[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16661616#comment-16661616
]
Chen Yufei commented on YARN-8513:
--
Our YARN cluster is now working fine. Thanks [~cheersyang] [~leftnoteasy], I'll
close this issue as fixed.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16654575#comment-16654575
]
Weiwei Yang commented on YARN-8513:
---
Thanks [~leftnoteasy], [~hustnn], I created YARN-8896 to track this change.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16652886#comment-16652886
]
niu commented on YARN-8513:
---
[~cheersyang] [~leftnoteasy]
Thanks.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16652645#comment-16652645
]
Wangda Tan commented on YARN-8513:
--
Sounds like a plan, default value set to 100 may make more sense. thanks
[~cheersyang]
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16651349#comment-16651349
]
Weiwei Yang commented on YARN-8513:
---
Hi [~leftnoteasy]/[~cyfdecyf]/[~hustnn]/[~Tao Yang]
When this issue created with the logs attached, it looks very suspicious there
is a bug in {{CS#allocateContainersToNode}}, causing it never breaks out from
following while loop,
{code:java}
while (canAllocateMore(...)) {...}
{code}
then I suggested [~cyfdecyf] to change the property
{{yarn.scheduler.capacity.per-node-heartbeat.maximum-container-assignments}}
from *-1* (the default value) to *10*. Which seems to work-around the issue. I
think we should change the default value to 10, to have this sort of infinite
greedy-lookup is not safe. I am going to submit a patch to change the default
value.
Thoughts?
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16651024#comment-16651024
]
Chen Yufei commented on YARN-8513:
--
Make some corrections to my previous comment.
The behavior of YARN resource manager is different this time. RM is not
allocating resources, but it's not flushing out log message like before, CPU
usage is relative low. So I guess it's another problem and not the same one as
I reported in this issue.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16649797#comment-16649797
]
Chen Yufei commented on YARN-8513:
--
[~Tao Yang] I encounter this problem again today. I see logs saying "we need to
unreserve to be able to allocate". I'm try to apply your patch in YARN-8771 and
see if it can fix the problem.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16623040#comment-16623040
]
Tao Yang commented on YARN-8513:
Hi, [~cyfdecyf]
Dose your cluster have empty resource type? This problem seems similar to
YARN-8771 whose problem is caused by wrong calculation for
needUnreservedResource in RegularContainerAllocator#assignContainer when
cluster has empty resource type.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16611742#comment-16611742
]
Chen Yufei commented on YARN-8513:
--
[~cheersyang] Thanks for your help. I haven't run into any problem since
applying your suggested config. But I don't know how these configurations
change the allocation logic thus avoids the issue.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16611677#comment-16611677
]
Weiwei Yang commented on YARN-8513:
---
Hi [~cyfdecyf]
Did the config changes I suggested last time
"yarn.scheduler.capacity.per-node-heartbeat.maximum-container-assignments"
help? No?
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16610141#comment-16610141
]
Chen Yufei commented on YARN-8513:
--
[~leftnoteasy] Thanks for taking time to investigate this.
The resource allocation scheme has no problem for me. I'm not able to turn on
preemption in our cluster for now, is there any other way to avoid RM not
responding to any other requests when the problem occurs?
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16608263#comment-16608263
]
niu commented on YARN-8513:
---
[~leftnoteasy] . No problem. Take your time. Let me try first.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16608213#comment-16608213
]
Wangda Tan commented on YARN-8513:
--
[~hustnn],
I agree that it is still a problem, but relatively minor because it won't
change how scheduler allocate resources between queues.
If you could help to give detailed config of the existing problem cluster,
include CS config, NM resources, size of containers / AM, user of each job ,
etc. I should be able to reproduce it by using UT, that should be more easier
to figure out issue and prevent regression in the future.
I will travel next week, please expect some delays of my responses.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16607901#comment-16607901
]
niu commented on YARN-8513:
---
Thanks [~leftnoteasy] for your effort to look at this problem.
In my attached debug log, the setting is we have 2 queues: root.dw and
root.dev. Capacity setting for dw and dev are dw(capacity: 68, max:100) and
dev(capacity:32, max:60), respectively. In this case, root almost fully
occupied by dw and only has 256000 resources for dev. Therefore, each
container request (360448) from dev will not be reserved according to the logic
in YARN-4280 as the the used+notallocated beyonds the capacity of root (parent
of dev) 's capacity.
It makes sense for the above scenario. However, I still feel there is some
problem. When I set the max capacity of dev from 60 to 100. Then, the problem
will not occur. The root also beyonds the limitation under this setting. How to
explain it ? I will attach the log next Monday.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16607495#comment-16607495
]
Wangda Tan commented on YARN-8513:
--
Spent good amount of time to check the issue.
I found scheduler tries to reserve containers on two nodes. What happens is:
1) For root queue, total resource = 1351680, used resource = 1095680, available
resource = 256000
2) The app which gets resource is running under dev queue, maximum resource =
8811008, used resource = 7168.
3) The app always get container reserved with size=360448, which is beyond
parent queue's available resource. So this request will be rejected by resource
committer.
In my mind, this is expected behavior, even though the resource proposal /
reject is not necessary. This behavior is in-line with YARN-4280, which we want
to keep under-utilized queue still get resources when resource request is large.
Let me use an example to explain this:
Scheduler has two queues, a and b, capacity of each queues are 0.5. max
capacity of a = 1.0, b=0.8. Assume cluster resource = 100.
There's an app running in a, which uses 75 resources, so a's absolute used
capacity = 0.75. There're still many pending resource request from a, size of
each = 1
And then user submit app to b. asking a single container, which has size = 30.
In that case, scheduler cannot allocate the container because cluster's total
available = 25.
If we give these resources to queue=a, queue=b can never get the available
resource, because smaller resource request will be always preferred.
Instead, the logic in YARN-4280 is: if queue b don't get resource because of
parent queue's resource limit. Instead of giving resources to other queues,
scheduler hold the resource. So you can see that there're 25 resources
available, but no one can get the resource.
The problem only occurs in a super busy cluster, with less node. To solve the
problem, turn on preemption can alleviate the issue a lot.
I prefer to close this as "no fix needed".
Thoughts?
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16602522#comment-16602522
]
niu commented on YARN-8513:
---
Debug dump:
{code:java}
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.rmnode.RMNodeImpl: Processing
shopee-test-cluster04:45454 of type STATUS_UPDATE
2018-09-03 11:44:11,175 DEBUG org.apache.hadoop.yarn.event.AsyncDispatcher:
Dispatching the event
org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.NodeUpdateSchedulerEvent.EventType:
NODE_UPDATE
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.AbstractYarnScheduler:
nodeUpdate: shopee-test-cluster04:45454 cluster capacity:
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.AbstractYarnScheduler:
Node being looked for scheduling shopee-test-cluster04:45454 availableResource:
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
Trying to schedule on node: shopee-test-cluster04, available:
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
Trying to assign containers to child-queue of root
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.AbstractCSQueue:
Check assign to queue: root nodePartition: , usedResources: , clusterResources: , currentUsedCapacity:
0.81060606, max-capacity: 1.0
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
printChildQueues - queue: root child-queues: root.dwusedCapacity=(1.1842697),
label=(*)root.devusedCapacity=(0.016571993), label=(*)
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
Trying to assign to queue: root.dev stats: dev: capacity=0.32,
absoluteCapacity=0.32, usedResources=,
usedCapacity=0.016571993, absoluteUsedCapacity=0.0053030304, numApps=1,
numContainers=1
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.LeafQueue:
assignContainers: partition= #applications=1
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.AbstractCSQueue:
Check assign to queue: dev nodePartition: , usedResources: , clusterResources: , currentUsedCapacity:
0.0053030304, max-capacity: 0.6
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.UsersManager:
userLimit is fetched. userLimit=,
userSpecificUserLimit=,
schedulingMode=RESPECT_PARTITION_EXCLUSIVITY, partition=
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.LeafQueue:
Headroom calculation for user work: userLimit=
queueMaxAvailRes= consumed=
partition=
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerApp:
pre-assignContainers for application application_1535930391687_0019
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.UsersManager:
userLimit is fetched. userLimit=,
userSpecificUserLimit=,
schedulingMode=RESPECT_PARTITION_EXCLUSIVITY, partition=
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.SchedulerApplicationAttempt:
showRequests: application=application_1535930391687_0019
headRoom= currentConsumption=7168
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.placement.LocalitySchedulingPlacementSet:
Request={AllocationRequestId: 0, Priority: 1, Capability:
, # Containers: 3, Location: *, Relax Locality: true,
Execution Type Request: null, Node Label Expression: }
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator:
assignContainers: node=shopee-test-cluster04
application=application_1535930391687_0019 priority=1
pendingAsk=,repeat=3>
type=OFF_SWITCH
2018-09-03 11:44:11,175 INFO
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
Reserved container application=application_1535930391687_0019
resource=
queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@65ed660
cluster=
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.LeafQueue:
post-assignContainers for application application_1535930391687_0019
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.UsersManager:
userLimit is fetched. userLimit=,
userSpecificUserLimit=,
schedulingMode=RESPECT_PARTITION_EXCLUSIVITY, partition=
2018-09-03 11:44:11,175 DEBUG
org.apache.hadoop.yarn.server.resourcemanager.scheduler.SchedulerApplicationAttempt:
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16602519#comment-16602519
]
niu commented on YARN-8513:
---
Attached the debug log.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16602507#comment-16602507
]
niu commented on YARN-8513:
---
I also tested. This happens when node label is not used. Does it cause by the
resource request which is already rejected but still recovered?
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16602326#comment-16602326
]
Wangda Tan commented on YARN-8513:
--
And btw, I found a comment in LeafQueue:
{code:java}
private void updateCurrentResourceLimits(
ResourceLimits currentResourceLimits, Resource clusterResource) {
// TODO: need consider non-empty node labels when resource limits supports
// node labels
// Even if ParentQueue will set limits respect child's max queue capacity,
// but when allocating reserved container, CapacityScheduler doesn't do
// this. So need cap limits by queue's max capacity here.
this.cachedResourceLimitsForHeadroom =
new ResourceLimits(currentResourceLimits.getLimit());
Resource queueMaxResource = getEffectiveMaxCapacityDown(
RMNodeLabelsManager.NO_LABEL, minimumAllocation);
this.cachedResourceLimitsForHeadroom.setLimit(Resources.min(
resourceCalculator, clusterResource, queueMaxResource,
currentResourceLimits.getLimit()));
}{code}
I can remember a little bit when I wrote the code: YARN-3243 fixed an issue
which ParentQueue's max capacity could be violated. I didn't consider node
label max capacity because at that time per-queue per-label capacities support
has some issues. I believe the issue should be fixed in later patches, but it
is worth to check if we need any other fixes.
[~Card], does this happen when node label is being used or not?
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16602323#comment-16602323
]
Wangda Tan commented on YARN-8513:
--
Interesting, it must be caused by CS allocation doesn't fully consider queue
maximum resource in some cases. Tried to look at related code, hasn't figured
out root case yet.
CS allocation phase relies on the logic of ResourceLimits passed by upper level
component (Parent of queues, queue of apps, etc.). Under some corner cases, the
ResourceLimits passed in could be larger than accurate.
[~Card], could u enable DEBUG log of
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity, and rerun the
test? (or you can click "dump DEBUG log" in CS web UI) It gonna be helpful if
you can get a few seconds DEBUG log for our troubleshooting when the infinite
loop happens.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16601223#comment-16601223
]
niu commented on YARN-8513:
---
Thank. I will try it tomorrow.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16601171#comment-16601171
]
Janus Chow commented on YARN-8513:
--
[~leftnoteasy] Can reproduce as follow:
1.Specify 2 queues:
- queue1:
- capacity: 68
- maximum-capacity: 100
- queue2:
- capacity: 32
- maximum-capacity: 60
2. Submit a big spark job, make sure this job use resource higher than queue's
capacity(68%), like 90% of the cluster's resource.
pyspark --num-executor=100 --executor-memory=20g --queue=queue1
3. Submit another big spark job.
pyspark --num-executor=100 --executor-memory=20g --queue=queue2
4.Then the resource manager log will show the "Failed to accept allocation
proposal".
And if I change the maximum-capacity of queue2 to 100 or -1, the log stops
flushing.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16601138#comment-16601138
]
Janus Chow commented on YARN-8513:
--
Set two queues:
- queue1:
- capacity: 50
- maximum-capacity: 100
- queue2:
- capacity: 50
- maximum-capacity: 100
First, submit a large job consuming 90% resource of the whole cluster resource
in queue1.
Then, submit another big job in queue2, after the AM started, the log started
to print "Failed to accept allocation proposal".
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16593141#comment-16593141
]
niu commented on YARN-8513:
---
OK. I will try.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16591878#comment-16591878
]
Wangda Tan commented on YARN-8513:
--
[~hustnn], what is the cause of "Failed to accept allocation proposal"? You
should be able to get this from DEBUG log. Once we understand why allocation
proposal got rejected, we can understand why such loop happens.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16591337#comment-16591337
]
niu commented on YARN-8513:
---
I also check the logic for printing out of the error "Failed to accept
allocation proposal".
allocateContainersToNode
-- if canAllocateMore // (decided by
yarn.scheduler.capacity.per-node-heartbeat.maximum-container-assignments)
-- allocateContainersToNode
-- allocateContainerOnSingleNode
-- allocateOrReserveNewContainers
-- submitResourceCommitRequest
-- tryCommit
-- LOG.info("Failed to accept allocation proposal");
It seems this error is indeed caused by infinit number of of container requests.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16588238#comment-16588238
]
Chen Yufei commented on YARN-8513:
--
[~leftnoteasy] My original config did not have the two config options specified
so should be using the default values.
Currently I have applied the configuration suggested by [~cheersyang], so
maximum-container-assignments is 10 now.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16586854#comment-16586854
]
Wangda Tan commented on YARN-8513:
--
Interesting, [~cheersyang],
I can only think about reservation allocation causes the issue:
{code}
// And it should not be a reserved container
if (assignment.getAssignmentInformation().getNumReservations() > 0) {
return false;
}
{code}
We should be able to see what kind of allocation causes the issue, or is it
possible that CSAssignment indicate allocation happens but actually it doesn't.
What is the {{maximum-container-assignments}} settings now?
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16586811#comment-16586811
]
Weiwei Yang commented on YARN-8513:
---
Discussed with [~cyfdecyf] in the slack channel, it looks like this issue was
caused by the greedy container assignments per HB mechanism, for some reason,
it never stop trying assign new containers for this particular node in a while
loop. I suggested to add following config to work-around
{noformat}
“yarn.scheduler.capacity.per-node-heartbeat.multiple-assignments-enable”=“true”
“yarn.scheduler.capacity.per-node-heartbeat.maximum-container-assignments”=“10”
{noformat}
At the mean time, [~cyfdecyf] is trying to apply this config changes to their
cluster to see if that helps, and I am trying to reproduce this issue locally.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16585359#comment-16585359
]
Weiwei Yang commented on YARN-8513:
---
Thanks for the info [~cyfdecyf]. To work more efficiently on this issue, lets
collaborate on Slack. I created a channel {{yarn-8513}} and please use [this
link|https://join.slack.com/t/yarn-group/shared_invite/enQtNDE5NTA0ODk2NTE3LTljNWE2MzdjMWU5NGUyYTJkYjI0YmJhNzc4NDg5MTMwMWRiZDI1OGE2YWI4ZTE1MzQ2MjBkYWRjNDk0MjJhMTY]
to join.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16585036#comment-16585036
]
Chen Yufei commented on YARN-8513:
--
[~cheersyang] Thanks for looking into this issue.
The log message file is truncated (because mosts are repeated), the whole log
size in a second is 60MB and contains about 5300 lines of "Trying to schedule
on node".
Cluster size:
* default partition: 79 NM
* sim: 45 NM
* gpu: 0 NM (we still have Hadoop 2.9.1 running and some nodes haven't join the
new version of Hadoop)
NM HB interval is not changed and is the default values 1000ms.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16584825#comment-16584825
]
Weiwei Yang commented on YARN-8513:
---
Hi [~cyfdecyf]
>From the RM log you uploaded, in 1 sec, there are 70 times of
{code:java}
Trying to schedule on node: rndcl58.rt.com, available:
{code}
>From jstack files, they both have
{code:java}
Thread 328918: (state = IN_JAVA)
...
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler.allocateContainersToNode(org.apache.hadoop.yarn.server.resourcemanager.scheduler.placement.CandidateNodeSet,
boolean) @bci=50, line=1647 (Compiled frame) -
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler.allocateContainersToNode(org.apache.hadoop.yarn.api.records.NodeId,
boolean) @bci=102, line=1417 (Compiled frame) -
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler.nodeUpdate(org.apache.hadoop.yarn.server.resourcemanager.rmnode.RMNode)
@bci=110, line=1258 (Compiled frame)
{code}
so they are triggered by nodeUpdate (HB). Looks like the RM dispatcher is
flooded with NM HBs? I think we need more info, what's the size of your cluster
and what is the time interval of the NM HB?
Thanks
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16584787#comment-16584787
]
Chen Yufei commented on YARN-8513:
--
New jstack/top and RM logs are uploaded and prefixed with yarn3. We upgraded to
Hadoop 3.1.1 yesterday and encounter this problem several times.
The problem seems reproducible when one queue is near fully utilized. Killing
current active RM can not solve the problem. We have to kill some jobs in the
fully utilized queue in order to submit new jobs.
Debug log shows that CapacityScheduler repeatedly trying to schedule on a
specific node, but as queue resource has exceeded resource limit, allocation
proposal won't be accepted. top command shows only one thread with near 100%
CPU usage, strace shows this thread is the one trying to do the allocation and
flushing out logs.
I've tried to dig into source code, but can't find out why RM repeatedly trying
to schedule on a specific node.
Some notes about our setup:
* 3 partitions: default, sim, gpu
* 4 queues: dev & mkt (10% capacity, max 90%), dev-daily & mkt-daily (40%
capacity, max 100%)
* preemption disabled
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log, yarn3-jstack1.log,
> yarn3-jstack2.log, yarn3-jstack3.log, yarn3-jstack4.log, yarn3-jstack5.log,
> yarn3-resourcemanager.log, yarn3-top
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16583270#comment-16583270
]
Chen Yufei commented on YARN-8513:
--
[~leftnoteasy] Thanks for your help. I'll enable capacity scheduler DEBUG log
and upload logs/jstacks when the problem occurs again.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16583026#comment-16583026
]
Wangda Tan commented on YARN-8513:
--
[~cyfdecyf],
Could u upload logs/jstacks for 3.1.0 deployment? We can help to take a look at
it. It will be better if you can enable DEBUG log of
{{org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity}} for a
short while.
cc: [~sunil.gov...@gmail.com], [~cheersyang].
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16582584#comment-16582584
]
Chen Yufei commented on YARN-8513:
--
[~leftnoteasy] We encounter the same problem twice today with Hadoop 3.1.0.
ResourceManager log messages are the same as before, flushing out 20k lines
every seconds during deadlock.
Our current setup enables ResourceManager and NameNode HA, killing the active
ResourceManager with SIGKILL will turn other standby RM to active state, and
the new active RM can work properly.
ApplicationMaster running on our cluster is a slightly modified version of
distributedshell (modified from Hadoop 2.7.1). When updating to Hadoop 3.1.0,
we fixed all deprecated APIs and didn't make other modifications. During the RM
deadlock, I killed the AM (having very low CPU usage) which is being reported
in the log message, but RM still flushes the same log messages. If the problem
is caused by our AM, killing the problematic AM may turn RM to a normal state.
But seems like it's the internal deadlock in RM causing the problem.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 3.1.0, 2.9.1
> Environment: Ubuntu 14.04.5 and 16.04.4
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16580862#comment-16580862
]
Chen Yufei commented on YARN-8513:
--
We got infinite loops two times recently with 2.9.1, restarting ResourceManager
fixed the issue again.
As the cause of the problem is still not clear, we have upgraded to Hadoop
3.1.0. I'll give further updates in case we encounter this issue again.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 2.9.1
> Environment: Ubuntu 14.04.5
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16552536#comment-16552536
]
niu commented on YARN-8513:
---
We also met this problem in 2.9.1.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 2.9.1
> Environment: Ubuntu 14.04.5
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16551551#comment-16551551
]
Yuanbo Liu commented on YARN-8513:
--
Sorry for the late response. Quite busy this week. I will go through the dump
files today
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 2.9.1
> Environment: Ubuntu 14.04.5
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16544742#comment-16544742
]
Chen Yufei commented on YARN-8513:
--
[~yuanbo] I've uploaded jstack and top log when the problem appears yesterday.
jstack log are captured for 5 times thus 5 log files.
[^top-during-lock.log] is captured when RM is not responding to requests.
[^top-when-normal.log] is captured today and RM is running normally.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 2.9.1
> Environment: Ubuntu 14.04.5
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
> Attachments: jstack-1.log, jstack-2.log, jstack-3.log, jstack-4.log,
> jstack-5.log, top-during-lock.log, top-when-normal.log
>
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16544234#comment-16544234
]
Chen Yufei commented on YARN-8513:
--
[~leftnoteasy] Thanks for your suggestions. I'll deploy a test environment with
3.1.0 and test it with some of our production workloads. This may take quite
some time but I'll report my findings when I've done the test.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 2.9.1
> Environment: Ubuntu 14.04.5
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16544231#comment-16544231
]
Chen Yufei commented on YARN-8513:
--
[~yuanbo] I'll capture those info when I encounter this problem again next
time. Currently I don't know the exact trigger condition for the problem, so
have to wait for some time.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 2.9.1
> Environment: Ubuntu 14.04.5
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16543689#comment-16543689
]
Wangda Tan commented on YARN-8513:
--
[~cyfdecyf], I couldn't find the error message on the latest codebase. Not sure
if this still a problem in latest release (3.1.0). We have many fixes in the
last several months for CapacityScheduler scheduling after YARN-5139, I believe
many of them are not backported to 2.9.1. Could u check if the problem still
exists in 3.1.0 if possible?
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 2.9.1
> Environment: Ubuntu 14.04.5
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Commented] (YARN-8513) CapacityScheduler infinite loop when queue is near fully utilized
[
https://issues.apache.org/jira/browse/YARN-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16542719#comment-16542719
]
Yuanbo Liu commented on YARN-8513:
--
[~cyfdecyf] Can you reproduce this issue and capture the stack of RM
# jstack -F pid
# top -H -p pid
then attach those info in this Jira so that we can figure out the cause of
infinite loop here.
> CapacityScheduler infinite loop when queue is near fully utilized
> -
>
> Key: YARN-8513
> URL: https://issues.apache.org/jira/browse/YARN-8513
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler, yarn
>Affects Versions: 2.9.1
> Environment: Ubuntu 14.04.5
> YARN is configured with one label and 5 queues.
>Reporter: Chen Yufei
>Priority: Major
>
> ResourceManager does not respond to any request when queue is near fully
> utilized sometimes. Sending SIGTERM won't stop RM, only SIGKILL can. After RM
> restart, it can recover running jobs and start accepting new ones.
>
> Seems like CapacityScheduler is in an infinite loop printing out the
> following log messages (more than 25,000 lines in a second):
>
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue:
> assignedContainer queue=root usedCapacity=0.99816763
> absoluteUsedCapacity=0.99816763 used=
> cluster=}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler:
> Failed to accept allocation proposal}}
> {{2018-07-10 17:16:29,227 INFO
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.AbstractContainerAllocator:
> assignedContainer application attempt=appattempt_1530619767030_1652_01
> container=null
> queue=org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.allocator.RegularContainerAllocator@14420943
> clusterResource= type=NODE_LOCAL
> requestedPartition=}}
>
> I encounter this problem several times after upgrading to YARN 2.9.1, while
> the same configuration works fine under version 2.7.3.
>
> YARN-4477 is an infinite loop bug in FairScheduler, not sure if this is a
> similar problem.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
