[Pacemaker] A demand to the process trouble.(OpenAIS/Corosync and Pacemaker)

[renayama19661014 at ybb.ne.jp]

Hi Andrew,


> If the crmd dies, then I (IIRC) the lrmd cancels all existing resource
> monitoring.
> However, when the crmd is recovered, it should setup the resource
> monitoring again.
> 
> Is the second part not happening?

There are some patterns in a problem. 

When crmd/stonithd restarts in a ACT node in a DC node, the monitor completely stops.
When others of lrmd/mgmtd restart in the STB node of the DC node, the monitor completely stops.

> > 1) When a process related to a monitor fell, a system reboots.(Emergency Reboot)
> Thats the lazy way out.

I think your opinion is right.
If a monitor is reopened definitely, I think that the Emergency reboot is not necessary.

Best Regards,
Hideo Yamauchi.


--- Andrew Beekhof <andrew at beekhof.net> wrote:

> 2009/7/17  <renayama19661014 at ybb.ne.jp>:
> > Hi Andrew,
> >
> >> What do you mean by monitor here?
> >> Do you mean that pacemaker would no longer detect if those two processes died?
> >
> > This monitor means the monitor of the resource.
> > When these processes fall, the monitor of the resource of lrmd/stonithd stops.
> > For example, the monitor of external/ssh and pgsql stops.
> >
> >> Do you mean the way heartbeat behaves with "crm on" instead of "crm respawn" ?
> >
> > I do not understand this meaning well.
> 
> That makes two of us :-)
> I'm not sure I really understand the problem here.
> 
> If the crmd dies, then I (IIRC) the lrmd cancels all existing resource
> monitoring.
> However, when the crmd is recovered, it should setup the resource
> monitoring again.
> 
> Is the second part not happening?
> 
> >
> > I think that the following approach is necessary.
> >
> > 1) When a process related to a monitor fell, a system reboots.(Emergency Reboot)
> 
> Thats the lazy way out.
> 
> > 2) When a process related to a monitor fell, a monitor does not stop even if it reboot.
> >
> > The first approach thinks that realization is simple.
> >
> > Best Regards,
> > Hideo Yamauchi.
> >
> > --- Andrew Beekhof <andrew at beekhof.net> wrote:
> >
> >> On Fri, Jul 17, 2009 at 3:34 AM, <renayama19661014 at ybb.ne.jp> wrote:
> >> > Hi,
> >> >
> >> > We began shift investigation to the combination of Pacemaker and corosync/openais now.
> >> >
> >> > We put Pacemaker and openais(whitetank) together and confirmed movement at the time of the
> >> process
> >> > trouble.
> >> > (This is the function that a reboot emergency occurred by a combination with Heartbeat.)
> >> >
> >> > I let a process of Pacemaker break down. (kill -9 pid)
> >> > The following behavior was seen then.
> >> >
> >> > * When crmd reboots on ACT node(Not DC), the monitor of the lrmd resource stops.
> >> >  And the monitor of the stonnith resource stops.
> >>
> >> What do you mean by monitor here?
> >> Do you mean that pacemaker would no longer detect if those two processes died?
> >>
> >> > * When stonithd reboots on ACT node(Not DC), the monitor of the stonnith resource stops.
> >> > * When crmd reboots on STB node(DC), the monitor of the stonnith resource stops.
> >> > * When pengine reboots on STB node(DC), the monitor of the stonnith resource stops.
> >> > * And more....
> >> >
> >> > We feel a problem for the stop of the monitor after the process reboot.
> >> > When we combined openais/corosync, we hope a function such as the urgent reboot of
> Heartbeat
> >> to be
> >> > included.
> >>
> >> Do you mean the way heartbeat behaves with "crm on" instead of "crm respawn" ?
> >> >
> >> > Best Regards,
> >> > Hideo Yamauchi.
> >> >
> >> >
> >> > _______________________________________________
> >> > Pacemaker mailing list
> >> > Pacemaker at oss.clusterlabs.org
> >> > http://oss.clusterlabs.org/mailman/listinfo/pacemaker
> >> >
> >>
> >> _______________________________________________
> >> Pacemaker mailing list
> >> Pacemaker at oss.clusterlabs.org
> >> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
> >>
> >
> >
> >
> > _______________________________________________
> > Pacemaker mailing list
> > Pacemaker at oss.clusterlabs.org
> > http://oss.clusterlabs.org/mailman/listinfo/pacemaker
> >
> >
> 
> _______________________________________________
> Pacemaker mailing list
> Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>