Hi Jose,

Apologies for the late reply.

On 14:34 Mon 14 Apr     , Jose A. Lopes wrote:
> On Apr 14 15:23, Apollon Oikonomopoulos wrote:
> > Hi Jose,
> > 
> > On 14:13 Mon 14 Apr     , Jose A. Lopes wrote:
> > > Hi Apollon,
> > > 
> > > Should we add support for the metadata and KVM daemons (i.e., metad
> > > and kvmd).  It is true that they are started automatically by Ganeti,
> > > although currently they do not call 'daemon-util'.  But it should
> > > possible to stop them in any case.
> > 
> > Hm, I missed metad (this job was mostly done before metad was merged to 
> > master, my bad!). I'll resubmit for that.
> > 
> > Regarding kvmd, yes, we should handle this as well. However, kvmd is 
> > currently handled in a different way than the rest of the daemons: it's 
> > not started via daemon-util, but directly from the node daemon. This 
> > means that we will have to either code systemd detection logic in the 
> > node daemon just for that, or change the node daemon to start kvmd via 
> > daemon-util. I'd prefer the latter, what do you think?
> 
> I agree, the latter sounds good.
> 
> But, aren't 'metad' and 'kvmd' in the same situation?  They are both
> started directly from the node daemon.  So, we have to change both of
> them to start through 'daemon-util', right?

Right, although I guess since they're meant to be loaded on demand, they 
shouldn't be started during start_all, but they should be stopped during 
stop_all. Also note that if we stop and start kvmd during updates, there 
will we a small time window where an instance stop will go undetected.  
Are we ok with that? If we agree on that, I can send the relevant 
patches.

Apollon

Reply via email to