[
https://issues.apache.org/jira/browse/YUNIKORN-513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17278024#comment-17278024
]
Kinga Marton commented on YUNIKORN-513:
---------------------------------------
[~rozhang], the steps during recovery is the following one:
* When the application is created in the core side, it will have the New state
* During the node recovery we recover the existing allocations as well.
* When the first allocation is recovered, the application will transit into
Starting state through the Accepted one. So if there is only one allocation to
recover the application will stay in the Starting state until it times out and
will auto-progress into the Running one
* When the second allocation is recovered, than the application will transit
into Running state
Actually the expected behaviour is exactly the same as in case of apps and
tasks submission. So in case of a 2-allocation application, (if both pods are
in running state when the recovery happens) is expected to be in Running state.
However I checked it now and it seems to be broken again. I think now the
locking has some issues. I suspect a deadlock, so I will reopen this issue.
[~rozhang] have you encountered any issues as well?
> ApplicationState remains in Accepted after recovery
> ---------------------------------------------------
>
> Key: YUNIKORN-513
> URL: https://issues.apache.org/jira/browse/YUNIKORN-513
> Project: Apache YuniKorn
> Issue Type: Bug
> Components: core - cache
> Affects Versions: 0.10
> Reporter: Kinga Marton
> Assignee: Kinga Marton
> Priority: Major
> Labels: pull-request-available
> Fix For: 0.10
>
>
> Steps to reproduce:
> * Start 2 sleep jobs
> * Wait for both to run and applicationState to be Running
> * Kill yunikorn
> * After 10 minutes, the rest call now shows both applicationState as
> accepted instead of running
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
