Skip to content

[core] avoid etcd updates when AWs do not have preemption condition generated #632

@asm582

Description

@asm582

Describe the Bug

AWs were sending etcd updates for every run of the preemption thread. The update should only be send the AWs have preempt condition. the side effect of this is that the current AW will be sent to back-off queue and wont be dispatched for next 20 seconds (default) backoff time.

Codeflare Stack Component Versions

Please specify the component versions in which you have encountered this bug.

Codeflare SDK:
MCAD: Release 1.34.1 and main

Steps to Reproduce the Bug

Submit an AW that takes a very long time to spawn pods, the AW will get preempted and will be sent to the backoff queue.

What Have You Already Tried to Debug the Issue?

Same as above

Expected Behavior

AWs should only be preempted and backed off when they have a valid preemption condition.

Screenshots, Console Output, Logs, etc.

NA

Affected Releases

Release 1.34.1

Additional Context

NA

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    Status

    No status

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions