flake: TestResponsesInjectedTool/blocking_tool_error

[flake-investigator]

CI Failure Details

CI Run: https://github.com/coder/coder/actions/runs/24848334727
Failed Job: https://github.com/coder/coder/actions/runs/24848334727/job/72744223912 (test-go-pg-17)
Commit: dbcc654d289f2dd2ed68ba737e635e1291224433 (Michael Suchacz)
Date: 2026-04-23

Failing Test

TestResponsesInjectedTool/blocking_tool_error in aibridge/internal/integrationtest/responses_test.go (around line 770).

Error Output

=== FAIL: aibridge/internal/integrationtest TestResponsesInjectedTool/blocking_tool_error (0.01s)
responses_test.go:949:
    Error:       Not equal:
    expected: recorder.TokenUsageRecord{MsgID:"resp_06e2afba24b6b2ad00696b774d1df0819eaf1ec802bc8a2ca9", Input:233, Output:119, ExtraTokenTypes:{"output_reasoning":70,"total_tokens":6496}}
    actual  : recorder.TokenUsageRecord{MsgID:"resp_06e2afba24b6b2ad00696b775044e8819ea14840698ef966e2", Input:395, Output:144, ExtraTokenTypes:{"output_reasoning":28,"total_tokens":6683}}

Error Analysis

The test asserts two token usage records for the blocking tool-error flow. In this run the first recorded token-usage entry mismatched the expected values (MsgID and token counts), indicating that token usage recording is non-deterministic or the fixture’s expected usage does not match the recorded response.

No data race warnings, panics, or OOM indicators were observed in the job logs.

Root Cause Assessment

Likely flaky test due to token usage record ordering/values not being stable in the injected-tool response flow, or the recorded usage not matching the static fixture expectations.

Assignment Analysis

git log --oneline -10 --follow aibridge/internal/integrationtest/responses_test.go shows the most recent meaningful change as:

e00e8576 chore: move aibridge library code into coder repo (Paweł Banaszewski)

Assigning to @pawbana as the most recent maintainer of this test file. (Line-level blame not available in current tooling.)

Related Issues

• flake: TestResponsesOutputMatchesUpstream (streaming_conversation) #1490 (other aibridge responses test flake; different test case)

Reproduction

go test ./aibridge/internal/integrationtest -run TestResponsesInjectedTool/blocking_tool_error -count=1