[
https://issues.apache.org/jira/browse/MESOS-262?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13443766#comment-13443766
]
Jie Yu commented on MESOS-262:
------------------------------
OK, one possible fix I can think of is: set resource limit in child process
before execl is called (after the download) in ExecutorLauncher::run(). That
may require to split ExecutorLauncher::run() into two methods: prepare() and
exec(), so that in the isolation module, we can do the following:
{code}
pid_t pid = fork();
if (pid) {
// parent
...
} else {
// child
...
launcher::ExecutorLauncher* launcher =
createExecutorLauncher(frameworkId,
frameworkInfo,
executorInfo,
directory);
launcher->prepare();
// set resource limit
resourcesChanged(...);
launcher->exec();
}
{code}
> Slave should not charge the resources required for launching a executor
> against the executor
> ---------------------------------------------------------------------------------------------
>
> Key: MESOS-262
> URL: https://issues.apache.org/jira/browse/MESOS-262
> Project: Mesos
> Issue Type: Bug
> Reporter: Vinod Kone
>
> This is exacerbated when using cgroups isolation module on the slave.
> At Twitter, we have seen this manifest as executors being killed by the
> cgroups isolation module. This happened because the high memory footprint of
> the hdfs download (~400MB) of the executor exceeds the memory requested by
> the executor (128MB) for itself.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
