Hi,

 

I can’t seem to be able to reproduce your scenario from the bug #1364129.

 

Stopped cluster on a controller then rebooted while tailing neutron server log and nova conductor and API logs on the other two nodes.

Did this with all 3 controllers and was clicking in Horizon while the server was rebooting.

Logs show connectivity lost and 4 dead neutron agents, but things re-connect upon reboot as expected.

Horizon works OK, when I rebooted second and third nodes, got kicked out of horizon and had to log in again.

Saw a warning in neutron server.log about keystonemiddleware.auth_token Usin the in-process toke cache is deprecated as of 4.2.0 release … (this error only appeared during two reboot tests, when I was kicked out of horizon)

 

Now, pcs status had recovered completely after all reboots, but ceph status shows HEALTH_WARN clock skew detected on <two of the nodes>

 

This got me looking into the clock synchronization.  TripleO installs and configured ntpd, but my tripleo-built images also have chronyd installed and enabled.  The result is that ntpd.service  configured with my NTP server is inactive (dead) and chronyd.service with default centos configuration is running.

 

I use the same package versions you’ve reported in the bug, could it be that nova/neutron/glance restarts you experience are related to cluster time sync problems?

 

Let me know if you’d like to compare our environments or if I can help in any other way.

 

Cheers,

Ilja

 

From: Pedro Sousa [mailto:pgsousa@gmail.com]
Sent: Thursday, August 04, 2016 12:02 PM
To: Ilja Maslov <imaslov@dispersivegroup.com>
Cc: Raoul Scarazzini <rasca@redhat.com>; rdo-list <rdo-list@redhat.com>
Subject: Re: [rdo-list] Overcloud pacemaker services restart behavior causes downtime

 

Hi,

 

I've deleted the nova and neutron services but the issue persists, so I guess it's not related.

 

Filing the sosreport.

 

Thanks

 

On Thu, Aug 4, 2016 at 4:32 PM, Ilja Maslov <imaslov@dispersivegroup.com> wrote:

Not on this fresh install, but what I saw few weeks back was that when controller nodes restart, I see services created with FQDN names that were up and I was able to safely clean the original services with short host names.  But I haven’t re-tested controller restarts afterwards.

 

With my fresh install, rabbitmq is not coming up upon reboot (‘unknown error’ (1)), so I need to fix this first before I’m able to proceed with testing.   I’ll let you know how it goes.

 

Ilja

 

From: Pedro Sousa [mailto:pgsousa@gmail.com]
Sent: Thursday, August 04, 2016 11:23 AM
To: Ilja Maslov <imaslov@dispersivegroup.com>
Cc: Raoul Scarazzini <rasca@redhat.com>; rdo-list <rdo-list@redhat.com>


Subject: Re: [rdo-list] Overcloud pacemaker services restart behavior causes downtime

 

Hi Ilja,

 

I noticed that too. Did you try to delete the services that are marked down and retest?

 

Thanks

 

On Thu, Aug 4, 2016 at 4:12 PM, Ilja Maslov <imaslov@dispersivegroup.com> wrote:

Hi,

I've noticed similar behavior on Mitaka installed from trunk/mitaka/passed-ci.  Appreciate if you could put me in CC.

Additional detail is that during initial deployment, nova services, neutron agents and heat engines are registered with the short hostnames and upon controller node restart, these will all show with state=down.  Probably because hosts files are re-written after the services had been started with FQDN as a first entry.  I do not know to what extent pacemaker resources are monitored, but it could be related to the problem you are reporting.

Cheers,
Ilja



-----Original Message-----
From: rdo-list-bounces@redhat.com [mailto:rdo-list-bounces@redhat.com] On Behalf Of Raoul Scarazzini
Sent: Thursday, August 04, 2016 9:31 AM
To: Pedro Sousa <pgsousa@gmail.com>
Cc: rdo-list <rdo-list@redhat.com>
Subject: Re: [rdo-list] Overcloud pacemaker services restart behavior causes downtime

That will be great, thank you, put me in CC so I can follow this.

Thanks,

--
Raoul Scarazzini
rasca@redhat.com

On 04/08/2016 15:29, Pedro Sousa wrote:
> Hi Raoul,
>
> this only happens when the node comes back online after booting. When I
> stop the node with "pcs cluster stop", everything works fine, even if
> VIP is active on that node.
>
> Anyway I will file a bugzilla.
>
> Thanks
>
>
>
>
> On Thu, Aug 4, 2016 at 1:51 PM, Raoul Scarazzini <rasca@redhat.com
> <mailto:rasca@redhat.com>> wrote:
>
>     Ok, so we are on mitaka. Here we have VIPs that are a (Optional)
>     dependency for haproxy, which is a (Mandatory) dependency for
>     openstack-core from which all the others (nova, neutron, cinder and so
>     on) depends.
>     This means that if you are rebooting a controller in which a VIP is
>     active you will NOT have a restart of openstack-core since haproxy will
>     not be restarted, because of the OPTIONAL constraint.
>     So the behavior you're describing is quite strange.
>     Maybe other components are in the game here. Can you open a bugzilla
>     with the exact steps you're using to reproduce the problem and share the
>     sosreports of your systems?
>
>     Thanks,
>
>     --
>     Raoul Scarazzini
>     rasca@redhat.com <mailto:rasca@redhat.com>
>
>     On 04/08/2016 12:34, Pedro Sousa wrote:
>     > Hi,
>     >
>     > I use mitaka from centos sig repos:
>     >
>     > Centos 7.2
>     > centos-release-openstack-mitaka-1-3.el7.noarch
>     > pacemaker-cli-1.1.13-10.el7_2.2.x86_64
>     > pacemaker-1.1.13-10.el7_2.2.x86_64
>     > pacemaker-remote-1.1.13-10.el7_2.2.x86_64
>     > pacemaker-cluster-libs-1.1.13-10.el7_2.2.x86_64
>     > pacemaker-libs-1.1.13-10.el7_2.2.x86_64
>     > corosynclib-2.3.4-7.el7_2.3.x86_64
>     > corosync-2.3.4-7.el7_2.3.x86_64
>     > resource-agents-3.9.5-54.el7_2.10.x86_64
>     >
>     > Let me know if you need more info.
>     >
>     > Thanks
>     >
>     >
>     >
>     > On Thu, Aug 4, 2016 at 11:21 AM, Raoul Scarazzini <rasca@redhat.com <mailto:rasca@redhat.com>
>     > <mailto:rasca@redhat.com <mailto:rasca@redhat.com>>> wrote:
>     >
>     >     Hi,
>     >     can you please give us more information about the environment you are
>     >     using? Release, package versions and so on.
>     >
>     >     --
>     >     Raoul Scarazzini
>     >     rasca@redhat.com <mailto:rasca@redhat.com>
>     <mailto:rasca@redhat.com <mailto:rasca@redhat.com>>
>     >
>     >     On 04/08/2016 11:34, Pedro Sousa wrote:
>     >     > Hi all,
>     >     >
>     >     > I have an overcloud with 3 controller nodes, everything is
>     working fine,
>     >     > the problem is when I reboot one of the controllers. When
>     the node comes
>     >     > online, all the services (nova-api, neutron-server) on the
>     other nodes
>     >     > are also restarted, causing a couple of minutes of downtime
>     until
>     >     > everything is recovered.
>     >     >
>     >     > In the example below I restarted controller2 and I see these
>     messages on
>     >     > controller0. My question is if this is the expected
>     behavior, because in
>     >     > my opinion it shouldn't happen.
>     >     >
>     >     > *Authorization Failed: Service Unavailable (HTTP 503)*
>     >     > *== Glance images ==*
>     >     > *Service Unavailable (HTTP 503)*
>     >     > *== Nova managed services ==*
>     >     > *No handlers could be found for logger
>     >     "keystoneauth.identity.generic.base"*
>     >     > *ERROR (ServiceUnavailable): Service Unavailable (HTTP 503)*
>     >     > *== Nova networks ==*
>     >     > *No handlers could be found for logger
>     >     "keystoneauth.identity.generic.base"*
>     >     > *ERROR (ServiceUnavailable): Service Unavailable (HTTP 503)*
>     >     > *== Nova instance flavors ==*
>     >     > *No handlers could be found for logger
>     >     "keystoneauth.identity.generic.base"*
>     >     > *ERROR (ServiceUnavailable): Service Unavailable (HTTP 503)*
>     >     > *== Nova instances ==*
>     >     > *No handlers could be found for logger
>     >     "keystoneauth.identity.generic.base"*
>     >     > *ERROR (ServiceUnavailable): Service Unavailable (HTTP 503)*
>     >     > *[root@overcloud-controller-0 ~]# openstack-status *
>     >     > *Broadcast message from
>     >     > systemd-journald@overcloud-controller-0.localdomain (Thu
>     2016-08-04
>     >     > 09:22:31 UTC):*
>     >     > *
>     >     > *
>     >     > *haproxy[2816]: proxy neutron has no server available!*
>     >     >
>     >     > Thanks,
>     >     > Pedro Sousa
>     >     >
>     >     >
>     >     >
>     >     >
>     >     > _______________________________________________
>     >     > rdo-list mailing list
>     >     > rdo-list@redhat.com <mailto:rdo-list@redhat.com>
>     <mailto:rdo-list@redhat.com <mailto:rdo-list@redhat.com>>
>     >     > https://www.redhat.com/mailman/listinfo/rdo-list
>     >     >
>     >     > To unsubscribe: rdo-list-unsubscribe@redhat.com <mailto:rdo-list-unsubscribe@redhat.com>
>     >     <mailto:rdo-list-unsubscribe@redhat.com
>     <mailto:rdo-list-unsubscribe@redhat.com>>
>     >     >
>     >
>     >
>
>

_______________________________________________
rdo-list mailing list
rdo-list@redhat.com
https://www.redhat.com/mailman/listinfo/rdo-list

To unsubscribe: rdo-list-unsubscribe@redhat.com