flake: TestTasksNotification/TemplateTaskWorkingFromIdle

[flake-investigator]

CI Failure Details

CI Run Link: https://github.com/coder/coder/actions/runs/20498829535

Failed Job: test-go-pg (windows-2022)

Commit Info:

• SHA: 05529139bc9b90f01fc0ef4f413adc2f3febdbb7
• Author: Danielle Maywood
• Date: 2025-12-24

Failure Signature

From the Windows job logs:

• Test: TestTasksNotification/TemplateTaskWorkingFromIdle
• Location: coderd/aitasks_test.go:1637
• Assertion:
  • Error: "[]" should have 1 item(s), but has 0

The failure occurs when the test expects exactly one notification to have been enqueued:

Error Trace: C:/actions-runner/coder/coder/coderd/aitasks_test.go:1637
Error:      "[]" should have 1 item(s), but has 0
Test:       TestTasksNotification/TemplateTaskWorkingFromIdle

Root Cause Hypothesis (most likely)

This test seeds multiple workspace_app_statuses rows in a tight loop using dbgen.WorkspaceAppStatus, which defaults CreatedAt to dbtime.Now():

CreatedAt: takeFirst(orig.CreatedAt, dbtime.Now())

If multiple inserts happen within the same clock tick (more likely on Windows due to coarser timestamp resolution), two seeded statuses can end up with identical created_at values.

The production query that determines the “latest” status is:

SELECT *
FROM workspace_app_statuses
WHERE app_id = @app_id::uuid
ORDER BY created_at DESC, id DESC
LIMIT 1;

When created_at ties, the tie-breaker becomes random-ish for test data because UUIDs are random. That can cause the handler to think the latest status was already working, so it treats the new working update as a no-op transition and skips notification enqueueing.

This explains why the test sometimes observes notifyEnq.Sent(...) == [].

Suggested Fix

Make the test’s seeded history deterministic by explicitly setting CreatedAt when inserting tc.latestAppStatuses so that the “latest” row is always the intended one.

Example approach:

• pick a base := dbtime.Now()
• for older statuses use base.Add(-2*time.Second), base.Add(-1*time.Second), etc.
• ensure the intended “latest” status has the greatest timestamp

Assignment Analysis

Most relevant recent change touching TestTasksNotification is:

• f2a1a7e8c3beda4f40a545beb4d62efa23b401d1 ("gate AI task notifications on agent ready state")

That commit modified TestTasksNotification and the related notification gating logic in coderd/workspaceagents.go, so assigning to the author of that change for triage.

Related Issues

• Not a duplicate of: flake: TestTasks/Logs/UpstreamError #1067 (different test: TestTasks/Logs/UpstreamError)