hadoop-mapreduce-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Tsuyoshi Ozawa (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (MAPREDUCE-6265) make INITIAL_POOL_SIZE in ContainerLauncherImpl configurable to better control the thread pool size to launch/kill containers.
Date Fri, 13 Mar 2015 16:59:39 GMT

    [ https://issues.apache.org/jira/browse/MAPREDUCE-6265?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14360641#comment-14360641
] 

Tsuyoshi Ozawa commented on MAPREDUCE-6265:
-------------------------------------------

[~zxu] looks good to me overall. Minor nits: I think the name of the property should be yarn.app.mapreduce.am.containerlauncher.*threadpool*-initial-size
and MR_AM_CONTAINERLAUNCHER_*THREADPOOL*_INITIAL_SIZE for accuracy since it's a bit difficult
to understand what "thread initial size" means. What do you think?

> make INITIAL_POOL_SIZE in ContainerLauncherImpl configurable to better control the thread
pool size to launch/kill containers.
> ------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: MAPREDUCE-6265
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-6265
>             Project: Hadoop Map/Reduce
>          Issue Type: Improvement
>          Components: mrv2
>            Reporter: zhihai xu
>            Assignee: zhihai xu
>         Attachments: MAPREDUCE-6265.000.patch
>
>
> make INITIAL_POOL_SIZE in ContainerLauncherImpl configurable to better control the thread
pool size to launch/kill containers
> Currently INITIAL_POOL_SIZE in ContainerLauncherImpl is hard-coded at 
> {code}
>   protected static final int INITIAL_POOL_SIZE = 10;
> {code}
> We should make it configurable because the thread pool size will be decided by INITIAL_POOL_SIZE,
limitOnPoolSize and number of node used by the AM.
> Since we already made limitOnPoolSize configurable, it make senses to also make INITIAL_POOL_SIZE
configurable to better manage the thread pool size.
> We saw some issue due to the small thread pool size when some node is down. The recovery
from a shutdown node take very long time due to all the ContainerLauncher threads are blocked
by IPC client connection to the shutdown node.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message