perf: reduce durable transaction round trips

Cargo.toml

@@ -8,12 +8,12 @@ repository = "https://github.com/crabbuild/beyonddb"
 rust-version = "1.97"
 
 [workspace.dependencies]
-cellule-app = { git = "https://github.com/crabbuild/cellule.git", rev = "70c3c218fafc815b951a40d4df1b50bba15b7d4c" }
-cellule-host = { git = "https://github.com/crabbuild/cellule.git", rev = "70c3c218fafc815b951a40d4df1b50bba15b7d4c" }
-cellule-peer-http = { git = "https://github.com/crabbuild/cellule.git", rev = "70c3c218fafc815b951a40d4df1b50bba15b7d4c" }
-cellule-runtime = { git = "https://github.com/crabbuild/cellule.git", rev = "70c3c218fafc815b951a40d4df1b50bba15b7d4c" }
-cellule-store = { git = "https://github.com/crabbuild/cellule.git", rev = "70c3c218fafc815b951a40d4df1b50bba15b7d4c" }
-cellule-ltx = { git = "https://github.com/crabbuild/cellule.git", rev = "70c3c218fafc815b951a40d4df1b50bba15b7d4c" }
+cellule-app = { git = "https://github.com/crabbuild/cellule.git", rev = "9e17746a633ca1046bd93866866074226091f81a" }
+cellule-host = { git = "https://github.com/crabbuild/cellule.git", rev = "9e17746a633ca1046bd93866866074226091f81a" }
+cellule-peer-http = { git = "https://github.com/crabbuild/cellule.git", rev = "9e17746a633ca1046bd93866866074226091f81a" }
+cellule-runtime = { git = "https://github.com/crabbuild/cellule.git", rev = "9e17746a633ca1046bd93866866074226091f81a" }
+cellule-store = { git = "https://github.com/crabbuild/cellule.git", rev = "9e17746a633ca1046bd93866866074226091f81a" }
+cellule-ltx = { git = "https://github.com/crabbuild/cellule.git", rev = "9e17746a633ca1046bd93866866074226091f81a" }
 async-trait = "0.1"
 blake3 = "1.8"
 futures-util = "0.3"
@@ -29,6 +29,7 @@ aws-credential-types = "1"
 aws-sdk-dynamodb = "1"
 ed25519-dalek = "2.2"
 object_store = { version = "0.14.1", default-features = false }
+reqwest = { version = "0.12", default-features = false, features = ["rustls-tls", "stream"] }
 tempfile = "3"
 
 [package]
@@ -60,6 +61,7 @@ extenddb-storage = { git = "https://github.com/crabbuild/extenddb.git", rev = "7
 futures-util.workspace = true
 fs4 = "0.13.1"
 getrandom = "0.3"
+reqwest.workspace = true
 serde.workspace = true
 serde_json.workspace = true
 sysinfo = { version = "0.38.4", default-features = false, features = ["system"] }

README.md

@@ -64,6 +64,12 @@ stream intent commit in one Cell command. The successful response follows
 durable publication. [PNG version](diagram/beyonddb-architecture/durable-write@2x.png) ·
 [Detailed architecture and recovery design](docs/architecture.md)
 
+The serving binary currently waits for object-store publication on each
+durable write. Cellule's follower-log mode is not enabled in BeyondDB. An
+opt-in persistent follower store and authenticated peer transport exist, but
+the durability provider is not installed in serving and successor recovery
+is unfinished. See the [follower durability design](docs/follower-durability.md).
+
 ## Current capability boundary
 
 | Area | Current state |

benchmarks/2026-09-29-cellule-main-attempt/README.md

@@ -0,0 +1,48 @@
+# Cellule `9e17746` release qualification attempt
+
+This is a **failed, incomplete performance qualification**, retained to make the
+result reproducible. BeyondDB was built with Cellule
+`9e17746a633ca1046bd93866866074226091f81a`; ExtendDB SQLite remained at
+`7eaa89b437feed0af0f05883d3f1493f86c6fc6d`. The BeyondDB binary SHA-256
+and the exact pinned RustFS image are in [meta.json](meta.json). Both backends
+used signed boto3 requests, a 1 KiB payload, 5 seconds per case, and one/eight
+clients. BeyondDB used four initial partitions, 12 SQL workers, and the auth
+cache. SQLite used its local file backend. See the fixture details in
+[meta.json](meta.json) and [sqlite-meta.json](sqlite-meta.json).
+
+The host had 12 logical CPUs. Load was **44.5** at pair start, **47.5** after
+BeyondDB, and **90.4** after SQLite. Other builds and virtual machines were
+active. These numbers cannot establish an intrinsic throughput difference or
+the impact of the Cellule update.
+
+| API | BeyondDB 1 / 8 clients (req/s) | SQLite 1 / 8 clients (req/s) |
+| --- | ---: | ---: |
+| GetItem | 125.85 / 283.41 | 260.56 / 482.32 |
+| Query | 153.15 / 228.51 | 224.34 / 538.40 |
+| Scan | 107.93 / 86.75 | 215.92 / 475.59 |
+| BatchGetItem | 30.78 / 191.86 | 115.40 / 353.96 |
+| TransactGetItems | 1.41 / 2.14 | 76.73 / 225.91 |
+
+The eight-client TransactGetItems case completed 21 requests and returned one
+`TransactionCanceledException` with a `ThrottlingError` cancellation reason.
+The BeyondDB server also logged deferred GSI projection and capacity sweeps,
+incomplete cross-Cell participant resolution, and a transaction participant
+capacity refusal from Cell mailbox-byte exhaustion. Zero errors in the first
+nine cases therefore do not mean background work settled.
+The fail-fast harness then stopped, leaving **10 of 24 BeyondDB cases** in
+[run.json](run.json). SQLite completed **24 of 24 cases with zero request
+errors** in [sqlite-run.json](sqlite-run.json). Full p50/p95/p99 latency,
+elapsed time, and error fields are in those raw files. The harness now offers
+`--continue-on-error` so the next overloaded attempt can collect all cases and
+still exit unsuccessfully when any request fails.
+
+The separate signed release smoke check wrote a 1 KiB item, killed the serving
+process, and read the same item after restart. All 17 library tests, formatting,
+and strict Clippy passed. The broader ignored SDK restart test failed earlier
+in GSI setup with HTTP 503 after Cell SQL deadlines on a host with load near
+100. That failure remains unresolved. Neither this attempt nor the smoke check
+proves the all-API SQLite performance objective.
+
+Run control, server warnings, and raw case output are in
+[pair-status.json](pair-status.json), [server.log](server.log), and
+[bench.log](bench.log).

benchmarks/2026-09-29-cellule-main-attempt/bench.log

@@ -0,0 +1,11 @@
+{"operation": "get", "items_per_request": 1, "clients": 1, "elapsed_s": 5.01, "completed": 630, "errors": 0, "first_error": null, "ops_per_s": 125.85, "items_per_s": 125.85, "mean_ms": 7.94, "p50_ms": 4.93, "p95_ms": 18.72, "p99_ms": 26.41}
+{"operation": "get", "items_per_request": 1, "clients": 8, "elapsed_s": 5.01, "completed": 1421, "errors": 0, "first_error": null, "ops_per_s": 283.41, "items_per_s": 283.41, "mean_ms": 28.11, "p50_ms": 21.15, "p95_ms": 64.72, "p99_ms": 120.36}
+{"operation": "query", "items_per_request": 1, "clients": 1, "elapsed_s": 5.05, "completed": 773, "errors": 0, "first_error": null, "ops_per_s": 153.15, "items_per_s": 153.15, "mean_ms": 6.53, "p50_ms": 4.61, "p95_ms": 16.61, "p99_ms": 28.73}
+{"operation": "query", "items_per_request": 1, "clients": 8, "elapsed_s": 5.01, "completed": 1146, "errors": 0, "first_error": null, "ops_per_s": 228.51, "items_per_s": 228.51, "mean_ms": 34.9, "p50_ms": 25.67, "p95_ms": 94.88, "p99_ms": 210.2}
+{"operation": "scan", "items_per_request": 1, "clients": 1, "elapsed_s": 5.0, "completed": 540, "errors": 0, "first_error": null, "ops_per_s": 107.93, "items_per_s": 107.93, "mean_ms": 9.26, "p50_ms": 5.22, "p95_ms": 25.83, "p99_ms": 77.64}
+{"operation": "scan", "items_per_request": 1, "clients": 8, "elapsed_s": 5.05, "completed": 438, "errors": 0, "first_error": null, "ops_per_s": 86.75, "items_per_s": 86.75, "mean_ms": 91.6, "p50_ms": 31.97, "p95_ms": 92.1, "p99_ms": 2885.71}
+{"operation": "batch_get", "items_per_request": 2, "clients": 1, "elapsed_s": 5.0, "completed": 154, "errors": 0, "first_error": null, "ops_per_s": 30.78, "items_per_s": 61.56, "mean_ms": 32.45, "p50_ms": 20.11, "p95_ms": 109.1, "p99_ms": 202.28}
+{"operation": "batch_get", "items_per_request": 2, "clients": 8, "elapsed_s": 5.03, "completed": 965, "errors": 0, "first_error": null, "ops_per_s": 191.86, "items_per_s": 383.72, "mean_ms": 41.49, "p50_ms": 29.77, "p95_ms": 94.39, "p99_ms": 225.75}
+{"operation": "transact_get", "items_per_request": 2, "clients": 1, "elapsed_s": 5.67, "completed": 8, "errors": 0, "first_error": null, "ops_per_s": 1.41, "items_per_s": 2.82, "mean_ms": 708.14, "p50_ms": 852.61, "p95_ms": 1047.76, "p99_ms": 1047.76}
+{"operation": "transact_get", "items_per_request": 2, "clients": 8, "elapsed_s": 9.83, "completed": 21, "errors": 1, "first_error": "TransactionCanceledException: An error occurred (TransactionCanceledException) when calling the TransactGetItems operation: Transaction cancelled, please refer cancellation reasons for specific reasons [ThrottlingError, None]", "ops_per_s": 2.14, "items_per_s": 4.27, "mean_ms": 3127.2, "p50_ms": 3610.88, "p95_ms": 5749.02, "p99_ms": 5749.02}
+stopped after request error

benchmarks/2026-09-29-cellule-main-attempt/beyond-runner.log

@@ -0,0 +1,2 @@
+run started; host load: (46.8740234375, 58.19580078125, 53.64794921875)
+run finished exit: 1 host load: (47.5185546875, 56.4794921875, 53.38232421875)

benchmarks/2026-09-29-cellule-main-attempt/meta.json

@@ -0,0 +1,29 @@
+{
+  "server_commit": "ad4af2efd775409c2820a19013e59c79807d7c20 + Cellule 9e17746 pin and recovery fix",
+  "binary_sha256": "6429879d281eb49c9dd5b9b361007d52002560e64ce43a5da5c5a34e05c12abf",
+  "rustfs_image": "ghcr.io/rustfs/rustfs:1.0.0-glibc@sha256:bffcab0c9d647aab0055d1c69d340b202d0909966b385932d4ead1aeb7602858",
+  "host_load_start": [
+    46.8740234375,
+    58.19580078125,
+    53.64794921875
+  ],
+  "logical_cpus": 12,
+  "boto3_version": "1.43.105",
+  "fixture": {
+    "initial_partitions": 4,
+    "sql_workers": 12,
+    "max_active_cells": 128,
+    "follower_store_bytes": 1073741824,
+    "auth_cache_enabled": true,
+    "server_object_publication": true
+  },
+  "endpoint": "http://127.0.0.1:53167",
+  "started_at_unix": 1790749922.373023,
+  "host_load_end": [
+    47.5185546875,
+    56.4794921875,
+    53.38232421875
+  ],
+  "ended_at_unix": 1790749990.2865338,
+  "bench_exit_code": 1
+}

benchmarks/2026-09-29-cellule-main-attempt/pair-status.json

@@ -0,0 +1,22 @@
+{
+  "root": "/var/folders/rn/k2b7mvbs2dgb__6ys6_h94pm0000gn/T/beyonddb-perf-current-zsbtxro3",
+  "started_at_unix": 1790749916.511093,
+  "host_load_start": [
+    44.51318359375,
+    57.92919921875,
+    53.5283203125
+  ],
+  "beyond_exit_code": 1,
+  "host_load_after_beyond": [
+    47.5185546875,
+    56.4794921875,
+    53.38232421875
+  ],
+  "sqlite_exit_code": 0,
+  "ended_at_unix": 1790750119.0698938,
+  "host_load_end": [
+    90.37353515625,
+    66.35498046875,
+    57.46142578125
+  ]
+}