CI Run Link: https://github.com/coder/coder/actions/runs/20029668075
Failing Job: ci / test-go-pg (macos-latest)
Completed at: 2025-12-08T13:29:57Z (within minutes of Slack alert)
Run attempt: 1

Commit Info:

• SHA: 52243557a26320ec21be0cf4cbc6be2d2135ac68
• Author: Ehab Younes
• Link: coder/coder@5224355

Root Cause Classification: Flaky test (timing-dependent lifecycle state sequence)

Failure Evidence (from macOS job logs):

--- FAIL: TestAgent_Lifecycle/ShutdownTimeout (4.36s)
    agent_test.go:1597:
        Error Trace: /Users/runner/work/coder/coder/agent/agent_test.go:1597
        Error:        Not equal:
                      expected: []codersdk.WorkspaceAgentLifecycle{"starting", "ready", "shutting_down", "shutdown_timeout"}
                      actual  : []codersdk.WorkspaceAgentLifecycle{"starting", "ready", "shutting_down", "shutting_down"}
        Test:         TestAgent_Lifecycle/ShutdownTimeout
...
2025-12-08 13:28:44.663 [warn]  agent: shutdown script(s) failed ...
        error= run agent script "00000000-0000-0000-0000-000000000000":
                  - script timed out

Relevant test code (agent/agent_test.go around the failing subtest):

• Subtest expects sequence: starting -> ready -> shutting_down -> shutdown_timeout
• The shutdown script intentionally times out (Timeout: 1ms; Script: "sleep 3").
• Assertion: require.Equal(t, want, got[:len(want)]) after waiting for ShutdownTimeout to appear in lifecycle states.

Data Race Check: No WARNING: DATA RACE or race detected during execution of test found in logs.
Process Crash/OOM Check: No panic, OOM, or resource exhaustion indicators found.
Matrix Cancellation Artifact: Not applicable. Only macOS job failed; other matrix jobs were cancelled (run_attempt = 1).

Comprehensive Duplicate Search (coder/internal):

• Query: "TestAgent_Lifecycle" -> Found flake: TestAgent_Lifecycle/ShutdownScriptOnce #576 (ShutdownScriptOnce) [closed]
• Query: "ShutdownTimeout" -> No results
• Query: "shutdown_timeout shutting_down" -> No results
• Query: "InmemoryListener is already closed" -> Related to shutdown errors in other tests, but not this specific subtest
  Conclusion: No existing issue for this specific flake.

Precise Assignment Analysis:

• Primary (git blame on specific function lines): Not available via automation here.
• Secondary (recent test file contributors): agent/agent_test.go has multiple recent, meaningful changes by component owners, including Spike Curtis (e.g., commit 5807fe01e4c46aeb4de5680f1e793a2b8d20914b: coder/coder@5807fe0), as well as Danielle/Mathias in adjacent agent lifecycle/devcontainer areas.
• Tertiary (component ownership): This is in the agent lifecycle area; assigning to Spike Curtis for triage.

Analysis:

• The lifecycle state remained "shutting_down" and did not progress to "shutdown_timeout" within the expected window, even though the shutdown script timed out. This suggests a timing/ordering flake in lifecycle state reporting on macOS.

Reproduction Hints:

• Re-run ci / test-go-pg (macos-latest) or locally run go test ./agent -run TestAgent_Lifecycle/ShutdownTimeout on macOS. Intermittent.

Next Steps:

• Investigate lifecycle transition from ShuttingDown -> ShutdownTimeout on macOS.
• Consider relaxing the equality assertion to tolerate transient duplicate states, or ensure the state machine reliably emits ShutdownTimeout before teardown.