Repository navigation
fix(provider): retry timed-out status probes - #22794
Closed
juliusmarminge wants to merge 3 commits into
Closed
juliusmarminge wants to merge 3 commits into
juliusmarminge wants to merge 3 commits into
Conversation
…at must recover (#7230) Takeover regression tests only; the fix lands separately on this branch. A probe that timed out must be retried before the next periodic refresh, and a probe that keeps timing out must not be retried without bound. Co-Authored-By: Claude Opus 5.5 <[email protected]>
github-actions
Bot
added
vouch:trusted
Contributor
Thread transfer impact✅ Thread transfer remains within every enforced ceiling.
Baseline: Scenario and decoded snapshot size10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.
Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed. |
…t periodic refresh (#7230) A probe that reached the executable but got no verdict (Claude's --version past its deadline, a crashed process) stayed authoritative as an error until the periodic refresh, five minutes by default, or a manual refresh. The managed provider now retries such a probe up to three times, thirty seconds apart, through the existing demand-aware serialized refresh path. A conclusive probe resets the budget; an exhausted budget leaves the periodic refresh in charge, so a provider that keeps failing is not probed without bound. Inconclusive is a structural check on the snapshot (enabled, installed, error status, unknown auth) rather than message text, so Codex's timeout snapshot qualifies too and an unauthenticated verdict never retries. Co-Authored-By: Claude Fable 5.1 <[email protected]>
…terval A health refresh interval of 0 turns background provider refreshes off, but the inconclusive-probe retry only checked refreshOnInterval, so a provider whose probe kept timing out was still probed four times in the first five minutes. The retry now decides under the refresh semaphore right before probing: it checks the effective interval (including a setting changed while it slept), and uses the failure count as a probe generation so a manual refresh that ran meanwhile makes the queued retry a no-op instead of a redundant probe. Also corrects the comment on the retry delay: 30s is shorter than the staged Claude probe and the ACP probe deadlines; overlap is prevented by the semaphore, not the delay. Co-Authored-By: Claude Fable 5.1 <[email protected]>
2 tasks done
Contributor
|
Related: #25499. A Codex status probe there failed because the app-server sent back an error ( |
2 tasks done
Member
Author
|
Closing this draft: it has sat unfinished and #7230, which it was for, is closed. |
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
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.
A single provider status timeout replaces the usable snapshot with an error that remains visible after the executable recovers. This fixes #7230.
Reproduced on 1dfb572871 using an isolated
vp run devserver and a Claude wrapper that sleeps 25 seconds on its first version probe, exceeding the current 20-second deadline. Settings reported Unavailable while later wrapper version calls succeeded; manual Refresh recovered readiness without credential changes.Draft: regression tests first; implementation and review are pending. Intended change: bounded timeout recovery through the existing serialized, demand-aware refresh path.
Evidence: baseline runtime transcript.
Unverified: automatic recovery, Windows, macOS, Codex and other providers, desktop shell and mobile.
Prepared with Claude Opus 5.5 and GPT-6.1-Sol through the T3 Code Codex harness.