<font size=2 face="sans-serif">+1</font>
<br>
<br><font size=2 face="sans-serif">Regards,</font>
<br><font size=2 face="sans-serif">Alex</font>
<br><font size=2 face="sans-serif"><br>
</font>
<br><tt><font size=2>Joshua Harlow <harlowja@yahoo-inc.com> wrote
on 26/10/2013 09:29:03 AM:<br>
> <br>
> An idea that others and I are having for a similar use case in <br>
> cinder (or it appears to be similar).<br>
> <br>
> If there was a well defined state machine/s in nova with well <br>
> defined and managed transitions between states then it seems like
<br>
> this state machine could resume on failure as well as be interrupted<br>
> when a "dueling" or preemptable operation arrives (a delete
while <br>
> being created for example). This way not only would it be very clear<br>
> the set of states and transitions but it would also be clear how <br>
> preemption occurs (and under what cases). <br>
> <br>
> Right now in nova there is a distributed and ad-hoc state machine
<br>
> which if it was more formalized it could inherit some if the <br>
> described useful capabilities. It would also be much more resilient
<br>
> to these types of locking problems that u described. <br>
> <br>
> IMHO that's the only way these types of problems will be fully be
<br>
> fixed, not by more queues or more periodic tasks, but by solidifying<br>
> & formalizing the state machines that compose the work nova does.<br>
> <br>
> Sent from my really tiny device...<br>
> <br>
> > On Oct 25, 2013, at 3:52 AM, "Day, Phil" <philip.day@hp.com>
wrote:<br>
> > <br>
> > Hi Folks,<br>
> > <br>
> > We're very occasionally seeing problems where a thread processing
<br>
> a create hangs (and we've seen when taking to Cinder and Glance).
<br>
> Whilst those issues need to be hunted down in their own rights, they<br>
> do show up what seems to me to be a weakness in the processing of
<br>
> delete requests that I'd like to get some feedback on.<br>
> > <br>
> > Delete is the one operation that is allowed regardless of the
<br>
> Instance state (since it's a one-way operation, and users should <br>
> always be able to free up their quota). However when we get
a <br>
> create thread hung in one of these states, the delete requests when
<br>
> they hit the manager will also block as they are synchronized on the<br>
> uuid. Because the user making the delete request doesn't see
<br>
> anything happen they tend to submit more delete requests. The
<br>
> Service is still up, so these go to the computer manager as well,
<br>
> and eventually all of the threads will be waiting for the lock, and
<br>
> the compute manager will stop consuming new messages.<br>
> > <br>
> > The problem isn't limited to deletes - although in most cases
the <br>
> change of state in the API means that you have to keep making <br>
> different calls to get past the state checker logic to do it with
an<br>
> instance stuck in another state. Users also seem to be more
<br>
> impatient with deletes, as they are trying to free up quota for other
things. <br>
> > <br>
> > So while I know that we should never get a thread into a hung
<br>
> state into the first place, I was wondering about one of the <br>
> following approaches to address just the delete case:<br>
> > <br>
> > i) Change the delete call on the manager so it doesn't wait for
<br>
> the uuid lock. Deletes should be coded so that they work regardless<br>
> of the state of the VM, and other actions should be able to cope <br>
> with a delete being performed from under them. There is of course
<br>
> no guarantee that the delete itself won't block as well. <br>
> > <br>
> > ii) Record in the API server that a delete has been started (maybe<br>
> enough to use the task state being set to DELETEING in the API if
<br>
> we're sure this doesn't get cleared), and add a periodic task in the<br>
> compute manager to check for and delete instances that are in a <br>
> "DELETING" state for more than some timeout. Then the API,
knowing <br>
> that the delete will be processes eventually can just no-op any <br>
> further delete requests.<br>
> > <br>
> > iii) Add some hook into the ServiceGroup API so that the timer
<br>
> could depend on getting a free thread from the compute manager pool
<br>
> (ie run some no-op task) - so that of there are no free threads then<br>
> the service becomes down. That would (eventually) stop the scheduler<br>
> from sending new requests to it, and make deleted be processed in
<br>
> the API server but won't of course help with commands for other <br>
> instances on the same host.<br>
> > <br>
> > iv) Move away from having a general topic and thread pool for
all <br>
> requests, and start a listener on an instance specific topic for <br>
> each running instance on a host (leaving the general topic and pool
<br>
> just for creates and other non-instance calls like the hypervisor
<br>
> API). Then a blocked task would only affect request for a specificinstance.<br>
> > <br>
> > I'm tending towards ii) as a simple and pragmatic solution in
the <br>
> near term, although I like both iii) and iv) as being both generally<br>
> good enhancments - but iv) in particular feels like a pretty seismic
change.<br>
> > <br>
> > Thoughts please,<br>
> > <br>
> > Phil <br>
> > <br>
> > _______________________________________________<br>
> > OpenStack-dev mailing list<br>
> > OpenStack-dev@lists.openstack.org<br>
> > </font></tt><a href="http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev"><tt><font size=2>http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev</font></tt></a><tt><font size=2><br>
> <br>
> _______________________________________________<br>
> OpenStack-dev mailing list<br>
> OpenStack-dev@lists.openstack.org<br>
> </font></tt><a href="http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev"><tt><font size=2>http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev</font></tt></a><tt><font size=2><br>
> <br>
</font></tt>