yugabyte-db
yugabyte-db copied to clipboard
[flaky test] Several org.yb.loadtester.* test suites
Jira Link: DB-3388
Description
org.yb.loadtester.TestClusterWithHighLoadAndSlowSync.testClusterFullMoveWithHighLoadAndSlowSync
org.yb.loadtester.TestFullMoveWithHeartBeatDelay.testClusterFullMoveWithHeartbeatDelay
org.yb.loadtester.TestMasterLeaderDecommission.testMasterLeaderDecommission
org.yb.loadtester.TestClusterFullMove.testClusterFullMove
org.yb.loadtester.TestRF1Cluster.testDefaultTTLWithChangedRF
org.yb.loadtester.TestRF1Cluster.testRF1toRF3
Looks like they all significantly increased the failure rate recently -- regression was introduced in this range for all of them: https://github.com/yugabyte/yugabyte-db/compare/34ca65fd605a95173ed7ffabdf2ee14e86757585~...c1d676580933d34bd34fcbb7e86099dcd9e0eb99
From the detective diff results looks like this diff might be the cause: https://github.com/yugabyte/yugabyte-db/commit/9c8e831302822a18170ed5f7c5a4350fb4472f32
This one seems to have been failing indeed for some time
https://detective-gcp.dev.yugabyte.com/stability/test?branch=master&build_type=all&class=org.yb.loadtester.TestClusterWithHighLoadAndSlowSync&fail_tag=all&name=testClusterFullMoveWithHighLoadAndSlowSync&platform=linux
There was a proposed change deadlock_detector.cc . However, it seems no current test goes through that code path.
Edit: I am looking closer at the tests since the test stability may have improved due to infra changes.
Not showing up in top 20% failure rate anymore