flake: TestSubscribeRelayReconnectsOnDrop

[flake-investigator]

New occurrence (2026-03-30):

• CI Run: https://github.com/coder/coder/actions/runs/23723910631
• Commit: f7aa46c4ba45ad75d74939531a5bbf24d9a27d4f (Ethan)
• Failed Job: test-go-pg-17
• Additional failing test: TestSubscribeRelayMultipleReconnects
• Error:
  • processor: failed to process chat ... error="stream response: unauthorized: Incorrect API key provided: test-key. You can find your API key at https://platform.openai.com/account/api-keys."
  • Condition never satisfied (chatd_test.go:239, :1285)

---

CI Run: https://github.com/coder/coder/actions/runs/23696295673
Commit: bfeb91d9cdce782bdb3721ed52f09fb34f5a489c (Michael Suchacz)
Failed Jobs: test-go-pg (ubuntu-latest), test-go-pg-17

Failure:

FAIL: enterprise/coderd/x/chatd TestSubscribeRelayReconnectsOnDrop
chatd_test.go:239: Condition never satisfied
processor: failed to process chat ... error="stream response: unauthorized: Incorrect API key provided: test-key. You can find your API key at https://platform.openai.com/account/api-keys."

Log snippet (test-go-pg ubuntu-latest):

=== FAIL: enterprise/coderd/x/chatd TestSubscribeRelayReconnectsOnDrop (15.35s)
    chatd_test.go:239:
        Error:      Condition never satisfied
    ...
    processor: failed to process chat  chat_id=20f4fe49-ef0d-4a46-a3c4-c8a123667881  error="stream response: unauthorized: Incorrect API key provided: test-key. You can find your API key at https://platform.openai.com/account/api-keys."

Error Analysis:

• Test expects relay reconnection to deliver a second message part after a simulated drop.
• The relay did not deliver the second event before timeout, and logs show the chat processor returned an OpenAI auth error for the test model ("test-key"). This suggests the reconnect path may be interrupted by provider auth errors or timing around relay/publisher handling.
• No panic/OOM/data race indicators observed in the job logs.

Root Cause Assessment: Flaky test / relay reconnect timing + provider error handling (OpenAI auth error surfaced during test) causing the expected second relay part not to arrive.

Assignment Analysis:

• git blame -L 145,230 enterprise/coderd/x/chatd/chatd_test.go
• Recent file history: 4d746030 (Ethan) "fix(coderd/x/chatd): respect provider Retry-After headers in chat retry loop" (most recent meaningful change touching the test file).

Assignee: @ethanndickson

Duplicate Search:

• "TestSubscribeRelayReconnectsOnDrop"
• "Incorrect API key provided"
• "chatd_test.go"
• "SubscribeRelay" / "relay reconnect"
  No existing issues found.

Reproduction:

• go test ./enterprise/coderd/x/chatd -run TestSubscribeRelayReconnectsOnDrop -count=1
  (Requires DB + pubsub test harness).