Description of problem: In case of etcd running on nodes with slow disks (IO/fsync), higher count of leader changes were spotted, that potentially disrupt the API server and other components trying to write/read from etcd. While several tweaks to etcd were made to mitigate this problem, we need better reporting from etcd operator that will provide detailed information about `etcd_disk_wal_fsync_duration_seconds_bucket` in case of leader changes are excessive. Version-Release number of selected component (if applicable): How reproducible: When leader changes are excessive (high count), we don't report any disk metrics in operator. Steps to Reproduce: (this is hard) 1. Make etcd perform leader changes (several in 5 minutes) Actual results: The alert will be fired for leader changes, but no info is given back to admins about disk usage of node. Expected results: When the leader changes occurs, the operator should fire a warning event automatically: Warningf("EtcdLeaderChangeMetrics", "Detected %s leader changes in last 5 minutes on %q disk metrics are: %s", leaderChanges, platformType, strings.Join(values, ",")) Additional info:
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://access.redhat.com/errata/RHBA-2020:2409