Skip to content

Retry retryable worker failures once before failing the session - #898

Merged
TheGreatAxios merged 1 commit into
mainfrom
cl-7677-retry-retryable-provider-failures-inside-runsubagent-before
Sep 11, 2026
Merged

Retry retryable worker failures once before failing the session#898
TheGreatAxios merged 1 commit into
mainfrom
cl-7677-retry-retryable-provider-failures-inside-runsubagent-before

Conversation

@TheGreatAxios

Copy link
Copy Markdown
Collaborator

Summary

  • A worker inference failure classified retryable is retried in place once with bounded exponential backoff before the session is marked failed
  • The retry budget is bounded to one outer retry behind a tool-replay guard, so post-tool failures never re-execute side effects
  • The fleet status stays running during retries with a progress notice carrying attempt count and delay; exhausting the budget rethrows the resolved failure unchanged

Verification

  • Targeted retry tests pass (14 pass, 0 fail in run-resolved-provider-failure.test.ts)
  • bun run typecheck, bun run lint, and bun run build pass
  • Full bun run check shows only 4 pre-existing failures in src/session tests, reproduced on the clean tree

Fixes CL-7677

@linear-code

linear-code Bot commented Sep 11, 2026

Copy link
Copy Markdown

CL-7677

The harness gives up after three attempts and the worker dies even though no tool ran yet. One outer retry with backoff recovers the flake without another orchestrator round-trip, guarded against tool replay and bounded so it cannot multiply sends.
@TheGreatAxios
TheGreatAxios force-pushed the cl-7677-retry-retryable-provider-failures-inside-runsubagent-before branch from 26ed4b5 to 42b4ea9 Compare September 11, 2026 03:09
@TheGreatAxios
TheGreatAxios merged commit f9eb3f7 into main Sep 11, 2026
9 checks passed
@TheGreatAxios
TheGreatAxios deleted the cl-7677-retry-retryable-provider-failures-inside-runsubagent-before branch September 11, 2026 03:17
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