[
https://issues.apache.org/jira/browse/HADOOP-18639?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sejin Hwang updated HADOOP-18639:
---------------------------------
Description:
YARN NodeManager's deletion service has two types of deletion tasks: the
FileDeletionTask for deleting log files and the DockerContainerDeletionTask for
deleting Docker containers.
The FileDeletionTask is removed from the statestore when the task is completed,
but the DockerContainerDeletionTask is not.
Therefore, the DockerContainerDeletionTask accumulates continuously in the
statestore.
This causes the NodeManager's deletion service to run the accumulated
DockerContainerDeletionTask in the statestore when the NodeManager restarts.
As a result, the FileDeletionTask and DockerContainerDeletionTask are delayed
unnecessarily while processing accumulated tasks, which can cause disk full
issues in environments where a large number of containers are allocated and
released.
I will attach a patch soon
was:
YARN NodeManager has two types of deletion tasks: the FileDeletionTask for
deleting log files and the DockerContainerDeletionTask for deleting Docker
containers.
The FileDeletionTask is removed from the statestore when the task is completed,
but the DockerContainerDeletionTask is not.
Therefore, the DockerContainerDeletionTask accumulates continuously in the
statestore.
This causes the NodeManager's deletion service to run the accumulated
DockerContainerDeletionTask in the statestore when the NodeManager restarts.
As a result, the FileDeletionTask and DockerContainerDeletionTask are delayed
unnecessarily while processing accumulated tasks, which can cause disk full
issues in environments where a large number of containers are allocated and
released.
I will attach a patch soon
> DockerContainerDeletionTask is not removed from the Nodemanager's statestore
> when the task is completed.
> --------------------------------------------------------------------------------------------------------
>
> Key: HADOOP-18639
> URL: https://issues.apache.org/jira/browse/HADOOP-18639
> Project: Hadoop Common
> Issue Type: Bug
> Reporter: Sejin Hwang
> Priority: Major
> Labels: pull-request-available
>
> YARN NodeManager's deletion service has two types of deletion tasks: the
> FileDeletionTask for deleting log files and the DockerContainerDeletionTask
> for deleting Docker containers.
>
> The FileDeletionTask is removed from the statestore when the task is
> completed, but the DockerContainerDeletionTask is not.
> Therefore, the DockerContainerDeletionTask accumulates continuously in the
> statestore.
>
> This causes the NodeManager's deletion service to run the accumulated
> DockerContainerDeletionTask in the statestore when the NodeManager restarts.
> As a result, the FileDeletionTask and DockerContainerDeletionTask are delayed
> unnecessarily while processing accumulated tasks, which can cause disk full
> issues in environments where a large number of containers are allocated and
> released.
> I will attach a patch soon
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
