u128::MAX as f64, 1.1).unwrap();
-    let mut rng = rand::rng();
-    for i in 0..100000 {
-        let mut key: TestKey = (rng.sample(distribution) as u128).into();
-
-        if rng.random_bool(0.10) {
-            key = TestKey::from(u128::from(&key) | 0xffffffff);
-        }
-
-        let op = TestOp(key, if rng.random_bool(0.75) { Some(i) } else { None });
-
-        apply_op(&op, &tree_writer, &mut shadow);
-
-        if i % 1000 == 0 {
-            eprintln!("{i} ops processed");
-            eprintln!("stats: {:?}", tree_writer.get_statistics());
-            test_iter(&tree_writer, &shadow);
-        }
-    }
-}

libs/pageserver_api/src/config.rs

@@ -5,7 +5,6 @@ mod tests;
 
 use const_format::formatcp;
 use posthog_client_lite::PostHogClientConfig;
-use utils::serde_percent::Percent;
 pub const DEFAULT_PG_LISTEN_PORT: u16 = 64000;
 pub const DEFAULT_PG_LISTEN_ADDR: &str = formatcp!("127.0.0.1:{DEFAULT_PG_LISTEN_PORT}");
 pub const DEFAULT_HTTP_LISTEN_PORT: u16 = 9898;
@@ -224,9 +223,8 @@ pub struct ConfigToml {
     pub metric_collection_bucket: Option<RemoteStorageConfig>,
     #[serde(with = "humantime_serde")]
     pub synthetic_size_calculation_interval: Duration,
-    pub disk_usage_based_eviction: DiskUsageEvictionTaskConfig,
+    pub disk_usage_based_eviction: Option<DiskUsageEvictionTaskConfig>,
     pub test_remote_failures: u64,
-    pub test_remote_failures_probability: u64,
     pub ondemand_download_behavior_treat_error_as_warn: bool,
     #[serde(with = "humantime_serde")]
     pub background_task_maximum_delay: Duration,
@@ -272,13 +270,9 @@ pub struct ConfigToml {
     pub timeline_import_config: TimelineImportConfig,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub basebackup_cache_config: Option<BasebackupCacheConfig>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    pub image_layer_generation_large_timeline_threshold: Option<u64>,
-    pub force_metric_collection_on_scrape: bool,
 }
 
 #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
-#[serde(default)]
 pub struct DiskUsageEvictionTaskConfig {
     pub max_usage_pct: utils::serde_percent::Percent,
     pub min_avail_bytes: u64,
@@ -289,21 +283,6 @@ pub struct DiskUsageEvictionTaskConfig {
     /// Select sorting for evicted layers
     #[serde(default)]
     pub eviction_order: EvictionOrder,
-    pub enabled: bool,
-}
-
-impl Default for DiskUsageEvictionTaskConfig {
-    fn default() -> Self {
-        Self {
-            max_usage_pct: Percent::new(80).unwrap(),
-            min_avail_bytes: 2_000_000_000,
-            period: Duration::from_secs(60),
-            #[cfg(feature = "testing")]
-            mock_statvfs: None,
-            eviction_order: EvictionOrder::default(),
-            enabled: true,
-        }
-    }
 }
 
 #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
@@ -564,11 +543,6 @@ pub struct TenantConfigToml {
     pub gc_period: Duration,
     // Delta layer churn threshold to create L1 image layers.
     pub image_creation_threshold: usize,
-    // HADRON
-    // When the timeout is reached, PageServer will (1) force compact any remaining L0 deltas and
-    // (2) create image layers if there are any L1 deltas.
-    #[serde(with = "humantime_serde")]
-    pub image_layer_force_creation_period: Option<Duration>,
     // Determines how much history is retained, to allow
     // branching and read replicas at an older point in time.
     // The unit is time.
@@ -764,10 +738,9 @@ impl Default for ConfigToml {
 
             metric_collection_bucket: (None),
 
-            disk_usage_based_eviction: DiskUsageEvictionTaskConfig::default(),
+            disk_usage_based_eviction: (None),
 
             test_remote_failures: (0),
-            test_remote_failures_probability: (100),
 
             ondemand_download_behavior_treat_error_as_warn: (false),
 
@@ -831,8 +804,6 @@ impl Default for ConfigToml {
             },
             basebackup_cache_config: None,
             posthog_config: None,
-            image_layer_generation_large_timeline_threshold: Some(2 * 1024 * 1024 * 1024),
-            force_metric_collection_on_scrape: true,
         }
     }
 }
@@ -926,7 +897,6 @@ impl Default for TenantConfigToml {
             gc_period: humantime::parse_duration(DEFAULT_GC_PERIOD)
                 .expect("cannot parse default gc period"),
             image_creation_threshold: DEFAULT_IMAGE_CREATION_THRESHOLD,
-            image_layer_force_creation_period: None,
             pitr_interval: humantime::parse_duration(DEFAULT_PITR_INTERVAL)
                 .expect("cannot parse default PITR interval"),
             walreceiver_connect_timeout: humantime::parse_duration(

libs/pageserver_api/src/models.rs

@@ -384,7 +384,7 @@ pub struct SafekeepersInfo {
     pub safekeepers: Vec<SafekeeperInfo>,
 }
 
-#[derive(Serialize, Deserialize, Clone, Debug)]
+#[derive(Serialize, Deserialize, Clone)]
 pub struct SafekeeperInfo {
     pub id: NodeId,
     pub hostname: String,
@@ -597,9 +597,6 @@ pub struct TenantConfigPatch {
     pub gc_period: FieldPatch<String>,
     #[serde(skip_serializing_if = "FieldPatch::is_noop")]
     pub image_creation_threshold: FieldPatch<usize>,
-    // HADRON
-    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
-    pub image_layer_force_creation_period: FieldPatch<String>,
     #[serde(skip_serializing_if = "FieldPatch::is_noop")]
     pub pitr_interval: FieldPatch<String>,
     #[serde(skip_serializing_if = "FieldPatch::is_noop")]
@@ -703,11 +700,6 @@ pub struct TenantConfig {
     #[serde(skip_serializing_if = "Option::is_none")]
     pub image_creation_threshold: Option<usize>,
 
-    // HADRON
-    #[serde(skip_serializing_if = "Option::is_none")]
-    #[serde(with = "humantime_serde")]
-    pub image_layer_force_creation_period: Option<Duration>,
-
     #[serde(skip_serializing_if = "Option::is_none")]
     #[serde(with = "humantime_serde")]
     pub pitr_interval: Option<Duration>,
@@ -806,7 +798,6 @@ impl TenantConfig {
             mut gc_horizon,
             mut gc_period,
             mut image_creation_threshold,
-            mut image_layer_force_creation_period,
             mut pitr_interval,
             mut walreceiver_connect_timeout,
             mut lagging_wal_timeout,
@@ -870,11 +861,6 @@ impl TenantConfig {
         patch
             .image_creation_threshold
             .apply(&mut image_creation_threshold);
-        // HADRON
-        patch
-            .image_layer_force_creation_period
-            .map(|v| humantime::parse_duration(&v))?
-            .apply(&mut image_layer_force_creation_period);
         patch
             .pitr_interval
             .map(|v| humantime::parse_duration(&v))?
@@ -956,7 +942,6 @@ impl TenantConfig {
             gc_horizon,
             gc_period,
             image_creation_threshold,
-            image_layer_force_creation_period,
             pitr_interval,
             walreceiver_connect_timeout,
             lagging_wal_timeout,
@@ -1031,9 +1016,6 @@ impl TenantConfig {
             image_creation_threshold: self
                 .image_creation_threshold
                 .unwrap_or(global_conf.image_creation_threshold),
-            image_layer_force_creation_period: self
-                .image_layer_force_creation_period
-                .or(global_conf.image_layer_force_creation_period),
             pitr_interval: self.pitr_interval.unwrap_or(global_conf.pitr_interval),
             walreceiver_connect_timeout: self
                 .walreceiver_connect_timeout

libs/pageserver_api/src/shard.rs

@@ -332,11 +332,7 @@ fn hash_combine(mut a: u32, mut b: u32) -> u32 {
 ///
 /// The mapping of key to shard is not stable across changes to ShardCount: this is intentional
 /// and will be handled at higher levels when shards are split.
-pub fn key_to_shard_number(
-    count: ShardCount,
-    stripe_size: ShardStripeSize,
-    key: &Key,
-) -> ShardNumber {
+fn key_to_shard_number(count: ShardCount, stripe_size: ShardStripeSize, key: &Key) -> ShardNumber {
     // Fast path for un-sharded tenants or broadcast keys
     if count < ShardCount(2) || key_is_shard0(key) {
         return ShardNumber(0);

libs/remote_storage/Cargo.toml

@@ -13,7 +13,6 @@ aws-smithy-async.workspace = true
 aws-smithy-types.workspace = true
 aws-config.workspace = true
 aws-sdk-s3.workspace = true
-base64.workspace = true
 bytes.workspace = true
 camino = { workspace = true, features = ["serde1"] }
 humantime-serde.workspace = true
@@ -42,9 +41,6 @@ http-body-util.workspace = true
 itertools.workspace = true
 sync_wrapper = { workspace = true, features = ["futures"] }
 
-byteorder = "1.4"
-rand = "0.8.5"
-
 [dev-dependencies]
 camino-tempfile.workspace = true
 test-context.workspace = true

libs/remote_storage/src/azure_blob.rs

@@ -14,25 +14,17 @@ use anyhow::{Context, Result, anyhow};
 use azure_core::request_options::{IfMatchCondition, MaxResults, Metadata, Range};
 use azure_core::{Continuable, HttpClient, RetryOptions, TransportOptions};
 use azure_storage::StorageCredentials;
-use azure_storage_blobs::blob::BlobBlockType;
-use azure_storage_blobs::blob::BlockList;
+use azure_storage_blobs::blob::operations::GetBlobBuilder;
 use azure_storage_blobs::blob::{Blob, CopyStatus};
 use azure_storage_blobs::container::operations::ListBlobsBuilder;
-use azure_storage_blobs::prelude::ClientBuilder;
-use azure_storage_blobs::{blob::operations::GetBlobBuilder, prelude::ContainerClient};
-use base64::{Engine as _, engine::general_purpose::URL_SAFE};
-use byteorder::{BigEndian, ByteOrder};
+use azure_storage_blobs::prelude::{ClientBuilder, ContainerClient};
 use bytes::Bytes;
-use camino::Utf8Path;
 use futures::FutureExt;
 use futures::future::Either;
 use futures::stream::Stream;
 use futures_util::{StreamExt, TryStreamExt};
 use http_types::{StatusCode, Url};
 use scopeguard::ScopeGuard;
-use tokio::fs::File;
-use tokio::io::AsyncReadExt;
-use tokio::io::AsyncSeekExt;
 use tokio_util::sync::CancellationToken;
 use tracing::debug;
 use utils::backoff;
@@ -59,9 +51,6 @@ pub struct AzureBlobStorage {
 
     // Alternative timeout used for metadata objects which are expected to be small
     pub small_timeout: Duration,
-    /* BEGIN_HADRON */
-    pub put_block_size_mb: Option<usize>,
-    /* END_HADRON */
 }
 
 impl AzureBlobStorage {
@@ -118,9 +107,6 @@ impl AzureBlobStorage {
             concurrency_limiter: ConcurrencyLimiter::new(azure_config.concurrency_limit.get()),
             timeout,
             small_timeout,
-            /* BEGIN_HADRON */
-            put_block_size_mb: azure_config.put_block_size_mb,
-            /* END_HADRON */
         })
     }
 
@@ -597,137 +583,31 @@ impl RemoteStorage for AzureBlobStorage {
 
         let started_at = start_measuring_requests(kind);
 
-        let mut metadata_map = metadata.unwrap_or([].into());
-        let timeline_file_path = metadata_map.0.remove("databricks_azure_put_block");
-
-        /* BEGIN_HADRON */
-        let op = async move {
+        let op = async {
             let blob_client = self.client.blob_client(self.relative_path_to_name(to));
-            let put_block_size = self.put_block_size_mb.unwrap_or(0) * 1024 * 1024;
-            if timeline_file_path.is_none() || put_block_size == 0 {
-                // Use put_block_blob directly.
-                let from: Pin<
-                    Box<dyn Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static>,
-                > = Box::pin(from);
-                let from = NonSeekableStream::new(from, data_size_bytes);
-                let body = azure_core::Body::SeekableStream(Box::new(from));
 
-                let mut builder = blob_client.put_block_blob(body);
-                if !metadata_map.0.is_empty() {
-                    builder = builder.metadata(to_azure_metadata(metadata_map));
-                }
-                let fut = builder.into_future();
-                let fut = tokio::time::timeout(self.timeout, fut);
-                let result = fut.await;
-                match result {
-                    Ok(Ok(_response)) => return Ok(()),
-                    Ok(Err(azure)) => return Err(azure.into()),
-                    Err(_timeout) => return Err(TimeoutOrCancel::Timeout.into()),
-                };
-            }
-            // Upload chunks concurrently using Put Block.
-            // Each PutBlock uploads put_block_size bytes of the file.
-            let mut upload_futures: Vec<tokio::task::JoinHandle<Result<(), azure_core::Error>>> =
-                vec![];
-            let mut block_list = BlockList::default();
-            let mut start_bytes = 0u64;
-            let mut remaining_bytes = data_size_bytes;
-            let mut block_list_count = 0;
+            let from: Pin<Box<dyn Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static>> =
+                Box::pin(from);
 
-            while remaining_bytes > 0 {
-                let block_size = std::cmp::min(remaining_bytes, put_block_size);
-                let end_bytes = start_bytes + block_size as u64;
-                let block_id = block_list_count;
-                let timeout = self.timeout;
-                let blob_client = blob_client.clone();
-                let timeline_file = timeline_file_path.clone().unwrap().clone();
+            let from = NonSeekableStream::new(from, data_size_bytes);
 
-                let mut encoded_block_id = [0u8; 8];
-                BigEndian::write_u64(&mut encoded_block_id, block_id);
-                URL_SAFE.encode(encoded_block_id);
+            let body = azure_core::Body::SeekableStream(Box::new(from));
 
-                // Put one block.
-                let part_fut = async move {
-                    let mut file = File::open(Utf8Path::new(&timeline_file.clone())).await?;
-                    file.seek(io::SeekFrom::Start(start_bytes)).await?;
-                    let limited_reader = file.take(block_size as u64);
-                    let file_chunk_stream =
-                        tokio_util::io::ReaderStream::with_capacity(limited_reader, 1024 * 1024);
-                    let file_chunk_stream_pin: Pin<
-                        Box<dyn Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static>,
-                    > = Box::pin(file_chunk_stream);
-                    let stream_wrapper = NonSeekableStream::new(file_chunk_stream_pin, block_size);
-                    let body = azure_core::Body::SeekableStream(Box::new(stream_wrapper));
-                    // Azure put block takes URL-encoded block ids and all blocks must have the same byte length.
-                    // https://learn.microsoft.com/en-us/rest/api/storageservices/put-block?tabs=microsoft-entra-id#uri-parameters
-                    let builder = blob_client.put_block(encoded_block_id.to_vec(), body);
-                    let fut = builder.into_future();
-                    let fut = tokio::time::timeout(timeout, fut);
-                    let result = fut.await;
-                    tracing::debug!(
-                        "azure put block id-{} size {} start {} end {} file {} response {:#?}",
-                        block_id,
-                        block_size,
-                        start_bytes,
-                        end_bytes,
-                        timeline_file,
-                        result
-                    );
-                    match result {
-                        Ok(Ok(_response)) => Ok(()),
-                        Ok(Err(azure)) => Err(azure),
-                        Err(_timeout) => Err(azure_core::Error::new(
-                            azure_core::error::ErrorKind::Io,
-                            std::io::Error::new(
-                                std::io::ErrorKind::TimedOut,
-                                "Operation timed out",
-                            ),
-                        )),
-                    }
-                };
-                upload_futures.push(tokio::spawn(part_fut));
+            let mut builder = blob_client.put_block_blob(body);
 
-                block_list_count += 1;
-                remaining_bytes -= block_size;
-                start_bytes += block_size as u64;
-
-                block_list
-                    .blocks
-                    .push(BlobBlockType::Uncommitted(encoded_block_id.to_vec().into()));
+            if let Some(metadata) = metadata {
+                builder = builder.metadata(to_azure_metadata(metadata));
             }
 
-            tracing::debug!(
-                "azure put blocks {} total MB: {} chunk size MB: {}",
-                block_list_count,
-                data_size_bytes / 1024 / 1024,
-                put_block_size / 1024 / 1024
-            );
-            // Wait for all blocks to be uploaded.
-            let upload_results = futures::future::try_join_all(upload_futures).await;
-            if upload_results.is_err() {
-                return Err(anyhow::anyhow!(format!(
-                    "Failed to upload all blocks {:#?}",
-                    upload_results.unwrap_err()
-                )));
-            }
-
-            // Commit the blocks.
-            let mut builder = blob_client.put_block_list(block_list);
-            if !metadata_map.0.is_empty() {
-                builder = builder.metadata(to_azure_metadata(metadata_map));
-            }
             let fut = builder.into_future();
             let fut = tokio::time::timeout(self.timeout, fut);
-            let result = fut.await;
-            tracing::debug!("azure put block list response {:#?}", result);
 
-            match result {
+            match fut.await {
                 Ok(Ok(_response)) => Ok(()),
                 Ok(Err(azure)) => Err(azure.into()),
                 Err(_timeout) => Err(TimeoutOrCancel::Timeout.into()),
             }
         };
-        /* END_HADRON */
 
         let res = tokio::select! {
             res = op => res,
@@ -742,6 +622,7 @@ impl RemoteStorage for AzureBlobStorage {
         crate::metrics::BUCKET_METRICS
             .req_seconds
             .observe_elapsed(kind, outcome, started_at);
+
         res
     }
 

libs/remote_storage/src/config.rs

@@ -195,19 +195,8 @@ pub struct AzureConfig {
     pub max_keys_per_list_response: Option<i32>,
     #[serde(default = "default_azure_conn_pool_size")]
     pub conn_pool_size: usize,
-    /* BEGIN_HADRON */
-    #[serde(default = "default_azure_put_block_size_mb")]
-    pub put_block_size_mb: Option<usize>,
-    /* END_HADRON */
 }
 
-/* BEGIN_HADRON */
-fn default_azure_put_block_size_mb() -> Option<usize> {
-    // Disable parallel upload by default.
-    Some(0)
-}
-/* END_HADRON */
-
 fn default_remote_storage_azure_concurrency_limit() -> NonZeroUsize {
     NonZeroUsize::new(DEFAULT_REMOTE_STORAGE_AZURE_CONCURRENCY_LIMIT).unwrap()
 }
@@ -224,9 +213,6 @@ impl Debug for AzureConfig {
                 "max_keys_per_list_response",
                 &self.max_keys_per_list_response,
             )
-            /* BEGIN_HADRON */
-            .field("put_block_size_mb", &self.put_block_size_mb)
-            /* END_HADRON */
             .finish()
     }
 }
@@ -366,7 +352,6 @@ timeout = '5s'";
     upload_storage_class = 'INTELLIGENT_TIERING'
     timeout = '7s'
     conn_pool_size = 8
-    put_block_size_mb = 1024
     ";
 
         let config = parse(toml).unwrap();
@@ -382,9 +367,6 @@ timeout = '5s'";
                     concurrency_limit: default_remote_storage_azure_concurrency_limit(),
                     max_keys_per_list_response: DEFAULT_MAX_KEYS_PER_LIST_RESPONSE,
                     conn_pool_size: 8,
-                    /* BEGIN_HADRON */
-                    put_block_size_mb: Some(1024),
-                    /* END_HADRON */
                 }),
                 timeout: Duration::from_secs(7),
                 small_timeout: RemoteStorageConfig::DEFAULT_SMALL_TIMEOUT

libs/remote_storage/src/lib.rs

@@ -732,15 +732,9 @@ impl GenericRemoteStorage {
         })
     }
 
-    /* BEGIN_HADRON */
-    pub fn unreliable_wrapper(s: Self, fail_first: u64, fail_probability: u64) -> Self {
-        Self::Unreliable(Arc::new(UnreliableWrapper::new(
-            s,
-            fail_first,
-            fail_probability,
-        )))
+    pub fn unreliable_wrapper(s: Self, fail_first: u64) -> Self {
+        Self::Unreliable(Arc::new(UnreliableWrapper::new(s, fail_first)))
     }
-    /* END_HADRON */
 
     /// See [`RemoteStorage::upload`], which this method calls with `None` as metadata.
     pub async fn upload_storage_object(

libs/remote_storage/src/simulate_failures.rs

@@ -1,8 +1,6 @@
 //! This module provides a wrapper around a real RemoteStorage implementation that
 //! causes the first N attempts at each upload or download operatio to fail. For
 //! testing purposes.
-use rand::Rng;
-use std::cmp;
 use std::collections::HashMap;
 use std::collections::hash_map::Entry;
 use std::num::NonZeroU32;
@@ -27,12 +25,6 @@ pub struct UnreliableWrapper {
 
     // Tracks how many failed attempts of each operation has been made.
     attempts: Mutex<HashMap<RemoteOp, u64>>,
-
-    /* BEGIN_HADRON */
-    // This the probability of failure for each operation, ranged from [0, 100].
-    // The probability is default to 100, which means that all operations will fail.
-    attempt_failure_probability: u64,
-    /* END_HADRON */
 }
 
 /// Used to identify retries of different unique operation.
@@ -48,11 +40,7 @@ enum RemoteOp {
 }
 
 impl UnreliableWrapper {
-    pub fn new(
-        inner: crate::GenericRemoteStorage,
-        attempts_to_fail: u64,
-        attempt_failure_probability: u64,
-    ) -> Self {
+    pub fn new(inner: crate::GenericRemoteStorage, attempts_to_fail: u64) -> Self {
         assert!(attempts_to_fail > 0);
         let inner = match inner {
             GenericRemoteStorage::AwsS3(s) => GenericRemoteStorage::AwsS3(s),
@@ -63,11 +51,9 @@ impl UnreliableWrapper {
                 panic!("Can't wrap unreliable wrapper unreliably")
             }
         };
-        let actual_attempt_failure_probability = cmp::min(attempt_failure_probability, 100);
         UnreliableWrapper {
             inner,
             attempts_to_fail,
-            attempt_failure_probability: actual_attempt_failure_probability,
             attempts: Mutex::new(HashMap::new()),
         }
     }
@@ -80,7 +66,6 @@ impl UnreliableWrapper {
     ///
     fn attempt(&self, op: RemoteOp) -> anyhow::Result<u64> {
         let mut attempts = self.attempts.lock().unwrap();
-        let mut rng = rand::thread_rng();
 
         match attempts.entry(op) {
             Entry::Occupied(mut e) => {
@@ -90,19 +75,15 @@ impl UnreliableWrapper {
                     *p
                 };
 
-                /* BEGIN_HADRON */
-                // If there are more attempts to fail, fail the request by probability.
-                if (attempts_before_this < self.attempts_to_fail)
-                    && (rng.gen_range(0..=100) < self.attempt_failure_probability)
-                {
+                if attempts_before_this >= self.attempts_to_fail {
+                    // let it succeed
+                    e.remove();
+                    Ok(attempts_before_this)
+                } else {
                     let error =
                         anyhow::anyhow!("simulated failure of remote operation {:?}", e.key());
                     Err(error)
-                } else {
-                    e.remove();
-                    Ok(attempts_before_this)
                 }
-                /* END_HADRON */
             }
             Entry::Vacant(e) => {
                 let error = anyhow::anyhow!("simulated failure of remote operation {:?}", e.key());

libs/remote_storage/tests/common/mod.rs

@@ -165,42 +165,10 @@ pub(crate) async fn upload_remote_data(
 
             let (data, data_len) =
                 upload_stream(format!("remote blob data {i}").into_bytes().into());
-
-            /* BEGIN_HADRON */
-            let mut metadata = None;
-            if matches!(&*task_client, GenericRemoteStorage::AzureBlob(_)) {
-                let file_path = "/tmp/dbx_upload_tmp_file.txt";
-                {
-                    // Open the file in append mode
-                    let mut file = std::fs::OpenOptions::new()
-                        .append(true)
-                        .create(true) // Create the file if it doesn't exist
-                        .open(file_path)?;
-                    // Append some bytes to the file
-                    std::io::Write::write_all(
-                        &mut file,
-                        &format!("remote blob data {i}").into_bytes(),
-                    )?;
-                    file.sync_all()?;
-                }
-                metadata = Some(remote_storage::StorageMetadata::from([(
-                    "databricks_azure_put_block",
-                    file_path,
-                )]));
-            }
-            /* END_HADRON */
-
             task_client
-                .upload(data, data_len, &blob_path, metadata, &cancel)
+                .upload(data, data_len, &blob_path, None, &cancel)
                 .await?;
 
-            // TODO: Check upload is using the put_block upload.
-            // We cannot consume data here since data is moved inside the upload.
-            // let total_bytes = data.fold(0, |acc, chunk| async move {
-            //     acc + chunk.map(|bytes| bytes.len()).unwrap_or(0)
-            // }).await;
-            // assert_eq!(total_bytes, data_len);
-
             Ok::<_, anyhow::Error>((blob_prefix, blob_path))
         });
     }

libs/remote_storage/tests/test_real_azure.rs

@@ -219,9 +219,6 @@ async fn create_azure_client(
             concurrency_limit: NonZeroUsize::new(100).unwrap(),
             max_keys_per_list_response,
             conn_pool_size: 8,
-            /* BEGIN_HADRON */
-            put_block_size_mb: Some(1),
-            /* END_HADRON */
         }),
         timeout: RemoteStorageConfig::DEFAULT_TIMEOUT,
         small_timeout: RemoteStorageConfig::DEFAULT_SMALL_TIMEOUT,

libs/utils/src/env.rs

@@ -44,62 +44,3 @@ where
         }
     }
 }
-
-/* BEGIN_HADRON */
-pub enum DeploymentMode {
-    Dev,
-    Staging,
-    Prod,
-}
-
-pub fn get_deployment_mode() -> Option<DeploymentMode> {
-    match std::env::var("DEPLOYMENT_MODE") {
-        Ok(env) => match env.as_str() {
-            "development" => Some(DeploymentMode::Dev),
-            "staging" => Some(DeploymentMode::Staging),
-            "production" => Some(DeploymentMode::Prod),
-            _ => {
-                tracing::error!("Unexpected DEPLOYMENT_MODE: {}", env);
-                None
-            }
-        },
-        Err(_) => {
-            tracing::error!("DEPLOYMENT_MODE not set");
-            None
-        }
-    }
-}
-
-pub fn is_dev_or_staging() -> bool {
-    matches!(
-        get_deployment_mode(),
-        Some(DeploymentMode::Dev) | Some(DeploymentMode::Staging)
-    )
-}
-
-pub enum TestingMode {
-    Chaos,
-    Stress,
-}
-
-pub fn get_test_mode() -> Option<TestingMode> {
-    match std::env::var("HADRON_TEST_MODE") {
-        Ok(env) => match env.as_str() {
-            "chaos" => Some(TestingMode::Chaos),
-            "stress" => Some(TestingMode::Stress),
-            _ => {
-                tracing::error!("Unexpected HADRON_TEST_MODE: {}", env);
-                None
-            }
-        },
-        Err(_) => {
-            tracing::error!("HADRON_TEST_MODE not set");
-            None
-        }
-    }
-}
-
-pub fn is_chaos_testing() -> bool {
-    matches!(get_test_mode(), Some(TestingMode::Chaos))
-}
-/* END_HADRON */

libs/utils/src/lib.rs

@@ -99,8 +99,6 @@ pub mod elapsed_accum;
 #[cfg(target_os = "linux")]
 pub mod linux_socket_ioctl;
 
-pub mod metrics_collector;
-
 // Re-export used in macro. Avoids adding git-version as dep in target crates.
 #[doc(hidden)]
 pub use git_version;

libs/utils/src/metrics_collector.rs

@@ -1,75 +0,0 @@
-use std::{
-    sync::{Arc, RwLock},
-    time::{Duration, Instant},
-};
-
-use metrics::{IntGauge, proto::MetricFamily, register_int_gauge};
-use once_cell::sync::Lazy;
-
-pub static METRICS_STALE_MILLIS: Lazy<IntGauge> = Lazy::new(|| {
-    register_int_gauge!(
-        "metrics_metrics_stale_milliseconds",
-        "The current metrics stale time in milliseconds"
-    )
-    .expect("failed to define a metric")
-});
-
-#[derive(Debug)]
-pub struct CollectedMetrics {
-    pub metrics: Vec<MetricFamily>,
-    pub collected_at: Instant,
-}
-
-impl CollectedMetrics {
-    fn new(metrics: Vec<MetricFamily>) -> Self {
-        Self {
-            metrics,
-            collected_at: Instant::now(),
-        }
-    }
-}
-
-#[derive(Debug)]
-pub struct MetricsCollector {
-    last_collected: RwLock<Arc<CollectedMetrics>>,
-}
-
-impl MetricsCollector {
-    pub fn new() -> Self {
-        Self {
-            last_collected: RwLock::new(Arc::new(CollectedMetrics::new(vec![]))),
-        }
-    }
-
-    #[tracing::instrument(name = "metrics_collector", skip_all)]
-    pub fn run_once(&self, cache_metrics: bool) -> Arc<CollectedMetrics> {
-        let started = Instant::now();
-        let metrics = metrics::gather();
-        let collected = Arc::new(CollectedMetrics::new(metrics));
-        if cache_metrics {
-            let mut guard = self.last_collected.write().unwrap();
-            *guard = collected.clone();
-        }
-        tracing::info!(
-            "Collected {} metric families in {} ms",
-            collected.metrics.len(),
-            started.elapsed().as_millis()
-        );
-        collected
-    }
-
-    pub fn last_collected(&self) -> Arc<CollectedMetrics> {
-        self.last_collected.read().unwrap().clone()
-    }
-}
-
-impl Default for MetricsCollector {
-    fn default() -> Self {
-        Self::new()
-    }
-}
-
-// Interval for metrics collection. Currently hard-coded to be the same as the metrics scape interval from the obs agent
-pub static METRICS_COLLECTION_INTERVAL: Duration = Duration::from_secs(30);
-
-pub static METRICS_COLLECTOR: Lazy<MetricsCollector> = Lazy::new(MetricsCollector::default);

libs/utils/src/shard.rs

@@ -171,12 +171,6 @@ impl std::fmt::Display for ShardNumber {
     }
 }
 
-impl std::fmt::Display for ShardCount {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        self.0.fmt(f)
-    }
-}
-
 impl std::fmt::Display for ShardSlug<'_> {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
         write!(

libs/walproposer/src/api_bindings.rs

@@ -428,12 +428,6 @@ pub fn empty_shmem() -> crate::bindings::WalproposerShmemState {
         shard_number: 0,
     };
 
-    let empty_wal_rate_limiter = crate::bindings::WalRateLimiter {
-        should_limit: crate::bindings::pg_atomic_uint32 { value: 0 },
-        sent_bytes: 0,
-        last_recorded_time_us: 0,
-    };
-
     crate::bindings::WalproposerShmemState {
         propEpochStartLsn: crate::bindings::pg_atomic_uint64 { value: 0 },
         donor_name: [0; 64],
@@ -447,7 +441,6 @@ pub fn empty_shmem() -> crate::bindings::WalproposerShmemState {
         num_shards: 0,
         replica_promote: false,
         min_ps_feedback: empty_feedback,
-        wal_rate_limiter: empty_wal_rate_limiter,
     }
 }
 

pageserver/Cargo.toml

@@ -54,7 +54,6 @@ pageserver_api.workspace = true
 pageserver_client.workspace = true # for ResponseErrorMessageExt TOOD refactor that
 pageserver_compaction.workspace = true
 pageserver_page_api.workspace = true
-peekable.workspace = true
 pem.workspace = true
 pin-project-lite.workspace = true
 postgres_backend.workspace = true
@@ -67,7 +66,6 @@ postgres-types.workspace = true
 posthog_client_lite.workspace = true
 pprof.workspace = true
 pq_proto.workspace = true
-prost.workspace = true
 rand.workspace = true
 range-set-blaze = { version = "0.1.16", features = ["alloc"] }
 regex.workspace = true
@@ -114,7 +112,6 @@ twox-hash.workspace = true
 procfs.workspace = true
 
 [dev-dependencies]
-base64.workspace = true
 criterion.workspace = true
 hex-literal.workspace = true
 tokio = { workspace = true, features = ["process", "sync", "fs", "rt", "io-util", "time", "test-util"] }

pageserver/client/src/mgmt_api.rs

@@ -1,4 +1,4 @@
-use std::collections::{BTreeMap, HashMap};
+use std::collections::HashMap;
 use std::error::Error as _;
 use std::time::Duration;
 
@@ -251,70 +251,6 @@ impl Client {
         Ok(())
     }
 
-    pub async fn tenant_timeline_compact(
-        &self,
-        tenant_shard_id: TenantShardId,
-        timeline_id: TimelineId,
-        force_image_layer_creation: bool,
-        must_force_image_layer_creation: bool,
-        scheduled: bool,
-        wait_until_done: bool,
-    ) -> Result<()> {
-        let mut path = reqwest::Url::parse(&format!(
-            "{}/v1/tenant/{tenant_shard_id}/timeline/{timeline_id}/compact",
-            self.mgmt_api_endpoint
-        ))
-        .expect("Cannot build URL");
-
-        if force_image_layer_creation {
-            path.query_pairs_mut()
-                .append_pair("force_image_layer_creation", "true");
-        }
-
-        if must_force_image_layer_creation {
-            path.query_pairs_mut()
-                .append_pair("must_force_image_layer_creation", "true");
-        }
-
-        if scheduled {
-            path.query_pairs_mut().append_pair("scheduled", "true");
-        }
-        if wait_until_done {
-            path.query_pairs_mut()
-                .append_pair("wait_until_scheduled_compaction_done", "true");
-            path.query_pairs_mut()
-                .append_pair("wait_until_uploaded", "true");
-        }
-        self.request(Method::PUT, path, ()).await?;
-        Ok(())
-    }
-
-    /* BEGIN_HADRON */
-    pub async fn tenant_timeline_describe(
-        &self,
-        tenant_shard_id: &TenantShardId,
-        timeline_id: &TimelineId,
-    ) -> Result<TimelineInfo> {
-        let mut path = reqwest::Url::parse(&format!(
-            "{}/v1/tenant/{tenant_shard_id}/timeline/{timeline_id}",
-            self.mgmt_api_endpoint
-        ))
-        .expect("Cannot build URL");
-        path.query_pairs_mut()
-            .append_pair("include-image-consistent-lsn", "true");
-
-        let response: reqwest::Response = self.request(Method::GET, path, ()).await?;
-        let body = response.json().await.map_err(Error::ReceiveBody)?;
-        Ok(body)
-    }
-
-    pub async fn list_tenant_visible_size(&self) -> Result<BTreeMap<TenantShardId, u64>> {
-        let uri = format!("{}/v1/list_tenant_visible_size", self.mgmt_api_endpoint);
-        let resp = self.get(&uri).await?;
-        resp.json().await.map_err(Error::ReceiveBody)
-    }
-    /* END_HADRON */
-
     pub async fn tenant_scan_remote_storage(
         &self,
         tenant_id: TenantId,

pageserver/client_grpc/Cargo.toml

@@ -1,24 +0,0 @@
-[package]
-name = "pageserver_client_grpc"
-version = "0.1.0"
-edition.workspace = true
-license.workspace = true
-
-[features]
-testing = ["pageserver_api/testing"]
-
-[dependencies]
-anyhow.workspace = true
-arc-swap.workspace = true
-bytes.workspace = true
-compute_api.workspace = true
-futures.workspace = true
-pageserver_api.workspace = true
-pageserver_page_api.workspace = true
-tokio.workspace = true
-tokio-stream.workspace = true
-tokio-util.workspace = true
-tonic.workspace = true
-tracing.workspace = true
-utils.workspace = true
-workspace_hack.workspace = true

pageserver/client_grpc/src/client.rs

@@ -1,543 +0,0 @@
-use std::collections::HashMap;
-use std::num::NonZero;
-use std::sync::Arc;
-
-use anyhow::anyhow;
-use arc_swap::ArcSwap;
-use futures::stream::FuturesUnordered;
-use futures::{FutureExt as _, StreamExt as _};
-use tonic::codec::CompressionEncoding;
-use tracing::instrument;
-
-use crate::pool::{ChannelPool, ClientGuard, ClientPool, StreamGuard, StreamPool};
-use crate::retry::Retry;
-use crate::split::GetPageSplitter;
-use compute_api::spec::PageserverProtocol;
-use pageserver_api::shard::ShardStripeSize;
-use pageserver_page_api as page_api;
-use utils::id::{TenantId, TimelineId};
-use utils::shard::{ShardCount, ShardIndex, ShardNumber};
-
-/// Max number of concurrent clients per channel (i.e. TCP connection). New channels will be spun up
-/// when full.
-///
-/// TODO: tune all of these constants, and consider making them configurable.
-/// TODO: consider separate limits for unary and streaming clients, so we don't fill up channels
-/// with only streams.
-const MAX_CLIENTS_PER_CHANNEL: NonZero<usize> = NonZero::new(16).unwrap();
-
-/// Max number of concurrent unary request clients per shard.
-const MAX_UNARY_CLIENTS: NonZero<usize> = NonZero::new(64).unwrap();
-
-/// Max number of concurrent GetPage streams per shard. The max number of concurrent GetPage
-/// requests is given by `MAX_STREAMS * MAX_STREAM_QUEUE_DEPTH`.
-const MAX_STREAMS: NonZero<usize> = NonZero::new(64).unwrap();
-
-/// Max number of pipelined requests per stream.
-const MAX_STREAM_QUEUE_DEPTH: NonZero<usize> = NonZero::new(2).unwrap();
-
-/// Max number of concurrent bulk GetPage streams per shard, used e.g. for prefetches. Because these
-/// are more throughput-oriented, we have a smaller limit but higher queue depth.
-const MAX_BULK_STREAMS: NonZero<usize> = NonZero::new(16).unwrap();
-
-/// Max number of pipelined requests per bulk stream. These are more throughput-oriented and thus
-/// get a larger queue depth.
-const MAX_BULK_STREAM_QUEUE_DEPTH: NonZero<usize> = NonZero::new(4).unwrap();
-
-/// A rich Pageserver gRPC client for a single tenant timeline. This client is more capable than the
-/// basic `page_api::Client` gRPC client, and supports:
-///
-/// * Sharded tenants across multiple Pageservers.
-/// * Pooling of connections, clients, and streams for efficient resource use.
-/// * Concurrent use by many callers.
-/// * Internal handling of GetPage bidirectional streams, with pipelining and error handling.
-/// * Automatic retries.
-/// * Observability.
-///
-/// TODO: this client does not support base backups or LSN leases, as these are only used by
-/// compute_ctl. Consider adding this, but LSN leases need concurrent requests on all shards.
-pub struct PageserverClient {
-    /// The tenant ID.
-    tenant_id: TenantId,
-    /// The timeline ID.
-    timeline_id: TimelineId,
-    /// The JWT auth token for this tenant, if any.
-    auth_token: Option<String>,
-    /// The compression to use, if any.
-    compression: Option<CompressionEncoding>,
-    /// The shards for this tenant.
-    shards: ArcSwap<Shards>,
-    /// The retry configuration.
-    retry: Retry,
-}
-
-impl PageserverClient {
-    /// Creates a new Pageserver client for a given tenant and timeline. Uses the Pageservers given
-    /// in the shard spec, which must be complete and must use gRPC URLs.
-    pub fn new(
-        tenant_id: TenantId,
-        timeline_id: TimelineId,
-        shard_spec: ShardSpec,
-        auth_token: Option<String>,
-        compression: Option<CompressionEncoding>,
-    ) -> anyhow::Result<Self> {
-        let shards = Shards::new(
-            tenant_id,
-            timeline_id,
-            shard_spec,
-            auth_token.clone(),
-            compression,
-        )?;
-        Ok(Self {
-            tenant_id,
-            timeline_id,
-            auth_token,
-            compression,
-            shards: ArcSwap::new(Arc::new(shards)),
-            retry: Retry,
-        })
-    }
-
-    /// Updates the shards from the given shard spec. In-flight requests will complete using the
-    /// existing shards, but may retry with the new shards if they fail.
-    ///
-    /// TODO: verify that in-flight requests are allowed to complete, and that the old pools are
-    /// properly spun down and dropped afterwards.
-    pub fn update_shards(&self, shard_spec: ShardSpec) -> anyhow::Result<()> {
-        // Validate the shard spec. We should really use `ArcSwap::rcu` for this, to avoid races
-        // with concurrent updates, but that involves creating a new `Shards` on every attempt,
-        // which spins up a bunch of Tokio tasks and such. These should already be checked elsewhere
-        // in the stack, and if they're violated then we already have problems elsewhere, so a
-        // best-effort but possibly-racy check is okay here.
-        let old = self.shards.load_full();
-        if shard_spec.count < old.count {
-            return Err(anyhow!(
-                "can't reduce shard count from {} to {}",
-                old.count,
-                shard_spec.count
-            ));
-        }
-        if !old.count.is_unsharded() && shard_spec.stripe_size != old.stripe_size {
-            return Err(anyhow!(
-                "can't change stripe size from {} to {}",
-                old.stripe_size,
-                shard_spec.stripe_size
-            ));
-        }
-
-        let shards = Shards::new(
-            self.tenant_id,
-            self.timeline_id,
-            shard_spec,
-            self.auth_token.clone(),
-            self.compression,
-        )?;
-        self.shards.store(Arc::new(shards));
-        Ok(())
-    }
-
-    /// Returns whether a relation exists.
-    #[instrument(skip_all, fields(rel=%req.rel, lsn=%req.read_lsn))]
-    pub async fn check_rel_exists(
-        &self,
-        req: page_api::CheckRelExistsRequest,
-    ) -> tonic::Result<page_api::CheckRelExistsResponse> {
-        self.retry
-            .with(async |_| {
-                // Relation metadata is only available on shard 0.
-                let mut client = self.shards.load_full().get_zero().client().await?;
-                client.check_rel_exists(req).await
-            })
-            .await
-    }
-
-    /// Returns the total size of a database, as # of bytes.
-    #[instrument(skip_all, fields(db_oid=%req.db_oid, lsn=%req.read_lsn))]
-    pub async fn get_db_size(
-        &self,
-        req: page_api::GetDbSizeRequest,
-    ) -> tonic::Result<page_api::GetDbSizeResponse> {
-        self.retry
-            .with(async |_| {
-                // Relation metadata is only available on shard 0.
-                let mut client = self.shards.load_full().get_zero().client().await?;
-                client.get_db_size(req).await
-            })
-            .await
-    }
-
-    /// Fetches pages. The `request_id` must be unique across all in-flight requests, and the
-    /// `attempt` must be 0 (incremented on retry). Automatically splits requests that straddle
-    /// shard boundaries, and assembles the responses.
-    ///
-    /// Unlike `page_api::Client`, this automatically converts `status_code` into `tonic::Status`
-    /// errors. All responses will have `GetPageStatusCode::Ok`.
-    #[instrument(skip_all, fields(
-        req_id = %req.request_id,
-        class = %req.request_class,
-        rel = %req.rel,
-        blkno = %req.block_numbers[0],
-        blks = %req.block_numbers.len(),
-        lsn = %req.read_lsn,
-    ))]
-    pub async fn get_page(
-        &self,
-        req: page_api::GetPageRequest,
-    ) -> tonic::Result<page_api::GetPageResponse> {
-        // Make sure we have at least one page.
-        if req.block_numbers.is_empty() {
-            return Err(tonic::Status::invalid_argument("no block number"));
-        }
-        // The request attempt must be 0. The client will increment it internally.
-        if req.request_id.attempt != 0 {
-            return Err(tonic::Status::invalid_argument("request attempt must be 0"));
-        }
-
-        // The shards may change while we're fetching pages. We execute the request using a stable
-        // view of the shards (especially important for requests that span shards), but retry the
-        // top-level (pre-split) request to pick up shard changes. This can lead to unnecessary
-        // retries and re-splits in some cases where requests span shards, but these are expected to
-        // be rare.
-        //
-        // TODO: the gRPC server and client doesn't yet properly support shard splits. Revisit this
-        // once we figure out how to handle these.
-        self.retry
-            .with(async |attempt| {
-                let mut req = req.clone();
-                req.request_id.attempt = attempt as u32;
-                Self::get_page_with_shards(req, &self.shards.load_full()).await
-            })
-            .await
-    }
-
-    /// Fetches pages using the given shards. This uses a stable view of the shards, regardless of
-    /// concurrent shard updates. Does not retry internally, but is retried by `get_page()`.
-    async fn get_page_with_shards(
-        req: page_api::GetPageRequest,
-        shards: &Shards,
-    ) -> tonic::Result<page_api::GetPageResponse> {
-        // Fast path: request is for a single shard.
-        if let Some(shard_id) =
-            GetPageSplitter::for_single_shard(&req, shards.count, shards.stripe_size)
-        {
-            return Self::get_page_with_shard(req, shards.get(shard_id)?).await;
-        }
-
-        // Request spans multiple shards. Split it, dispatch concurrent per-shard requests, and
-        // reassemble the responses.
-        let mut splitter = GetPageSplitter::split(req, shards.count, shards.stripe_size);
-
-        let mut shard_requests = FuturesUnordered::new();
-        for (shard_id, shard_req) in splitter.drain_requests() {
-            let future = Self::get_page_with_shard(shard_req, shards.get(shard_id)?)
-                .map(move |result| result.map(|resp| (shard_id, resp)));
-            shard_requests.push(future);
-        }
-
-        while let Some((shard_id, shard_response)) = shard_requests.next().await.transpose()? {
-            splitter.add_response(shard_id, shard_response)?;
-        }
-
-        splitter.get_response()
-    }
-
-    /// Fetches pages on the given shard. Does not retry internally.
-    async fn get_page_with_shard(
-        req: page_api::GetPageRequest,
-        shard: &Shard,
-    ) -> tonic::Result<page_api::GetPageResponse> {
-        let stream = shard.stream(req.request_class.is_bulk()).await;
-        let resp = stream.send(req.clone()).await?;
-
-        // Convert per-request errors into a tonic::Status.
-        if resp.status_code != page_api::GetPageStatusCode::Ok {
-            return Err(tonic::Status::new(
-                resp.status_code.into(),
-                resp.reason.unwrap_or_else(|| String::from("unknown error")),
-            ));
-        }
-
-        // Check that we received the expected pages.
-        if req.rel != resp.rel {
-            return Err(tonic::Status::internal(format!(
-                "shard {} returned wrong relation, expected {} got {}",
-                shard.id, req.rel, resp.rel
-            )));
-        }
-        if !req
-            .block_numbers
-            .iter()
-            .copied()
-            .eq(resp.pages.iter().map(|p| p.block_number))
-        {
-            return Err(tonic::Status::internal(format!(
-                "shard {} returned wrong pages, expected {:?} got {:?}",
-                shard.id,
-                req.block_numbers,
-                resp.pages
-                    .iter()
-                    .map(|page| page.block_number)
-                    .collect::<Vec<_>>()
-            )));
-        }
-
-        Ok(resp)
-    }
-
-    /// Returns the size of a relation, as # of blocks.
-    #[instrument(skip_all, fields(rel=%req.rel, lsn=%req.read_lsn))]
-    pub async fn get_rel_size(
-        &self,
-        req: page_api::GetRelSizeRequest,
-    ) -> tonic::Result<page_api::GetRelSizeResponse> {
-        self.retry
-            .with(async |_| {
-                // Relation metadata is only available on shard 0.
-                let mut client = self.shards.load_full().get_zero().client().await?;
-                client.get_rel_size(req).await
-            })
-            .await
-    }
-
-    /// Fetches an SLRU segment.
-    #[instrument(skip_all, fields(kind=%req.kind, segno=%req.segno, lsn=%req.read_lsn))]
-    pub async fn get_slru_segment(
-        &self,
-        req: page_api::GetSlruSegmentRequest,
-    ) -> tonic::Result<page_api::GetSlruSegmentResponse> {
-        self.retry
-            .with(async |_| {
-                // SLRU segments are only available on shard 0.
-                let mut client = self.shards.load_full().get_zero().client().await?;
-                client.get_slru_segment(req).await
-            })
-            .await
-    }
-}
-
-/// Shard specification for a PageserverClient.
-pub struct ShardSpec {
-    /// Maps shard indices to gRPC URLs.
-    ///
-    /// INVARIANT: every shard 0..count is present, and shard 0 is always present.
-    /// INVARIANT: every URL is valid and uses grpc:// scheme.
-    urls: HashMap<ShardIndex, String>,
-    /// The shard count.
-    ///
-    /// NB: this is 0 for unsharded tenants, following `ShardIndex::unsharded()` convention.
-    count: ShardCount,
-    /// The stripe size for these shards.
-    stripe_size: ShardStripeSize,
-}
-
-impl ShardSpec {
-    /// Creates a new shard spec with the given URLs and stripe size. All shards must be given.
-    /// The stripe size may be omitted for unsharded tenants.
-    pub fn new(
-        urls: HashMap<ShardIndex, String>,
-        stripe_size: Option<ShardStripeSize>,
-    ) -> anyhow::Result<Self> {
-        // Compute the shard count.
-        let count = match urls.len() {
-            0 => return Err(anyhow!("no shards provided")),
-            1 => ShardCount::new(0), // NB: unsharded tenants use 0, like `ShardIndex::unsharded()`
-            n if n > u8::MAX as usize => return Err(anyhow!("too many shards: {n}")),
-            n => ShardCount::new(n as u8),
-        };
-
-        // Determine the stripe size. It doesn't matter for unsharded tenants.
-        if stripe_size.is_none() && !count.is_unsharded() {
-            return Err(anyhow!("stripe size must be given for sharded tenants"));
-        }
-        let stripe_size = stripe_size.unwrap_or_default();
-
-        // Validate the shard spec.
-        for (shard_id, url) in &urls {
-            // The shard index must match the computed shard count, even for unsharded tenants.
-            if shard_id.shard_count != count {
-                return Err(anyhow!("invalid shard index {shard_id}, expected {count}"));
-            }
-            // The shard index' number and count must be consistent.
-            if !shard_id.is_unsharded() && shard_id.shard_number.0 >= shard_id.shard_count.0 {
-                return Err(anyhow!("invalid shard index {shard_id}"));
-            }
-            // The above conditions guarantee that we have all shards 0..count: len() matches count,
-            // shard number < count, and numbers are unique (via hashmap).
-
-            // Validate the URL.
-            if PageserverProtocol::from_connstring(url)? != PageserverProtocol::Grpc {
-                return Err(anyhow!("invalid shard URL {url}: must use gRPC"));
-            }
-        }
-
-        Ok(Self {
-            urls,
-            count,
-            stripe_size,
-        })
-    }
-}
-
-/// Tracks the tenant's shards.
-struct Shards {
-    /// Shards by shard index.
-    ///
-    /// INVARIANT: every shard 0..count is present.
-    /// INVARIANT: shard 0 is always present.
-    by_index: HashMap<ShardIndex, Shard>,
-    /// The shard count.
-    ///
-    /// NB: this is 0 for unsharded tenants, following `ShardIndex::unsharded()` convention.
-    count: ShardCount,
-    /// The stripe size. Only used for sharded tenants.
-    stripe_size: ShardStripeSize,
-}
-
-impl Shards {
-    /// Creates a new set of shards based on a shard spec.
-    fn new(
-        tenant_id: TenantId,
-        timeline_id: TimelineId,
-        shard_spec: ShardSpec,
-        auth_token: Option<String>,
-        compression: Option<CompressionEncoding>,
-    ) -> anyhow::Result<Self> {
-        // NB: the shard spec has already been validated when constructed.
-        let mut shards = HashMap::with_capacity(shard_spec.urls.len());
-        for (shard_id, url) in shard_spec.urls {
-            shards.insert(
-                shard_id,
-                Shard::new(
-                    url,
-                    tenant_id,
-                    timeline_id,
-                    shard_id,
-                    auth_token.clone(),
-                    compression,
-                )?,
-            );
-        }
-
-        Ok(Self {
-            by_index: shards,
-            count: shard_spec.count,
-            stripe_size: shard_spec.stripe_size,
-        })
-    }
-
-    /// Looks up the given shard.
-    #[allow(clippy::result_large_err)] // TODO: check perf impact
-    fn get(&self, shard_id: ShardIndex) -> tonic::Result<&Shard> {
-        self.by_index
-            .get(&shard_id)
-            .ok_or_else(|| tonic::Status::not_found(format!("unknown shard {shard_id}")))
-    }
-
-    /// Returns shard 0.
-    fn get_zero(&self) -> &Shard {
-        self.get(ShardIndex::new(ShardNumber(0), self.count))
-            .expect("always present")
-    }
-}
-
-/// A single shard. Uses dedicated resource pools with the following structure:
-///
-/// * Channel pool: unbounded.
-///   * Unary client pool: MAX_UNARY_CLIENTS.
-///   * Stream client pool: unbounded.
-///     * Stream pool: MAX_STREAMS and MAX_STREAM_QUEUE_DEPTH.
-/// * Bulk channel pool: unbounded.
-///   * Bulk client pool: unbounded.
-///     * Bulk stream pool: MAX_BULK_STREAMS and MAX_BULK_STREAM_QUEUE_DEPTH.
-struct Shard {
-    /// The shard ID.
-    id: ShardIndex,
-    /// Unary gRPC client pool.
-    client_pool: Arc<ClientPool>,
-    /// GetPage stream pool.
-    stream_pool: Arc<StreamPool>,
-    /// GetPage stream pool for bulk requests, e.g. prefetches.
-    bulk_stream_pool: Arc<StreamPool>,
-}
-
-impl Shard {
-    /// Creates a new shard. It has its own dedicated resource pools.
-    fn new(
-        url: String,
-        tenant_id: TenantId,
-        timeline_id: TimelineId,
-        shard_id: ShardIndex,
-        auth_token: Option<String>,
-        compression: Option<CompressionEncoding>,
-    ) -> anyhow::Result<Self> {
-        // Common channel pool for unary and stream requests. Bounded by client/stream pools.
-        let channel_pool = ChannelPool::new(url.clone(), MAX_CLIENTS_PER_CHANNEL)?;
-
-        // Client pool for unary requests.
-        let client_pool = ClientPool::new(
-            channel_pool.clone(),
-            tenant_id,
-            timeline_id,
-            shard_id,
-            auth_token.clone(),
-            compression,
-            Some(MAX_UNARY_CLIENTS),
-        );
-
-        // GetPage stream pool. Uses a dedicated client pool to avoid starving out unary clients,
-        // but shares a channel pool with it (as it's unbounded).
-        let stream_pool = StreamPool::new(
-            ClientPool::new(
-                channel_pool.clone(),
-                tenant_id,
-                timeline_id,
-                shard_id,
-                auth_token.clone(),
-                compression,
-                None, // unbounded, limited by stream pool
-            ),
-            Some(MAX_STREAMS),
-            MAX_STREAM_QUEUE_DEPTH,
-        );
-
-        // Bulk GetPage stream pool, e.g. for prefetches. Uses dedicated channel/client/stream pools
-        // to avoid head-of-line blocking of latency-sensitive requests.
-        let bulk_stream_pool = StreamPool::new(
-            ClientPool::new(
-                ChannelPool::new(url, MAX_CLIENTS_PER_CHANNEL)?,
-                tenant_id,
-                timeline_id,
-                shard_id,
-                auth_token,
-                compression,
-                None, // unbounded, limited by stream pool
-            ),
-            Some(MAX_BULK_STREAMS),
-            MAX_BULK_STREAM_QUEUE_DEPTH,
-        );
-
-        Ok(Self {
-            id: shard_id,
-            client_pool,
-            stream_pool,
-            bulk_stream_pool,
-        })
-    }
-
-    /// Returns a pooled client for this shard.
-    async fn client(&self) -> tonic::Result<ClientGuard> {
-        self.client_pool
-            .get()
-            .await
-            .map_err(|err| tonic::Status::internal(format!("failed to get client: {err}")))
-    }
-
-    /// Returns a pooled stream for this shard. If `bulk` is `true`, uses the dedicated bulk stream
-    /// pool (e.g. for prefetches).
-    async fn stream(&self, bulk: bool) -> StreamGuard {
-        match bulk {
-            false => self.stream_pool.get().await,
-            true => self.bulk_stream_pool.get().await,
-        }
-    }
-}

pageserver/client_grpc/src/lib.rs

@@ -1,7 +0,0 @@
-mod client;
-mod pool;
-mod retry;
-mod split;
-
-pub use client::{PageserverClient, ShardSpec};
-pub use pageserver_api::shard::ShardStripeSize; // used in ShardSpec

pageserver/client_grpc/src/pool.rs

@@ -1,779 +0,0 @@
-//! This module provides various Pageserver gRPC client resource pools.
-//!
-//! These pools are designed to reuse gRPC resources (connections, clients, and streams) across
-//! multiple concurrent callers (i.e. Postgres backends). This avoids the resource cost and latency
-//! of creating dedicated TCP connections and server tasks for every Postgres backend.
-//!
-//! Each resource has its own, nested pool. The pools are custom-built for the properties of each
-//! resource -- they are different enough that a generic pool isn't suitable.
-//!
-//! * ChannelPool: manages gRPC channels (TCP connections) to a single Pageserver. Multiple clients
-//!   can acquire and use the same channel concurrently (via HTTP/2 stream multiplexing), up to a
-//!   per-channel client limit. Channels may be closed when they are no longer used by any clients.
-//!
-//! * ClientPool: manages gRPC clients for a single tenant shard. Each client acquires a (shared)
-//!   channel from the ChannelPool for the client's lifetime. A client can only be acquired by a
-//!   single caller at a time, and is returned to the pool when dropped. Idle clients may be removed
-//!   from the pool after some time, to free up the channel.
-//!
-//! * StreamPool: manages bidirectional gRPC GetPage streams. Each stream acquires a client from the
-//!   ClientPool for the stream's lifetime. Internal streams are not exposed to callers; instead, it
-//!   returns a guard that can be used to send a single request, to properly enforce queue depth and
-//!   route responses. Internally, the pool will reuse or spin up a suitable stream for the request,
-//!   possibly pipelining multiple requests from multiple callers on the same stream (up to some
-//!   queue depth). Idle streams may be removed from the pool after a while to free up the client.
-//!
-//! Each channel corresponds to one TCP connection. Each client unary request and each stream
-//! corresponds to one HTTP/2 stream and server task.
-//!
-//! TODO: error handling (including custom error types).
-//! TODO: observability.
-
-use std::collections::{BTreeMap, HashMap};
-use std::num::NonZero;
-use std::ops::{Deref, DerefMut};
-use std::sync::atomic::{AtomicUsize, Ordering};
-use std::sync::{Arc, Mutex, Weak};
-use std::time::{Duration, Instant};
-
-use futures::StreamExt as _;
-use tokio::sync::mpsc::{Receiver, Sender};
-use tokio::sync::{OwnedSemaphorePermit, Semaphore, mpsc, oneshot};
-use tokio_util::sync::CancellationToken;
-use tonic::codec::CompressionEncoding;
-use tonic::transport::{Channel, Endpoint};
-use tracing::{error, warn};
-
-use pageserver_page_api as page_api;
-use utils::id::{TenantId, TimelineId};
-use utils::shard::ShardIndex;
-
-/// Reap channels/clients/streams that have been idle for this long.
-///
-/// TODO: this is per-pool. For nested pools, it can take up to 3x as long for a TCP connection to
-/// be reaped. First, we must wait for an idle stream to be reaped, which marks its client as idle.
-/// Then, we must wait for the idle client to be reaped, which marks its channel as idle. Then, we
-/// must wait for the idle channel to be reaped. Is that a problem? Maybe not, we just have to
-/// account for it when setting the reap threshold. Alternatively, we can immediately reap empty
-/// channels, and/or stream pool clients.
-const REAP_IDLE_THRESHOLD: Duration = match cfg!(any(test, feature = "testing")) {
-    false => Duration::from_secs(180),
-    true => Duration::from_secs(1), // exercise reaping in tests
-};
-
-/// Reap idle resources with this interval.
-const REAP_IDLE_INTERVAL: Duration = match cfg!(any(test, feature = "testing")) {
-    false => Duration::from_secs(10),
-    true => Duration::from_secs(1), // exercise reaping in tests
-};
-
-/// A gRPC channel pool, for a single Pageserver. A channel is shared by many clients (via HTTP/2
-/// stream multiplexing), up to `clients_per_channel` -- a new channel will be spun up beyond this.
-/// The pool does not limit the number of channels, and instead relies on `ClientPool` or
-/// `StreamPool` to limit the number of concurrent clients.
-///
-/// The pool is always wrapped in an outer `Arc`, to allow long-lived guards across tasks/threads.
-///
-/// TODO: consider prewarming a set of channels, to avoid initial connection latency.
-/// TODO: consider adding a circuit breaker for errors and fail fast.
-pub struct ChannelPool {
-    /// Pageserver endpoint to connect to.
-    endpoint: Endpoint,
-    /// Max number of clients per channel. Beyond this, a new channel will be created.
-    max_clients_per_channel: NonZero<usize>,
-    /// Open channels.
-    channels: Mutex<BTreeMap<ChannelID, ChannelEntry>>,
-    /// Reaps idle channels.
-    idle_reaper: Reaper,
-    /// Channel ID generator.
-    next_channel_id: AtomicUsize,
-}
-
-type ChannelID = usize;
-
-struct ChannelEntry {
-    /// The gRPC channel (i.e. TCP connection). Shared by multiple clients.
-    channel: Channel,
-    /// Number of clients using this channel.
-    clients: usize,
-    /// The channel has been idle (no clients) since this time. None if channel is in use.
-    /// INVARIANT: Some if clients == 0, otherwise None.
-    idle_since: Option<Instant>,
-}
-
-impl ChannelPool {
-    /// Creates a new channel pool for the given Pageserver endpoint.
-    pub fn new<E>(endpoint: E, max_clients_per_channel: NonZero<usize>) -> anyhow::Result<Arc<Self>>
-    where
-        E: TryInto<Endpoint> + Send + Sync + 'static,
-        <E as TryInto<Endpoint>>::Error: std::error::Error + Send + Sync,
-    {
-        let pool = Arc::new(Self {
-            endpoint: endpoint.try_into()?,
-            max_clients_per_channel,
-            channels: Mutex::default(),
-            idle_reaper: Reaper::new(REAP_IDLE_THRESHOLD, REAP_IDLE_INTERVAL),
-            next_channel_id: AtomicUsize::default(),
-        });
-        pool.idle_reaper.spawn(&pool);
-        Ok(pool)
-    }
-
-    /// Acquires a gRPC channel for a client. Multiple clients may acquire the same channel.
-    ///
-    /// This never blocks (except for mutex acquisition). The channel is connected lazily on first
-    /// use, and the `ChannelPool` does not have a channel limit. Channels will be re-established
-    /// automatically on failure (TODO: verify).
-    ///
-    /// Callers should not clone the returned channel, and must hold onto the returned guard as long
-    /// as the channel is in use. It is unfortunately not possible to enforce this: the Protobuf
-    /// client requires an owned `Channel` and we don't have access to the channel's internal
-    /// refcount.
-    ///
-    /// This is not performance-sensitive. It is only called when creating a new client, and clients
-    /// are pooled and reused by `ClientPool`. The total number of channels will also be small. O(n)
-    /// performance is therefore okay.
-    pub fn get(self: &Arc<Self>) -> ChannelGuard {
-        let mut channels = self.channels.lock().unwrap();
-
-        // Try to find an existing channel with available capacity. We check entries in BTreeMap
-        // order, to fill up the lower-ordered channels first. The ClientPool also prefers clients
-        // with lower-ordered channel IDs first. This will cluster clients in lower-ordered
-        // channels, and free up higher-ordered channels such that they can be reaped.
-        for (&id, entry) in channels.iter_mut() {
-            assert!(
-                entry.clients <= self.max_clients_per_channel.get(),
-                "channel overflow"
-            );
-            assert_eq!(
-                entry.idle_since.is_some(),
-                entry.clients == 0,
-                "incorrect channel idle state"
-            );
-            if entry.clients < self.max_clients_per_channel.get() {
-                entry.clients += 1;
-                entry.idle_since = None;
-                return ChannelGuard {
-                    pool: Arc::downgrade(self),
-                    id,
-                    channel: Some(entry.channel.clone()),
-                };
-            }
-        }
-
-        // Create a new channel. We connect lazily on first use, such that we don't block here and
-        // other clients can join onto the same channel while it's connecting.
-        let channel = self.endpoint.connect_lazy();
-
-        let id = self.next_channel_id.fetch_add(1, Ordering::Relaxed);
-        let entry = ChannelEntry {
-            channel: channel.clone(),
-            clients: 1, // account for the guard below
-            idle_since: None,
-        };
-        channels.insert(id, entry);
-
-        ChannelGuard {
-            pool: Arc::downgrade(self),
-            id,
-            channel: Some(channel),
-        }
-    }
-}
-
-impl Reapable for ChannelPool {
-    /// Reaps channels that have been idle since before the cutoff.
-    fn reap_idle(&self, cutoff: Instant) {
-        self.channels.lock().unwrap().retain(|_, entry| {
-            let Some(idle_since) = entry.idle_since else {
-                assert_ne!(entry.clients, 0, "empty channel not marked idle");
-                return true;
-            };
-            assert_eq!(entry.clients, 0, "idle channel has clients");
-            idle_since >= cutoff
-        })
-    }
-}
-
-/// Tracks a channel acquired from the pool. The owned inner channel can be obtained with `take()`,
-/// since the gRPC client requires an owned `Channel`.
-pub struct ChannelGuard {
-    pool: Weak<ChannelPool>,
-    id: ChannelID,
-    channel: Option<Channel>,
-}
-
-impl ChannelGuard {
-    /// Returns the inner owned channel. Panics if called more than once. The caller must hold onto
-    /// the guard as long as the channel is in use, and should not clone it.
-    pub fn take(&mut self) -> Channel {
-        self.channel.take().expect("channel already taken")
-    }
-}
-
-/// Returns the channel to the pool.
-impl Drop for ChannelGuard {
-    fn drop(&mut self) {
-        let Some(pool) = self.pool.upgrade() else {
-            return; // pool was dropped
-        };
-
-        let mut channels = pool.channels.lock().unwrap();
-        let entry = channels.get_mut(&self.id).expect("unknown channel");
-        assert!(entry.idle_since.is_none(), "active channel marked idle");
-        assert!(entry.clients > 0, "channel underflow");
-        entry.clients -= 1;
-        if entry.clients == 0 {
-            entry.idle_since = Some(Instant::now()); // mark channel as idle
-        }
-    }
-}
-
-/// A pool of gRPC clients for a single tenant shard. Each client acquires a channel from the inner
-/// `ChannelPool`. A client is only given out to single caller at a time. The pool limits the total
-/// number of concurrent clients to `max_clients` via semaphore.
-///
-/// The pool is always wrapped in an outer `Arc`, to allow long-lived guards across tasks/threads.
-pub struct ClientPool {
-    /// Tenant ID.
-    tenant_id: TenantId,
-    /// Timeline ID.
-    timeline_id: TimelineId,
-    /// Shard ID.
-    shard_id: ShardIndex,
-    /// Authentication token, if any.
-    auth_token: Option<String>,
-    /// Compression to use.
-    compression: Option<CompressionEncoding>,
-    /// Channel pool to acquire channels from.
-    channel_pool: Arc<ChannelPool>,
-    /// Limits the max number of concurrent clients for this pool. None if the pool is unbounded.
-    limiter: Option<Arc<Semaphore>>,
-    /// Idle pooled clients. Acquired clients are removed from here and returned on drop.
-    ///
-    /// The first client in the map will be acquired next. The map is sorted by client ID, which in
-    /// turn is sorted by its channel ID, such that we prefer acquiring idle clients from
-    /// lower-ordered channels. This allows us to free up and reap higher-numbered channels as idle
-    /// clients are reaped.
-    idle: Mutex<BTreeMap<ClientID, ClientEntry>>,
-    /// Reaps idle clients.
-    idle_reaper: Reaper,
-    /// Unique client ID generator.
-    next_client_id: AtomicUsize,
-}
-
-type ClientID = (ChannelID, usize);
-
-struct ClientEntry {
-    /// The pooled gRPC client.
-    client: page_api::Client,
-    /// The channel guard for the channel used by the client.
-    channel_guard: ChannelGuard,
-    /// The client has been idle since this time. All clients in `ClientPool::idle` are idle by
-    /// definition, so this is the time when it was added back to the pool.
-    idle_since: Instant,
-}
-
-impl ClientPool {
-    /// Creates a new client pool for the given tenant shard. Channels are acquired from the given
-    /// `ChannelPool`, which must point to a Pageserver that hosts the tenant shard. Allows up to
-    /// `max_clients` concurrent clients, or unbounded if None.
-    pub fn new(
-        channel_pool: Arc<ChannelPool>,
-        tenant_id: TenantId,
-        timeline_id: TimelineId,
-        shard_id: ShardIndex,
-        auth_token: Option<String>,
-        compression: Option<CompressionEncoding>,
-        max_clients: Option<NonZero<usize>>,
-    ) -> Arc<Self> {
-        let pool = Arc::new(Self {
-            tenant_id,
-            timeline_id,
-            shard_id,
-            auth_token,
-            compression,
-            channel_pool,
-            idle: Mutex::default(),
-            idle_reaper: Reaper::new(REAP_IDLE_THRESHOLD, REAP_IDLE_INTERVAL),
-            limiter: max_clients.map(|max| Arc::new(Semaphore::new(max.get()))),
-            next_client_id: AtomicUsize::default(),
-        });
-        pool.idle_reaper.spawn(&pool);
-        pool
-    }
-
-    /// Gets a client from the pool, or creates a new one if necessary. Connections are established
-    /// lazily and do not block, but this call can block if the pool is at `max_clients`. The client
-    /// is returned to the pool when the guard is dropped.
-    ///
-    /// This is moderately performance-sensitive. It is called for every unary request, but these
-    /// establish a new gRPC stream per request so they're already expensive. GetPage requests use
-    /// the `StreamPool` instead.
-    pub async fn get(self: &Arc<Self>) -> anyhow::Result<ClientGuard> {
-        // Acquire a permit if the pool is bounded.
-        let mut permit = None;
-        if let Some(limiter) = self.limiter.clone() {
-            permit = Some(limiter.acquire_owned().await.expect("never closed"));
-        }
-
-        // Fast path: acquire an idle client from the pool.
-        if let Some((id, entry)) = self.idle.lock().unwrap().pop_first() {
-            return Ok(ClientGuard {
-                pool: Arc::downgrade(self),
-                id,
-                client: Some(entry.client),
-                channel_guard: Some(entry.channel_guard),
-                permit,
-            });
-        }
-
-        // Slow path: construct a new client.
-        let mut channel_guard = self.channel_pool.get();
-        let client = page_api::Client::new(
-            channel_guard.take(),
-            self.tenant_id,
-            self.timeline_id,
-            self.shard_id,
-            self.auth_token.clone(),
-            self.compression,
-        )?;
-
-        Ok(ClientGuard {
-            pool: Arc::downgrade(self),
-            id: (
-                channel_guard.id,
-                self.next_client_id.fetch_add(1, Ordering::Relaxed),
-            ),
-            client: Some(client),
-            channel_guard: Some(channel_guard),
-            permit,
-        })
-    }
-}
-
-impl Reapable for ClientPool {
-    /// Reaps clients that have been idle since before the cutoff.
-    fn reap_idle(&self, cutoff: Instant) {
-        self.idle
-            .lock()
-            .unwrap()
-            .retain(|_, entry| entry.idle_since >= cutoff)
-    }
-}
-
-/// A client acquired from the pool. The inner client can be accessed via Deref. The client is
-/// returned to the pool when dropped.
-pub struct ClientGuard {
-    pool: Weak<ClientPool>,
-    id: ClientID,
-    client: Option<page_api::Client>,     // Some until dropped
-    channel_guard: Option<ChannelGuard>,  // Some until dropped
-    permit: Option<OwnedSemaphorePermit>, // None if pool is unbounded
-}
-
-impl Deref for ClientGuard {
-    type Target = page_api::Client;
-
-    fn deref(&self) -> &Self::Target {
-        self.client.as_ref().expect("not dropped")
-    }
-}
-
-impl DerefMut for ClientGuard {
-    fn deref_mut(&mut self) -> &mut Self::Target {
-        self.client.as_mut().expect("not dropped")
-    }
-}
-
-/// Returns the client to the pool.
-impl Drop for ClientGuard {
-    fn drop(&mut self) {
-        let Some(pool) = self.pool.upgrade() else {
-            return; // pool was dropped
-        };
-
-        let entry = ClientEntry {
-            client: self.client.take().expect("dropped once"),
-            channel_guard: self.channel_guard.take().expect("dropped once"),
-            idle_since: Instant::now(),
-        };
-        pool.idle.lock().unwrap().insert(self.id, entry);
-
-        _ = self.permit; // returned on drop, referenced for visibility
-    }
-}
-
-/// A pool of bidirectional gRPC streams. Currently only used for GetPage streams. Each stream
-/// acquires a client from the inner `ClientPool` for the stream's lifetime.
-///
-/// Individual streams are not exposed to callers -- instead, the returned guard can be used to send
-/// a single request and await the response. Internally, requests are multiplexed across streams and
-/// channels. This allows proper queue depth enforcement and response routing.
-///
-/// TODO: consider making this generic over request and response types; not currently needed.
-pub struct StreamPool {
-    /// The client pool to acquire clients from. Must be unbounded.
-    client_pool: Arc<ClientPool>,
-    /// All pooled streams.
-    ///
-    /// Incoming requests will be sent over an existing stream with available capacity. If all
-    /// streams are full, a new one is spun up and added to the pool (up to `max_streams`). Each
-    /// stream has an associated Tokio task that processes requests and responses.
-    streams: Mutex<HashMap<StreamID, StreamEntry>>,
-    /// The max number of concurrent streams, or None if unbounded.
-    max_streams: Option<NonZero<usize>>,
-    /// The max number of concurrent requests per stream.
-    max_queue_depth: NonZero<usize>,
-    /// Limits the max number of concurrent requests, given by `max_streams * max_queue_depth`.
-    /// None if the pool is unbounded.
-    limiter: Option<Arc<Semaphore>>,
-    /// Reaps idle streams.
-    idle_reaper: Reaper,
-    /// Stream ID generator.
-    next_stream_id: AtomicUsize,
-}
-
-type StreamID = usize;
-type RequestSender = Sender<(page_api::GetPageRequest, ResponseSender)>;
-type RequestReceiver = Receiver<(page_api::GetPageRequest, ResponseSender)>;
-type ResponseSender = oneshot::Sender<tonic::Result<page_api::GetPageResponse>>;
-
-struct StreamEntry {
-    /// Sends caller requests to the stream task. The stream task exits when this is dropped.
-    sender: RequestSender,
-    /// Number of in-flight requests on this stream.
-    queue_depth: usize,
-    /// The time when this stream went idle (queue_depth == 0).
-    /// INVARIANT: Some if queue_depth == 0, otherwise None.
-    idle_since: Option<Instant>,
-}
-
-impl StreamPool {
-    /// Creates a new stream pool, using the given client pool. It will send up to `max_queue_depth`
-    /// concurrent requests on each stream, and use up to `max_streams` concurrent streams.
-    ///
-    /// The client pool must be unbounded. The stream pool will enforce its own limits, and because
-    /// streams are long-lived they can cause persistent starvation if they exhaust the client pool.
-    /// The stream pool should generally have its own dedicated client pool (but it can share a
-    /// channel pool with others since these are always unbounded).
-    pub fn new(
-        client_pool: Arc<ClientPool>,
-        max_streams: Option<NonZero<usize>>,
-        max_queue_depth: NonZero<usize>,
-    ) -> Arc<Self> {
-        assert!(client_pool.limiter.is_none(), "bounded client pool");
-        let pool = Arc::new(Self {
-            client_pool,
-            streams: Mutex::default(),
-            limiter: max_streams.map(|max_streams| {
-                Arc::new(Semaphore::new(max_streams.get() * max_queue_depth.get()))
-            }),
-            max_streams,
-            max_queue_depth,
-            idle_reaper: Reaper::new(REAP_IDLE_THRESHOLD, REAP_IDLE_INTERVAL),
-            next_stream_id: AtomicUsize::default(),
-        });
-        pool.idle_reaper.spawn(&pool);
-        pool
-    }
-
-    /// Acquires an available stream from the pool, or spins up a new stream async if all streams
-    /// are full. Returns a guard that can be used to send a single request on the stream and await
-    /// the response, with queue depth quota already acquired. Blocks if the pool is at capacity
-    /// (i.e. `CLIENT_LIMIT * STREAM_QUEUE_DEPTH` requests in flight).
-    ///
-    /// This is very performance-sensitive, as it is on the GetPage hot path.
-    ///
-    /// TODO: this must do something more sophisticated for performance. We want:
-    ///
-    /// * Cheap, concurrent access in the common case where we can use a pooled stream.
-    /// * Quick acquisition of pooled streams with available capacity.
-    /// * Prefer streams that belong to lower-numbered channels, to reap idle channels.
-    /// * Prefer filling up existing streams' queue depth before spinning up new streams.
-    /// * Don't hold a lock while spinning up new streams.
-    /// * Allow concurrent clients to join onto streams while they're spun up.
-    /// * Allow spinning up multiple streams concurrently, but don't overshoot limits.
-    ///
-    /// For now, we just do something simple but inefficient (linear scan under mutex).
-    pub async fn get(self: &Arc<Self>) -> StreamGuard {
-        // Acquire a permit if the pool is bounded.
-        let mut permit = None;
-        if let Some(limiter) = self.limiter.clone() {
-            permit = Some(limiter.acquire_owned().await.expect("never closed"));
-        }
-        let mut streams = self.streams.lock().unwrap();
-
-        // Look for a pooled stream with available capacity.
-        for (&id, entry) in streams.iter_mut() {
-            assert!(
-                entry.queue_depth <= self.max_queue_depth.get(),
-                "stream queue overflow"
-            );
-            assert_eq!(
-                entry.idle_since.is_some(),
-                entry.queue_depth == 0,
-                "incorrect stream idle state"
-            );
-            if entry.queue_depth < self.max_queue_depth.get() {
-                entry.queue_depth += 1;
-                entry.idle_since = None;
-                return StreamGuard {
-                    pool: Arc::downgrade(self),
-                    id,
-                    sender: entry.sender.clone(),
-                    permit,
-                };
-            }
-        }
-
-        // No available stream, spin up a new one. We install the stream entry in the pool first and
-        // return the guard, while spinning up the stream task async. This allows other callers to
-        // join onto this stream and also create additional streams concurrently if this fills up.
-        let id = self.next_stream_id.fetch_add(1, Ordering::Relaxed);
-        let (req_tx, req_rx) = mpsc::channel(self.max_queue_depth.get());
-        let entry = StreamEntry {
-            sender: req_tx.clone(),
-            queue_depth: 1, // reserve quota for this caller
-            idle_since: None,
-        };
-        streams.insert(id, entry);
-
-        if let Some(max_streams) = self.max_streams {
-            assert!(streams.len() <= max_streams.get(), "stream overflow");
-        };
-
-        let client_pool = self.client_pool.clone();
-        let pool = Arc::downgrade(self);
-
-        tokio::spawn(async move {
-            if let Err(err) = Self::run_stream(client_pool, req_rx).await {
-                error!("stream failed: {err}");
-            }
-            // Remove stream from pool on exit. Weak reference to avoid holding the pool alive.
-            if let Some(pool) = pool.upgrade() {
-                let entry = pool.streams.lock().unwrap().remove(&id);
-                assert!(entry.is_some(), "unknown stream ID: {id}");
-            }
-        });
-
-        StreamGuard {
-            pool: Arc::downgrade(self),
-            id,
-            sender: req_tx,
-            permit,
-        }
-    }
-
-    /// Runs a stream task. This acquires a client from the `ClientPool` and establishes a
-    /// bidirectional GetPage stream, then forwards requests and responses between callers and the
-    /// stream. It does not track or enforce queue depths -- that's done by `get()` since it must be
-    /// atomic with pool stream acquisition.
-    ///
-    /// The task exits when the request channel is closed, or on a stream error. The caller is
-    /// responsible for removing the stream from the pool on exit.
-    async fn run_stream(
-        client_pool: Arc<ClientPool>,
-        mut caller_rx: RequestReceiver,
-    ) -> anyhow::Result<()> {
-        // Acquire a client from the pool and create a stream.
-        let mut client = client_pool.get().await?;
-
-        // NB: use an unbounded channel such that the stream send never blocks. Otherwise, we could
-        // theoretically deadlock if both the client and server block on sends (since we're not
-        // reading responses while sending). This is unlikely to happen due to gRPC/TCP buffers and
-        // low queue depths, but it was seen to happen with the libpq protocol so better safe than
-        // sorry. It should never buffer more than the queue depth anyway, but using an unbounded
-        // channel guarantees that it will never block.
-        let (req_tx, req_rx) = mpsc::unbounded_channel();
-        let req_stream = tokio_stream::wrappers::UnboundedReceiverStream::new(req_rx);
-        let mut resp_stream = client.get_pages(req_stream).await?;
-
-        // Track caller response channels by request ID. If the task returns early, these response
-        // channels will be dropped and the waiting callers will receive an error.
-        //
-        // NB: this will leak entries if the server doesn't respond to a request (by request ID).
-        // It shouldn't happen, and if it does it will often hold onto queue depth quota anyway and
-        // block further use. But we could consider reaping closed channels after some time.
-        let mut callers = HashMap::new();
-
-        // Process requests and responses.
-        loop {
-            tokio::select! {
-                // Receive requests from callers and send them to the stream.
-                req = caller_rx.recv() => {
-                    // Shut down if request channel is closed.
-                    let Some((req, resp_tx)) = req else {
-                        return Ok(());
-                    };
-
-                    // Store the response channel by request ID.
-                    if callers.contains_key(&req.request_id) {
-                        // Error on request ID duplicates. Ignore callers that went away.
-                        _ = resp_tx.send(Err(tonic::Status::invalid_argument(
-                            format!("duplicate request ID: {}", req.request_id),
-                        )));
-                        continue;
-                    }
-                    callers.insert(req.request_id, resp_tx);
-
-                    // Send the request on the stream. Bail out if the stream is closed.
-                    req_tx.send(req).map_err(|_| {
-                        tonic::Status::unavailable("stream closed")
-                    })?;
-                }
-
-                // Receive responses from the stream and send them to callers.
-                resp = resp_stream.next() => {
-                    // Shut down if the stream is closed, and bail out on stream errors.
-                    let Some(resp) = resp.transpose()? else {
-                        return Ok(())
-                    };
-
-                    // Send the response to the caller. Ignore errors if the caller went away.
-                    let Some(resp_tx) = callers.remove(&resp.request_id) else {
-                        warn!("received response for unknown request ID: {}", resp.request_id);
-                        continue;
-                    };
-                    _ = resp_tx.send(Ok(resp));
-                }
-            }
-        }
-    }
-}
-
-impl Reapable for StreamPool {
-    /// Reaps streams that have been idle since before the cutoff.
-    fn reap_idle(&self, cutoff: Instant) {
-        self.streams.lock().unwrap().retain(|_, entry| {
-            let Some(idle_since) = entry.idle_since else {
-                assert_ne!(entry.queue_depth, 0, "empty stream not marked idle");
-                return true;
-            };
-            assert_eq!(entry.queue_depth, 0, "idle stream has requests");
-            idle_since >= cutoff
-        });
-    }
-}
-
-/// A pooled stream reference. Can be used to send a single request, to properly enforce queue
-/// depth. Queue depth is already reserved and will be returned on drop.
-pub struct StreamGuard {
-    pool: Weak<StreamPool>,
-    id: StreamID,
-    sender: RequestSender,
-    permit: Option<OwnedSemaphorePermit>, // None if pool is unbounded
-}
-
-impl StreamGuard {
-    /// Sends a request on the stream and awaits the response. Consumes the guard, since it's only
-    /// valid for a single request (to enforce queue depth). This also drops the guard on return and
-    /// returns the queue depth quota to the pool.
-    ///
-    /// The `GetPageRequest::request_id` must be unique across in-flight requests.
-    ///
-    /// NB: errors are often returned as `GetPageResponse::status_code` instead of `tonic::Status`
-    /// to avoid tearing down the stream for per-request errors. Callers must check this.
-    pub async fn send(
-        self,
-        req: page_api::GetPageRequest,
-    ) -> tonic::Result<page_api::GetPageResponse> {
-        let (resp_tx, resp_rx) = oneshot::channel();
-
-        self.sender
-            .send((req, resp_tx))
-            .await
-            .map_err(|_| tonic::Status::unavailable("stream closed"))?;
-
-        resp_rx
-            .await
-            .map_err(|_| tonic::Status::unavailable("stream closed"))?
-    }
-}
-
-impl Drop for StreamGuard {
-    fn drop(&mut self) {
-        let Some(pool) = self.pool.upgrade() else {
-            return; // pool was dropped
-        };
-
-        // Release the queue depth reservation on drop. This can prematurely decrement it if dropped
-        // before the response is received, but that's okay.
-        //
-        // TODO: actually, it's probably not okay. Queue depth release should be moved into the
-        // stream task, such that it continues to account for the queue depth slot until the server
-        // responds. Otherwise, if a slow request times out and keeps blocking the stream, the
-        // server will keep waiting on it and we can pile on subsequent requests (including the
-        // timeout retry) in the same stream and get blocked. But we may also want to avoid blocking
-        // requests on e.g. LSN waits and layer downloads, instead returning early to free up the
-        // stream. Or just scale out streams with a queue depth of 1 to sidestep all head-of-line
-        // blocking. TBD.
-        let mut streams = pool.streams.lock().unwrap();
-        let entry = streams.get_mut(&self.id).expect("unknown stream");
-        assert!(entry.idle_since.is_none(), "active stream marked idle");
-        assert!(entry.queue_depth > 0, "stream queue underflow");
-        entry.queue_depth -= 1;
-        if entry.queue_depth == 0 {
-            entry.idle_since = Some(Instant::now()); // mark stream as idle
-        }
-
-        _ = self.permit; // returned on drop, referenced for visibility
-    }
-}
-
-/// Periodically reaps idle resources from a pool.
-struct Reaper {
-    /// The task check interval.
-    interval: Duration,
-    /// The threshold for reaping idle resources.
-    threshold: Duration,
-    /// Cancels the reaper task. Cancelled when the reaper is dropped.
-    cancel: CancellationToken,
-}
-
-impl Reaper {
-    /// Creates a new reaper.
-    pub fn new(threshold: Duration, interval: Duration) -> Self {
-        Self {
-            cancel: CancellationToken::new(),
-            threshold,
-            interval,
-        }
-    }
-
-    /// Spawns a task to periodically reap idle resources from the given task pool. The task is
-    /// cancelled when the reaper is dropped.
-    pub fn spawn(&self, pool: &Arc<impl Reapable>) {
-        // NB: hold a weak pool reference, otherwise the task will prevent dropping the pool.
-        let pool = Arc::downgrade(pool);
-        let cancel = self.cancel.clone();
-        let (interval, threshold) = (self.interval, self.threshold);
-
-        tokio::spawn(async move {
-            loop {
-                tokio::select! {
-                    _ = tokio::time::sleep(interval) => {
-                        let Some(pool) = pool.upgrade() else {
-                            return; // pool was dropped
-                        };
-                        pool.reap_idle(Instant::now() - threshold);
-                    }
-
-                    _ = cancel.cancelled() => return,
-                }
-            }
-        });
-    }
-}
-
-impl Drop for Reaper {
-    fn drop(&mut self) {
-        self.cancel.cancel(); // cancel reaper task
-    }
-}
-
-/// A reapable resource pool.
-trait Reapable: Send + Sync + 'static {
-    /// Reaps resources that have been idle since before the given cutoff.
-    fn reap_idle(&self, cutoff: Instant);
-}

pageserver/client_grpc/src/retry.rs

@@ -1,154 +0,0 @@
-use std::time::Duration;
-
-use tokio::time::Instant;
-use tracing::{error, info, warn};
-
-use utils::backoff::exponential_backoff_duration;
-
-/// A retry handler for Pageserver gRPC requests.
-///
-/// This is used instead of backoff::retry for better control and observability.
-pub struct Retry;
-
-impl Retry {
-    /// The per-request timeout.
-    // TODO: tune these, and/or make them configurable. Should we retry forever?
-    const REQUEST_TIMEOUT: Duration = Duration::from_secs(10);
-    /// The total timeout across all attempts
-    const TOTAL_TIMEOUT: Duration = Duration::from_secs(60);
-    /// The initial backoff duration.
-    const BASE_BACKOFF: Duration = Duration::from_millis(10);
-    /// The maximum backoff duration.
-    const MAX_BACKOFF: Duration = Duration::from_secs(10);
-    /// If true, log successful requests. For debugging.
-    const LOG_SUCCESS: bool = false;
-
-    /// Runs the given async closure with timeouts and retries (exponential backoff), passing the
-    /// attempt number starting at 0. Logs errors, using the current tracing span for context.
-    ///
-    /// Only certain gRPC status codes are retried, see [`Self::should_retry`]. For default
-    /// timeouts, see [`Self::REQUEST_TIMEOUT`] and [`Self::TOTAL_TIMEOUT`].
-    pub async fn with<T, F, O>(&self, mut f: F) -> tonic::Result<T>
-    where
-        F: FnMut(usize) -> O, // takes attempt number, starting at 0
-        O: Future<Output = tonic::Result<T>>,
-    {
-        let started = Instant::now();
-        let deadline = started + Self::TOTAL_TIMEOUT;
-        let mut last_error = None;
-        let mut retries = 0;
-        loop {
-            // Set up a future to wait for the backoff (if any) and run the request with a timeout.
-            let backoff_and_try = async {
-                // NB: sleep() always sleeps 1ms, even when given a 0 argument. See:
-                // https://github.com/tokio-rs/tokio/issues/6866
-                if let Some(backoff) = Self::backoff_duration(retries) {
-                    tokio::time::sleep(backoff).await;
-                }
-
-                let request_started = Instant::now();
-                tokio::time::timeout(Self::REQUEST_TIMEOUT, f(retries))
-                    .await
-                    .map_err(|_| {
-                        tonic::Status::deadline_exceeded(format!(
-                            "request timed out after {:.3}s",
-                            request_started.elapsed().as_secs_f64()
-                        ))
-                    })?
-            };
-
-            // Wait for the backoff and request, or bail out if the total timeout is exceeded.
-            let result = tokio::select! {
-                result = backoff_and_try => result,
-
-                _ = tokio::time::sleep_until(deadline) => {
-                    let last_error = last_error.unwrap_or_else(|| {
-                        tonic::Status::deadline_exceeded(format!(
-                            "request timed out after {:.3}s",
-                            started.elapsed().as_secs_f64()
-                        ))
-                    });
-                    error!(
-                        "giving up after {:.3}s and {retries} retries, last error {:?}: {}",
-                        started.elapsed().as_secs_f64(), last_error.code(), last_error.message(),
-                    );
-                    return Err(last_error);
-                }
-            };
-
-            match result {
-                // Success, return the result.
-                Ok(result) => {
-                    if retries > 0 || Self::LOG_SUCCESS {
-                        info!(
-                            "request succeeded after {retries} retries in {:.3}s",
-                            started.elapsed().as_secs_f64(),
-                        );
-                    }
-
-                    return Ok(result);
-                }
-
-                // Error, retry or bail out.
-                Err(status) => {
-                    let (code, message) = (status.code(), status.message());
-                    let attempt = retries + 1;
-
-                    if !Self::should_retry(code) {
-                        // NB: include the attempt here too. This isn't necessarily the first
-                        // attempt, because the error may change between attempts.
-                        error!(
-                            "request failed with {code:?}: {message}, not retrying (attempt {attempt})"
-                        );
-                        return Err(status);
-                    }
-
-                    warn!("request failed with {code:?}: {message}, retrying (attempt {attempt})");
-
-                    retries += 1;
-                    last_error = Some(status);
-                }
-            }
-        }
-    }
-
-    /// Returns the backoff duration for the given retry attempt, or None for no backoff.
-    fn backoff_duration(retry: usize) -> Option<Duration> {
-        let backoff = exponential_backoff_duration(
-            retry as u32,
-            Self::BASE_BACKOFF.as_secs_f64(),
-            Self::MAX_BACKOFF.as_secs_f64(),
-        );
-        (!backoff.is_zero()).then_some(backoff)
-    }
-
-    /// Returns true if the given status code should be retries.
-    fn should_retry(code: tonic::Code) -> bool {
-        match code {
-            tonic::Code::Ok => panic!("unexpected Ok status code"),
-
-            // These codes are transient, so retry them.
-            tonic::Code::Aborted => true,
-            tonic::Code::Cancelled => true,
-            tonic::Code::DeadlineExceeded => true, // maybe transient slowness
-            tonic::Code::ResourceExhausted => true,
-            tonic::Code::Unavailable => true,
-
-            // The following codes will like continue to fail, so don't retry.
-            tonic::Code::AlreadyExists => false,
-            tonic::Code::DataLoss => false,
-            tonic::Code::FailedPrecondition => false,
-            // NB: don't retry Internal. It is intended for serious errors such as invariant
-            // violations, and is also used for client-side invariant checks that would otherwise
-            // result in retry loops.
-            tonic::Code::Internal => false,
-            tonic::Code::InvalidArgument => false,
-            tonic::Code::NotFound => false,
-            tonic::Code::OutOfRange => false,
-            tonic::Code::PermissionDenied => false,
-            tonic::Code::Unauthenticated => false,
-            tonic::Code::Unimplemented => false,
-            tonic::Code::Unknown => false,
-        }
-    }
-}

pageserver/client_grpc/src/split.rs

@@ -1,209 +0,0 @@
-use std::collections::HashMap;
-
-use bytes::Bytes;
-
-use pageserver_api::key::rel_block_to_key;
-use pageserver_api::shard::{ShardStripeSize, key_to_shard_number};
-use pageserver_page_api as page_api;
-use utils::shard::{ShardCount, ShardIndex, ShardNumber};
-
-/// Splits GetPageRequests that straddle shard boundaries and assembles the responses.
-/// TODO: add tests for this.
-pub struct GetPageSplitter {
-    /// Split requests by shard index.
-    requests: HashMap<ShardIndex, page_api::GetPageRequest>,
-    /// The response being assembled. Preallocated with empty pages, to be filled in.
-    response: page_api::GetPageResponse,
-    /// Maps the offset in `request.block_numbers` and `response.pages` to the owning shard. Used
-    /// to assemble the response pages in the same order as the original request.
-    block_shards: Vec<ShardIndex>,
-}
-
-impl GetPageSplitter {
-    /// Checks if the given request only touches a single shard, and returns the shard ID. This is
-    /// the common case, so we check first in order to avoid unnecessary allocations and overhead.
-    pub fn for_single_shard(
-        req: &page_api::GetPageRequest,
-        count: ShardCount,
-        stripe_size: ShardStripeSize,
-    ) -> Option<ShardIndex> {
-        // Fast path: unsharded tenant.
-        if count.is_unsharded() {
-            return Some(ShardIndex::unsharded());
-        }
-
-        // Find the first page's shard, for comparison. If there are no pages, just return the first
-        // shard (caller likely checked already, otherwise the server will reject it).
-        let Some(&first_page) = req.block_numbers.first() else {
-            return Some(ShardIndex::new(ShardNumber(0), count));
-        };
-        let key = rel_block_to_key(req.rel, first_page);
-        let shard_number = key_to_shard_number(count, stripe_size, &key);
-
-        req.block_numbers
-            .iter()
-            .skip(1) // computed above
-            .all(|&blkno| {
-                let key = rel_block_to_key(req.rel, blkno);
-                key_to_shard_number(count, stripe_size, &key) == shard_number
-            })
-            .then_some(ShardIndex::new(shard_number, count))
-    }
-
-    /// Splits the given request.
-    pub fn split(
-        req: page_api::GetPageRequest,
-        count: ShardCount,
-        stripe_size: ShardStripeSize,
-    ) -> Self {
-        // The caller should make sure we don't split requests unnecessarily.
-        debug_assert!(
-            Self::for_single_shard(&req, count, stripe_size).is_none(),
-            "unnecessary request split"
-        );
-
-        // Split the requests by shard index.
-        let mut requests = HashMap::with_capacity(2); // common case
-        let mut block_shards = Vec::with_capacity(req.block_numbers.len());
-        for &blkno in &req.block_numbers {
-            let key = rel_block_to_key(req.rel, blkno);
-            let shard_number = key_to_shard_number(count, stripe_size, &key);
-            let shard_id = ShardIndex::new(shard_number, count);
-
-            requests
-                .entry(shard_id)
-                .or_insert_with(|| page_api::GetPageRequest {
-                    request_id: req.request_id,
-                    request_class: req.request_class,
-                    rel: req.rel,
-                    read_lsn: req.read_lsn,
-                    block_numbers: Vec::new(),
-                })
-                .block_numbers
-                .push(blkno);
-            block_shards.push(shard_id);
-        }
-
-        // Construct a response to be populated by shard responses. Preallocate empty page slots
-        // with the expected block numbers.
-        let response = page_api::GetPageResponse {
-            request_id: req.request_id,
-            status_code: page_api::GetPageStatusCode::Ok,
-            reason: None,
-            rel: req.rel,
-            pages: req
-                .block_numbers
-                .into_iter()
-                .map(|block_number| {
-                    page_api::Page {
-                        block_number,
-                        image: Bytes::new(), // empty page slot to be filled in
-                    }
-                })
-                .collect(),
-        };
-
-        Self {
-            requests,
-            response,
-            block_shards,
-        }
-    }
-
-    /// Drains the per-shard requests, moving them out of the splitter to avoid extra allocations.
-    pub fn drain_requests(
-        &mut self,
-    ) -> impl Iterator<Item = (ShardIndex, page_api::GetPageRequest)> {
-        self.requests.drain()
-    }
-
-    /// Adds a response from the given shard. The response must match the request ID and have an OK
-    /// status code. A response must not already exist for the given shard ID.
-    #[allow(clippy::result_large_err)]
-    pub fn add_response(
-        &mut self,
-        shard_id: ShardIndex,
-        response: page_api::GetPageResponse,
-    ) -> tonic::Result<()> {
-        // The caller should already have converted status codes into tonic::Status.
-        if response.status_code != page_api::GetPageStatusCode::Ok {
-            return Err(tonic::Status::internal(format!(
-                "unexpected non-OK response for shard {shard_id}: {} {}",
-                response.status_code,
-                response.reason.unwrap_or_default()
-            )));
-        }
-
-        if response.request_id != self.response.request_id {
-            return Err(tonic::Status::internal(format!(
-                "response ID mismatch for shard {shard_id}: expected {}, got {}",
-                self.response.request_id, response.request_id
-            )));
-        }
-
-        // Place the shard response pages into the assembled response, in request order.
-        let mut pages = response.pages.into_iter();
-
-        for (i, &s) in self.block_shards.iter().enumerate() {
-            if shard_id != s {
-                continue;
-            }
-
-            let Some(slot) = self.response.pages.get_mut(i) else {
-                return Err(tonic::Status::internal(format!(
-                    "no block_shards slot {i} for shard {shard_id}"
-                )));
-            };
-            let Some(page) = pages.next() else {
-                return Err(tonic::Status::internal(format!(
-                    "missing page {} in shard {shard_id} response",
-                    slot.block_number
-                )));
-            };
-            if page.block_number != slot.block_number {
-                return Err(tonic::Status::internal(format!(
-                    "shard {shard_id} returned wrong page at index {i}, expected {} got {}",
-                    slot.block_number, page.block_number
-                )));
-            }
-            if !slot.image.is_empty() {
-                return Err(tonic::Status::internal(format!(
-                    "shard {shard_id} returned duplicate page {} at index {i}",
-                    slot.block_number
-                )));
-            }
-
-            *slot = page;
-        }
-
-        // Make sure we've consumed all pages from the shard response.
-        if let Some(extra_page) = pages.next() {
-            return Err(tonic::Status::internal(format!(
-                "shard {shard_id} returned extra page: {}",
-                extra_page.block_number
-            )));
-        }
-
-        Ok(())
-    }
-
-    /// Fetches the final, assembled response.
-    #[allow(clippy::result_large_err)]
-    pub fn get_response(self) -> tonic::Result<page_api::GetPageResponse> {
-        // Check that the response is complete.
-        for (i, page) in self.response.pages.iter().enumerate() {
-            if page.image.is_empty() {
-                return Err(tonic::Status::internal(format!(
-                    "missing page {} for shard {}",
-                    page.block_number,
-                    self.block_shards
-                        .get(i)
-                        .map(|s| s.to_string())
-                        .unwrap_or_else(|| "?".to_string())
-                )));
-            }
-        }
-
-        Ok(self.response)
-    }
-}

pageserver/ctl/Cargo.toml

@@ -17,7 +17,6 @@ pageserver = { path = ".." }
 pageserver_api.workspace = true
 remote_storage = { path = "../../libs/remote_storage" }
 postgres_ffi.workspace = true
-serde.workspace = true
 thiserror.workspace = true
 tokio.workspace = true
 tokio-util.workspace = true

pageserver/ctl/src/download_remote_object.rs

@@ -1,85 +0,0 @@
-use camino::Utf8PathBuf;
-use clap::Parser;
-use tokio_util::sync::CancellationToken;
-
-/// Download a specific object from remote storage to a local file.
-///
-/// The remote storage configuration is supplied via the `REMOTE_STORAGE_CONFIG` environment
-/// variable, in the same TOML format that the pageserver itself understands. This allows the
-/// command to work with any cloud supported by the `remote_storage` crate (currently AWS S3,
-/// Azure Blob Storage and local files), as long as the credentials are available via the
-/// standard environment variables expected by the underlying SDKs.
-///
-/// Examples for setting the environment variable:
-///
-/// ```bash
-/// # AWS S3 (region can also be provided via AWS_REGION)
-/// export REMOTE_STORAGE_CONFIG='remote_storage = { bucket_name = "my-bucket", bucket_region = "us-east-2" }'
-///
-/// # Azure Blob Storage (account key picked up from AZURE_STORAGE_ACCOUNT_KEY)
-/// export REMOTE_STORAGE_CONFIG='remote_storage = { container = "my-container", account = "my-account" }'
-/// ```
-#[derive(Parser)]
-pub(crate) struct DownloadRemoteObjectCmd {
-    /// Key / path of the object to download (relative to the remote storage prefix).
-    ///
-    /// Examples:
-    ///   "wal/3aa8f.../00000001000000000000000A"
-    ///   "pageserver/v1/tenants/<tenant_id>/timelines/<timeline_id>/layer_12345"
-    pub remote_path: String,
-
-    /// Path of the local file to create. Existing file will be overwritten.
-    ///
-    /// Examples:
-    ///   "./segment"
-    ///   "/tmp/layer_12345.parquet"
-    pub output_file: Utf8PathBuf,
-}
-
-pub(crate) async fn main(cmd: &DownloadRemoteObjectCmd) -> anyhow::Result<()> {
-    use remote_storage::{DownloadOpts, GenericRemoteStorage, RemotePath, RemoteStorageConfig};
-
-    // Fetch remote storage configuration from the environment
-    let config_str = std::env::var("REMOTE_STORAGE_CONFIG").map_err(|_| {
-        anyhow::anyhow!(
-            "'REMOTE_STORAGE_CONFIG' environment variable must be set to a valid remote storage TOML config"
-        )
-    })?;
-
-    let config = RemoteStorageConfig::from_toml_str(&config_str)?;
-
-    // Initialise remote storage client
-    let storage = GenericRemoteStorage::from_config(&config).await?;
-
-    // RemotePath must be relative – leading slashes confuse the parser.
-    let remote_path_str = cmd.remote_path.trim_start_matches('/');
-    let remote_path = RemotePath::from_string(remote_path_str)?;
-
-    let cancel = CancellationToken::new();
-
-    println!(
-        "Downloading '{remote_path}' from remote storage bucket {:?} ...",
-        config.storage.bucket_name()
-    );
-
-    // Start the actual download
-    let download = storage
-        .download(&remote_path, &DownloadOpts::default(), &cancel)
-        .await?;
-
-    // Stream to file
-    let mut reader = tokio_util::io::StreamReader::new(download.download_stream);
-    let tmp_path = cmd.output_file.with_extension("tmp");
-    let mut file = tokio::fs::File::create(&tmp_path).await?;
-    tokio::io::copy(&mut reader, &mut file).await?;
-    file.sync_all().await?;
-    // Atomically move into place
-    tokio::fs::rename(&tmp_path, &cmd.output_file).await?;
-
-    println!(
-        "Downloaded to '{}'. Last modified: {:?}, etag: {}",
-        cmd.output_file, download.last_modified, download.etag
-    );
-
-    Ok(())
-}

pageserver/ctl/src/index_part.rs

@@ -1,180 +1,10 @@
-use std::str::FromStr;
-
-use anyhow::{Context, Ok};
+use anyhow::Context;
 use camino::Utf8PathBuf;
-use pageserver::tenant::{
-    IndexPart,
-    layer_map::{LayerMap, SearchResult},
-    remote_timeline_client::{index::LayerFileMetadata, remote_layer_path},
-    storage_layer::{LayerName, LayerVisibilityHint, PersistentLayerDesc, ReadableLayerWeak},
-};
-use pageserver_api::key::Key;
-use serde::Serialize;
-use std::collections::BTreeMap;
-use utils::{
-    id::{TenantId, TimelineId},
-    lsn::Lsn,
-    shard::TenantShardId,
-};
+use pageserver::tenant::IndexPart;
 
 #[derive(clap::Subcommand)]
 pub(crate) enum IndexPartCmd {
-    Dump {
-        path: Utf8PathBuf,
-    },
-    /// Find all layers that need to be searched to construct the given page at the given LSN.
-    Search {
-        #[arg(long)]
-        tenant_id: String,
-        #[arg(long)]
-        timeline_id: String,
-        #[arg(long)]
-        path: Utf8PathBuf,
-        #[arg(long)]
-        key: String,
-        #[arg(long)]
-        lsn: String,
-    },
-    /// List all visible delta and image layers at the latest LSN.
-    ListVisibleLayers {
-        #[arg(long)]
-        path: Utf8PathBuf,
-    },
-}
-
-fn create_layer_map_from_index_part(
-    index_part: &IndexPart,
-    tenant_shard_id: TenantShardId,
-    timeline_id: TimelineId,
-) -> LayerMap {
-    let mut layer_map = LayerMap::default();
-    {
-        let mut updates = layer_map.batch_update();
-        for (key, value) in index_part.layer_metadata.iter() {
-            updates.insert_historic(PersistentLayerDesc::from_filename(
-                tenant_shard_id,
-                timeline_id,
-                key.clone(),
-                value.file_size,
-            ));
-        }
-    }
-    layer_map
-}
-
-async fn search_layers(
-    tenant_id: &str,
-    timeline_id: &str,
-    path: &Utf8PathBuf,
-    key: &str,
-    lsn: &str,
-) -> anyhow::Result<()> {
-    let tenant_id = TenantId::from_str(tenant_id).unwrap();
-    let tenant_shard_id = TenantShardId::unsharded(tenant_id);
-    let timeline_id = TimelineId::from_str(timeline_id).unwrap();
-    let index_json = {
-        let bytes = tokio::fs::read(path).await?;
-        IndexPart::from_json_bytes(&bytes).unwrap()
-    };
-    let layer_map = create_layer_map_from_index_part(&index_json, tenant_shard_id, timeline_id);
-    let key = Key::from_hex(key)?;
-
-    let lsn = Lsn::from_str(lsn).unwrap();
-    let mut end_lsn = lsn;
-    loop {
-        let result = layer_map.search(key, end_lsn);
-        match result {
-            Some(SearchResult { layer, lsn_floor }) => {
-                let disk_layer = match layer {
-                    ReadableLayerWeak::PersistentLayer(layer) => layer,
-                    ReadableLayerWeak::InMemoryLayer(_) => {
-                        anyhow::bail!("unexpected in-memory layer")
-                    }
-                };
-
-                let metadata = index_json
-                    .layer_metadata
-                    .get(&disk_layer.layer_name())
-                    .unwrap();
-                println!(
-                    "{}",
-                    remote_layer_path(
-                        &tenant_id,
-                        &timeline_id,
-                        metadata.shard,
-                        &disk_layer.layer_name(),
-                        metadata.generation
-                    )
-                );
-                end_lsn = lsn_floor;
-            }
-            None => break,
-        }
-    }
-    Ok(())
-}
-
-#[derive(Debug, Clone, Serialize)]
-struct VisibleLayers {
-    pub total_images: u64,
-    pub total_image_bytes: u64,
-    pub total_deltas: u64,
-    pub total_delta_bytes: u64,
-    pub layer_metadata: BTreeMap<LayerName, LayerFileMetadata>,
-}
-
-impl VisibleLayers {
-    pub fn new() -> Self {
-        Self {
-            layer_metadata: BTreeMap::new(),
-            total_images: 0,
-            total_image_bytes: 0,
-            total_deltas: 0,
-            total_delta_bytes: 0,
-        }
-    }
-
-    pub fn add_layer(&mut self, name: LayerName, layer: LayerFileMetadata) {
-        match name {
-            LayerName::Image(_) => {
-                self.total_images += 1;
-                self.total_image_bytes += layer.file_size;
-            }
-            LayerName::Delta(_) => {
-                self.total_deltas += 1;
-                self.total_delta_bytes += layer.file_size;
-            }
-        }
-        self.layer_metadata.insert(name, layer);
-    }
-}
-
-async fn list_visible_layers(path: &Utf8PathBuf) -> anyhow::Result<()> {
-    let tenant_id = TenantId::generate();
-    let tenant_shard_id = TenantShardId::unsharded(tenant_id);
-    let timeline_id = TimelineId::generate();
-
-    let bytes = tokio::fs::read(path).await.context("read file")?;
-    let index_part = IndexPart::from_json_bytes(&bytes).context("deserialize")?;
-    let layer_map = create_layer_map_from_index_part(&index_part, tenant_shard_id, timeline_id);
-    let mut visible_layers = VisibleLayers::new();
-    let (layers, _key_space) = layer_map.get_visibility(Vec::new());
-    for (layer, visibility) in layers {
-        if visibility == LayerVisibilityHint::Visible {
-            visible_layers.add_layer(
-                layer.layer_name(),
-                index_part
-                    .layer_metadata
-                    .get(&layer.layer_name())
-                    .unwrap()
-                    .clone(),
-            );
-        }
-    }
-    let output = serde_json::to_string_pretty(&visible_layers).context("serialize output")?;
-    println!("{output}");
-
-    Ok(())
+    Dump { path: Utf8PathBuf },
 }
 
 pub(crate) async fn main(cmd: &IndexPartCmd) -> anyhow::Result<()> {
@@ -186,13 +16,5 @@ pub(crate) async fn main(cmd: &IndexPartCmd) -> anyhow::Result<()> {
             println!("{output}");
             Ok(())
         }
-        IndexPartCmd::Search {
-            tenant_id,
-            timeline_id,
-            path,
-            key,
-            lsn,
-        } => search_layers(tenant_id, timeline_id, path, key, lsn).await,
-        IndexPartCmd::ListVisibleLayers { path } => list_visible_layers(path).await,
     }
 }

pageserver/ctl/src/main.rs

@@ -4,7 +4,6 @@
 //!
 //! Separate, `metadata` subcommand allows to print and update pageserver's metadata file.
 
-mod download_remote_object;
 mod draw_timeline_dir;
 mod index_part;
 mod key;
@@ -17,7 +16,6 @@ use std::time::{Duration, SystemTime};
 
 use camino::{Utf8Path, Utf8PathBuf};
 use clap::{Parser, Subcommand};
-use download_remote_object::DownloadRemoteObjectCmd;
 use index_part::IndexPartCmd;
 use layers::LayerCmd;
 use page_trace::PageTraceCmd;
@@ -65,7 +63,6 @@ enum Commands {
     /// Debug print a hex key found from logs
     Key(key::DescribeKeyCommand),
     PageTrace(PageTraceCmd),
-    DownloadRemoteObject(DownloadRemoteObjectCmd),
 }
 
 /// Read and update pageserver metadata file
@@ -188,9 +185,6 @@ async fn main() -> anyhow::Result<()> {
         }
         Commands::Key(dkc) => dkc.execute(),
         Commands::PageTrace(cmd) => page_trace::main(&cmd)?,
-        Commands::DownloadRemoteObject(cmd) => {
-            download_remote_object::main(&cmd).await?;
-        }
     };
     Ok(())
 }

pageserver/page_api/proto/page_service.proto

@@ -153,7 +153,7 @@ message GetDbSizeResponse {
 message GetPageRequest {
   // A request ID. Will be included in the response. Should be unique for
   // in-flight requests on the stream.
-  RequestID request_id = 1;
+  uint64 request_id = 1;
   // The request class.
   GetPageClass request_class = 2;
   // The LSN to read at.
@@ -177,14 +177,6 @@ message GetPageRequest {
   repeated uint32 block_number = 5;
 }
 
-// A Request ID. Should be unique for in-flight requests on a stream. Included in the response.
-message RequestID {
-  // The base request ID.
-  uint64 id = 1;
-  // The request attempt. Starts at 0, incremented on each retry.
-  uint32 attempt = 2;
-}
-
 // A GetPageRequest class. Primarily intended for observability, but may also be
 // used for prioritization in the future.
 enum GetPageClass {
@@ -207,26 +199,13 @@ enum GetPageClass {
 // the entire batch is ready, so no one can make use of the individual pages.
 message GetPageResponse {
   // The original request's ID.
-  RequestID request_id = 1;
-  // The response status code. If not OK, the rel and page fields will be empty.
+  uint64 request_id = 1;
+  // The response status code.
   GetPageStatusCode status_code = 2;
   // A string describing the status, if any.
   string reason = 3;
-  // The relation that the pages belong to.
-  RelTag rel = 4;
-  // The page(s), in the same order as the request.
-  repeated Page page = 5;
-}
-
-// A page.
-//
-// TODO: it would be slightly more efficient (but less convenient) to have separate arrays of block
-// numbers and images, but given the 8KB page size it's probably negligible. Benchmark it anyway.
-message Page {
-  // The page number.
-  uint32 block_number = 1;
-  // The materialized page image, as an 8KB byte vector.
-  bytes image = 2;
+  // The 8KB page images, in the same order as the request. Empty if status_code != OK.
+  repeated bytes page_image = 4;
 }
 
 // A GetPageResponse status code.

pageserver/page_api/src/client.rs

@@ -1,152 +1,23 @@
-use anyhow::Context as _;
-use futures::future::ready;
+use anyhow::Result;
 use futures::{Stream, StreamExt as _, TryStreamExt as _};
 use tokio::io::AsyncRead;
 use tokio_util::io::StreamReader;
-use tonic::codec::CompressionEncoding;
 use tonic::metadata::AsciiMetadataValue;
-use tonic::service::Interceptor;
-use tonic::service::interceptor::InterceptedService;
-use tonic::transport::{Channel, Endpoint};
+use tonic::metadata::errors::InvalidMetadataValue;
+use tonic::transport::Channel;
+use tonic::{Request, Streaming};
 
-use utils::id::{TenantId, TimelineId};
+use utils::id::TenantId;
+use utils::id::TimelineId;
 use utils::shard::ShardIndex;
 
-use crate::model::*;
+use crate::model;
 use crate::proto;
 
-/// A basic Pageserver gRPC client, for a single tenant shard. This API uses native Rust domain
-/// types from `model` rather than generated Protobuf types.
-pub struct Client {
-    inner: proto::PageServiceClient<InterceptedService<Channel, AuthInterceptor>>,
-}
-
-impl Client {
-    /// Connects to the given gRPC endpoint.
-    pub async fn connect<E>(
-        endpoint: E,
-        tenant_id: TenantId,
-        timeline_id: TimelineId,
-        shard_id: ShardIndex,
-        auth_token: Option<String>,
-        compression: Option<CompressionEncoding>,
-    ) -> anyhow::Result<Self>
-    where
-        E: TryInto<Endpoint> + Send + Sync + 'static,
-        <E as TryInto<Endpoint>>::Error: std::error::Error + Send + Sync,
-    {
-        let endpoint: Endpoint = endpoint.try_into().context("invalid endpoint")?;
-        let channel = endpoint.connect().await?;
-        Self::new(
-            channel,
-            tenant_id,
-            timeline_id,
-            shard_id,
-            auth_token,
-            compression,
-        )
-    }
-
-    /// Creates a new client using the given gRPC channel.
-    pub fn new(
-        channel: Channel,
-        tenant_id: TenantId,
-        timeline_id: TimelineId,
-        shard_id: ShardIndex,
-        auth_token: Option<String>,
-        compression: Option<CompressionEncoding>,
-    ) -> anyhow::Result<Self> {
-        let auth = AuthInterceptor::new(tenant_id, timeline_id, shard_id, auth_token)?;
-        let mut inner = proto::PageServiceClient::with_interceptor(channel, auth);
-
-        if let Some(compression) = compression {
-            // TODO: benchmark this (including network latency).
-            inner = inner
-                .accept_compressed(compression)
-                .send_compressed(compression);
-        }
-
-        Ok(Self { inner })
-    }
-
-    /// Returns whether a relation exists.
-    pub async fn check_rel_exists(
-        &mut self,
-        req: CheckRelExistsRequest,
-    ) -> tonic::Result<CheckRelExistsResponse> {
-        let req = proto::CheckRelExistsRequest::from(req);
-        let resp = self.inner.check_rel_exists(req).await?.into_inner();
-        Ok(resp.into())
-    }
-
-    /// Fetches a base backup.
-    pub async fn get_base_backup(
-        &mut self,
-        req: GetBaseBackupRequest,
-    ) -> tonic::Result<impl AsyncRead + use<>> {
-        let req = proto::GetBaseBackupRequest::from(req);
-        let chunks = self.inner.get_base_backup(req).await?.into_inner();
-        Ok(StreamReader::new(
-            chunks
-                .map_ok(|resp| resp.chunk)
-                .map_err(std::io::Error::other),
-        ))
-    }
-
-    /// Returns the total size of a database, as # of bytes.
-    pub async fn get_db_size(&mut self, req: GetDbSizeRequest) -> tonic::Result<GetDbSizeResponse> {
-        let req = proto::GetDbSizeRequest::from(req);
-        let resp = self.inner.get_db_size(req).await?.into_inner();
-        Ok(resp.into())
-    }
-
-    /// Fetches pages.
-    ///
-    /// This is implemented as a bidirectional streaming RPC for performance. Per-request errors are
-    /// typically returned as status_code instead of errors, to avoid tearing down the entire stream
-    /// via a tonic::Status error.
-    pub async fn get_pages(
-        &mut self,
-        reqs: impl Stream<Item = GetPageRequest> + Send + 'static,
-    ) -> tonic::Result<impl Stream<Item = tonic::Result<GetPageResponse>> + Send + 'static> {
-        let reqs = reqs.map(proto::GetPageRequest::from);
-        let resps = self.inner.get_pages(reqs).await?.into_inner();
-        Ok(resps.and_then(|resp| ready(GetPageResponse::try_from(resp).map_err(|err| err.into()))))
-    }
-
-    /// Returns the size of a relation, as # of blocks.
-    pub async fn get_rel_size(
-        &mut self,
-        req: GetRelSizeRequest,
-    ) -> tonic::Result<GetRelSizeResponse> {
-        let req = proto::GetRelSizeRequest::from(req);
-        let resp = self.inner.get_rel_size(req).await?.into_inner();
-        Ok(resp.into())
-    }
-
-    /// Fetches an SLRU segment.
-    pub async fn get_slru_segment(
-        &mut self,
-        req: GetSlruSegmentRequest,
-    ) -> tonic::Result<GetSlruSegmentResponse> {
-        let req = proto::GetSlruSegmentRequest::from(req);
-        let resp = self.inner.get_slru_segment(req).await?.into_inner();
-        Ok(resp.try_into()?)
-    }
-
-    /// Acquires or extends a lease on the given LSN. This guarantees that the Pageserver won't
-    /// garbage collect the LSN until the lease expires. Must be acquired on all relevant shards.
-    ///
-    /// Returns the lease expiration time, or a FailedPrecondition status if the lease could not be
-    /// acquired because the LSN has already been garbage collected.
-    pub async fn lease_lsn(&mut self, req: LeaseLsnRequest) -> tonic::Result<LeaseLsnResponse> {
-        let req = proto::LeaseLsnRequest::from(req);
-        let resp = self.inner.lease_lsn(req).await?.into_inner();
-        Ok(resp.try_into()?)
-    }
-}
-
-/// Adds authentication metadata to gRPC requests.
+///
+/// AuthInterceptor adds tenant, timeline, and auth header to the channel. These
+/// headers are required at the pageserver.
+///
 #[derive(Clone)]
 struct AuthInterceptor {
     tenant_id: AsciiMetadataValue,
@@ -159,29 +30,174 @@ impl AuthInterceptor {
     fn new(
         tenant_id: TenantId,
         timeline_id: TimelineId,
-        shard_id: ShardIndex,
         auth_token: Option<String>,
-    ) -> anyhow::Result<Self> {
+        shard_id: ShardIndex,
+    ) -> Result<Self, InvalidMetadataValue> {
+        let tenant_ascii: AsciiMetadataValue = tenant_id.to_string().try_into()?;
+        let timeline_ascii: AsciiMetadataValue = timeline_id.to_string().try_into()?;
+        let shard_ascii: AsciiMetadataValue = shard_id.to_string().try_into()?;
+
+        let auth_header: Option<AsciiMetadataValue> = match auth_token {
+            Some(token) => Some(format!("Bearer {token}").try_into()?),
+            None => None,
+        };
+
         Ok(Self {
-            tenant_id: tenant_id.to_string().try_into()?,
-            timeline_id: timeline_id.to_string().try_into()?,
-            shard_id: shard_id.to_string().try_into()?,
-            auth_header: auth_token
-                .map(|token| format!("Bearer {token}").try_into())
-                .transpose()?,
+            tenant_id: tenant_ascii,
+            shard_id: shard_ascii,
+            timeline_id: timeline_ascii,
+            auth_header,
         })
     }
 }
 
-impl Interceptor for AuthInterceptor {
-    fn call(&mut self, mut req: tonic::Request<()>) -> tonic::Result<tonic::Request<()>> {
-        let metadata = req.metadata_mut();
-        metadata.insert("neon-tenant-id", self.tenant_id.clone());
-        metadata.insert("neon-timeline-id", self.timeline_id.clone());
-        metadata.insert("neon-shard-id", self.shard_id.clone());
-        if let Some(ref auth_header) = self.auth_header {
-            metadata.insert("authorization", auth_header.clone());
+impl tonic::service::Interceptor for AuthInterceptor {
+    fn call(&mut self, mut req: tonic::Request<()>) -> Result<tonic::Request<()>, tonic::Status> {
+        req.metadata_mut()
+            .insert("neon-tenant-id", self.tenant_id.clone());
+        req.metadata_mut()
+            .insert("neon-shard-id", self.shard_id.clone());
+        req.metadata_mut()
+            .insert("neon-timeline-id", self.timeline_id.clone());
+        if let Some(auth_header) = &self.auth_header {
+            req.metadata_mut()
+                .insert("authorization", auth_header.clone());
         }
         Ok(req)
     }
 }
+
+#[derive(Clone)]
+pub struct Client {
+    client: proto::PageServiceClient<
+        tonic::service::interceptor::InterceptedService<Channel, AuthInterceptor>,
+    >,
+}
+
+impl Client {
+    pub async fn new<T: TryInto<tonic::transport::Endpoint> + Send + Sync + 'static>(
+        into_endpoint: T,
+        tenant_id: TenantId,
+        timeline_id: TimelineId,
+        shard_id: ShardIndex,
+        auth_header: Option<String>,
+        compression: Option<tonic::codec::CompressionEncoding>,
+    ) -> anyhow::Result<Self> {
+        let endpoint: tonic::transport::Endpoint = into_endpoint
+            .try_into()
+            .map_err(|_e| anyhow::anyhow!("failed to convert endpoint"))?;
+        let channel = endpoint.connect().await?;
+        let auth = AuthInterceptor::new(tenant_id, timeline_id, auth_header, shard_id)
+            .map_err(|e| anyhow::anyhow!(e.to_string()))?;
+        let mut client = proto::PageServiceClient::with_interceptor(channel, auth);
+
+        if let Some(compression) = compression {
+            // TODO: benchmark this (including network latency).
+            client = client
+                .accept_compressed(compression)
+                .send_compressed(compression);
+        }
+
+        Ok(Self { client })
+    }
+
+    /// Returns whether a relation exists.
+    pub async fn check_rel_exists(
+        &mut self,
+        req: model::CheckRelExistsRequest,
+    ) -> Result<model::CheckRelExistsResponse, tonic::Status> {
+        let proto_req = proto::CheckRelExistsRequest::from(req);
+
+        let response = self.client.check_rel_exists(proto_req).await?;
+
+        let proto_resp = response.into_inner();
+        Ok(proto_resp.into())
+    }
+
+    /// Fetches a base backup.
+    pub async fn get_base_backup(
+        &mut self,
+        req: model::GetBaseBackupRequest,
+    ) -> Result<impl AsyncRead + use<>, tonic::Status> {
+        let req = proto::GetBaseBackupRequest::from(req);
+        let chunks = self.client.get_base_backup(req).await?.into_inner();
+        let reader = StreamReader::new(
+            chunks
+                .map_ok(|resp| resp.chunk)
+                .map_err(std::io::Error::other),
+        );
+        Ok(reader)
+    }
+
+    /// Returns the total size of a database, as # of bytes.
+    pub async fn get_db_size(
+        &mut self,
+        req: model::GetDbSizeRequest,
+    ) -> Result<u64, tonic::Status> {
+        let proto_req = proto::GetDbSizeRequest::from(req);
+
+        let response = self.client.get_db_size(proto_req).await?;
+        Ok(response.into_inner().into())
+    }
+
+    /// Fetches pages.
+    ///
+    /// This is implemented as a bidirectional streaming RPC for performance.
+    /// Per-request errors are often returned as status_code instead of errors,
+    /// to avoid tearing down the entire stream via tonic::Status.
+    pub async fn get_pages<ReqSt>(
+        &mut self,
+        inbound: ReqSt,
+    ) -> Result<
+        impl Stream<Item = Result<model::GetPageResponse, tonic::Status>> + Send + 'static,
+        tonic::Status,
+    >
+    where
+        ReqSt: Stream<Item = model::GetPageRequest> + Send + 'static,
+    {
+        let outbound_proto = inbound.map(|domain_req| domain_req.into());
+
+        let req_new = Request::new(outbound_proto);
+
+        let response_stream: Streaming<proto::GetPageResponse> =
+            self.client.get_pages(req_new).await?.into_inner();
+
+        let domain_stream = response_stream.map_ok(model::GetPageResponse::from);
+
+        Ok(domain_stream)
+    }
+
+    /// Returns the size of a relation, as # of blocks.
+    pub async fn get_rel_size(
+        &mut self,
+        req: model::GetRelSizeRequest,
+    ) -> Result<model::GetRelSizeResponse, tonic::Status> {
+        let proto_req = proto::GetRelSizeRequest::from(req);
+        let response = self.client.get_rel_size(proto_req).await?;
+        let proto_resp = response.into_inner();
+        Ok(proto_resp.into())
+    }
+
+    /// Fetches an SLRU segment.
+    pub async fn get_slru_segment(
+        &mut self,
+        req: model::GetSlruSegmentRequest,
+    ) -> Result<model::GetSlruSegmentResponse, tonic::Status> {
+        let proto_req = proto::GetSlruSegmentRequest::from(req);
+        let response = self.client.get_slru_segment(proto_req).await?;
+        Ok(response.into_inner().try_into()?)
+    }
+
+    /// Acquires or extends a lease on the given LSN. This guarantees that the Pageserver won't
+    /// garbage collect the LSN until the lease expires. Must be acquired on all relevant shards.
+    ///
+    /// Returns the lease expiration time, or a FailedPrecondition status if the lease could not be
+    /// acquired because the LSN has already been garbage collected.
+    pub async fn lease_lsn(
+        &mut self,
+        req: model::LeaseLsnRequest,
+    ) -> Result<model::LeaseLsnResponse, tonic::Status> {
+        let req = proto::LeaseLsnRequest::from(req);
+        Ok(self.client.lease_lsn(req).await?.into_inner().try_into()?)
+    }
+}

pageserver/page_api/src/model.rs

@@ -33,8 +33,6 @@ pub enum ProtocolError {
     Invalid(&'static str, String),
     #[error("required field '{0}' is missing")]
     Missing(&'static str),
-    #[error("invalid combination of not_modified_lsn '{0}' and request_lsn '{1}'")]
-    InvalidLsns(Lsn, Lsn),
 }
 
 impl ProtocolError {
@@ -87,9 +85,9 @@ impl TryFrom<proto::ReadLsn> for ReadLsn {
             return Err(ProtocolError::invalid("request_lsn", pb.request_lsn));
         }
         if pb.not_modified_since_lsn > pb.request_lsn {
-            return Err(ProtocolError::InvalidLsns(
-                Lsn(pb.not_modified_since_lsn),
-                Lsn(pb.request_lsn),
+            return Err(ProtocolError::invalid(
+                "not_modified_since_lsn",
+                pb.not_modified_since_lsn,
             ));
         }
         Ok(Self {
@@ -358,10 +356,7 @@ impl TryFrom<proto::GetPageRequest> for GetPageRequest {
             return Err(ProtocolError::Missing("block_number"));
         }
         Ok(Self {
-            request_id: pb
-                .request_id
-                .ok_or(ProtocolError::Missing("request_id"))?
-                .into(),
+            request_id: pb.request_id,
             request_class: pb.request_class.into(),
             read_lsn: pb
                 .read_lsn
@@ -376,7 +371,7 @@ impl TryFrom<proto::GetPageRequest> for GetPageRequest {
 impl From<GetPageRequest> for proto::GetPageRequest {
     fn from(request: GetPageRequest) -> Self {
         Self {
-            request_id: Some(request.request_id.into()),
+            request_id: request.request_id,
             request_class: request.request_class.into(),
             read_lsn: Some(request.read_lsn.into()),
             rel: Some(request.rel.into()),
@@ -385,54 +380,11 @@ impl From<GetPageRequest> for proto::GetPageRequest {
     }
 }
 
-/// A GetPage request ID and retry attempt. Should be unique for in-flight requests on a stream.
-#[derive(Clone, Copy, Debug, Default, PartialEq, Eq, Hash, PartialOrd, Ord)]
-pub struct RequestID {
-    /// The base request ID.
-    pub id: u64,
-    // The request attempt. Starts at 0, incremented on each retry.
-    pub attempt: u32,
-}
-
-impl RequestID {
-    /// Creates a new RequestID with the given ID and an initial attempt of 0.
-    pub fn new(id: u64) -> Self {
-        Self { id, attempt: 0 }
-    }
-}
-
-impl Display for RequestID {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        write!(f, "{}.{}", self.id, self.attempt)
-    }
-}
-
-impl From<proto::RequestId> for RequestID {
-    fn from(pb: proto::RequestId) -> Self {
-        Self {
-            id: pb.id,
-            attempt: pb.attempt,
-        }
-    }
-}
-
-impl From<u64> for RequestID {
-    fn from(id: u64) -> Self {
-        Self::new(id)
-    }
-}
-
-impl From<RequestID> for proto::RequestId {
-    fn from(request_id: RequestID) -> Self {
-        Self {
-            id: request_id.id,
-            attempt: request_id.attempt,
-        }
-    }
-}
+/// A GetPage request ID.
+pub type RequestID = u64;
 
 /// A GetPage request class.
-#[derive(Clone, Copy, Debug, strum_macros::Display)]
+#[derive(Clone, Copy, Debug)]
 pub enum GetPageClass {
     /// Unknown class. For backwards compatibility: used when an older client version sends a class
     /// that a newer server version has removed.
@@ -445,19 +397,6 @@ pub enum GetPageClass {
     Background,
 }
 
-impl GetPageClass {
-    /// Returns true if this is considered a bulk request (i.e. more throughput-oriented rather than
-    /// latency-sensitive).
-    pub fn is_bulk(&self) -> bool {
-        match self {
-            Self::Unknown => false,
-            Self::Normal => false,
-            Self::Prefetch => true,
-            Self::Background => true,
-        }
-    }
-}
-
 impl From<proto::GetPageClass> for GetPageClass {
     fn from(pb: proto::GetPageClass) -> Self {
         match pb {
@@ -504,41 +443,32 @@ impl From<GetPageClass> for i32 {
 pub struct GetPageResponse {
     /// The original request's ID.
     pub request_id: RequestID,
-    /// The response status code. If not OK, the `rel` and `pages` fields will be empty.
+    /// The response status code.
     pub status_code: GetPageStatusCode,
     /// A string describing the status, if any.
     pub reason: Option<String>,
-    /// The relation that the pages belong to.
-    pub rel: RelTag,
-    // The page(s), in the same order as the request.
-    pub pages: Vec<Page>,
+    /// The 8KB page images, in the same order as the request. Empty if status != OK.
+    pub page_images: Vec<Bytes>,
 }
 
-impl TryFrom<proto::GetPageResponse> for GetPageResponse {
-    type Error = ProtocolError;
-
-    fn try_from(pb: proto::GetPageResponse) -> Result<Self, ProtocolError> {
-        Ok(Self {
-            request_id: pb
-                .request_id
-                .ok_or(ProtocolError::Missing("request_id"))?
-                .into(),
+impl From<proto::GetPageResponse> for GetPageResponse {
+    fn from(pb: proto::GetPageResponse) -> Self {
+        Self {
+            request_id: pb.request_id,
             status_code: pb.status_code.into(),
             reason: Some(pb.reason).filter(|r| !r.is_empty()),
-            rel: pb.rel.ok_or(ProtocolError::Missing("rel"))?.try_into()?,
-            pages: pb.page.into_iter().map(Page::from).collect(),
-        })
+            page_images: pb.page_image,
+        }
     }
 }
 
 impl From<GetPageResponse> for proto::GetPageResponse {
     fn from(response: GetPageResponse) -> Self {
         Self {
-            request_id: Some(response.request_id.into()),
+            request_id: response.request_id,
             status_code: response.status_code.into(),
             reason: response.reason.unwrap_or_default(),
-            rel: Some(response.rel.into()),
-            page: response.pages.into_iter().map(proto::Page::from).collect(),
+            page_image: response.page_images,
         }
     }
 }
@@ -571,39 +501,11 @@ impl GetPageResponse {
             request_id,
             status_code,
             reason: Some(status.message().to_string()),
-            rel: RelTag::default(),
-            pages: Vec::new(),
+            page_images: Vec::new(),
         })
     }
 }
 
-// A page.
-#[derive(Clone, Debug)]
-pub struct Page {
-    /// The page number.
-    pub block_number: u32,
-    /// The materialized page image, as an 8KB byte vector.
-    pub image: Bytes,
-}
-
-impl From<proto::Page> for Page {
-    fn from(pb: proto::Page) -> Self {
-        Self {
-            block_number: pb.block_number,
-            image: pb.image,
-        }
-    }
-}
-
-impl From<Page> for proto::Page {
-    fn from(page: Page) -> Self {
-        Self {
-            block_number: page.block_number,
-            image: page.image,
-        }
-    }
-}
-
 /// A GetPage response status code.
 ///
 /// These are effectively equivalent to gRPC statuses. However, we use a bidirectional stream
@@ -700,21 +602,6 @@ impl TryFrom<tonic::Code> for GetPageStatusCode {
     }
 }
 
-impl From<GetPageStatusCode> for tonic::Code {
-    fn from(status_code: GetPageStatusCode) -> Self {
-        use tonic::Code;
-
-        match status_code {
-            GetPageStatusCode::Unknown => Code::Unknown,
-            GetPageStatusCode::Ok => Code::Ok,
-            GetPageStatusCode::NotFound => Code::NotFound,
-            GetPageStatusCode::InvalidRequest => Code::InvalidArgument,
-            GetPageStatusCode::InternalError => Code::Internal,
-            GetPageStatusCode::SlowDown => Code::ResourceExhausted,
-        }
-    }
-}
-
 // Fetches the size of a relation at a given LSN, as # of blocks. Only valid on shard 0, other
 // shards will error.
 #[derive(Clone, Copy, Debug)]

pageserver/pagebench/Cargo.toml

@@ -24,15 +24,11 @@ tracing.workspace = true
 tokio.workspace = true
 tokio-stream.workspace = true
 tokio-util.workspace = true
-axum.workspace = true
-http.workspace = true
-metrics.workspace = true
 tonic.workspace = true
 url.workspace = true
 
-pageserver_api.workspace = true
 pageserver_client.workspace = true
-pageserver_client_grpc.workspace = true
+pageserver_api.workspace = true
 pageserver_page_api.workspace = true
 utils = { path = "../../libs/utils/" }
 workspace_hack = { version = "0.1", path = "../../workspace_hack" }

pageserver/pagebench/src/cmd/basebackup.rs

@@ -326,7 +326,7 @@ impl GrpcClient {
         ttid: TenantTimelineId,
         compression: bool,
     ) -> anyhow::Result<Self> {
-        let inner = page_api::Client::connect(
+        let inner = page_api::Client::new(
             connstring.to_string(),
             ttid.tenant_id,
             ttid.timeline_id,

pageserver/pagebench/src/cmd/getpage_latest_lsn.rs

@@ -10,14 +10,12 @@ use anyhow::Context;
 use async_trait::async_trait;
 use bytes::Bytes;
 use camino::Utf8PathBuf;
-use futures::stream::FuturesUnordered;
 use futures::{Stream, StreamExt as _};
 use pageserver_api::key::Key;
 use pageserver_api::keyspace::KeySpaceAccum;
 use pageserver_api::pagestream_api::{PagestreamGetPageRequest, PagestreamRequest};
 use pageserver_api::reltag::RelTag;
 use pageserver_api::shard::TenantShardId;
-use pageserver_client_grpc::{self as client_grpc, ShardSpec};
 use pageserver_page_api as page_api;
 use rand::prelude::*;
 use tokio::task::JoinSet;
@@ -34,19 +32,11 @@ use crate::util::{request_stats, tokio_thread_local_stats};
 /// GetPage@LatestLSN, uniformly distributed across the compute-accessible keyspace.
 #[derive(clap::Parser)]
 pub(crate) struct Args {
-    #[clap(long, default_value = "false")]
-    grpc: bool,
-    #[clap(long, default_value = "false")]
-    grpc_stream: bool,
     #[clap(long, default_value = "http://localhost:9898")]
     mgmt_api_endpoint: String,
     /// Pageserver connection string. Supports postgresql:// and grpc:// protocols.
     #[clap(long, default_value = "postgres://postgres@localhost:64000")]
     page_service_connstring: String,
-    /// Use the rich gRPC Pageserver client `client_grpc::PageserverClient`, rather than the basic
-    /// no-frills `page_api::Client`. Only valid with grpc:// connstrings.
-    #[clap(long)]
-    rich_client: bool,
     #[clap(long)]
     pageserver_jwt: Option<String>,
     #[clap(long, default_value = "1")]
@@ -82,9 +72,6 @@ pub(crate) struct Args {
     #[clap(long)]
     set_io_mode: Option<pageserver_api::models::virtual_file::IoMode>,
 
-    #[clap(long)]
-    only_relnode: Option<u32>,
-
     /// Queue depth generated in each client.
     #[clap(long, default_value = "1")]
     queue_depth: NonZeroUsize,
@@ -99,31 +86,10 @@ pub(crate) struct Args {
     #[clap(long, default_value = "1")]
     batch_size: NonZeroUsize,
 
+    #[clap(long)]
+    only_relnode: Option<u32>,
+
     targets: Option<Vec<TenantTimelineId>>,
-
-    #[clap(long, default_value = "100")]
-    pool_max_consumers: NonZeroUsize,
-
-    #[clap(long, default_value = "5")]
-    pool_error_threshold: NonZeroUsize,
-
-    #[clap(long, default_value = "5000")]
-    pool_connect_timeout: NonZeroUsize,
-
-    #[clap(long, default_value = "1000")]
-    pool_connect_backoff: NonZeroUsize,
-
-    #[clap(long, default_value = "60000")]
-    pool_max_idle_duration: NonZeroUsize,
-
-    #[clap(long, default_value = "0")]
-    max_delay_ms: usize,
-
-    #[clap(long, default_value = "0")]
-    percent_drops: usize,
-
-    #[clap(long, default_value = "0")]
-    percent_hangs: usize,
 }
 
 /// State shared by all clients
@@ -180,6 +146,7 @@ pub(crate) fn main(args: Args) -> anyhow::Result<()> {
         main_impl(args, thread_local_stats)
     })
 }
+
 async fn main_impl(
     args: Args,
     all_thread_local_stats: AllThreadLocalStats<request_stats::Stats>,
@@ -344,7 +311,6 @@ async fn main_impl(
     let rps_period = args
         .per_client_rate
         .map(|rps_limit| Duration::from_secs_f64(1.0 / (rps_limit as f64)));
-
     let make_worker: &dyn Fn(WorkerId) -> Pin<Box<dyn Send + Future<Output = ()>>> = &|worker_id| {
         let ss = shared_state.clone();
         let cancel = cancel.clone();
@@ -366,7 +332,6 @@ async fn main_impl(
             let client: Box<dyn Client> = match scheme.as_str() {
                 "postgresql" | "postgres" => {
                     assert!(!args.compression, "libpq does not support compression");
-                    assert!(!args.rich_client, "rich client requires grpc://");
                     Box::new(
                         LibpqClient::new(&args.page_service_connstring, worker_id.timeline)
                             .await
@@ -374,16 +339,6 @@ async fn main_impl(
                     )
                 }
 
-                "grpc" if args.rich_client => Box::new(
-                    RichGrpcClient::new(
-                        &args.page_service_connstring,
-                        worker_id.timeline,
-                        args.compression,
-                    )
-                    .await
-                    .unwrap(),
-                ),
-
                 "grpc" => Box::new(
                     GrpcClient::new(
                         &args.page_service_connstring,
@@ -670,7 +625,7 @@ impl GrpcClient {
         ttid: TenantTimelineId,
         compression: bool,
     ) -> anyhow::Result<Self> {
-        let mut client = page_api::Client::connect(
+        let mut client = page_api::Client::new(
             connstring.to_string(),
             ttid.tenant_id,
             ttid.timeline_id,
@@ -702,7 +657,7 @@ impl Client for GrpcClient {
         blks: Vec<u32>,
     ) -> anyhow::Result<()> {
         let req = page_api::GetPageRequest {
-            request_id: req_id.into(),
+            request_id: req_id,
             request_class: page_api::GetPageClass::Normal,
             read_lsn: page_api::ReadLsn {
                 request_lsn: req_lsn,
@@ -722,79 +677,6 @@ impl Client for GrpcClient {
             "unexpected status code: {}",
             resp.status_code,
         );
-        Ok((
-            resp.request_id.id,
-            resp.pages.into_iter().map(|p| p.image).collect(),
-        ))
-    }
-}
-
-/// A rich gRPC Pageserver client.
-struct RichGrpcClient {
-    inner: Arc<client_grpc::PageserverClient>,
-    requests: FuturesUnordered<
-        Pin<Box<dyn Future<Output = anyhow::Result<page_api::GetPageResponse>> + Send>>,
-    >,
-}
-
-impl RichGrpcClient {
-    async fn new(
-        connstring: &str,
-        ttid: TenantTimelineId,
-        compression: bool,
-    ) -> anyhow::Result<Self> {
-        let inner = Arc::new(client_grpc::PageserverClient::new(
-            ttid.tenant_id,
-            ttid.timeline_id,
-            ShardSpec::new(
-                [(ShardIndex::unsharded(), connstring.to_string())].into(),
-                None,
-            )?,
-            None,
-            compression.then_some(tonic::codec::CompressionEncoding::Zstd),
-        )?);
-        Ok(Self {
-            inner,
-            requests: FuturesUnordered::new(),
-        })
-    }
-}
-
-#[async_trait]
-impl Client for RichGrpcClient {
-    async fn send_get_page(
-        &mut self,
-        req_id: u64,
-        req_lsn: Lsn,
-        mod_lsn: Lsn,
-        rel: RelTag,
-        blks: Vec<u32>,
-    ) -> anyhow::Result<()> {
-        let req = page_api::GetPageRequest {
-            request_id: req_id.into(),
-            request_class: page_api::GetPageClass::Normal,
-            read_lsn: page_api::ReadLsn {
-                request_lsn: req_lsn,
-                not_modified_since_lsn: Some(mod_lsn),
-            },
-            rel,
-            block_numbers: blks,
-        };
-        let inner = self.inner.clone();
-        self.requests.push(Box::pin(async move {
-            inner
-                .get_page(req)
-                .await
-                .map_err(|err| anyhow::anyhow!("{err}"))
-        }));
-        Ok(())
-    }
-
-    async fn recv_get_page(&mut self) -> anyhow::Result<(u64, Vec<Bytes>)> {
-        let resp = self.requests.next().await.unwrap()?;
-        Ok((
-            resp.request_id.id,
-            resp.pages.into_iter().map(|p| p.image).collect(),
-        ))
+        Ok((resp.request_id, resp.page_images))
     }
 }

pageserver/src/bin/pageserver.rs

@@ -29,8 +29,8 @@ use pageserver::task_mgr::{
 };
 use pageserver::tenant::{TenantSharedResources, mgr, secondary};
 use pageserver::{
-    CancellableTask, ConsumptionMetricsTasks, HttpEndpointListener, HttpsEndpointListener,
-    MetricsCollectionTask, http, page_cache, page_service, task_mgr, virtual_file,
+    CancellableTask, ConsumptionMetricsTasks, HttpEndpointListener, HttpsEndpointListener, http,
+    page_cache, page_service, task_mgr, virtual_file,
 };
 use postgres_backend::AuthType;
 use remote_storage::GenericRemoteStorage;
@@ -41,7 +41,6 @@ use tracing_utils::OtelGuard;
 use utils::auth::{JwtAuth, SwappableJwtAuth};
 use utils::crashsafe::syncfs;
 use utils::logging::TracingErrorLayerEnablement;
-use utils::metrics_collector::{METRICS_COLLECTION_INTERVAL, METRICS_COLLECTOR};
 use utils::sentry_init::init_sentry;
 use utils::{failpoint_support, logging, project_build_tag, project_git_version, tcp_listener};
 
@@ -764,41 +763,6 @@ fn start_pageserver(
         (http_task, https_task)
     };
 
-    /* BEGIN_HADRON */
-    let metrics_collection_task = {
-        let cancel = shutdown_pageserver.child_token();
-        let task = crate::BACKGROUND_RUNTIME.spawn({
-            let cancel = cancel.clone();
-            let background_jobs_barrier = background_jobs_barrier.clone();
-            async move {
-                if conf.force_metric_collection_on_scrape {
-                    return;
-                }
-
-                // first wait until background jobs are cleared to launch.
-                tokio::select! {
-                    _ = cancel.cancelled() => { return; },
-                    _ = background_jobs_barrier.wait() => {}
-                };
-                let mut interval = tokio::time::interval(METRICS_COLLECTION_INTERVAL);
-                loop {
-                    tokio::select! {
-                        _ = cancel.cancelled() => {
-                            tracing::info!("cancelled metrics collection task, exiting...");
-                             break;
-                        },
-                        _ = interval.tick() => {}
-                    }
-                    tokio::task::spawn_blocking(|| {
-                        METRICS_COLLECTOR.run_once(true);
-                    });
-                }
-            }
-        });
-        MetricsCollectionTask(CancellableTask { task, cancel })
-    };
-    /* END_HADRON */
-
     let consumption_metrics_tasks = {
         let cancel = shutdown_pageserver.child_token();
         let task = crate::BACKGROUND_RUNTIME.spawn({
@@ -880,7 +844,6 @@ fn start_pageserver(
             https_endpoint_listener,
             page_service,
             page_service_grpc,
-            metrics_collection_task,
             consumption_metrics_tasks,
             disk_usage_eviction_task,
             &tenant_manager,
@@ -926,11 +889,8 @@ async fn create_remote_storage_client(
             "Simulating remote failures for first {} attempts of each op",
             conf.test_remote_failures
         );
-        remote_storage = GenericRemoteStorage::unreliable_wrapper(
-            remote_storage,
-            conf.test_remote_failures,
-            conf.test_remote_failures_probability,
-        );
+        remote_storage =
+            GenericRemoteStorage::unreliable_wrapper(remote_storage, conf.test_remote_failures);
     }
 
     Ok(remote_storage)

pageserver/src/config.rs

@@ -28,6 +28,7 @@ use reqwest::Url;
 use storage_broker::Uri;
 use utils::id::{NodeId, TimelineId};
 use utils::logging::{LogFormat, SecretString};
+use utils::serde_percent::Percent;
 
 use crate::tenant::storage_layer::inmemory_layer::IndexEntry;
 use crate::tenant::{TENANTS_SEGMENT_NAME, TIMELINES_SEGMENT_NAME};
@@ -145,13 +146,9 @@ pub struct PageServerConf {
     pub metric_collection_bucket: Option<RemoteStorageConfig>,
     pub synthetic_size_calculation_interval: Duration,
 
-    pub disk_usage_based_eviction: DiskUsageEvictionTaskConfig,
+    pub disk_usage_based_eviction: Option<DiskUsageEvictionTaskConfig>,
 
-    // The number of allowed failures in remote storage operations.
     pub test_remote_failures: u64,
-    // The probability of failure in remote storage operations. Only works when test_remote_failures > 1.
-    // Use 100 for 100% failure, 0 for no failure.
-    pub test_remote_failures_probability: u64,
 
     pub ondemand_download_behavior_treat_error_as_warn: bool,
 
@@ -252,14 +249,6 @@ pub struct PageServerConf {
     pub timeline_import_config: pageserver_api::config::TimelineImportConfig,
 
     pub basebackup_cache_config: Option<pageserver_api::config::BasebackupCacheConfig>,
-
-    /// Defines what is a big tenant for the purpose of image layer generation.
-    /// See Timeline::should_check_if_image_layers_required
-    pub image_layer_generation_large_timeline_threshold: Option<u64>,
-
-    /// Controls whether to collect all metrics on each scrape or to return potentially stale
-    /// results.
-    pub force_metric_collection_on_scrape: bool,
 }
 
 /// Token for authentication to safekeepers
@@ -404,7 +393,6 @@ impl PageServerConf {
             synthetic_size_calculation_interval,
             disk_usage_based_eviction,
             test_remote_failures,
-            test_remote_failures_probability,
             ondemand_download_behavior_treat_error_as_warn,
             background_task_maximum_delay,
             control_plane_api,
@@ -440,8 +428,6 @@ impl PageServerConf {
             posthog_config,
             timeline_import_config,
             basebackup_cache_config,
-            image_layer_generation_large_timeline_threshold,
-            force_metric_collection_on_scrape,
         } = config_toml;
 
         let mut conf = PageServerConf {
@@ -474,9 +460,17 @@ impl PageServerConf {
             metric_collection_endpoint,
             metric_collection_bucket,
             synthetic_size_calculation_interval,
-            disk_usage_based_eviction,
+            disk_usage_based_eviction: Some(disk_usage_based_eviction.unwrap_or(
+                DiskUsageEvictionTaskConfig {
+                    max_usage_pct: Percent::new(80).unwrap(),
+                    min_avail_bytes: 2_000_000_000,
+                    period: Duration::from_secs(60),
+                    #[cfg(feature = "testing")]
+                    mock_statvfs: None,
+                    eviction_order: Default::default(),
+                },
+            )),
             test_remote_failures,
-            test_remote_failures_probability,
             ondemand_download_behavior_treat_error_as_warn,
             background_task_maximum_delay,
             control_plane_api: control_plane_api
@@ -500,8 +494,6 @@ impl PageServerConf {
             dev_mode,
             timeline_import_config,
             basebackup_cache_config,
-            image_layer_generation_large_timeline_threshold,
-            force_metric_collection_on_scrape,
 
             // ------------------------------------------------------------
             // fields that require additional validation or custom handling
@@ -643,7 +635,7 @@ impl PageServerConf {
     pub fn dummy_conf(repo_dir: Utf8PathBuf) -> Self {
         let pg_distrib_dir = Utf8PathBuf::from(env!("CARGO_MANIFEST_DIR")).join("../pg_install");
 
-        let mut config_toml = pageserver_api::config::ConfigToml {
+        let config_toml = pageserver_api::config::ConfigToml {
             wait_lsn_timeout: Duration::from_secs(60),
             wal_redo_timeout: Duration::from_secs(60),
             pg_distrib_dir: Some(pg_distrib_dir),
@@ -655,15 +647,6 @@ impl PageServerConf {
             control_plane_api: Some(Url::parse("http://localhost:6666").unwrap()),
             ..Default::default()
         };
-
-        // Test authors tend to forget about the default 10min initial lease deadline
-        // when writing tests, which turns their immediate gc requests via mgmt API
-        // into no-ops. Override the binary default here, such that there is no initial
-        // lease deadline by default in tests. Tests that care can always override it
-        // themselves.
-        // Cf https://databricks.atlassian.net/browse/LKB-92?focusedCommentId=6722329
-        config_toml.tenant_config.lsn_lease_length = Duration::from_secs(0);
-
         PageServerConf::parse_and_validate(NodeId(0), config_toml, &repo_dir).unwrap()
     }
 }
@@ -727,9 +710,8 @@ mod tests {
     use std::time::Duration;
 
     use camino::Utf8PathBuf;
-    use pageserver_api::config::{DiskUsageEvictionTaskConfig, EvictionOrder};
     use rstest::rstest;
-    use utils::{id::NodeId, serde_percent::Percent};
+    use utils::id::NodeId;
 
     use super::PageServerConf;
 
@@ -829,69 +811,19 @@ mod tests {
             .expect("parse_and_validate");
     }
 
-    #[rstest]
-    #[
-        case::omit_the_whole_config(
-            DiskUsageEvictionTaskConfig {
-                max_usage_pct: Percent::new(80).unwrap(),
-                min_avail_bytes: 2_000_000_000,
-                period: Duration::from_secs(60),
-                eviction_order: Default::default(),
-                #[cfg(feature = "testing")]
-                mock_statvfs: None,
-                enabled: true,
-            },
-        r#"
+    #[test]
+    fn test_config_disk_usage_based_eviction_is_valid() {
+        let input = r#"
             control_plane_api = "http://localhost:6666"
-        "#,
-    )]
-    #[
-        case::omit_enabled_field(
-            DiskUsageEvictionTaskConfig {
-                max_usage_pct: Percent::new(80).unwrap(),
-                min_avail_bytes: 1_000_000_000,
-                period: Duration::from_secs(60),
-                eviction_order: EvictionOrder::RelativeAccessed {
-                    highest_layer_count_loses_first: true,
-                },
-                #[cfg(feature = "testing")]
-                mock_statvfs: None,
-                enabled: true,
-            },
-        r#"
-            control_plane_api = "http://localhost:6666"
-            disk_usage_based_eviction = { max_usage_pct = 80, min_avail_bytes = 1000000000, period = "60s" }
-        "#,
-    )]
-    #[case::disabled(
-        DiskUsageEvictionTaskConfig {
-            max_usage_pct: Percent::new(80).unwrap(),
-            min_avail_bytes: 2_000_000_000,
-            period: Duration::from_secs(60),
-            eviction_order: EvictionOrder::RelativeAccessed {
-                highest_layer_count_loses_first: true,
-            },
-            #[cfg(feature = "testing")]
-            mock_statvfs: None,
-            enabled: false,
-        },
-        r#"
-            control_plane_api = "http://localhost:6666"
-            disk_usage_based_eviction = { enabled = false }
-        "#
-    )]
-    fn test_config_disk_usage_based_eviction_is_valid(
-        #[case] expected_disk_usage_based_eviction: DiskUsageEvictionTaskConfig,
-        #[case] input: &str,
-    ) {
+        "#;
         let config_toml = toml_edit::de::from_str::<pageserver_api::config::ConfigToml>(input)
             .expect("disk_usage_based_eviction is valid");
         let workdir = Utf8PathBuf::from("/nonexistent");
         let config = PageServerConf::parse_and_validate(NodeId(0), config_toml, &workdir).unwrap();
-        let disk_usage_based_eviction = config.disk_usage_based_eviction;
-        assert_eq!(
-            expected_disk_usage_based_eviction,
-            disk_usage_based_eviction
-        );
+        let disk_usage_based_eviction = config.disk_usage_based_eviction.unwrap();
+        assert_eq!(disk_usage_based_eviction.max_usage_pct.get(), 80);
+        assert_eq!(disk_usage_based_eviction.min_avail_bytes, 2_000_000_000);
+        assert_eq!(disk_usage_based_eviction.period, Duration::from_secs(60));
+        assert_eq!(disk_usage_based_eviction.eviction_order, Default::default());
     }
 }

pageserver/src/disk_usage_eviction_task.rs

@@ -171,8 +171,7 @@ pub fn launch_disk_usage_global_eviction_task(
     tenant_manager: Arc<TenantManager>,
     background_jobs_barrier: completion::Barrier,
 ) -> Option<DiskUsageEvictionTask> {
-    let task_config = &conf.disk_usage_based_eviction;
-    if !task_config.enabled {
+    let Some(task_config) = &conf.disk_usage_based_eviction else {
         info!("disk usage based eviction task not configured");
         return None;
     };
@@ -459,9 +458,6 @@ pub(crate) async fn disk_usage_eviction_task_iteration_impl<U: Usage>(
                 match next {
                     Ok(Ok(file_size)) => {
                         METRICS.layers_evicted.inc();
-                        /*BEGIN_HADRON */
-                        METRICS.bytes_evicted.inc_by(file_size);
-                        /*END_HADRON */
                         usage_assumed.add_available_bytes(file_size);
                     }
                     Ok(Err((
@@ -1269,7 +1265,6 @@ mod filesystem_level_usage {
                 #[cfg(feature = "testing")]
                 mock_statvfs: None,
                 eviction_order: pageserver_api::config::EvictionOrder::default(),
-                enabled: true,
             },
             total_bytes: 100_000,
             avail_bytes: 0,

pageserver/src/feature_resolver.rs

@@ -1,8 +1,4 @@
-use std::{
-    collections::HashMap,
-    sync::{Arc, atomic::AtomicBool},
-    time::Duration,
-};
+use std::{collections::HashMap, sync::Arc, time::Duration};
 
 use arc_swap::ArcSwap;
 use pageserver_api::config::NodeMetadata;
@@ -359,17 +355,11 @@ impl PerTenantProperties {
     }
 }
 
+#[derive(Clone)]
 pub struct TenantFeatureResolver {
     inner: FeatureResolver,
     tenant_id: TenantId,
-    cached_tenant_properties: ArcSwap<HashMap<String, PostHogFlagFilterPropertyValue>>,
-
-    // Add feature flag on the critical path below.
-    //
-    // If a feature flag will be used on the critical path, we will update it in the tenant housekeeping loop insetad of
-    // resolving directly by calling `evaluate_multivariate` or `evaluate_boolean`. Remember to update the flag in the
-    // housekeeping loop. The user should directly read this atomic flag instead of using the set of evaluate functions.
-    pub feature_test_remote_size_flag: AtomicBool,
+    cached_tenant_properties: Arc<ArcSwap<HashMap<String, PostHogFlagFilterPropertyValue>>>,
 }
 
 impl TenantFeatureResolver {
@@ -377,8 +367,7 @@ impl TenantFeatureResolver {
         Self {
             inner,
             tenant_id,
-            cached_tenant_properties: ArcSwap::new(Arc::new(HashMap::new())),
-            feature_test_remote_size_flag: AtomicBool::new(false),
+            cached_tenant_properties: Arc::new(ArcSwap::new(Arc::new(HashMap::new()))),
         }
     }
 
@@ -407,14 +396,12 @@ impl TenantFeatureResolver {
         self.inner.is_feature_flag_boolean(flag_key)
     }
 
-    /// Refresh the cached properties and flags on the critical path.
-    pub fn refresh_properties_and_flags(&self, tenant_shard: &TenantShard) {
-        let mut remote_size_mb = Some(0.0);
+    pub fn update_cached_tenant_properties(&self, tenant_shard: &TenantShard) {
+        let mut remote_size_mb = None;
         for timeline in tenant_shard.list_timelines() {
             let size = timeline.metrics.resident_physical_size_get();
             if size == 0 {
                 remote_size_mb = None;
-                break;
             }
             if let Some(ref mut remote_size_mb) = remote_size_mb {
                 *remote_size_mb += size as f64 / 1024.0 / 1024.0;
@@ -423,12 +410,5 @@ impl TenantFeatureResolver {
         self.cached_tenant_properties.store(Arc::new(
             PerTenantProperties { remote_size_mb }.into_posthog_properties(),
         ));
-
-        // BEGIN: Update the feature flag on the critical path.
-        self.feature_test_remote_size_flag.store(
-            self.evaluate_boolean("test-remote-size-flag").is_ok(),
-            std::sync::atomic::Ordering::Relaxed,
-        );
-        // END: Update the feature flag on the critical path.
     }
 }