You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "胡振宇 (Jira)" <ji...@apache.org> on 2021/09/02 06:33:00 UTC
[jira] [Updated] (SPARK-36584) ExecutorMonitor#onBlockUpdated will
receive event from driver
[ https://issues.apache.org/jira/browse/SPARK-36584?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
胡振宇 updated SPARK-36584:
------------------------
Description:
When driver broadcast object, it will send the [SparkListenerBlockUpdated|[https://github.com/apache/spark/blob/df0ec56723f0b47c3629055fa7a8c63bb4285147/core/src/main/scala/org/apache/spark/scheduler/SparkListener.scala#L228]] event. [[ExecutorMonitor#onBlockUpdated|#onBlockUpdated]|[https://github.com/apache/spark/blob/df0ec56723f0b47c3629055fa7a8c63bb4285147/core/src/main/scala/org/apache/spark/scheduler/dynalloc/ExecutorMonitor.scala#L380]] receives and handles the event, in this method,
it calls [ensureExecutorIsTracked|[https://github.com/apache/spark/blob/df0ec56723f0b47c3629055fa7a8c63bb4285147/core/src/main/scala/org/apache/spark/scheduler/dynalloc/ExecutorMonitor.scala#L489]]
to put driver in `executors` variable with `UNKNOWN_RESOURCE_PROFILE_ID`. In my understanding, `ExecutorMonitor` should only monitor Executor. Although this will not cause any problems at the moment because UNKNOWN_RESOURCE_PROFILE_ID will be filtered out, but I think this is a potential risk
was:
When driver broadcast object,
it will send the [SparkListenerBlockUpdated](https://github.com/apache/spark/blob/df0ec56723f0b47c3629055fa7a8c63bb4285147/core/src/main/scala/org/apache/spark/scheduler/SparkListener.scala#L228) event.
[ExecutorMonitor#onBlockUpdated](https://github.com/apache/spark/blob/df0ec56723f0b47c3629055fa7a8c63bb4285147/core/src/main/scala/org/apache/spark/scheduler/dynalloc/ExecutorMonitor.scala#L380)
receives and handles the event, in this method,
it calls [ensureExecutorIsTracked](https://github.com/apache/spark/blob/df0ec56723f0b47c3629055fa7a8c63bb4285147/core/src/main/scala/org/apache/spark/scheduler/dynalloc/ExecutorMonitor.scala#L489)
to put driver in `executors` variable, but in my understanding, `ExecutorMonitor` should only monitor Executor, not Driver.
Moreover, adding a `driver` to the `executors` will affect the calculation of
[ExecutorAllocationManager#removeExecutors](https://github.com/apache/spark/blob/df0ec56723f0b47c3629055fa7a8c63bb4285147/core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala#L552), and the driver will occupy the count of `executors`
Issue Type: Question (was: Bug)
Priority: Minor (was: Major)
> ExecutorMonitor#onBlockUpdated will receive event from driver
> -------------------------------------------------------------
>
> Key: SPARK-36584
> URL: https://issues.apache.org/jira/browse/SPARK-36584
> Project: Spark
> Issue Type: Question
> Components: Spark Core
> Affects Versions: 3.1.2
> Environment: Spark 3.1.2
> Reporter: 胡振宇
> Priority: Minor
> Original Estimate: 1h
> Remaining Estimate: 1h
>
> When driver broadcast object, it will send the [SparkListenerBlockUpdated|[https://github.com/apache/spark/blob/df0ec56723f0b47c3629055fa7a8c63bb4285147/core/src/main/scala/org/apache/spark/scheduler/SparkListener.scala#L228]] event. [[ExecutorMonitor#onBlockUpdated|#onBlockUpdated]|[https://github.com/apache/spark/blob/df0ec56723f0b47c3629055fa7a8c63bb4285147/core/src/main/scala/org/apache/spark/scheduler/dynalloc/ExecutorMonitor.scala#L380]] receives and handles the event, in this method,
> it calls [ensureExecutorIsTracked|[https://github.com/apache/spark/blob/df0ec56723f0b47c3629055fa7a8c63bb4285147/core/src/main/scala/org/apache/spark/scheduler/dynalloc/ExecutorMonitor.scala#L489]]
> to put driver in `executors` variable with `UNKNOWN_RESOURCE_PROFILE_ID`. In my understanding, `ExecutorMonitor` should only monitor Executor. Although this will not cause any problems at the moment because UNKNOWN_RESOURCE_PROFILE_ID will be filtered out, but I think this is a potential risk
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org