Boundary Condition Considerations in Gang Plugin's OnSessionClose #4011
Labels
area/performance
Issues or PRs related to performance
kind/feature
Categorizes issue or PR as related to a new feature.
Comments
Wang-Kai
added
the
kind/feature
6 tasks
|
/area performance |
Wang-Kai
added a commit
to Wang-Kai/volcano
that referenced
this issue
Feb 14, 2025
Signed-off-by: 王凯 <[email protected]>
Wang-Kai
added a commit
to Wang-Kai/volcano
that referenced
this issue
Feb 17, 2025
Signed-off-by: 王凯 <[email protected]>
mahdikhashan
added a commit
to mahdikhashan/volcano
that referenced
this issue
Feb 21, 2025
Signed-off-by: mahdikhashan <[email protected]> skip unavailable jobs during the close session step volcano-sh#4011 Signed-off-by: 王凯 <[email protected]>
mahdikhashan
pushed a commit
to mahdikhashan/volcano
that referenced
this issue
Feb 24, 2025
Signed-off-by: 王凯 <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
What is the problem you're trying to solve
The Gang Plugin's OnSessionClose updates the conditions of the PodGroup (PG) based on whether the current scheduling Job is Ready. In certain scenarios, I believe this behavior is unnecessary.
Scenario:
When a Volcano Job (VJ) is deleted, the PodGroup (PG) and Pods are cascadingly deleted. Since the Pod deletion event is observed earlier than the PG deletion event,
vj.pg != nilwhilelen(vj.tasks) == 0at this point. This causes the Gang Plugin to unnecessarily update the PG conditions in this scenario.Describe the solution you'd like
Proposed Improvement:
The fundamental premise of Gang scheduling is that there is at least one Pod to schedule. If vj.tasks is empty (i.e., no tasks require scheduling), the plugin should skip updating the PG conditions.
Additional context
No response
The text was updated successfully, but these errors were encountered: