Query deadlines can be disabled by `0`, are not capped by the run budget, and are not cancelled on session abort
I maintainer di solito rispondono entro 1 giorno
Nessuno ha ancora preso questa issue.
Valutazione
- Difficoltà
- 4/5
- Tempo stimato
- 3-5 giorni
- Idoneità per principianti
- 18/100
- Tipo di issue
- Bug
- Chiarezza
- Abbastanza chiara
- Stato di attività
- Attiva
- Stack tecnologico
- mongodb, sql, sqlite, typescript
- Ambito
- backend, databases, performance
Direzione di ricerca
Start in packages/drivers/src/query-timeout.ts (parse() maps 0 to disabled) and the timeout message in packages/drivers/src/types.ts, then trace sql-materialize.ts where the deadline race and the missing ctx.abort handling live. Check packages/drivers/src/mongodb.ts for the cursor loop with no abort check. Done means a session abort returns within 2 s with the cursor closed, and per-call 0 cannot exceed the remaining budget.
Scritto dal modello di indicizzazione a partire dal testo della issue.
Descrizione
Problem
A single long tool call can outlive the run's soft and hard time budgets. The runner then kills the process and no answer or trace span survives.
Mechanism
0disables the deadline and the error text recommends it.parse()maps0to{ disabled: true }(packages/drivers/src/query-timeout.ts:36-40); the timeout error says "set query_timeout_ms ... (0 disables the deadline)" (packages/drivers/src/types.ts:74-75);sql_execute,sql_materializeandsql_checkadvertise it (sql-execute.ts:57-66,sql-materialize.ts:557). Clients with strict JSON schemas also send0as a filler value for unset optional fields.- No cap by remaining budget. Per-call deadlines are not clamped to the run's remaining time minus a finalization reserve.
- No cancellation.
sql-materialize.tsnever readsctx.abort;executeSourcepasses no signal (:239-260);sql-execute.ts:149-155passes none toDispatcher.call. The run's hard abort therefore cannot stop a running read. - Abandoned reads keep running. For remote sources the tool races the read against
deadline + 5 s(sql-materialize.ts:252-263) and rejects withcancelled: false, but the read continues (packages/drivers/src/mongodb.ts:379has no abort check; the cursor is closed only after the loop at:388). The error text then suggestsquery_timeout_ms: 0, so models start a second full read that competes with the first. Local SQLite/DuckDB engines get no outer timer at all (:247-249). - The runner's kill comes 60 s after the hard deadline (840 s inside 900 s), less startup time.
Evidence
- Retry reads with
query_timeout_ms: 0took 659.8, 658.9 and 674.8 s, started at about 143 s and ran through the soft deadline (672 s). - Reads that follow an abandoned attempt took a median 549 s versus 238 s for clean reads (37 vs 43 reads).
- In 5 sessions the hard deadline fired at 841-843 s, the abort never completed and the runner killed the run; in 2 more the abort arrived 35-53 s late.
- 19 sessions across recorded runs ended with no answer after a timeout of this shape; a scratch query with
query_timeout_ms: 0ran 428.6 s. - 1,095 of 2,685 (41%) of one client's SQL tool calls carried
query_timeout_ms: 0.
Impact
Answer-losing every time it triggers (a total loss, not a degraded answer). Estimated recoverable: 0-3 sessions per 270, more when combined with the Mongo cursor fix (#1479).
Proposed fix
- Clamp every tool deadline to
min(requested, remaining budget - finalization reserve); treat per-call0as "use default", reserving "disable" for connection configuration. - Thread an
AbortSignal(session abort plus deadline) throughDispatcherinto drivers; on timeout/abort close cursors (cursor.close()), cancel statements and kill the local helper. - Remove "(0 disables the deadline)" from agent-facing messages.
- Leave at least about 90 s between the hard deadline and the runner's kill.
Acceptance check
- Test: a session abort during a long
sql_materializereturns within 2 s and the cursor is closed. - Test:
query_timeout_ms: 0on a call does not exceed the remaining budget. - Test: after a timeout, no further documents are fetched from the abandoned cursor.
Related: #1479 (Mongo byte-bounded reads fetched one document per round trip; fix in #1480).
- Lingua principale
- TypeScript
- Stelle
- 810
- Fork
- 121
- Merge medio
- 1g 12h
- PR unite (30g)
- 94
Preparare l'ambiente
Avvia il container di sviluppo del progetto nel browser, con il tuo account GitHub.
- Nessun Dockerfile né file Docker Compose
- Ha un modello di pull request
- Leggi la guida per i contributori
Come iniziare
- Leggi tutta la issue e poi la guida ai contributi del progetto.
- Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
- Fai un fork del repository e lavora su un branch.
- Apri una pull request che faccia riferimento al numero della issue.
Altre issue di AltimateAI/altimate-code
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 68/100
AltimateAI/altimate-code#1454 ·
I maintainer di solito rispondono entro 1 giorno
-
test: MCP tests fail when the developer's ~/.claude.json has MCP servers (HOME is not sandboxed)Aperta
Difficoltà 2/5 1-3 ore Idoneità per principianti 78/100
AltimateAI/altimate-code#1386 ·
I maintainer di solito rispondono entro 1 giorno
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 84/100
AltimateAI/altimate-code#1384 ·
I maintainer di solito rispondono entro 1 giorno
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 84/100
AltimateAI/altimate-code#1323 ·
I maintainer di solito rispondono entro 1 giorno
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 86/100
AltimateAI/altimate-code#1288 ·
I maintainer di solito rispondono entro 1 giorno
Tutte le issue di AltimateAI/altimate-code
Issue simili
-
docs(api): openapi.yaml still documents gif-tools optimize colors as 2-256 instead of 2, 4, 16, 256Apertaarea: backend documentation priority: low
Difficoltà 2/5 1-3 ore Idoneità per principianti 84/100
snapotter-hq/SnapOtter#2919 ·
I maintainer di solito rispondono entro 1 giorno
-
accepted :+1: bug db/redis
Difficoltà 2/5 1-3 ore Idoneità per principianti 76/100
beekeeper-studio/beekeeper-studio#4882 · 1 commento ·
I maintainer di solito rispondono entro 1 giorno
-
ready-for-triage
Difficoltà 2/5 1-3 ore Idoneità per principianti 72/100
konflux-ci/konflux-ui#1630 · 1 commento ·
I maintainer di solito rispondono entro 2 giorni
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 67/100
ahmedkhaleel2004/gitdiagram#233 ·
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 72/100
simkimsia/netlify-axi#19 ·
I maintainer di solito rispondono entro 1 giorno