Craig Condit created YUNIKORN-2915:
--------------------------------------
Summary: Scheduling latency metric should update even if no
scheduling occurs
Key: YUNIKORN-2915
URL: https://issues.apache.org/jira/browse/YUNIKORN-2915
Project: Apache YuniKorn
Issue Type: Improvement
Components: core - scheduler
Reporter: Craig Condit
Assignee: Craig Condit
The scheduler metric scheduling_latency_milliseconds is currently only updated
if an allocation actually occurs. This is not particularly useful, as latency
could be quite long but in the case where no scheduling was possible after
traversing all queues, no reporting is done, so the visible latency is 0. This
makes the metric difficult to use for analysis, as in a busy cluster,
scheduling can be very slow, but not show up on the histogram at all.
We should move the reporting of scheduling latency outside the check for an
allocation and report even scheduling runs with no result.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
