[jira] [Commented] (HDDS-8902) Close open container immediately on ICR of unhealthy replica

["Stephen O'Donnell (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HDDS-8902?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17757954#comment-17757954 ] 

Stephen O'Donnell commented on HDDS-8902:
-----------------------------------------

I think this makes sense. Infact the code change is very small to do it. This only applies to containers in the OPEN state, and we really want them to transition to and get handled as normal. Inside AbstractContainerReportHandler, the current code looks like:

{code}
    switch (container.getState()) {
    case OPEN:
      /*
       * If the state of a container is OPEN, datanodes cannot report
       * any other state.
       */
      if (replica.getState() != State.OPEN) {
        logger.warn("Container {} is in OPEN state, but the datanode {} " +
            "reports an {} replica.", containerId,
            datanode, replica.getState());
        // Should we take some action?
      }
      break;
{code}

Notice that someone added a comment thinking about taking action, but never followed up.

I think we can simply add a call like:

{code}
        containerManager.updateContainerState(containerId,
            LifeCycleEvent.FINALIZE);
{code}

And then SCM will stop allocating to the container, and RM will clean things up by sending any close commands etc on its next run.

> Close open container immediately on ICR of unhealthy replica
> ------------------------------------------------------------
>
>                 Key: HDDS-8902
>                 URL: https://issues.apache.org/jira/browse/HDDS-8902
>             Project: Apache Ozone
>          Issue Type: Bug
>            Reporter: Ethan Rose
>            Assignee: Siddhant Sangwan
>            Priority: Major
>
> Currently when a datanode's replica moves from open to unhealthy, it will send an ICR to SCM. This is processed on the next run of the SCM replication manager, which could be up to 5 minutes by default. In the mean time, SCM will continue to send writes to this open container, which will fail at the datanode. The unhealthy replica could be handled quicker if the container is closed when the unhealthy ICR is processed in [AbstractContainerReportHandler#updateState|https://github.com/apache/ozone/blob/0fcfe212e18efc620260f733b14db43e63e2ea08/hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/container/AbstractContainerReportHandler.java#L262]



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@ozone.apache.org
For additional commands, e-mail: issues-help@ozone.apache.org