fix(strix): keep orchestrator model preflight timeout positive - #1600
fix(strix): keep orchestrator model preflight timeout positive#1600seonghobae wants to merge 5 commits into
Conversation
Preserve protected-main Noema fixture repairs while carrying only #1597's verified Strix LLM timeout correction and regression forward. The temporary one-shot workflow is absent from the resulting tree.
|
Warning Review limit reachedNext included review available in 41 minutes. View limit detailsLimit details: You’ve used the included review currently available. You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Review configuration: ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Team Run ID: 📒 Files selected for processing (2)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
| IS_PR_EVIDENCE_RUN: ${{ (github.event_name == 'pull_request_target' || github.event.client_payload.pr_number != '') && 'true' || 'false' }} | ||
| run: | | ||
| export LLM_TIMEOUT=0 | ||
| export LLM_TIMEOUT=300 |
There was a problem hiding this comment.
🟡 Five-minute preflight rejects healthy models
When model preflight exceeds five minutes, LLM_TIMEOUT cancels it despite the unbounded inference contract. Slow healthy routes fail central Strix checks.
Prompt for agents
Restore the repository's unbounded Strix inference contract without reintroducing Strix 1.5.3's immediate cancellation when LLM_TIMEOUT is zero. The current dependency feeds zero into asyncio.wait_for, while a positive value imposes a forbidden fixed deadline on preflight and request settings. Patch or upgrade the pinned Strix integration so the disabled value maps to no asyncio deadline, then keep .github/workflows/strix.yml, scripts/ci/test_strix_quick_gate.sh, and the Python timeout contract test aligned. Verify both that zero/disabled no longer cancels immediately and that a model call lasting beyond five minutes is not cancelled.
Was this helpful? React with 👍 or 👎 to provide feedback.
| workflow, | ||
| ) | ||
|
|
||
| assert configured_timeouts == ["300"] |
There was a problem hiding this comment.
🔍 Regression contracts disagree
The new test requires LLM_TIMEOUT=300, while assert_strix_workflow_pr_trigger_hardened requires zero. The changed-path quality check runs both and fails.
Was this helpful? React with 👍 or 👎 to provide feedback.
|
Superseded by merged #1601, which landed the same verified two-file current-main Strix timeout repair on protected main ( |
1 participant
Current-main Ready successor to closed Draft #1597. The previous branch was concurrently reset/closed after the verified repair had already produced the corrected workflow blob, so this successor is rebuilt non-destructively from protected
main@547fcc875d70a9489b30a6863692c977f1444fe2plus exactly the two intended #1597 files.Root cause: Strix 1.5.3 feeds
LLM_TIMEOUTinto both request settings andasyncio.wait_for;LLM_TIMEOUT=0cancels model preflight immediately instead of disabling a timeout. The exact production repair changes onlyexport LLM_TIMEOUT=0toexport LLM_TIMEOUT=300, matching the existing RED regression contract. A hosted one-shot run installed the repository hash-pinned test tooling and proved the focused regression GREEN (1 passed); its later GitHub-App workflow-permission error occurred only while trying to publish the workflow-file commit. The transformed workflow blob was nevertheless captured and is the exact blob carried here. The temporary source-fix workflow is absent from this tree.No provider bypass, fallback-model policy, retry count, finding threshold, fail-closed classification, or process/total timeout is changed. Fresh exact-head review and hosted evidence on this successor are authoritative; predecessor PR evidence does not transfer.