Atin Mukherjee
2015-Sep-01 16:21 UTC
[Gluster-users] Why is it not possible to mount a replicated gluster volume with one Gluster server?
-Atin Sent from one plus one On Sep 1, 2015 9:39 PM, "Joe Julian" <joe at julianfamily.org> wrote:> > > > On 09/01/2015 02:59 AM, Atin Mukherjee wrote: >> >> >> On 09/01/2015 02:34 PM, Joe Julian wrote: >>> >>> >>> On 08/31/2015 09:03 PM, Atin Mukherjee wrote: >>>> >>>> On 09/01/2015 01:00 AM, Merlin Morgenstern wrote: >>>>> >>>>> this all makes sense and sounds a bit like a solr setup :-) >>>>> >>>>> I have now added the third node as a peer >>>>> sudo gluster peer probe gs3 >>>>> >>>>> That indeed allow me to mount the share manually on node2 even if >>>>> node1 is >>>>> down. >>>>> >>>>> BUT: It does not mount on reboot! It only successfully mounts if >>>>> node1 is >>>>> up. I need to do a manual: sudo mount -a >>>> >>>> You would need to ensure that at least two of the nodes are up in the >>>> cluster in this case. >>> >>> Atin, why? I've never had that restriction. >>> >>> It sounds to me like the mount's trying to happen before any bricks are >>> available and/or glusterd is listening. >> >> In a 3 node cluster if two of the nodes are already down and the other >> is rebooted then the GlusterD instance wouldn't start the brick process >> until it receives the first handshake from one of its peer and for that >> you would need your 2nd node to be up as well. This why its recommended >> to have a 3rd dummy node (without any bricks) added to a existing 2 node >> cluster. > > > Again I ask, is this a departure from prior behavior?Not really, its there from long time. IIRC, this change was done when quorum feature was introduced. KP can correct me.> > >>>>> Is there a particular reason for this, or is it a misconfiguration? >>>>> >>>>> >>>>> >>>>> 2015-08-31 21:01 GMT+02:00 Joe Julian <joe at julianfamily.org>: >>>>> >>>>>> On 08/31/2015 10:41 AM, Vijay Bellur wrote: >>>>>> >>>>>>> On Monday 31 August 2015 10:42 PM, Atin Mukherjee wrote: >>>>>>> >>>>>>>> > 2. Server2 dies. Server1 has to reboot. >>>>>>>> > >>>>>>>> > In this case the service stays down. It is inpossible to >>>>>>>> remount the >>>>>>>> share without Server1. This is not acceptable for a HighAvailability>>>>>>>> System and I believe also not intended, but a misconfiguration or >>>>>>>> bug. >>>>>>>> This is exactly what I gave as an example in the thread (pleaseread>>>>>>>> again). GlusterD is not supposed to start brick process if itsother>>>>>>>> counter part hasn't come up yet in a 2 node setup. The reason ithas>>>>>>>> been designed in this way is to block GlusterD on operating on a >>>>>>>> volume >>>>>>>> which could be stale as the node was down and cluster wasoperational>>>>>>>> earlier. >>>>>>>> >>>>>>> For two node deployments, a third dummy node is recommended toensure>>>>>>> that quorum is maintained when one of the nodes is down. >>>>>>> >>>>>>> Regards, >>>>>>> Vijay >>>>>>> >>>>>> Have the settings changed to enable server quorum by default? >>>>>> >>>>>> _______________________________________________ >>>>>> Gluster-users mailing list >>>>>> Gluster-users at gluster.org >>>>>> http://www.gluster.org/mailman/listinfo/gluster-users >>>>>> >>>>> >>>>> _______________________________________________ >>>>> Gluster-users mailing list >>>>> Gluster-users at gluster.org >>>>> http://www.gluster.org/mailman/listinfo/gluster-users >>>>> >>> > > _______________________________________________ > Gluster-users mailing list > Gluster-users at gluster.org > http://www.gluster.org/mailman/listinfo/gluster-users-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://www.gluster.org/pipermail/gluster-users/attachments/20150901/9dbf813b/attachment.html>
Joe Julian
2015-Sep-01 16:21 UTC
[Gluster-users] Why is it not possible to mount a replicated gluster volume with one Gluster server?
On 09/01/2015 09:21 AM, Atin Mukherjee wrote:> > -Atin > Sent from one plus one > On Sep 1, 2015 9:39 PM, "Joe Julian" <joe at julianfamily.org > <mailto:joe at julianfamily.org>> wrote: > > > > > > > > On 09/01/2015 02:59 AM, Atin Mukherjee wrote: > >> > >> > >> On 09/01/2015 02:34 PM, Joe Julian wrote: > >>> > >>> > >>> On 08/31/2015 09:03 PM, Atin Mukherjee wrote: > >>>> > >>>> On 09/01/2015 01:00 AM, Merlin Morgenstern wrote: > >>>>> > >>>>> this all makes sense and sounds a bit like a solr setup :-) > >>>>> > >>>>> I have now added the third node as a peer > >>>>> sudo gluster peer probe gs3 > >>>>> > >>>>> That indeed allow me to mount the share manually on node2 even if > >>>>> node1 is > >>>>> down. > >>>>> > >>>>> BUT: It does not mount on reboot! It only successfully mounts if > >>>>> node1 is > >>>>> up. I need to do a manual: sudo mount -a > >>>> > >>>> You would need to ensure that at least two of the nodes are up in the > >>>> cluster in this case. > >>> > >>> Atin, why? I've never had that restriction. > >>> > >>> It sounds to me like the mount's trying to happen before any > bricks are > >>> available and/or glusterd is listening. > >> > >> In a 3 node cluster if two of the nodes are already down and the other > >> is rebooted then the GlusterD instance wouldn't start the brick process > >> until it receives the first handshake from one of its peer and for that > >> you would need your 2nd node to be up as well. This why its recommended > >> to have a 3rd dummy node (without any bricks) added to a existing 2 > node > >> cluster. > > > > > > Again I ask, is this a departure from prior behavior? > Not really, its there from long time. IIRC, this change was done when > quorum feature was introduced. KP can correct me. >Right, so unless server quorum is enabled, this shouldn't be the problem.> > > > > >>>>> Is there a particular reason for this, or is it a misconfiguration? > >>>>> > >>>>> > >>>>> > >>>>> 2015-08-31 21:01 GMT+02:00 Joe Julian <joe at julianfamily.org > <mailto:joe at julianfamily.org>>: > >>>>> > >>>>>> On 08/31/2015 10:41 AM, Vijay Bellur wrote: > >>>>>> > >>>>>>> On Monday 31 August 2015 10:42 PM, Atin Mukherjee wrote: > >>>>>>> > >>>>>>>> > 2. Server2 dies. Server1 has to reboot. > >>>>>>>> > > >>>>>>>> > In this case the service stays down. It is inpossible to > >>>>>>>> remount the > >>>>>>>> share without Server1. This is not acceptable for a High > Availability > >>>>>>>> System and I believe also not intended, but a misconfiguration or > >>>>>>>> bug. > >>>>>>>> This is exactly what I gave as an example in the thread > (please read > >>>>>>>> again). GlusterD is not supposed to start brick process if > its other > >>>>>>>> counter part hasn't come up yet in a 2 node setup. The reason > it has > >>>>>>>> been designed in this way is to block GlusterD on operating on a > >>>>>>>> volume > >>>>>>>> which could be stale as the node was down and cluster was > operational > >>>>>>>> earlier. > >>>>>>>> > >>>>>>> For two node deployments, a third dummy node is recommended to > ensure > >>>>>>> that quorum is maintained when one of the nodes is down. > >>>>>>> > >>>>>>> Regards, > >>>>>>> Vijay > >>>>>>> > >>>>>> Have the settings changed to enable server quorum by default? > >>>>>> > >>>>>> _______________________________________________ > >>>>>> Gluster-users mailing list > >>>>>> Gluster-users at gluster.org <mailto:Gluster-users at gluster.org> > >>>>>> http://www.gluster.org/mailman/listinfo/gluster-users > >>>>>> > >>>>> > >>>>> _______________________________________________ > >>>>> Gluster-users mailing list > >>>>> Gluster-users at gluster.org <mailto:Gluster-users at gluster.org> > >>>>> http://www.gluster.org/mailman/listinfo/gluster-users > >>>>> > >>> > > > > _______________________________________________ > > Gluster-users mailing list > > Gluster-users at gluster.org <mailto:Gluster-users at gluster.org> > > http://www.gluster.org/mailman/listinfo/gluster-users >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://www.gluster.org/pipermail/gluster-users/attachments/20150901/7764bf9d/attachment.html>