[jira] [Commented] (SPARK-20286) dynamicAllocation.executorIdleTimeout is ignored after unpersist
[
https://issues.apache.org/jira/browse/SPARK-20286?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16770344#comment-16770344
]
t oo commented on SPARK-20286:
--
gentle ping
> dynamicAllocation.executorIdleTimeout is ignored after unpersist
>
>
> Key: SPARK-20286
> URL: https://issues.apache.org/jira/browse/SPARK-20286
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
>Affects Versions: 2.0.1
>Reporter: Miguel Pérez
>Priority: Major
>
> With dynamic allocation enabled, it seems that executors with cached data
> which are unpersisted are still being killed using the
> {{dynamicAllocation.cachedExecutorIdleTimeout}} configuration, instead of
> {{dynamicAllocation.executorIdleTimeout}}. Assuming the default configuration
> ({{dynamicAllocation.cachedExecutorIdleTimeout = Infinity}}), an executor
> with unpersisted data won't be released until the job ends.
> *How to reproduce*
> - Set different values for {{dynamicAllocation.executorIdleTimeout}} and
> {{dynamicAllocation.cachedExecutorIdleTimeout}}
> - Load a file into a RDD and persist it
> - Execute an action on the RDD (like a count) so some executors are activated.
> - When the action has finished, unpersist the RDD
> - The application UI removes correctly the persisted data from the *Storage*
> tab, but if you look in the *Executors* tab, you will find that the executors
> remain *active* until ({{dynamicAllocation.cachedExecutorIdleTimeout}} is
> reached.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-20286) dynamicAllocation.executorIdleTimeout is ignored after unpersist
[
https://issues.apache.org/jira/browse/SPARK-20286?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16570754#comment-16570754
]
Apache Spark commented on SPARK-20286:
--
User 'dhruve' has created a pull request for this issue:
https://github.com/apache/spark/pull/22015
> dynamicAllocation.executorIdleTimeout is ignored after unpersist
>
>
> Key: SPARK-20286
> URL: https://issues.apache.org/jira/browse/SPARK-20286
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
>Affects Versions: 2.0.1
>Reporter: Miguel Pérez
>Priority: Major
>
> With dynamic allocation enabled, it seems that executors with cached data
> which are unpersisted are still being killed using the
> {{dynamicAllocation.cachedExecutorIdleTimeout}} configuration, instead of
> {{dynamicAllocation.executorIdleTimeout}}. Assuming the default configuration
> ({{dynamicAllocation.cachedExecutorIdleTimeout = Infinity}}), an executor
> with unpersisted data won't be released until the job ends.
> *How to reproduce*
> - Set different values for {{dynamicAllocation.executorIdleTimeout}} and
> {{dynamicAllocation.cachedExecutorIdleTimeout}}
> - Load a file into a RDD and persist it
> - Execute an action on the RDD (like a count) so some executors are activated.
> - When the action has finished, unpersist the RDD
> - The application UI removes correctly the persisted data from the *Storage*
> tab, but if you look in the *Executors* tab, you will find that the executors
> remain *active* until ({{dynamicAllocation.cachedExecutorIdleTimeout}} is
> reached.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-20286) dynamicAllocation.executorIdleTimeout is ignored after unpersist
[
https://issues.apache.org/jira/browse/SPARK-20286?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15972253#comment-15972253
]
Miguel Pérez commented on SPARK-20286:
--
Thank you! I'll check it again and close the issue if I cannot reproduce the
problem in 2.1.0.
> dynamicAllocation.executorIdleTimeout is ignored after unpersist
>
>
> Key: SPARK-20286
> URL: https://issues.apache.org/jira/browse/SPARK-20286
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
>Affects Versions: 2.0.1
>Reporter: Miguel Pérez
>
> With dynamic allocation enabled, it seems that executors with cached data
> which are unpersisted are still being killed using the
> {{dynamicAllocation.cachedExecutorIdleTimeout}} configuration, instead of
> {{dynamicAllocation.executorIdleTimeout}}. Assuming the default configuration
> ({{dynamicAllocation.cachedExecutorIdleTimeout = Infinity}}), an executor
> with unpersisted data won't be released until the job ends.
> *How to reproduce*
> - Set different values for {{dynamicAllocation.executorIdleTimeout}} and
> {{dynamicAllocation.cachedExecutorIdleTimeout}}
> - Load a file into a RDD and persist it
> - Execute an action on the RDD (like a count) so some executors are activated.
> - When the action has finished, unpersist the RDD
> - The application UI removes correctly the persisted data from the *Storage*
> tab, but if you look in the *Executors* tab, you will find that the executors
> remain *active* until ({{dynamicAllocation.cachedExecutorIdleTimeout}} is
> reached.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-20286) dynamicAllocation.executorIdleTimeout is ignored after unpersist
[
https://issues.apache.org/jira/browse/SPARK-20286?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15972231#comment-15972231
]
Umesh Chaudhary commented on SPARK-20286:
-
Yep, +1 to the UI changes. However, I tested the behaviour by following the
steps mentioned by you and this seems to be fixed in 2.1.0.
{code}
spark2-shell --conf spark.dynamicAllocation.executorIdleTimeout=7s --conf
spark.dynamicAllocation.cachedExecutorIdleTimeout=20s --conf
spark.dynamicAllocation.minExecutors=2
scala> sc.setLogLevel("INFO")
scala> val rdd=sc.textFile("/tmp/config.txt")
..
17/04/17 23:46:58 INFO spark.SparkContext: Created broadcast 0 from textFile at
:24
rdd: org.apache.spark.rdd.RDD[String] = /tmp/config.txt MapPartitionsRDD[1] at
textFile at :24
..
..
scala> rdd.collect
17/04/17 23:47:10 INFO mapred.FileInputFormat: Total input paths to process : 1
17/04/17 23:47:10 INFO spark.SparkContext: Starting job: collect at :27
..
..
17/04/17 23:47:13 INFO scheduler.DAGScheduler: Job 0 finished: collect at
:27, took 3.457215 s
..
scala> 17/04/17 23:47:20 INFO spark.ExecutorAllocationManager: Request to
remove executorIds: 1
{code}
As expected executors remove request was received above after 7 secs. Then I
tested behaviour for persist:
{code}
scala> rdd.persist
res2: rdd.type = /tmp/config.txt MapPartitionsRDD[1] at textFile at :24
scala> rdd.count
17/04/17 23:47:45 INFO spark.SparkContext: Starting job: count at :27
..
17/04/17 23:47:45 INFO scheduler.DAGScheduler: Job 1 finished: count at
:27, took 0.293053 s
..
scala> 17/04/17 23:48:05 INFO spark.ExecutorAllocationManager: Request to
remove executorIds: 1
{code}
In this case also request for removing the executor was received after 20 secs.
Afterwords I tested unpersist:
{code}
scala> rdd.unpersist(true)
17/04/17 23:50:22 INFO rdd.MapPartitionsRDD: Removing RDD 1 from persistence
list
17/04/17 23:50:22 INFO storage.BlockManager: Removing RDD 1
res7: rdd.type = /tmp/config.txt MapPartitionsRDD[1] at textFile at :24
scala> rdd.count
17/04/17 23:50:31 INFO spark.SparkContext: Starting job: count at :27
..
17/04/17 23:50:31 INFO scheduler.DAGScheduler: Job 3 finished: count at
:27, took 0.219764 s
res8: Long = 100
..
scala> 17/04/17 23:50:38 INFO spark.ExecutorAllocationManager: Request to
remove executorIds: 1
{code}
This time remove request was received after 7 secs.
> dynamicAllocation.executorIdleTimeout is ignored after unpersist
>
>
> Key: SPARK-20286
> URL: https://issues.apache.org/jira/browse/SPARK-20286
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
>Affects Versions: 2.0.1
>Reporter: Miguel Pérez
>
> With dynamic allocation enabled, it seems that executors with cached data
> which are unpersisted are still being killed using the
> {{dynamicAllocation.cachedExecutorIdleTimeout}} configuration, instead of
> {{dynamicAllocation.executorIdleTimeout}}. Assuming the default configuration
> ({{dynamicAllocation.cachedExecutorIdleTimeout = Infinity}}), an executor
> with unpersisted data won't be released until the job ends.
> *How to reproduce*
> - Set different values for {{dynamicAllocation.executorIdleTimeout}} and
> {{dynamicAllocation.cachedExecutorIdleTimeout}}
> - Load a file into a RDD and persist it
> - Execute an action on the RDD (like a count) so some executors are activated.
> - When the action has finished, unpersist the RDD
> - The application UI removes correctly the persisted data from the *Storage*
> tab, but if you look in the *Executors* tab, you will find that the executors
> remain *active* until ({{dynamicAllocation.cachedExecutorIdleTimeout}} is
> reached.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-20286) dynamicAllocation.executorIdleTimeout is ignored after unpersist
[
https://issues.apache.org/jira/browse/SPARK-20286?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15969944#comment-15969944
]
Miguel Pérez commented on SPARK-20286:
--
My supposition is that {{onExecutorIdle}} is only called when a task ends, so
it's already idle when you call {{unpersist}}. I'm not sure how to test this
though. Also, it will be great if the UI can show an "idle" status for the
executors. Currently, they're shown as "Active" until they're killed and then
shown as "Dead".
> dynamicAllocation.executorIdleTimeout is ignored after unpersist
>
>
> Key: SPARK-20286
> URL: https://issues.apache.org/jira/browse/SPARK-20286
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
>Affects Versions: 2.0.1
>Reporter: Miguel Pérez
>
> With dynamic allocation enabled, it seems that executors with cached data
> which are unpersisted are still being killed using the
> {{dynamicAllocation.cachedExecutorIdleTimeout}} configuration, instead of
> {{dynamicAllocation.executorIdleTimeout}}. Assuming the default configuration
> ({{dynamicAllocation.cachedExecutorIdleTimeout = Infinity}}), an executor
> with unpersisted data won't be released until the job ends.
> *How to reproduce*
> - Set different values for {{dynamicAllocation.executorIdleTimeout}} and
> {{dynamicAllocation.cachedExecutorIdleTimeout}}
> - Load a file into a RDD and persist it
> - Execute an action on the RDD (like a count) so some executors are activated.
> - When the action has finished, unpersist the RDD
> - The application UI removes correctly the persisted data from the *Storage*
> tab, but if you look in the *Executors* tab, you will find that the executors
> remain *active* until ({{dynamicAllocation.cachedExecutorIdleTimeout}} is
> reached.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-20286) dynamicAllocation.executorIdleTimeout is ignored after unpersist
[
https://issues.apache.org/jira/browse/SPARK-20286?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15969924#comment-15969924
]
Umesh Chaudhary commented on SPARK-20286:
-
While looking at ExecutorAllocationManager.onExecutorIdle, there is a condition
which checks whether executor has CachedBlocks or not , if it has cached blocks
then it uses cachedExecutorIdleTimeoutS and if no cached blocks it uses
executorIdleTimeoutS.
Still not sure why even after calling unpersist it is behaving like this. One
possibility: there might be some cached data on executors which is not reported
to the BlockManager and it is causing executor to follow
cachedExecutorIdleTimeout instead of executorIdleTimeout.
Need some thoughts though cc: [~joshrosen], [~rxin].
> dynamicAllocation.executorIdleTimeout is ignored after unpersist
>
>
> Key: SPARK-20286
> URL: https://issues.apache.org/jira/browse/SPARK-20286
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
>Affects Versions: 2.0.1
>Reporter: Miguel Pérez
>
> With dynamic allocation enabled, it seems that executors with cached data
> which are unpersisted are still being killed using the
> {{dynamicAllocation.cachedExecutorIdleTimeout}} configuration, instead of
> {{dynamicAllocation.executorIdleTimeout}}. Assuming the default configuration
> ({{dynamicAllocation.cachedExecutorIdleTimeout = Infinity}}), an executor
> with unpersisted data won't be released until the job ends.
> *How to reproduce*
> - Set different values for {{dynamicAllocation.executorIdleTimeout}} and
> {{dynamicAllocation.cachedExecutorIdleTimeout}}
> - Load a file into a RDD and persist it
> - Execute an action on the RDD (like a count) so some executors are activated.
> - When the action has finished, unpersist the RDD
> - The application UI removes correctly the persisted data from the *Storage*
> tab, but if you look in the *Executors* tab, you will find that the executors
> remain *active* until ({{dynamicAllocation.cachedExecutorIdleTimeout}} is
> reached.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-20286) dynamicAllocation.executorIdleTimeout is ignored after unpersist
[
https://issues.apache.org/jira/browse/SPARK-20286?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15965556#comment-15965556
]
Miguel Pérez commented on SPARK-20286:
--
I'm zero familiar with the code, but I'll try to send a merge request when I
have time to look it up. I'll be grateful for any kind of help. Thank you.
> dynamicAllocation.executorIdleTimeout is ignored after unpersist
>
>
> Key: SPARK-20286
> URL: https://issues.apache.org/jira/browse/SPARK-20286
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
>Affects Versions: 2.0.1
>Reporter: Miguel Pérez
>
> With dynamic allocation enabled, it seems that executors with cached data
> which are unpersisted are still being killed using the
> {{dynamicAllocation.cachedExecutorIdleTimeout}} configuration, instead of
> {{dynamicAllocation.executorIdleTimeout}}. Assuming the default configuration
> ({{dynamicAllocation.cachedExecutorIdleTimeout = Infinity}}), an executor
> with unpersisted data won't be released until the job ends.
> *How to reproduce*
> - Set different values for {{dynamicAllocation.executorIdleTimeout}} and
> {{dynamicAllocation.cachedExecutorIdleTimeout}}
> - Load a file into a RDD and persist it
> - Execute an action on the RDD (like a count) so some executors are activated.
> - When the action has finished, unpersist the RDD
> - The application UI removes correctly the persisted data from the *Storage*
> tab, but if you look in the *Executors* tab, you will find that the executors
> remain *active* until ({{dynamicAllocation.cachedExecutorIdleTimeout}} is
> reached.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-20286) dynamicAllocation.executorIdleTimeout is ignored after unpersist
[
https://issues.apache.org/jira/browse/SPARK-20286?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15965411#comment-15965411
]
Saisai Shao commented on SPARK-20286:
-
bq. Maybe the best approach is to change from cachedExecutorIdleTimeout to
executorIdleTimeout on all cached executors when the last RDD has been
unpersisted, and then restart the time counter (unpersist will then count as an
action).
Yes, I think this is a feasible solution. I can help out it if you're not
familiar with that code.
> dynamicAllocation.executorIdleTimeout is ignored after unpersist
>
>
> Key: SPARK-20286
> URL: https://issues.apache.org/jira/browse/SPARK-20286
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
>Affects Versions: 2.0.1
>Reporter: Miguel Pérez
>
> With dynamic allocation enabled, it seems that executors with cached data
> which are unpersisted are still being killed using the
> {{dynamicAllocation.cachedExecutorIdleTimeout}} configuration, instead of
> {{dynamicAllocation.executorIdleTimeout}}. Assuming the default configuration
> ({{dynamicAllocation.cachedExecutorIdleTimeout = Infinity}}), an executor
> with unpersisted data won't be released until the job ends.
> *How to reproduce*
> - Set different values for {{dynamicAllocation.executorIdleTimeout}} and
> {{dynamicAllocation.cachedExecutorIdleTimeout}}
> - Load a file into a RDD and persist it
> - Execute an action on the RDD (like a count) so some executors are activated.
> - When the action has finished, unpersist the RDD
> - The application UI removes correctly the persisted data from the *Storage*
> tab, but if you look in the *Executors* tab, you will find that the executors
> remain *active* until ({{dynamicAllocation.cachedExecutorIdleTimeout}} is
> reached.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-20286) dynamicAllocation.executorIdleTimeout is ignored after unpersist
[
https://issues.apache.org/jira/browse/SPARK-20286?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15965408#comment-15965408
]
Miguel Pérez commented on SPARK-20286:
--
Yes. I suppose this is only checked when an executors passes from active to
idle status (maybe when the last action has finished), and it's not changed
when unpersist is called. But this is problematic, for example, if the job is
interactive. Imagine you have a Zeppelin notebook or a Spark shell. The same
driver could have different purposes and the session could be opened a long
time. Once you have called a single persist, all these executors will have to
wait {{dynamicAllocation.cachedExecutorIdleTimeout}} (which is Infinity by
default) or the user executing another action (executors will become active
again and then idle, but this time without cached data).
I don't know the solution. Maybe the best approach is to change from
cachedExecutorIdleTimeout to executorIdleTimeout on all cached executors when
the last RDD has been unpersisted, and then restart the time counter (unpersist
will then count as an action).
> dynamicAllocation.executorIdleTimeout is ignored after unpersist
>
>
> Key: SPARK-20286
> URL: https://issues.apache.org/jira/browse/SPARK-20286
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
>Affects Versions: 2.0.1
>Reporter: Miguel Pérez
>
> With dynamic allocation enabled, it seems that executors with cached data
> which are unpersisted are still being killed using the
> {{dynamicAllocation.cachedExecutorIdleTimeout}} configuration, instead of
> {{dynamicAllocation.executorIdleTimeout}}. Assuming the default configuration
> ({{dynamicAllocation.cachedExecutorIdleTimeout = Infinity}}), an executor
> with unpersisted data won't be released until the job ends.
> *How to reproduce*
> - Set different values for {{dynamicAllocation.executorIdleTimeout}} and
> {{dynamicAllocation.cachedExecutorIdleTimeout}}
> - Load a file into a RDD and persist it
> - Execute an action on the RDD (like a count) so some executors are activated.
> - When the action has finished, unpersist the RDD
> - The application UI removes correctly the persisted data from the *Storage*
> tab, but if you look in the *Executors* tab, you will find that the executors
> remain *active* until ({{dynamicAllocation.cachedExecutorIdleTimeout}} is
> reached.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-20286) dynamicAllocation.executorIdleTimeout is ignored after unpersist
[
https://issues.apache.org/jira/browse/SPARK-20286?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15965306#comment-15965306
]
Saisai Shao commented on SPARK-20286:
-
So the fix is that if there's no RDD get persisted, executors idle time should
be changed to {{dynamicAllocation.executorIdleTimeout}}.
> dynamicAllocation.executorIdleTimeout is ignored after unpersist
>
>
> Key: SPARK-20286
> URL: https://issues.apache.org/jira/browse/SPARK-20286
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
>Affects Versions: 2.0.1
>Reporter: Miguel Pérez
>
> With dynamic allocation enabled, it seems that executors with cached data
> which are unpersisted are still being killed using the
> {{dynamicAllocation.cachedExecutorIdleTimeout}} configuration, instead of
> {{dynamicAllocation.executorIdleTimeout}}. Assuming the default configuration
> ({{dynamicAllocation.cachedExecutorIdleTimeout = Infinity}}), an executor
> with unpersisted data won't be released until the job ends.
> *How to reproduce*
> - Set different values for {{dynamicAllocation.executorIdleTimeout}} and
> {{dynamicAllocation.cachedExecutorIdleTimeout}}
> - Load a file into a RDD and persist it
> - Execute an action on the RDD (like a count) so some executors are activated.
> - When the action has finished, unpersist the RDD
> - The application UI removes correctly the persisted data from the *Storage*
> tab, but if you look in the *Executors* tab, you will find that the executors
> remain *active* until ({{dynamicAllocation.cachedExecutorIdleTimeout}} is
> reached.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
