Conversation
- F1: syncIndexes only adopts idx:* indexes; foreign vector indexes on a shared Redis no longer abort startup on metric mismatch - F2: embedding provider output enforces the Float32 bound (502) — 1e39 no longer silently encodes to Infinity bytes via /upsert-data, /query-data, or embedded /update - F3: SCAN iteration cap now fails loudly on reset, delete (prefix/filter), fetch prefix, random, and namespace enumeration — matching range.ts; silent truncation reported success while keys survived - F4: /update and /upsert-data enforce the /upsert metadata/data budgets - F6: FT.SEARCH projects only needed fields via RETURN (NOCONTENT unsafe: parser reads _score from attributes) — ~1GB transfers at topK=1000 gone - F8: fetch/range/random project fields via HMGET/EXISTS instead of HGETALL - F10: lift the 1000-ids cap on /fetch and /delete (Upstash documents no cap for explicit ids) - F11: prefix globs are escaped into SCAN patterns instead of rejected (Upstash parity; ids may contain glob metacharacters) New tests: sync-indexes, query-projection, entry-point-limits, plus extensions to route-validation, keys, embedding suites. 465 tests green (331/60/74).
- docs/audits/2026-09-06-deep-audit.md: full finder/refuter/synthesis record - README: 465-test counts, batch-query cap, unbounded explicit-id fetch, empty-filter deviation row - RUNBOOK: baseline 465, deep-audit section now points at the packaged procedure (docs/workflows/deep-audit.md) instead of duplicating it - CLAUDE.md: counts 444 -> 465, audit pointer
The 50ms wall-clock assertion on the 100k-char pathological glob passed locally but blew up on CI runners. The property under test is absence of catastrophic backtracking — which manifests in seconds-to-minutes, not milliseconds — so the bound moves to 2s: still fails hard on any backtracking regression, tolerates noisy shared runners.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Second deep-audit pass, now executed from the packaged deterministic procedure (
docs/workflows/deep-audit.md).Pipeline. 5 mutually-blind finder agents (protocol, security, resource, parity, concurrency) → 17 raw findings → 17 default-refute skeptics → 12 survivors → orchestrator re-verification → 8 fixes landed, 3 issues filed, 5 refuted (all on record). Full record:
docs/audits/2026-09-06-deep-audit.md.Fixes (each with a regression test that fails pre-fix):
RETURNclause; previously every KNN candidate shipped its full hash (_vec+ metadata + data), ~1.2GB of RESP3 parsing per topK=1000 query.syncIndexesno longer adopts non-idx:*indexes; a shared Redis with someone else's vector index no longer crashes our startup on metric mismatch./updateand/upsert-datanow enforce the same metadata/data byte budgets as/upsert.MAX_SCAN_ITERATIONS).doc[1]are now reachable by prefix.Refuted on the record (5): metrics-miss-error-responses (hono compose catches at the throwing frame — verified against installed source), timer-leak shutdown impact, re-init null window (unconstructible), range O(N) (committed deferred redesign), empty-filter 400 (deliberate, now documented in README).
Open, tracked (
deep-auditlabel): #22 chunked slow-drip buffering, #23/infodimension 0 cold state, #24 reset dropIndex race (needs a sticky-dimension product decision).Also:
docs/workflows/deep-audit.mdis the single-source audit procedure — re-runnable by any agent harness or a future CI agent job. Doc truth reconciled (465 tests: 331/60/74;bun auditclean).Gate:
./scripts/test-all.shgreen — 331 unit / 60 integration / 74 SDK compatibility.