fix(db): self-heal under sustained memory pressure, VACUUM gate, WAL housekeeping by HouMinXi · Pull Request #13011 · cheaper-inference/OmniRoute

HouMinXi · 2026-09-08T03:56:20Z

What happened

On 2026-09-07 the production container stopped serving for 36 minutes and then wedged entirely until a manual restart. What I found on the host:

The container's cgroup working set sat at the 5 GiB cap, so admission control returned 503 for every request, including /api/health/ping.
The V8 heap was only ~300 MB. RSS was dominated by ~1.9 GB of glibc main-heap (native) memory, on a host with 7.5 GB total.
storage.sqlite-wal had grown to 2.98 GB, the size of the main DB. That is the high-water mark a VACUUM leaves behind: the startup cleanup ran VACUUM after deleting ~125 rows, rewriting the whole 3 GB database into the WAL. Auto-checkpoint (default 1000 pages) never shrinks the file, and the only TRUNCATE ran every 6 hours.

better-sqlite3 is synchronous, so once the WAL is huge and memory is tight, every query blocks the event loop long enough to stall the process.

Changes

Self-restart circuit (open-sse/utils/resourcePressure.ts). When pressure stays critical for OMNIROUTE_PRESSURE_SELF_RESTART_AFTER_MS (default 120s), the process logs its counters and exits, so the supervisor (systemd Restart=always, Docker restart policy) brings back a clean process in seconds. Off by default (OMNIROUTE_PRESSURE_SELF_RESTART). An unref'd 1-10s driver re-arms sampling so the circuit still works when no requests arrive, and a throwing exit hook re-arms the circuit instead of bricking it. Entering critical now logs a structured one-line diagnostic (heap/RSS/cgroup/PSI) whether or not self-restart is enabled, so the next incident starts with evidence.
Post-cleanup VACUUM gate (src/lib/db/cleanup.ts). VACUUM rewrites the whole database file; the 6h scheduler used to run it after every cleanup, including ones that freed ~100 rows. It now runs only when at least OMNIROUTE_VACUUM_MIN_DELETED_ROWS (default 1000) rows were freed. 0 means always vacuum. The weekly scheduled VACUUM from the Storage page still reclaims space on its own cadence.
WAL housekeeping (src/lib/db/core.ts). A 5-minute wal_checkpoint(PASSIVE) keeps the pending-frames count small so TRUNCATE never copies a multi-GB backlog on the main thread (OMNIROUTE_WAL_PASSIVE_INTERVAL_MS, 0 disables). If the WAL file exceeds OMNIROUTE_WAL_GUARD_MAX_MB (default 256), the tick escalates to TRUNCATE immediately instead of waiting for the 6h tick. The 6h TRUNCATE now logs duration, WAL size before/after, and busy/log/checkpointed frame counts.

Tests

50 unit tests across the six affected suites pass; typecheck:core is clean.
New tests cover the sustained-critical exit and its traffic-free driver, re-arm on a throwing exit hook, dispose cleanup, the VACUUM gate thresholds (including the 0 case), and the WAL scheduler wiring.
Each new behavior was verified by bug injection: break the condition, watch the test fail, revert, watch it pass.

Notes

Self-restart is opt-in; default behavior changes only in the WAL/VACUUM housekeeping and one extra log line on critical transitions.
All new knobs are documented in docs/reference/ENVIRONMENT.md.

On 2026-09-07 the production container stopped serving for 36 minutes and then wedged completely until a manual restart: the cgroup working set sat at the 5 GiB cap (admission shed every request), and a 2.98 GB storage.sqlite-wal left over from a startup VACUUM amplified the memory churn (glibc main heap reached 1.9 GB; V8 heap was only ~300 MB, so this was never a JS leak). Three changes make the failure self-limiting instead of fatal: 1. Self-restart circuit (open-sse/utils/resourcePressure.ts). When severity stays critical for OMNIROUTE_PRESSURE_SELF_RESTART_AFTER_MS (default 120s), the process logs its counters and exits; the container supervisor pulls a clean process up in seconds. Off by default (OMNIROUTE_PRESSURE_SELF_RESTART). An unref'd 1-10s driver re-arms sampling so the circuit advances even when no requests arrive, and a throwing exit hook re-arms the circuit instead of bricking it. Entering critical now logs a structured one-line diagnostic (heap/RSS/cgroup/PSI numbers) so the next incident starts with evidence. 2. Post-cleanup VACUUM gate (src/lib/db/cleanup.ts). VACUUM rewrites the whole database file; the 6h scheduler used to run it even when cleanup freed ~125 rows, spiking I/O and page cache every cycle. It now runs only when at least OMNIROUTE_VACUUM_MIN_DELETED_ROWS (default 1000) rows were freed; the weekly scheduled VACUUM (cheaper-inference#4437) still reclaims space on its own cadence. 3. WAL housekeeping (src/lib/db/core.ts). A 5-minute passive checkpoint (OMNIROUTE_WAL_PASSIVE_INTERVAL_MS, 0 disables) keeps the hot path drained; if the WAL file exceeds OMNIROUTE_WAL_GUARD_MAX_MB (default 256) the tick escalates to TRUNCATE immediately instead of waiting up to 6h. The existing 6h TRUNCATE tick now logs duration, WAL size before/after, and busy/log/ checkpointed frame counts. Tests: 39 unit tests around the circuit (fake-clock drive, re-arm on throwing exit hook, real-time driver advance, dispose cleanup), the VACUUM gate, and the WAL wiring; each new behavior was bug-injected to red and reverted to green. Signed-off-by: Minxi Hou <houminxi@gmail.com>

Signed-off-by: Minxi Hou <houminxi@gmail.com>

HouMinXi · 2026-09-11T15:49:23Z

CI red on this PR matches current release/v3.8.51, not the unique files in the diff.

Measured on tip af49d4972 (#12925) against the HouMinXi open set:

Docs Gates: live provider modules 358 vs docs/SVG still saying 356; cli-tunnel skill dry-run out of date.
Fast Quality: open-sse/services/autoCombo/__tests__/autoCombo.test.ts TS2739 vs frozen baseline 0; pack-policy follows that.
ESLint: tests/unit/volcengine-plan-binding-upsert.test.ts @typescript-eslint/no-explicit-any (file is on tip).
Unit shards: chat-rate-limit-body-lock, antigravity-missing-project-chat, chat-rejects-image-only-model, stream-handler-public-error-boundary (and the glm GLM_STREAM_BUFFER_BYTES assert, which fix(stream): accept the buffer size glm.ts has been passing since #12179 #12925 already landed on tip).

Unique diff does not touch those files. Contributor-only; not merging.

@HouMinXi

…2339) (#13038) Right call not to reuse the combo loop's `fallbackCount`: it only increments after a leg fails or is skipped, so the second target would still report 0 at dispatch. Stamping the ordered index (or round-robin offset) at the gate is the only place the number is actually known. This PR also carries the batch's file-size rebaseline, since it merges first and the ceiling has to cover every intermediate state. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (#13035 + #13011) and the `chatHelpers.ts` import block (#12975 on the tip + #13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on #13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: #12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

… refresh on non-streaming leg (#13043) (#13069) This is the one that mattered most. The non-streaming leg returning error outcomes without classification meant 402/429 never reached model lockout or rate-limit bookkeeping, and a 401 on a refreshable provider failed instead of refreshing — on a leg that serves real traffic. Extracting `applyProviderFailureClassification` and wiring both legs through it is the right shape: the asymmetry was the bug, so the fix has to remove the asymmetry rather than patch one side. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (#13035 + #13011) and the `chatHelpers.ts` import block (#12975 on the tip + #13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on #13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: #12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…#13061) The live evidence makes this: four Grok Build logins at 32-93% weekly remaining, skipped as "model locked by resilience" because the one login at 1% answered 402 and `passthroughModels: true` made that read as a per-model billing miss. A shared weekly wallet is a connection-wide signal, not a model verdict — `isSharedWalletCredits402` puts the scope where the biller put it. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (#13035 + #13011) and the `chatHelpers.ts` import block (#12975 on the tip + #13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on #13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: #12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…3060) Two real defects in one report, both correctly separated: the missing phrase in `OAUTH_INVALID_TOKEN_SIGNALS` made a recoverable Cline 401 terminal even though `refreshClineToken` already existed, and the cooling panel printing "429 (rate-limit)" for every future cooldown hid what actually happened. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (#13035 + #13011) and the `chatHelpers.ts` import block (#12975 on the tip + #13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on #13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: #12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

) (#13035) Correct root cause chain, and it explains why this only bites global installs: npm 11 skips `optionalDependencies` install scripts, the server child runs with `cwd: <packageRoot>/dist`, and `sql.js` sits at `<packageRoot>/node_modules`. Probing the parent directory is the missing rung. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (#13035 + #13011) and the `chatHelpers.ts` import block (#12975 on the tip + #13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on #13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: #12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

#13006) Both gaps are real and they compound: `executeTargetAttempt` already classified the 402 through `isQuotaExhaustionResponse` and then dropped it, so the only writer into the quota cache was the 429 path in `chat.ts`. A snapshot reading `remaining=1%, is_exhausted=0` five hours stale is then exactly what quota-weighted routing will keep picking. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (#13035 + #13011) and the `chatHelpers.ts` import block (#12975 on the tip + #13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on #13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: #12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…cs (#13136) Security-relevant and the right fix. A Basic gate on the whole `/docs` path was a deploy bandage that also locked LAN operators out of `https://<lan>:20128/docs`; dropping the four operator-internal writeups from the fumadocs glob and the Docker image removes the reason for the gate instead of papering over it. Keeping them in git and citing repo paths from the public pages is the right trade. The `.dockerignore` matcher walking rules in file order with last-match-wins is a genuine correctness fix in the test's own matcher. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (#13035 + #13011) and the `chatHelpers.ts` import block (#12975 on the tip + #13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on #13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: #12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

A catalog drop with no DELETE path is a real dead end for operators — no card, no route, no way out. Good that by-provider delete now bumps the proxy generation and purges synced models the way single-row delete already did, and that a cleanup error there does not turn a successful row delete into a 500. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (#13035 + #13011) and the `chatHelpers.ts` import block (#12975 on the tip + #13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on #13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: #12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…13120) Live-verified end to end: 1.5 gone from `GET /v1/models` with a 503, 3.0 answering 200 on a pong probe, and the CN-region host difference across 10 accounts. Aliasing 1.5 to 3.0 through `BUILT_IN_ALIASES` rather than leaving a dead id is the right retirement path, and noting that the #11503 as-is guard must not skip the rewrite is the detail that makes it work. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (#13035 + #13011) and the `chatHelpers.ts` import block (#12975 on the tip + #13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on #13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: #12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

Straightforward and well-scoped: a live catalog model falling through to the conservative 16384 clamp because `MODEL_SPECS` never got a 3.8 entry. Keeping the fallback for genuinely unknown ids is right, and the note on how this relates to #12663, #12499 and #12672 saved the review. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (#13035 + #13011) and the `chatHelpers.ts` import block (#12975 on the tip + #13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on #13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: #12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…objects (#13107) Both halves are real. A quota fetcher that needs console cookies with no way to enter them is unusable on any headless or remote install, and `new Error($'{'}object{'}'}` rendering `[object Object]` in a toast is exactly what makes a route-guard rejection unreadable. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (#13035 + #13011) and the `chatHelpers.ts` import block (#12975 on the tip + #13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on #13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: #12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

#13042) `Array.every` on modality arrays meant one unknown target wiped `input_modalities` for the entire combo — degrade-not-wipe is the correct semantic for a least-common-denominator. Upserting OpenRouter's live `architecture.input_modalities` without a full-table replace, and only invalidating cache when SQLite reports a change, are both the careful version. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (#13035 + #13011) and the `chatHelpers.ts` import block (#12975 on the tip + #13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on #13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: #12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…13026) Three related things that genuinely belong together: the registry and suffix splitter not knowing Astra while the live catalog serves it, the Codex client pin Astra requires, and dropping `next/font/google` so a production image build does not reach fonts.googleapis.com. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (#13035 + #13011) and the `chatHelpers.ts` import block (#12975 on the tip + #13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on #13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: #12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…#13001) Correct diagnosis on both counts. A probe that asks a reasoning model to add two numbers with a 2048-token budget will spend the budget thinking and return no text, and `Promise.all` across ten steps against one free-tier pool is a self-inflicted burst. The dashboard reporting `error` on a combo that serves real traffic fine is worse than no probe. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (#13035 + #13011) and the `chatHelpers.ts` import block (#12975 on the tip + #13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on #13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: #12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…12974) `parseModel` splitting on the first `/` is the trap, and recovering the custom row for all three id forms without changing `parseModel` is the right way around it — that function has too many callers to widen. Lifting the lookup into `customVisionOverride.ts` because `models.ts` is frozen is the correct response to the cap, not a workaround. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (#13035 + #13011) and the `chatHelpers.ts` import block (#12975 on the tip + #13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on #13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: #12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

… credential health (#12272) (#13034) Filling 20 sentinels across 39 locales, keeping `{minutes}` intact, and deliberately not rerunning `sync-ui-keys.mjs` because it restamps the sentinel — that last point is the one that usually gets this kind of PR wrong. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (#13035 + #13011) and the `chatHelpers.ts` import block (#12975 on the tip + #13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on #13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: #12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

…nce#13011 row-count gate Merge origin/release/v3.8.51 into this branch and replace the row-count-based vacuumAfterCleanup()/OMNIROUTE_VACUUM_MIN_DELETED_ROWS gate (merged by cheaper-inference#13011 after this PR's base) with reclaimFreedPages(), which reads the real SQLite freelist via PRAGMA instead of approximating via deleted-row count. Removes the now-orphaned gate functions, their dedicated test file, and the env var from .env.example/ENVIRONMENT.md. Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>

…uite regressions Split the cheaper-inference#12821 incremental-vacuum reclamation logic out of cleanup.ts into src/lib/db/reclaimFreedPages.ts (re-exported for callers/tests) so cleanup.ts stays under the file-size cap after the cheaper-inference#13011 reconciliation merge grew it past the 1200-line threshold. Also fixes two full-suite failures surfaced by running the cleanup/vacuumScheduler/db-health suite post-merge (not just this PR's own 3 test files, per the plan-file's mandatory item): - tests/unit/cleanup-column-fix.test.mjs scanned cleanup.ts's raw source for the PRAGMA incremental_vacuum invariant, which now lives in the extracted module — updated to scan both files. - tests/unit/db/cleanup-reclaim-freed-pages.test.ts asserted the freelist count is byte-for-byte unchanged when auto_vacuum=NONE. The tip's runAutoCleanup() now also runs cleanupCompressionRunTelemetry(), which lazily creates its table on first use (ensureCompressionRunTelemetryTable) — a legitimate one-time page cost from a freshly migrated DB, unrelated to reclaimFreedPages()'s own behavior. Loosened the assertion to a small tolerance while keeping the page_count assertion that actually guards against a full rebuild. Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>

@HouMinXi

…eaper-inference#12339) (cheaper-inference#13038) Right call not to reuse the combo loop's `fallbackCount`: it only increments after a leg fails or is skipped, so the second target would still report 0 at dispatch. Stamping the ordered index (or round-robin offset) at the gate is the only place the number is actually known. This PR also carries the batch's file-size rebaseline, since it merges first and the ceiling has to cover every intermediate state. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

… refresh on non-streaming leg (cheaper-inference#13043) (cheaper-inference#13069) This is the one that mattered most. The non-streaming leg returning error outcomes without classification meant 402/429 never reached model lockout or rate-limit bookkeeping, and a 401 on a refreshable provider failed instead of refreshing — on a leg that serves real traffic. Extracting `applyProviderFailureClassification` and wiring both legs through it is the right shape: the asymmetry was the bug, so the fix has to remove the asymmetry rather than patch one side. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…cheaper-inference#13061) The live evidence makes this: four Grok Build logins at 32-93% weekly remaining, skipped as "model locked by resilience" because the one login at 1% answered 402 and `passthroughModels: true` made that read as a per-model billing miss. A shared weekly wallet is a connection-wide signal, not a model verdict — `isSharedWalletCredits402` puts the scope where the biller put it. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…erence#12594) (cheaper-inference#13060) Two real defects in one report, both correctly separated: the missing phrase in `OAUTH_INVALID_TOKEN_SIGNALS` made a recoverable Cline 401 terminal even though `refreshClineToken` already existed, and the cooling panel printing "429 (rate-limit)" for every future cooldown hid what actually happened. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…aper-inference#12960) (cheaper-inference#13035) Correct root cause chain, and it explains why this only bites global installs: npm 11 skips `optionalDependencies` install scripts, the server child runs with `cwd: <packageRoot>/dist`, and `sql.js` sits at `<packageRoot>/node_modules`. Probing the parent directory is the missing rung. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

cheaper-inference#13006) Both gaps are real and they compound: `executeTargetAttempt` already classified the 402 through `isQuotaExhaustionResponse` and then dropped it, so the only writer into the quota cache was the 429 path in `chat.ts`. A snapshot reading `remaining=1%, is_exhausted=0` five hours stale is then exactly what quota-weighted routing will keep picking. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…cs (cheaper-inference#13136) Security-relevant and the right fix. A Basic gate on the whole `/docs` path was a deploy bandage that also locked LAN operators out of `https://<lan>:20128/docs`; dropping the four operator-internal writeups from the fumadocs glob and the Docker image removes the reason for the gate instead of papering over it. Keeping them in git and citing repo paths from the public pages is the right trade. The `.dockerignore` matcher walking rules in file order with last-match-wins is a genuine correctness fix in the test's own matcher. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…rence#13197) A catalog drop with no DELETE path is a real dead end for operators — no card, no route, no way out. Good that by-provider delete now bumps the proxy generation and purges synced models the way single-row delete already did, and that a cleanup error there does not turn a successful row delete into a 500. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…heaper-inference#13120) Live-verified end to end: 1.5 gone from `GET /v1/models` with a 503, 3.0 answering 200 on a pong probe, and the CN-region host difference across 10 accounts. Aliasing 1.5 to 3.0 through `BUILT_IN_ALIASES` rather than leaving a dead id is the right retirement path, and noting that the cheaper-inference#11503 as-is guard must not skip the rewrite is the detail that makes it work. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…ence#13195) Straightforward and well-scoped: a live catalog model falling through to the conservative 16384 clamp because `MODEL_SPECS` never got a 3.8 entry. Keeping the fallback for genuinely unknown ids is right, and the note on how this relates to cheaper-inference#12663, cheaper-inference#12499 and cheaper-inference#12672 saved the review. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…objects (cheaper-inference#13107) Both halves are real. A quota fetcher that needs console cookies with no way to enter them is unusable on any headless or remote install, and `new Error($'{'}object{'}'}` rendering `[object Object]` in a toast is exactly what makes a route-guard rejection unreadable. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…r-inference#12613) (cheaper-inference#13042) `Array.every` on modality arrays meant one unknown target wiped `input_modalities` for the entire combo — degrade-not-wipe is the correct semantic for a least-common-denominator. Upserting OpenRouter's live `architecture.input_modalities` without a full-table replace, and only invalidating cache when SQLite reports a change, are both the careful version. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…heaper-inference#13026) Three related things that genuinely belong together: the registry and suffix splitter not knowing Astra while the live catalog serves it, the Codex client pin Astra requires, and dropping `next/font/google` so a production image build does not reach fonts.googleapis.com. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…cheaper-inference#13001) Correct diagnosis on both counts. A probe that asks a reasoning model to add two numbers with a 2048-token budget will spend the budget thinking and return no text, and `Promise.all` across ten steps against one free-tier pool is a self-inflicted burst. The dashboard reporting `error` on a combo that serves real traffic fine is worse than no probe. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…heaper-inference#12974) `parseModel` splitting on the first `/` is the trap, and recovering the custom row for all three id forms without changing `parseModel` is the right way around it — that function has too many callers to widen. Lifting the lookup into `customVisionOverride.ts` because `models.ts` is frozen is the correct response to the cap, not a workaround. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

… credential health (cheaper-inference#12272) (cheaper-inference#13034) Filling 20 sentinels across 39 locales, keeping `{minutes}` intact, and deliberately not rerunning `sync-ui-keys.mjs` because it restamps the sentinel — that last point is the one that usually gets this kind of PR wrong. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…aper-inference#12561) (cheaper-inference#13027) Good archaeology: cheaper-inference#12178 added seven strings and only `de` and `vi` ever got them, with `pt-BR` left holding placeholders. Restoring the feature-flag description missing from `pt.json` alone is the kind of thing only a full-catalog diff finds. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…housekeeping (cheaper-inference#13011) A 36-minute outage with a full host-level write-up is the best kind of bug report. The WAL at 2.98 GB being the high-water mark a VACUUM leaves behind — rewriting a 3 GB database after deleting ~125 rows, with auto-checkpoint never shrinking the file and TRUNCATE only every 6 hours — is exactly the kind of thing that is invisible from inside the process, where the V8 heap read 300 MB while RSS was dominated by 1.9 GB of glibc main-heap. Reconciled against the tip after the batch landed: the `ENVIRONMENT.md` table conflicted with cheaper-inference#13035's `OMNIROUTE_SQLJS_WASM_PATH` row and both sides were kept. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…aper-inference#13017) The one-shot framing is what makes this safe: a pin is an explicit operator act, so taking the inactive row only for that request, with siblings out of the pool and a 60s per-connection storm gate, keeps the blast radius at one request. Dashboard deactivate staying off is the right carve-out. Reconciled against the tip after the batch landed: the `chatHelpers.ts` import block conflicted with `buildExhaustionOptions` (cheaper-inference#12975) and both imports were kept. 41/41 across the probe suites afterwards. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…aper-inference#13050) Nasty shape: HTTP 200, upstream billed, and Codex times out waiting for `response.completed` that never comes because the body is JSON. Good catch that wiring the wrap in `handleResponsesCore` would not have fixed it — production goes through `handleChat`, and that helper is tests-only. Stamping the client's SSE intent before flipping `stream` off is the right place to remember it. Reconciled against the tip after the batch landed: only the file-size baseline conflicted, resolved to the batch's rebaselined value. 5/5 afterwards. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

…UM in the cleanup scheduler (#12821) (#12830) * fix(db): reclaim freed pages incrementally instead of a blocking VACUUM in the cleanup scheduler (#12821) startCleanupScheduler() ran a synchronous whole-database VACUUM on the event loop whenever a cleanup pass deleted at least one row - 30 s after every start and every 6 h. With node:sqlite that blocks every route (/healthz included) for the duration: 7 min 55 s on a 540 MB storage.sqlite to reclaim six rows. It also bypassed vacuumScheduler, the app-level owner of full VACUUMs and the operator's scheduledVacuum / vacuumHour settings. cleanup.ts no longer issues a full VACUUM. After each pass reclaimFreedPages() branches on PRAGMA auto_vacuum: - INCREMENTAL: drain the freelist with PRAGMA incremental_vacuum(N) in ~1 MiB batches (N from page_size), pausing between batches for as long as the last one took (<=250 ms), PASSIVE checkpoint every 64 batches and a TRUNCATE checkpoint at the end so the main file shrinks in WAL mode; hard caps of 2048 batches / 30 s per pass, the remainder waits for the next pass. - FULL: nothing to do, SQLite reclaims on commit. - NONE: incremental_vacuum is a no-op, so record a request via the new vacuumScheduler.requestFullVacuum(); the rebuild runs in the configured window (or via the Storage page button). scheduledVacuum=never is honored. vacuumScheduler persists fullVacuumRequestedAt / fullVacuumRequestReason, clears them on the next successful runNow(), and hydrates from key_value before an early request so it cannot clobber a persisted lastRunAt. Loop robustness: db.exec() rather than pragma() (bun:sqlite's all() steps a zero-column pragma once), SQLITE_BUSY/LOCKED and a handle closed under the pass stop it quietly, other errors stop it with partial progress logged. Also drops the duplicate cleanupProxyLogs() call in the scheduled pass - runAutoCleanup() already covers proxy_logs. Tests: new tests/unit/db/cleanup-reclaim-freed-pages.test.ts (INCREMENTAL drain/pause/checkpoint, page_size-derived batch, caps, FULL no-op, NONE defers and leaves page_count untouched, runScheduledCleanupPass() path); vacuum-scheduler.test.ts covers requestFullVacuum persistence, restart survival and clearing; cleanup-column-fix.test.mjs now asserts incremental_vacuum and the absence of a full VACUUM statement. * chore(changelog): name the #12821 fragment after its PR (#12830) * fix(db): extract reclaimFreedPages into its own module and fix full-suite regressions Split the #12821 incremental-vacuum reclamation logic out of cleanup.ts into src/lib/db/reclaimFreedPages.ts (re-exported for callers/tests) so cleanup.ts stays under the file-size cap after the #13011 reconciliation merge grew it past the 1200-line threshold. Also fixes two full-suite failures surfaced by running the cleanup/vacuumScheduler/db-health suite post-merge (not just this PR's own 3 test files, per the plan-file's mandatory item): - tests/unit/cleanup-column-fix.test.mjs scanned cleanup.ts's raw source for the PRAGMA incremental_vacuum invariant, which now lives in the extracted module — updated to scan both files. - tests/unit/db/cleanup-reclaim-freed-pages.test.ts asserted the freelist count is byte-for-byte unchanged when auto_vacuum=NONE. The tip's runAutoCleanup() now also runs cleanupCompressionRunTelemetry(), which lazily creates its table on first use (ensureCompressionRunTelemetryTable) — a legitimate one-time page cost from a freshly migrated DB, unrelated to reclaimFreedPages()'s own behavior. Loosened the assertion to a small tolerance while keeping the page_count assertion that actually guards against a full rebuild. Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> * chore(db): drop the reclaimable-bytes VACUUM gate test superseded by incremental reclaim tests/unit/vacuum-reclaimable-threshold.test.ts pinned cleanup.ts's vacuumAfterCleanup()/getReclaimableBytes()/getVacuumMinReclaimableBytes() (#13079). This branch removes the inline post-cleanup full VACUUM entirely in favour of reclaimFreedPages() (#12821), which reads the same freelist_count / page_size signal and defers a full VACUUM to the vacuum scheduler when auto_vacuum=NONE. With those three exports gone the file cannot compile, and the behaviour it guarded no longer exists. --------- Co-authored-by: insoln <is@careerum.com> Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>

@HouMinXi

…eaper-inference#12339) (cheaper-inference#13038) Right call not to reuse the combo loop's `fallbackCount`: it only increments after a leg fails or is skipped, so the second target would still report 0 at dispatch. Stamping the ordered index (or round-robin offset) at the gate is the only place the number is actually known. This PR also carries the batch's file-size rebaseline, since it merges first and the ceiling has to cover every intermediate state. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

… refresh on non-streaming leg (cheaper-inference#13043) (cheaper-inference#13069) This is the one that mattered most. The non-streaming leg returning error outcomes without classification meant 402/429 never reached model lockout or rate-limit bookkeeping, and a 401 on a refreshable provider failed instead of refreshing — on a leg that serves real traffic. Extracting `applyProviderFailureClassification` and wiring both legs through it is the right shape: the asymmetry was the bug, so the fix has to remove the asymmetry rather than patch one side. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

@HouMinXi

…cheaper-inference#13061) The live evidence makes this: four Grok Build logins at 32-93% weekly remaining, skipped as "model locked by resilience" because the one login at 1% answered 402 and `passthroughModels: true` made that read as a per-model billing miss. A shared weekly wallet is a connection-wide signal, not a model verdict — `isSharedWalletCredits402` puts the scope where the biller put it. --- Validated in one consolidated worktree cut from `release/v3.8.51`, boarded with the other 19 PRs of this batch. Two in-batch conflicts, both additive and resolved by keeping each side: the `ENVIRONMENT.md` table (cheaper-inference#13035 + cheaper-inference#13011) and the `chatHelpers.ts` import block (cheaper-inference#12975 on the tip + cheaper-inference#13017). - `typecheck:core` clean; `check:dashboard-typecheck` OK (206 pre-existing, within baseline); `check:changelog-integrity` OK; `check:docs-counts` migrations ✓ - complexity 2816 / baseline 3218 and cognitive-complexity 1271 / baseline 1437 — both under baseline - 531 of 532 focused assertions green across the batch's 46 test files - `check-file-size` rebaselined for the batch's real growth (annotation `_rebaseline_2026_09_11_mergebatch_v3851_houminxi`, landed on cheaper-inference#13038), attributed per PR The single red is **not this batch**: `tests/unit/combo/quota-weighted-strategy.test.ts` → "A/B isolation: 7 hard-empty + 2 at 0.5% + 1 at 40%, floor=1" asserts an order between two connections of identical weight and flakes on the pure tip too — 2 failures in 4 runs at `origin/release/v3.8.51` with nothing from this batch applied. ⚠️ base-red inherited: cheaper-inference#12732 — `Docs Gates`, `Merge integrity`, `No new ESLint warnings`, `Unit Tests fast-path` and `Fast Quality Gates` reproduce on the pure tip (provider count 356 vs the 358 the modules define, SKILL.md drift, and `open-sse/utils/stream.ts` at 3115 > frozen 3098, untouched here). Thanks @HouMinXi — the live evidence on these (X500 logs, `storage.sqlite` state, real `/v1/models` probes, the 36-minute outage write-up) is what let a 20-PR batch be reviewed as a unit.

HouMinXi requested a review from diegosouzapw as a code owner September 8, 2026 03:56

HouMinXi force-pushed the fix/p0-memory-self-heal branch from a2795e3 to 46632a5 Compare September 8, 2026 17:51

HouMinXi force-pushed the fix/p0-memory-self-heal branch from 46632a5 to 29767d4 Compare September 10, 2026 11:20

HouMinXi force-pushed the fix/p0-memory-self-heal branch from 29767d4 to c7d1569 Compare September 10, 2026 13:03

HouMinXi mentioned this pull request Sep 11, 2026

docs(backend): 8 GiB host sqlite growth vs RSS vs cgroup cap (measured inventory) #13325

Closed

HouMinXi added 2 commits September 11, 2026 04:17

changelog: entry for cheaper-inference#13011

7e807a8

Signed-off-by: Minxi Hou <houminxi@gmail.com>

HouMinXi force-pushed the fix/p0-memory-self-heal branch from c7d1569 to 7e807a8 Compare September 11, 2026 08:32

HouMinXi mentioned this pull request Sep 11, 2026

fix(db): gate the auto-cleanup VACUUM on reclaimable space, not row count #13079

Merged

diegosouzapw mentioned this pull request Sep 15, 2026

fix(db): reclaim freed pages incrementally instead of a blocking VACUUM in the cleanup scheduler (#12821) #12830

Merged

HouMinXi deleted the fix/p0-memory-self-heal branch September 16, 2026 14:06

rapha4lx mentioned this pull request Sep 23, 2026

fix(resilience): sample recovery in background driver when selfRestart is disabled #14576

Merged

5 tasks done

xiaoyaner0201 mentioned this pull request Oct 10, 2026

fix(resilience): RSS-only admission pressure never reaches self-restart observer #16259

Closed

Navigation Menu

Repository navigation

fix(db): self-heal under sustained memory pressure, VACUUM gate, WAL housekeeping - #13011

fix(db): self-heal under sustained memory pressure, VACUUM gate, WAL housekeeping#13011
diegosouzapw merged 3 commits into
cheaper-inference:release/v3.8.51from
HouMinXi:fix/p0-memory-self-heal

HouMinXi commented Sep 8, 2026

Uh oh!

HouMinXi commented Sep 11, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

HouMinXi commented Sep 8, 2026

What happened

Changes

Tests

Notes

Uh oh!

HouMinXi commented Sep 11, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants