[
https://issues.apache.org/jira/browse/YUNIKORN-2930?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Craig Condit updated YUNIKORN-2930:
-----------------------------------
Target Version: 1.8.0 (was: 1.7.0)
> Enhance Metrics for Scheduling Cycle Latency
> --------------------------------------------
>
> Key: YUNIKORN-2930
> URL: https://issues.apache.org/jira/browse/YUNIKORN-2930
> Project: Apache YuniKorn
> Issue Type: Improvement
> Components: core - scheduler
> Reporter: Mit Desai
> Assignee: Mit Desai
> Priority: Major
>
> Update scheduler metrics to provide a comprehensive view of scheduling cycle
> latency, including cycles where no pods were scheduled and more granular
> details on node evaluation times.
> The current metrics for the scheduler do not provide a complete picture.
> Specifically:
> * {*}Scheduling Latency{*}: Only shows the latency for pods that were
> scheduled. If no pod was scheduled in a cycle, the latency metric is not
> emitted.
> * {*}tryNodeLatency{*}: Shows the time taken to evaluate a single node, but
> does not capture the total time taken to find a node for a pod.
> To improve the metrics we should add additional metrics:
> # {*}Scheduling Cycle Metric{*}: Emit a metric for each scheduling cycle,
> regardless of whether a pod was scheduled. This metric will show the total
> time taken for each scheduling cycle.
> # {*}Node Evaluation Metric{*}: Add a new metric that shows the total time
> taken to find a node for a given pod. For instance, if a pod was scheduled
> after evaluating five nodes, this new metric will be the sum of the times for
> those five nodes.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
