Rerun
gstack 1.91.163.0 reruns a failed eval test once instead of the whole gate
When one test turns gstack's eval gate red, the gate now reruns just that test, once. If it passes, the gate goes green and marks the test as flaky.
Twice
A test that fails its rerun is a real red
The rerun starts fresh at the same commit, and a behavior test reruns its whole three-trial panel. If the test fails a second time, the report calls it failed twice, and the gate stays red.
By hand
Rerunning failed jobs by hand follows the same rule
Rerunning failed jobs by hand follows the same rule. An attempt that passes every earlier red is green. A red that first appears in a later attempt stays red until its own rerun passes.
Reuse
The next push reuses a flaky green instead of paying for a full gate
Before this release, a push after a flaky run still paid for a full gate. Now the next push reuses that green result, and the log and the pull request comment name the flaky tests.
Limits
Some reds never get a rerun
CHANGELOG.mdthe full rerun rulesHarness failures, infrastructure failures and incomplete runs never rerun, and thresholds and trial counts stay the same. Which test flakes most often in your setup?
