Fix Daily Go Test Parallelizer failures: drop preview cloud-hypervisor runtime, restore daily schedule

[Copilot]

The Daily Go Test Parallelizer workflow was reported at a 42.86% success rate costing ~$274/day for 3 safe items. Investigation shows both symptoms have causes different from those hypothesized in the issue.

Failure root cause

Failures cluster contiguously (runs #63–#68, #53–#55), not randomly. Every failed run aborts in the agent job's Execute GitHub Copilot CLI step with the same signature (verified on runs 32447678169 and 32404370749):

[WARN] [cloud-hypervisor] stage=guest-connectivity status=failed: Cloud Hypervisor guest connectivity probe failed with exit code 4
[ERROR] Fatal error: Error: Cloud Hypervisor guest connectivity probe failed with exit code 4
##[error]Agent execution exited with code 1

The guest VM's eth0 comes up NO-CARRIER/DOWN and the probe fails with fallback=disabled. This is preview-runtime infrastructure, not a flaky test shard or aggregation error. Failed runs die ~2 min into the agent job and never invoke the model — they consume no agent tokens, so the issue's premise that failures are expensive does not hold, and neither a fail-fast smoke step nor shard-scoped retry would help.

Cost root cause

The schedule was every 2h — 12 runs/day — while the workflow is named "Daily", described as "daily round-robin analysis", and its body references "the next daily run". Spend was driven by successful runs firing 12× more often than intended.

Changes

• sandbox.agent.runtime: cloud-hypervisor removed — falls back to the default rootless Docker sandbox. Also reduces privilege surface; the repo linter independently flags this runtime as "a privileged KVM preview path… require a human security review".
• schedule: every 2h → schedule: daily — restores intended cadence, ~12× cost reduction. Round-robin advance is unaffected; each run still persists last_file to cache memory.
• Recompiled daily-go-test-parallelizer.lock.yml.

 on:
-  schedule: every 2h
+  schedule: daily
...
-sandbox:
-  agent:
-    runtime: cloud-hypervisor
 tools:

No retry logic or pre-flight smoke step was added; both were suggested in the issue but neither addresses the actual failure.

Follow-up

~40 other workflows still pin cloud-hypervisor and are exposed to the same intermittent probe failure. Out of scope here, but likely worth its own tracking issue.

• Fixes [aw] Daily Go Test Parallelizer: 43% success rate, high-cost failures #54542

[github-actions]

PR Triage

• Category: bug
• Risk: low
• Priority score: 55/100 (impact 35, urgency 5, quality 15)
• Recommended action: defer
• CI status: no_checks

Generated by 🔧 PR Triage Agent · auto · 82.9 AIC · ⌖ 6.65 AIC · ⊞ 8.3K · ◷

[github-actions]

PR Triage

Category: bug | Risk: low | Priority: medium | Score: 68/100
Recommended action: defer

Draft but high-value: fixes 42.86% CI success rate / ~$274/day cost issue on Daily Go Test Parallelizer. Should be undrafted to fast_track.

_{Automated triage — run 32572524009}

Generated by 🔧 PR Triage Agent · auto · 65.4 AIC · ⌖ 6.04 AIC · ⊞ 8.3K · ◷