Skip to content

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

Description

@github-actions

Daily Go Test Parallelizer — 43% run success rate, expensive failures

Detected by: Agent Performance Analyzer — 2026-08-21 daily run

Evidence (from metrics/latest.json, 24h window 2026-08-20 → 2026-08-21)

  • Runs: 7 total, 3 successful, 4 failed (success rate 42.86%)
  • Avg duration: 774s (~12.9 min)
  • Total tokens: 965,732
  • Total cost: $274.02/day
  • Safe items produced: only 3 across all 7 runs (low output relative to token spend)

No existing open issue was found referencing "Daily Go Test Parallelizer" or "Go Test Parallelizer" after searching open issues in this repository, so this is filed as a new tracking issue rather than a duplicate.

Why this matters

Nearly 60% of runs fail, but the workflow still consumes significant tokens/cost even on failing runs (avg duration is high regardless of outcome), suggesting failures occur late in the run (e.g., during parallel test execution or result aggregation) rather than failing fast. This is a costly reliability pattern: burning ~$274/day for only 3 successful outputs.

Suggested investigation

  1. Pull recent failed run logs for "Daily Go Test Parallelizer" (gh run list / workflow logs) to identify the failure signature (timeout, OOM, flaky parallel test runner, tool/API error).
  2. Check whether failures cluster around a specific time window or Go test shard.
  3. Consider adding a fail-fast check (e.g., quick smoke/lint step) before the expensive parallel test execution to avoid burning tokens on runs that are doomed to fail early.
  4. If failures are due to test flakiness rather than the workflow itself, consider retry/backoff logic scoped to flaky shards only, not the entire run.

Data caveat

This finding is based on aggregate workflow_runs metrics only (no direct log/content sampling was performed this run, per token-budget guidance). Please verify against actual run logs before implementing a fix.

Generated by ⚡ Agent Performance Analyzer - Meta-Orchestrator · auto · 55.7 AIC · ⌖ 8.73 AIC · ⊞ 16.8K ·

  • expires on Aug 23, 2026, 5:04 AM UTC-08:00

Metadata

Metadata

Labels

cookieIssue Monster Loves Cookies!

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions