[ 
https://issues.apache.org/jira/browse/MESOS-2462?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14389203#comment-14389203
 ] 

Ian Downes commented on MESOS-2462:
-----------------------------------

It should be the actual process, not the nanny's pid, because we need to be 
able to kill/signal the process. Killing the nanny won't propagate to the 
actual process (that's the point of this functionality!) but the nanny would 
die if it detects the child exit (in addition to killing the child and dying if 
the parent exits).

[I question the need to implement this for non linux anyway, given how 
restrictive process management is anyway]

> Add option for Subprocess to set a death signal for the forked child
> --------------------------------------------------------------------
>
>                 Key: MESOS-2462
>                 URL: https://issues.apache.org/jira/browse/MESOS-2462
>             Project: Mesos
>          Issue Type: Improvement
>          Components: isolation
>    Affects Versions: 0.21.1
>            Reporter: Ian Downes
>            Assignee: Jie Yu
>            Priority: Minor
>              Labels: twitter
>
> Currently, children forked by the slave, including those through Subprocess, 
> will continue running if the slave exits. For some processes, including 
> helper processes like the fetcher, du, or perf, we'd like them to be 
> terminated when the slave exits.
> Add support to Subprocess to optionally set a DEATHSIG for the child, e.g., 
> setting SIGTERM would mean the child would get SIGTERM when the slave 
> terminates.
> This can be done (*after forking*) with PR_SET_DEATHSIG. See "man prctl". It 
> is preserved through an exec call.



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

Reply via email to