On Tue, May 13, 2014 at 05:06:48PM -0700, Adam Bordelon wrote:
> Chengwei,
> 
> You can restart the leading master (with new roles/config), and running
> frameworks/slaves will re-register with it. Tasks will keep running. You
> probably want to restart the non-leading masters first with the new config, so
> that it will be picked up on HA failover.
> You can also restart the slave processes with the new resource-roles config 
> and
> running tasks will continue running, as long as you have slave checkpointing
> enabled.

Thanks Adam, generally I did the same thing to update configuration of
my cluster, it seems works fine.

However, I have one thing needs a confirm.

After resources configuration changed for a slave, it has to remove
/tmp/mesos/meta/slaves/latest and then start the slave.

Due to it can not checkpoint from different resources configuration.

I'm not sure if this has any side-effect, can you confirm?

--
Thanks,
Chengwei

> You also need to restart the framework-scheduler so that it can reregister 
> with
> the new role (in its FrameworkInfo), and get new offers appropriate for that
> role. Tasks will keep running.
> 
> You need to restart the master (with new roles config) before restarting the
> framework (with new role in its FrameworkInfo), or else the framework will 
> fail
> to reregister ("Role 'foo' is not valid."). I would recommend restarting the
> master first, then the slave(s), then the framework. Currently running tasks
> will continue running, regardless of role configuration. The new role(s) will
> only apply to new resource offers.
> 
> Hope that helps,
> -Adam-
> 
> 
> On Tue, May 13, 2014 at 2:31 PM, Vinod Kone <[email protected]> wrote:
> 
>     Hey Chengwei,
> 
>     Mesos doesn't allow online update of its configuration. The only 
> exception,
>     so far, has been the VLOG  level.
> 
>     To update resources, you should roll the slave with new flags.
> 
> 
>     On Sun, May 11, 2014 at 12:02 AM, Chengwei Yang 
> <[email protected]
>     > wrote:
> 
>         Hi List,
> 
>         Generally I have a question: does mesos cluster can reload
>         configuration
>         on the fly? Without side-effect to on-top frameworks, jobs?
> 
>         I have read the upgrad guide,
>         http://mesos.apache.org/documentation/latest/upgrades/
> 
>         and slave recovery guide,
>         http://mesos.apache.org/documentation/latest/slave-recovery/
> 
>         However, I didn't see any statement says about does the upgrade hurt
>         on-top frameworks, jobs.
> 
>         My situation is: I have a mesos 0.16 cluster up and running, some
>         frameworks and jobs running, and I want to add another framework, and
>         like to add a dedicate role for it, so change should made to mesos
>         master, and resolve some resources for it, so change should made to
>         mesos slaves.
> 
>         Could I do this seamlessly? Thanks in advance!
> 
>         --
>         Thanks,
>         Chengwei
> 
> 
> 
> 

Reply via email to