[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Get list of restarted services during upgrade



On Sun, 30 Jun 2013 00:21:51 -0600
Bob Proulx <bob@proulx.com> wrote:

> > I'm running a Cluster that has to be set in Maintenance-Mode if
> > monitored services are restarted.  

Hi Bob,

> I assume from this that services are being monitored.  If they are
> restarted and during the restart the monitoring detects that the
> service is offline then it triggers an alert.  And whatever the

First it will try to restart the service, if this fails the service
will be migrated to another node.

> setting of "Maintenance-Mode" does basically tells the monitoring to
> ignore any problem for that time period.  Is that about right?

Yes. But it will disable the monitor for all services on all nodes,
which isn't always needed. 

> I see the problem for you now.  Because various packages such as
> phpmyadmin will run the restarts in the postinst scripts.  Therefore
> in order to really determine what would be restarted it would be
> necessary to walk through the postinst scripts.  And that is an open
> ended basically practically impossible problem.

Correct.
 
> > At the moment I try out update-rc.d to avoid any service to be
> > restarted during upgrade, but I wonder if there might be a better
> > solution.

> > It's policy-rc.d, of course, not update-rc.  

> That is an idea.  Every use of a postinst script to restart a daemon
> should be gated by policy-rc.d.  And you could always approve the
> restart but that script could at that time put the node into your
> maintenance-mode right then.  Don't know what you would use to return
> it to regular service afterward.  Perhaps after a timeout.

That is a nice idea, thanks. At the moment policy-rc.d just exit with
101. 

> Note that I (on this mailing list) learned recently that while
> policy-rc.d does gate all uses of postinst it does nothing at boot
> time.  I think that is a hole in the feature set.  During boot time
> the daemon is started regardless of any policy-rc.d configuration.

That's fine. The init-Scripts for the monitored services are disabled
at boot time and the Cluster Resource Manager will start them. I think
that policy-rc.d is ignored at boot is more of a feature (at least in
some use cases).

I'll check out your idea and think it will work quite fine. Thanks
again.

Best regards
Denis Witt


Reply to: