mesos-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Cody Maloney (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (MESOS-1992) Support launching executors with configured systemd
Date Fri, 24 Oct 2014 21:47:36 GMT

    [ https://issues.apache.org/jira/browse/MESOS-1992?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14183583#comment-14183583
] 

Cody Maloney commented on MESOS-1992:
-------------------------------------

So a couple things that using systemd as an executor/containerizer/isolator gives us:

# Much more extensible startup. Systemd has a lot more experience launching software, and
a great many knobs to be able to launch anything which currently exists in Linux
See: http://www.freedesktop.org/software/systemd/man/systemd.service.html
# Structured logging which provides easier access to both syslogs of a individual task, as
well as stdout and stderr
http://0pointer.de/blog/projects/journal-submit.html
# Highly configurable isolation, isolating things like /tmp if desired, read-only mounts of
most the filesystem
# No lost executors / tasks. Systemd is/runs as /sbin/init, which means it always is the parent
of all tasks and can see how they move/regroup, as well as kill them cleanly

> Support launching executors with configured systemd 
> ----------------------------------------------------
>
>                 Key: MESOS-1992
>                 URL: https://issues.apache.org/jira/browse/MESOS-1992
>             Project: Mesos
>          Issue Type: Improvement
>          Components: slave
>            Reporter: Timothy Chen
>
> Currently running mesos-slave in docker with systemd, the mesos-slave container cannot
be upgraded while keeping all the tasks running since killing the docker container will kill
all the processes that is launched with the mesos containerizer.
> If we can let the executor to be launched with systemd outside of the docker container,
then we can let the tasks remain running and recover them when the slave is upgraded.



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

Mime
View raw message