[Fleet Monitor] petry-projects/.github-private — .github/workflows/auto-rebase.yml #1473

Activity

1. don-petry

   added

   health-checkAutomated health check reportAutomated health check report

   dev-leadFor dev-lead agent pickupFor dev-lead agent pickup

   fleet-trackerTracked by the Actions Fleet MonitorTracked by the Actions Fleet Monitor

   on Aug 7, 2026
2. don-petry commented on Aug 7, 2026

   

   don-petry

   on Aug 7, 2026

   CollaboratorAuthor

   More actions

   Dev-Lead Implementation Plan

   Issue: #1473 — [Fleet Monitor] petry-projects/.github-private — .github/workflows/auto-rebase.yml

   Root cause

   The single failing run (31119528520) failed in "Set up job" with Failed to resolve action download info. Error: Service Unavailable — a transient GitHub action-resolution outage (503), not a config/code defect in the thin caller stub (which AGENTS.md forbids editing anyway).

   The repo-specific self-heal handler auto-rebase-retry.yml (AGENTS.md exception; exists precisely to recover "transient API error, rate-limit, infra hiccup") did fire (31119777561) but hung ~15 min and was cancelled — because its own "Set up job" must resolve actions/checkout@<sha> during the very same action-resolution outage. So the mitigation shares the failure it's meant to heal, and the auto-rebase run stayed failed → DEGRADED metric.

   Scope

   Harden auto-rebase-retry.yml so the self-heal handler no longer depends on resolving any marketplace action during "Set up job": drop actions/checkout, fetch the retry script via the pre-installed gh CLI (REST contents API, a different service than action resolution), and run it. No change to the tested scripts/auto-rebase-retry.sh logic, and no change to the frozen auto-rebase.yml caller stub.

   Implementation Checklist

   • .github/workflows/auto-rebase-retry.yml — remove actions/checkout step; add contents: read permission; fetch scripts/auto-rebase-retry.sh via gh api ... raw and execute it
   • tests/dev-lead/integration/test_auto_rebase_retry_no_action_resolution.py — new regression guard
   • .github/workflows/test-dev-lead.yml — add a job to run the new test

   Tests to Write

   • Assert the retry job declares no step with a uses: key (so "Set up job" resolves zero actions — immune to the [Fleet Monitor] petry-projects/.github-private — .github/workflows/auto-rebase.yml #1473 outage class)
   • Assert the retry job still invokes scripts/auto-rebase-retry.sh
   • Assert the retry job keeps actions: write and adds contents: read

   Test Command

   bats tests/dev-lead/unit/ && python3 tests/dev-lead/integration/test_auto_rebase_retry_no_action_resolution.py (plus bash .dev-lead/scripts/dev-lead-lint.sh)
3. don-petry commented on Aug 7, 2026

   

   don-petry

   on Aug 7, 2026

   CollaboratorAuthor

   More actions

   Dev-Lead: Implementation Complete

   Root cause (why auto-rebase.yml went DEGRADED)

   The single failing run (31119528520) failed in "Set up job" with
   Failed to resolve action download info. Error: Service Unavailable — a transient
   GitHub action-resolution outage (503), not a defect in the (frozen) auto-rebase.yml
   caller stub. The repo-specific self-heal handler auto-rebase-retry.yml did fire
   (31119777561) but hung ~15 min and was cancelled, because its own "Set up job"
   had to resolve actions/checkout during the same outage. So the mitigation shared the
   failure it exists to heal, the auto-rebase run stayed failed, and 1/3 runs → 33.3% DEGRADED.

   Fix

   Removed the retry handler's dependency on marketplace-action resolution so it can run
   during the exact outage class it heals: dropped actions/checkout, added contents: read,
   and self-fetch scripts/auto-rebase-retry.sh via the pre-installed gh CLI (REST contents
   API — a different service than action resolution) before executing it. A job with zero
   uses: steps skips "Prepare all required actions" entirely. The tested retry logic in
   scripts/auto-rebase-retry.sh is unchanged; the frozen auto-rebase.yml stub is untouched.

   Plan Execution

   • Diagnosed the failing run + the cancelled self-heal retry
   • Added failing regression test (red)
   • Hardened auto-rebase-retry.yml (green)
   • Wired the new test into test-dev-lead.yml
   • Full suite + dev-lead-lint.sh pass

   Test Results

   $ python3 tests/dev-lead/integration/test_auto_rebase_retry_no_action_resolution.py
   PASS: .github/workflows/auto-rebase-retry.yml retry job resolves no actions and self-fetches scripts/auto-rebase-retry.sh
   
   $ python3 tests/dev-lead/integration/test_auto_rebase_stub.py
   PASS: auto-rebase.yml correctly references ...auto-rebase-reusable.yml@auto-rebase/v2-next
   
   $ bats tests/dev-lead/unit/test_auto_rebase_retry.bats
   1..11  (all ok)
   
   $ bats tests/dev-lead/unit/        → 764 ok, 0 not ok
   $ bash .dev-lead/scripts/dev-lead-lint.sh
     [lint] shellcheck on 102 file(s)... / validating 4 agent profile(s)... / all checks passed
   

   Files Changed

   • .github/workflows/auto-rebase-retry.yml: drop actions/checkout; add contents: read; self-fetch retry script via gh api ... raw and run it (immune to action-resolution outages).
   • tests/dev-lead/integration/test_auto_rebase_retry_no_action_resolution.py: new regression guard — retry job must resolve no actions, must still invoke the retry script, and must declare actions: write + contents: read.
   • .github/workflows/test-dev-lead.yml: new auto-rebase-retry-resilience job running the guard.

   Notes

   • The original failure was pure transient infra; there is nothing to "fix" in the caller stub itself (and AGENTS.md forbids editing it). The durable fix is making the documented self-heal path actually recover this class — so a future 503 during "Set up job" gets re-run to success and the metric no longer degrades.
   • Edge case: if the REST contents API is also unavailable, the step fails fast under set -euo pipefail (never executes a partial script) — the pre-existing "a handler failure is just noise" posture, no worse than before.
4. don-petry

   added a commit that references this issue on Aug 7, 2026

   feat: implement issue #1473 — [Fleet Monitor] petry-projects/.github-…

   ...

   dc378f1
5. don-petry

   closed this as completedin #1476on Aug 7, 2026
6. don-petry

   added a commit that references this issue on Aug 8, 2026

   feat: implement issue #1473 — [Fleet Monitor] petry-projects/.github-…

   ...

   05e1674
7. don-petry

   added a commit that references this issue on Aug 18, 2026

   feat: implement issue #1473 — [Fleet Monitor] petry-projects/.github-…

   ...

   015b3bc