Pranith Kumar Karampuri
2017-Apr-22 07:18 UTC
[Gluster-users] Quorum replica 2 and arbiter
We are evolving a document to answer this question, the wip document is at https://github.com/karthik-us/glusterdocs/blob/1c97001d482923c2e6a9c566b3faf89d0c32b269/Administrator%20Guide/Split%20brain%20and%20ways%20to%20deal%20with%20it.md Let us know if you still have doubts after reading this. On Tue, Apr 18, 2017 at 8:41 AM, Mahdi Adnan <mahdi.adnan at outlook.com> wrote:> Hi, > > We have a replica 2 volume and we have issue with setting proper quorum. > > The volumes used as datastore for vmware/ovirt, the current settings for > the quorum are: > > > > cluster.quorum-type: auto > > cluster.server-quorum-type: server > > cluster.server-quorum-ratio: 51% > > > > Losing the first node which hosting the first bricks will take the storage > domain in ovirt offline but in the FUSE mount point works fine "read/write" > > Losing the second node or any other node that hosts only the second bricks > of the replication will not affect ovirt storage domain i.e 2nd or 4th > ndoes. > > As i understand, losing the first brick in replica 2 volumes will render > the volume to read only, but how FUSE mount works in read write ? > > Also, can we add an arbiter node to the current replica 2 volume without > losing data ? if yes, does the re-balance bug "Bug 1440635" affect this > process ? > > And what happen if we set "cluster.quorum-type: none" and the first node > goes offline ? > > Thank you. > > Get Outlook for Android <https://aka.ms/ghei36> > > > _______________________________________________ > Gluster-users mailing list > Gluster-users at gluster.org > http://lists.gluster.org/mailman/listinfo/gluster-users >-- Pranith -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.gluster.org/pipermail/gluster-users/attachments/20170422/c82725a6/attachment.html>
Pranith Kumar Karampuri
2017-Apr-22 07:24 UTC
[Gluster-users] Quorum replica 2 and arbiter
On Sat, Apr 22, 2017 at 12:48 PM, Pranith Kumar Karampuri < pkarampu at redhat.com> wrote:> We are evolving a document to answer this question, the wip document is at > https://github.com/karthik-us/glusterdocs/blob/ > 1c97001d482923c2e6a9c566b3faf89d0c32b269/Administrator% > 20Guide/Split%20brain%20and%20ways%20to%20deal%20with%20it.md > > Let us know if you still have doubts after reading this. > > On Tue, Apr 18, 2017 at 8:41 AM, Mahdi Adnan <mahdi.adnan at outlook.com> > wrote: > >> Hi, >> >> We have a replica 2 volume and we have issue with setting proper quorum. >> >> The volumes used as datastore for vmware/ovirt, the current settings for >> the quorum are: >> >> >> >> cluster.quorum-type: auto >> >> cluster.server-quorum-type: server >> >> cluster.server-quorum-ratio: 51% >> >> >> >> Losing the first node which hosting the first bricks will take the >> storage domain in ovirt offline but in the FUSE mount point works fine >> "read/write" >> >> Losing the second node or any other node that hosts only the second >> bricks of the replication will not affect ovirt storage domain i.e 2nd or >> 4th ndoes. >> >> As i understand, losing the first brick in replica 2 volumes will render >> the volume to read only, but how FUSE mount works in read write ? >> >> Also, can we add an arbiter node to the current replica 2 volume without >> losing data ? if yes, does the re-balance bug "Bug 1440635" affect this >> process ? >> > I remember we had one user in Redhat who wanted to do the same procedureon his production workload with 3.8.x upstream bits and it worked fine. I forget which version it was. Ravi, do you remember?> And what happen if we set "cluster.quorum-type: none" and the first node >> goes offline ? >> > cluster.quorum-type none is equivalent to no quorum, so the file can gointo split-brains/ it can serve stale reads. Please don't do this on 3-way replica/arbiter volumes unless you know what you are doing.> Thank you. >> >> Get Outlook for Android <https://aka.ms/ghei36> >> >> >> _______________________________________________ >> Gluster-users mailing list >> Gluster-users at gluster.org >> http://lists.gluster.org/mailman/listinfo/gluster-users >> > > > > -- > Pranith >-- Pranith -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.gluster.org/pipermail/gluster-users/attachments/20170422/ece06966/attachment.html>