Skip to content

docs: triage Qwen worker budget failures (#371) - #373

Open
bearmug wants to merge 6 commits into
mainfrom
fix/issue-371-qwen-tool-budget
Open

docs: triage Qwen worker budget failures (#371)#373
bearmug wants to merge 6 commits into
mainfrom
fix/issue-371-qwen-tool-budget

Conversation

@bearmug

@bearmug bearmug commented Jul 17, 2026

Copy link
Copy Markdown
Contributor

Summary

  • document the observed Qwen worker failure as a pi-subagents tool-budget outcome, not a turns quota or current llama.cpp health failure
  • add health/model/smoke-test triage commands and bounded-worker operating guidance
  • link the runbook from the README

Closes #371

Validation

  • git diff --check
  • local Qwen /health
  • exact /v1/models model and 131072 context ✅
  • chat smoke test returned OK
  • npm run check ⚠️ blocked in the fresh worktree because dependencies are not installed (tsc: command not found)
  • fresh reviewer: no blockers/majors; docs-only change
  • codex review --base origin/main: no actionable findings

Scope

This does not change llama.cpp settings, runner budgets, fallback behavior, or server lifecycle. The machine-readable resumable outcome remains an upstream pi-subagents follow-up tracked by this issue.

@bearmug

bearmug commented Jul 17, 2026

Copy link
Copy Markdown
Contributor Author

Pre-merge validation reconciliation: commit 57eed67 had a CI test failure in the repository binary-dataset gate (Binary dataset is weak-label research only (0/2 reviewed)), unrelated to this documentation-only change. A transient gh pr checks query also returned no checks while the replacement run was queued. After the final evidence correction at commit 1fd2046, CI run 29545289289 completed successfully: type/validation coverage, tests, and packed minimum-runtime durability all passed. Current PR head is 1fd2046. No merge performed.

@bearmug

bearmug commented Jul 17, 2026

Copy link
Copy Markdown
Contributor Author

Feature-loop completion record for Issue #371: investigation completed from the Qwen child session and runner metadata; diagnosis documented as a 600-second runner timeout after soft tool-budget pressure (turn quota and hard tool limit ruled out); resolution artifacts are in docs/qwen-worker-failure-triage.md and docs/evidence/qwen-worker-tool-budget-2026-07-17.json; pre-merge validation completed successfully on head 1fd2046 (run 29545289289, attempt 2). Issue: #371. PR: #373. No merge performed.

@bearmug

bearmug commented Jul 17, 2026

Copy link
Copy Markdown
Contributor Author

Final validation reconciliation: valid gh pr checks 373 returned check (22.19.0) pass for head 1fd2046. Valid gh run view 29545289289 --json headSha,status,conclusion,attempt,jobs returned head 1fd2046, attempt 2, conclusion success, and job check (22.19.0) completed successfully. The earlier failed validation was a CLI read/query error using unsupported runAttempt, not a failed PR check. PR remains open and mergeable; no merge performed.

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.

fix(orchestration): surface Qwen tool-budget exhaustion as resumable worker outcome

1 participant