Skip to content

docs: drop stale engine-timeout rationales - #112

Merged
possiblyneal merged 2 commits into
mainfrom
fix/stale-replay-timeout-comment
Oct 3, 2026
Merged

possiblyneal merged 2 commits into
mainfrom
fix/stale-replay-timeout-comment

Conversation

@possiblyneal

@possiblyneal possiblyneal commented Oct 3, 2026 •

Copy link
Copy Markdown
Owner

Problem

Eight comments justify a bound with an engine invocation timeout: "the 30s function timeout", the 30s state::set timeout, or the iii invocation timeout or deadline. The in-process Engine has no invocation timeout (ADR 0001), so those rationales are stale, in the same way as the one #52 fixed.

Change

Comments only. Each bound still holds, because the work it limits is slow when left unbounded. The comments now give that as the reason:

  • src/functions/replay.ts: traversalCap stops a tree full of non-jsonl files from stalling the request that walks it.
  • src/index.ts: the unhandledRejection safety net is kept, so that one stray rejection cannot terminate the service.
  • src/state/schema.ts, src/types.ts, src/functions/graph.ts: the graph snapshot and the live-enumeration budget exist because a full kv.list over a large graph is slow. Removed: the "8s engine invocation deadline" headroom note.
  • src/functions/graph-import.ts: the import bounds stop a very large graph from stalling the request.
  • src/triggers/api.ts, src/viewer/index.html: export pagination and the viewer's memory cap.

Upstream issue references on the lines this PR touches are now written as rohitg00/agentmemory#N.

grep -rniE "invocation (timeout|ceiling|budget|window|deadline)|iii (engine )?invocation|30s timeout" src now finds only src/engine/inproc/sdk.ts:8, which records that the ceiling was removed. The remaining iii mentions are not about timeouts, so they are left as they are:

  • graph.ts:1258 and graph.ts:1328 describe a past "Invocation stopped" failure.
  • obsidian-export.ts:284 is about error serialization.

Verification

  • npm run typecheck: 0 errors.
  • npm test: 2113 passed, 1 skipped.

🤖 Generated with Claude Code

neal added 2 commits October 3, 2026 16:02
The in-process Engine has no invocation timeout (ADR 0001); the traversal
cap bounds how long a request spends walking a tree, which still holds.

Generated-By: Claude Opus 5.5 (1M context)
Seven more comments justified a bound by an iii invocation timeout or
the 30s state::set timeout. The in-process Engine has neither (ADR
0001); each bound still holds because the unbounded work is slow, so
the comments now say that. Upstream issue references on the touched
lines are written as rohitg00/agentmemory#N.

Generated-By: Claude Opus 5.5 (1M context)
@possiblyneal possiblyneal changed the title docs(replay): drop the stale 30s function timeout rationale docs: drop stale engine-timeout rationales Oct 3, 2026
@possiblyneal

Copy link
Copy Markdown
Owner Author

Code review: standard scan

Axes: Standards, Spec, Correctness/Logic. Every hard violation is fixed and every judgement call is resolved. Each axis signed off on the fixes to its own findings.

Reviewed-Head: 4805335

@possiblyneal
possiblyneal merged commit 73ddb4b into main Oct 3, 2026
6 checks passed
@possiblyneal
possiblyneal deleted the fix/stale-replay-timeout-comment branch October 19, 2026 12:59
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant