flake: TestOpenAIReasoningWithWebSearchRoundTripStoreFalse

[flake-investigator]

CI Run

• https://github.com/coder/coder/actions/runs/23734771864
• Failed job: test-go-pg (ubuntu-latest)

Commit

• SHA: bc5e4b5d54a399427cf87863dcaf5126051f07d0
• Author: Atif Ali

Failure

=== FAIL: coderd/x/chatd TestOpenAIReasoningWithWebSearchRoundTripStoreFalse (0.75s)
    integration_test.go:856:
        Error:       Not equal:
                     expected: 2
                     actual  : 1
        Messages:    expected exactly two streamed OpenAI responses
    t.go:111: ... coderd.chatd.processor: chat interrupted ...

Error Analysis

The test expects two streaming OpenAI requests (initial + follow-up). The failure indicates only one streaming request was captured (streamRequestCount == 1). Logs show the follow-up stream ended early with coderd.chatd.processor: chat interrupted, suggesting a timing/race in the follow-up streaming path for store=false behavior.

Root Cause

Likely flaky test in the OpenAI reasoning + web search round-trip follow-up flow. The follow-up stream occasionally does not trigger a second streaming request in the test harness.

Assignment Analysis

• git log --oneline --follow coderd/x/chatd/integration_test.go shows the failing test was added in commit 02356c61 by Michael Suchacz (added the OpenAI store=true/false round-trip tests).
• Latest commit 3220d1d5 only changes env var names and does not affect this assertion.

Assigning to Michael Suchacz as the most recent non-trivial modifier of the failing test.

Related Issues

• Searched: TestOpenAIReasoningWithWebSearchRoundTripStoreFalse, TestOpenAIReasoningWithWebSearchRoundTrip, expected exactly two streamed OpenAI responses, coderd/x/chatd/integration_test.go, chatd "web search"
• Found: Rename OPENAI_API_KEY to CODER_TEST_OPENAI_API_KEY in integration tests and improve test strategy #1425 (env var rename, not this failure)

Reproduction (if needed)

go test ./coderd/x/chatd -run TestOpenAIReasoningWithWebSearchRoundTripStoreFalse -count=1