mesos-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ian Downes (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (MESOS-923) Containerizers should persist their own recovery state.
Date Fri, 14 Feb 2014 19:40:19 GMT

    [ https://issues.apache.org/jira/browse/MESOS-923?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13901840#comment-13901840
] 

Ian Downes commented on MESOS-923:
----------------------------------

The interface would remain the 'same': Containerizer::recover(Option<SlaveState>) but
the Option<pid_t> in RunState would be removed and wouldn't be recovered as part of
the slave recover.

Instead, Containerizer implementations should persist and recover their own state they require.
The current MesosContainerizer would persist the pid but, for example, a docker specific containerizer
may not need this because it can query the daemon.

> Containerizers should persist their own recovery state.
> -------------------------------------------------------
>
>                 Key: MESOS-923
>                 URL: https://issues.apache.org/jira/browse/MESOS-923
>             Project: Mesos
>          Issue Type: Improvement
>          Components: isolation
>            Reporter: Ian Downes
>             Fix For: 0.19.0
>
>
> In the current containerizer refactor (and previous process_isolator and cgroups_isolator)
the forked executor pid is persisted by the MesosContainerizer but recovered with the rest
of the slave state. This coupling is undesirable as containerizer implementations may or may
not use the pid for recovery and may require additional, specific state for recovery.
> Instead each containerizer implementation should manage persistence of whatever state
they require for recovery.



--
This message was sent by Atlassian JIRA
(v6.1.5#6160)

Mime
View raw message