My understanding was a little incorrect. AFR (the glusterfs translator which actually handles replication), works in the following way.
1. AFR gets an operation 2. AFR sends the operation to all available replicas. This means only those replicas which are online. If none are online, AFR returns with error. 3. AFR waits for replicas to return with success or failure. 4. If all replicas to which the operation was sent return with failure, AFR returns with failure. Even if one replica returns with success, AFR returns success. So, to answer your question gluster return an error only if the write fails on all available replicas or if none of the replicas are available. If the write succeeds on even one replica, but fails on others gluster will return success. Hope this clears your doubts. ~kaushal On Wed, Apr 30, 2014 at 7:40 AM, 可樂我 <[email protected]> wrote: > Thanks Kaushal M. > > When I write file into glusterfs replication volume, > if one replica write failure(brink server offline or brick mount point has > deleted), > it will tell me write failure, this is write operation failure > > Cloud glusterfs do this? > Or I have only way to find it in log file?? > > Thank you very much!!! ^^ > > > > 2014-04-30 0:35 GMT+08:00 Kaushal M <[email protected]>: > > A write will return after it completes on all of the replicas that are >> up. In your example, you have replica 2, but only one of the bricks is >> up. So write will return after it finishes on the up brick. The data >> that was written to just one replica will be copied over, or healed to >> the other replica when it comes back online. This healing is done via >> the self-heal mechanism. >> >> On Tue, Apr 29, 2014 at 8:25 PM, 可樂我 <[email protected]> wrote: >> > Thank your help, Kaushal M >> > >> > but I have some question about it >> > >> > I create glusterfs replica volume in different brick in different node >> > like this >> > #gluster vol create test-vol node-A:/brick1 node-B:/brick2 >> > >> > I mount this volume on /mnt >> > #mount.glusterfs node-A:/test-vol /mnt >> > >> > I write file by using dd cmd >> > #dd if=/dev/zero of=/mnt/test-file bs=1M count=2048 >> > >> > i shudown node-B when the dd cmd is executing >> > >> > wirte test-file is still finish without any error or report any error >> msg >> > >> > if it return only after it is completed on all the replicas. >> > I think it would occur some error >> > >> > I don't understand why it can still finish without error >> > >> > Thanks!! >> > >> > >> > 2014-04-29 22:38 GMT+08:00 Kaushal M <[email protected]>: >> > >> >> For a replica volume a write will return only after it is completed on >> >> all the replicas. >> >> >> >> On Tue, Apr 29, 2014 at 8:06 PM, 可樂我 <[email protected]> wrote: >> >> > Hi everyone, >> >> > I have some questions about replication >> >> > >> >> > when i write file into glusterfs volume, it is 2 replica volume. >> >> > does it return ok when one of the replica write operation complete or >> >> > both >> >> > replica write operation complete? >> >> > >> >> > Can I require it return ok when all replica write complete? >> >> > >> >> > sorry~ my English is poor >> >> > I wish you can understand my qeustions >> >> > Thank you very much! >> >> > Thanks! >> >> > >> >> > >> >> > >> >> > _______________________________________________ >> >> > Gluster-users mailing list >> >> > [email protected] >> >> > http://supercolony.gluster.org/mailman/listinfo/gluster-users >> > >> > >> > >
_______________________________________________ Gluster-users mailing list [email protected] http://supercolony.gluster.org/mailman/listinfo/gluster-users
