Skip to content

fix(dreaming): configurable consolidation timeout and run-scoped fallback after the first hang (T-1793) - #306

Merged
wolverin0 merged 6 commits into
mainfrom
lane/t1793
Oct 25, 2026
Merged

wolverin0 merged 6 commits into
mainfrom
lane/t1793

Conversation

@wolverin0

Copy link
Copy Markdown
Owner

T-1793 (AO-2): the antigravity consolidation stage tripped consolidate_breaker = TimeoutExpired 13 times in the last 14 days.

Measurement (read-only, live capture-control database, 2026-10-25, new scripts/measure_consolidate_latency.py): 238 antigravity consolidation calls since 2026-10-11, 214 ok / 24 error, p50 10.2 s, p95 26.3 s, 24 calls with unknown duration (the hung ones), 13 runs with the TimeoutExpired breaker. So the failures are hangs, not slow calls: a bigger budget alone would not fix them.

Change: MEMORYMASTER_DREAM_CONSOLIDATE_TIMEOUT_SECONDS (60..1800, default 300; existing MEMORYMASTER_AGY_TIMEOUT still honoured) and, after the first subprocess timeout in a run, the remaining consolidation calls of that run go to the configured extraction provider (recorded as consolidate_fallback); the timed-out batch waits for a later run. Without a consolidation-capable extractor the existing breaker applies. Model selections are untouched.

Evidence (lane-run outside the worker sandbox): tests/test_dreaming_worker.py + tests/test_dreaming_providers.py + tests/test_public_demo.py = 144 passed. Worker: ruff, release-truth --check, contract-witness passed. Follow-up row: 7 days with zero TimeoutExpired breaker runs after install.

🤖 Generated with Claude Code

@wolverin0
wolverin0 merged commit 1b51cfc into main Oct 25, 2026
11 of 12 checks passed
@wolverin0
wolverin0 deleted the lane/t1793 branch October 25, 2026 22:17
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant