Repository navigation
Conversation
ci/test-duration-budget.toml listed every test that took over 1.33 s in any of 24 CI runs, so that an unlisted test kept a third of the limit as headroom. 170 of the 449 never reached 1.8 s in any run that had them, at least 19 of 28 runs. Four of those runs came after the list was made, and none of the 170 reached 2 s in them; the slowest took 1.73 s. This removes their exceptions, so they are held to the 2 s limit like any other test, and lists a test from 1.8 s. Replaying the new budget over the 28 runs fails only the five tests the earlier changes in this stack make fast. 279 exceptions remain.
|
No actionable comments were generated in the recent review. Walkthrough
Merge Risk:
|
Summary
This removes 170 of the 449 exceptions in
ci/test-duration-budget.toml. The tests they covered already run under the 2 s limit, so from now on the limit applies to them like any other test. 279 exceptions remain.#1408 listed every test that took over 1.33 s in any of 24 CI runs, so that an unlisted test kept a third of the limit as headroom. That threshold listed many tests that never come near 2 s:
This PR keeps listing tests from 1.8 s. It removes an exception when the test stayed under 1.8 s in every run that had it, which is at least 19 of 28 runs. Four of those runs came after #1408's list was made: the Rust tests jobs of #1407, #1408 and #1414. None of the 170 reached 2 s in them; the slowest took 1.73 s.
The budget's header now states the 1.8 s rule.
Stack
Validation
Severity and risk
Severity: does not apply; this is CI policy.
Risk: an unlisted test now has about 10% headroom under 2 s instead of 33%. A test that sometimes runs near 2 s on a loaded runner could fail the check. The four later runs show no such case among the 170. If one appears, the fix is to make the test faster; the header still forbids new exceptions. To roll back, revert this PR.
AI assistance: Claude Code drafted the change and this description.
🤖 Generated with Claude Code
Summary by CodeRabbit