[
https://issues.apache.org/jira/browse/YUNIKORN-3131?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Peter Bacsko updated YUNIKORN-3131:
-----------------------------------
Description:
When the daemonset preemptor cannot find a victim pod, this is only visible in
the state dump.
To enhance supportability, we can:
# Generate a message just like we removed in YUNIKORN-2980, but limit the
amount of message that goes to the console with a rate limited logger (also
print out how many pods are part of a daemon set and how many have higher
priority)
# -Send a K8s event to the daemon set pod that preemption does not help.-
(update: this was introduced in YUNIKORN-
When the preemption takes place, it makes sense to send a K8s event to the
victim pod. For regular preemption, we already do this (see YUNIKORN-2982).
was:
When the daemonset preemptor cannot find a victim pod, this is only visible in
the state dump.
To enhance supportability, we can:
# Generate a message just like we removed in YUNIKORN-2980, but limit the
amount of message that goes to the console with a rate limited logger (also
print out how many pods are part of a daemon set and how many have higher
priority)
# Send a K8s event to the daemon set pod that preemption does not help.
When the preemption takes place, it makes sense to send a K8s event to the
victim pod. For regular preemption, we already do this (see YUNIKORN-2982).
> Improve DaemonSet preemption diagnostic messages
> ------------------------------------------------
>
> Key: YUNIKORN-3131
> URL: https://issues.apache.org/jira/browse/YUNIKORN-3131
> Project: Apache YuniKorn
> Issue Type: Improvement
> Components: core - scheduler
> Reporter: Peter Bacsko
> Assignee: Peter Bacsko
> Priority: Major
>
> When the daemonset preemptor cannot find a victim pod, this is only visible
> in the state dump.
> To enhance supportability, we can:
> # Generate a message just like we removed in YUNIKORN-2980, but limit the
> amount of message that goes to the console with a rate limited logger (also
> print out how many pods are part of a daemon set and how many have higher
> priority)
> # -Send a K8s event to the daemon set pod that preemption does not help.-
> (update: this was introduced in YUNIKORN-
> When the preemption takes place, it makes sense to send a K8s event to the
> victim pod. For regular preemption, we already do this (see YUNIKORN-2982).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
