zmodem wrote: I wish we could make the tests actually non-flaky instead, maybe by relaxing the expectations. Re-running sounds like it just changes the flakiness rate. https://github.com/llvm/llvm-project/pull/138613