IT: fix flaky HeartbeatTests.HeartbeatFailed test#424
Conversation
d6c3616 to
e88160b
Compare
There was a problem hiding this comment.
Explanations from the cover letter and the commit message are different, although not mutually exclusive. I'm confused.
Cover letter mentions completing TCP handshakes to SIGSTOP'ed nodes, while the commit message mentions additional connections opened asynchronously to shards.
e88160b to
4fd4dc2
Compare
Use 1 connection per host instead of the default per-shard pool to avoid asynchronous shard-aware connection establishment racing with the baseline connection count read. Fixes #395
4fd4dc2 to
10bcf2e
Compare
@dkropachev can you explain? |
Forgot to update it, I have changed code twice since PR description was created. |
Summary
HeartbeatTests.Integration_Cassandra_HeartbeatFailedtest by using 1 connection per host (with_core_connections_per_host(1)) instead of the default per-shard pooltotal_connectionscount read, making the count non-deterministic and the assertion flakyFixes #395
Test plan
HeartbeatTests.Integration_Cassandra_HeartbeatFailedrepeatedly to verify it no longer flakes