A build on main failed in its test job, while every platform build passed. A test checked too early that stopping an agent also stops the command it runs: on a busy CI machine, the command can still be around for a moment after the agent itself has exited. The test now waits briefly for it to go and still fails if it really keeps running.
A flaky CI test made reliable
Shared by Daniel Sadilek on October 21, 2026