[
https://issues.apache.org/jira/browse/SPARK-49783?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Junfan Zhang updated SPARK-49783:
---------------------------------
Description:
When the target < running containers number, the assigned containers from the
resource manager will be skipped, but these containers are not released by
invoking the amClient.releaseAssignedContainer , that will make these
containers reserved into the Yarn resourceManager at least 10 minutes. And so,
the cluster resource will be wasted at a high ratio.
And this will reflect that the vcore * seconds statistics from yarn side will
be greater than the result from the spark event logs.
>From my statistics, the cluster resource waste ratio is ~25% if the spark jobs
>are exclusive in this cluster.
The more details could be found in this blog:
[https://zuston.vercel.app/publish/resource-leak-of-spark-yarn-allocator/]
was:
When the target < running containers number, the assigned containers from the
resource manager will be skipped, but these containers are not released by
invoking the amClient.releaseAssignedContainer , that will make these
containers reserved into the Yarn resourceManager at least 10 minutes. And so,
the cluster resource will be wasted at a high ratio.
Another problem will happen that the vcore * seconds statistics from yarn side
will be greater than the result from the spark event logs.
>From my statistics, the cluster resource waste ratio is ~25% if the spark jobs
>are exclusive in this cluster.
The more details could be found in this blog:
https://zuston.vercel.app/publish/resource-leak-of-spark-yarn-allocator/
> Resource leak of spark yarn allocator
> -------------------------------------
>
> Key: SPARK-49783
> URL: https://issues.apache.org/jira/browse/SPARK-49783
> Project: Spark
> Issue Type: Bug
> Components: YARN
> Affects Versions: 3.5.0, 4.0.0
> Reporter: Junfan Zhang
> Priority: Major
> Labels: pull-request-available
>
> When the target < running containers number, the assigned containers from the
> resource manager will be skipped, but these containers are not released by
> invoking the amClient.releaseAssignedContainer , that will make these
> containers reserved into the Yarn resourceManager at least 10 minutes. And
> so, the cluster resource will be wasted at a high ratio.
> And this will reflect that the vcore * seconds statistics from yarn side will
> be greater than the result from the spark event logs.
> From my statistics, the cluster resource waste ratio is ~25% if the spark
> jobs are exclusive in this cluster.
>
> The more details could be found in this blog:
> [https://zuston.vercel.app/publish/resource-leak-of-spark-yarn-allocator/]
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
