[
https://issues.apache.org/jira/browse/MESOS-1610?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14064189#comment-14064189
]
Jie Yu commented on MESOS-1610:
-------------------------------
OK, after discussed with [~idownes], we agreed that the isolator should
tolerate child process termination while calling 'isolate' because the child
process can be killed while the isolator is calling 'isolate'. Nothing prevent
that from happening.
Also, the 'cleanup' should properly handle partial isolated environment so that
all related links/filters/etc will be cleaned up properly without leaking any
thing.
That means we need to adjust the current port mapping network isolator.
> Mesos containerizer should not call isolate if the child process already died.
> ------------------------------------------------------------------------------
>
> Key: MESOS-1610
> URL: https://issues.apache.org/jira/browse/MESOS-1610
> Project: Mesos
> Issue Type: Bug
> Reporter: Jie Yu
>
> With the recent refactor of the mesos containerizer
> (https://reviews.apache.org/r/22852/), it's likely that a child process dies
> before 'isolate' is called. For example, we may see this if someone
> accidentally removes the mesos rpm (or shared library):
> {noformat}
> mesos-containerizer: error while loading shared libraries:
> libmesos-0.20.0-jyu9.so: cannot open shared object file: No such file or
> directory
> {noformat}
> That's gonna cause some strange error messages being printed out because the
> isolators always assume the pid exists when calling 'isolate()'.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
