#KeyPerf collector not respecting latency_io_reqd?

1 messages · Page 1 of 1 (latest)

warm trail
#

I upgraded to 26.05.0 which is my first release where KeyPerf is being used for volume latency. I'm now seeing really spiky latency on my dashboards from volumes that are effectively idle. Under defaults in harvest.yml I specify:

latency_io_reqd: 500

But I'm still seeing latency reported on volumes with 1 or less IOs. Am I missing something, or does KeyPerf not respect that parameter?

opaque bane
#

@warm trail This setting needs to be done in the default.yaml of the relevant collector as in screenshot. The KeyPerf collector supports this parameter.

Please note that the value you set gets multiplied by the collection interval. For example, if you set the value to 500 and the polling interval is 60 seconds, then the latency calculation will be skipped for IOPS less than 500 * 60 = 30,000. Please set this value accordingly.

image.png
warm trail
#

Got it, thanks! It wasn't clear to me from the documentation how to specify it within the collector's default.yaml. I'll try your syntax

opaque bane
#

This setting was changed from 10 to 0 in release 25.11 as it didn't seem needed. Are you saying that you see a spike in latency when IOPS are low? What sort of spikes are we talking about?