CONTAP-519736: synchronous SnapMirror High CIFS IO latency after upgrading to 9.14.1
Issue
- Increase in latency reported on multiple volumes with synchronous snap mirror running and Oplock enabled on the CIFS shares impacted.
- The latency reported can reach upward of 2-3 seconds in Graphana or AIQUM but not reported in
QOS statistic volume latency show
command. - Performance archives data covering the issue shows the high latency in the workload graph view with low IOPs and high concurrency on other IO.
- packet traces captured during the issue shows high SRT's from the storage with File Lock Break not registering on the clients.
- Multiple
repl.switch.to.seq.split
events with high operation latency reported in EMS during the issue:
Mon Aug 11 08:07:19 -0400 [Node-Name: repl_Handle_sync_resl: repl.switch.to.seq.split:notice]: Operation Latency has exceeded the threshold value "2890" in the parallel split mode for sync relationship with UUID "11111111111111111".
- The latency drops down to expected values when synchronous SnapMirror of the volume impacted is disabled.