Hello, I've been reading a lot of posts of dovecot implementation on clustered filesystems, while my first option was DRDB, I was suggested to take a look at glusterfs for my scenario. Basically I'm considering a two server active-active scenario (or active-passive at least), with mailboxes data replication. We have actually 6,000 mailboxes and almost 300GB of data (and growing). I'm wondering...how had been your experience with glusterfs + dovecot/postfix? I would appreciate any hints. Thanks, Tiago -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20120315/118d8837/attachment.html>
I am running gluster with dovecot and postfix on two active/active servers and it is working relatively OK. Sometimes I get an index corrupted message in the mail.err log, but it's nothing serious. I've had an issue once when files were marked as bad and got tons of input/output errors, but I managed to solve it by removing the files from the background storage on one node and deleting extented attributes on the other node. Jure On 15.3.2012 13:47, Tiago A. Pe?anha wrote:> Hello, > I've been reading a lot of posts of dovecot implementation on > clustered filesystems, while my first option was DRDB, I was suggested > to take a look at glusterfs for my scenario. > Basically I'm considering a two server active-active scenario (or > active-passive at least), with mailboxes data replication. We have > actually 6,000 mailboxes and almost 300GB of data (and growing). > I'm wondering...how had been your experience with glusterfs + > dovecot/postfix? > I would appreciate any hints. > Thanks, > Tiago > > > _______________________________________________ > Gluster-users mailing list > Gluster-users at gluster.org > http://gluster.org/cgi-bin/mailman/listinfo/gluster-users-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20120315/15506e7e/attachment.html>
On Tue, Apr 10, 2012 at 3:42 AM, Jure Koren <jure.koren+gluster at uninet.si>wrote:> On 09/04/12 16:22, James Devine wrote: > >> What mail storage type were you using? I've been testing mdbox on >> gluster and when I put a lot of load on it using more than one thread I see >> lots of corruption problems, both in the index files which store flag >> information and in the data storage also. This doesn't happen on a local >> filesystem but I can duplicate it with one dovecot server talking to one >> gluster server. >> > I am using maildir, as it is easier, beacuse each server writes new files > into unique filenames so there aren't many locking issues. I have also > enabled all the options for nfs storage in dovecots settings. > > The mail files are always ok, but I sometimes get corruptions in dovecot's > temporary index files. When it happens I just delete those files as dovecot > will recreate them. > > But I started to have some issues lately (with 2.3.4-2.3.6) on directories > and I don't know the cause. I am currently testing a ospf solution so the > connection between the nodes would be more stable. I have added a virtual > ethernet adapter with a private ip address and network to both nodes in my > afr setup and ospf makes sure that those are always accessible. > > >Yeah, maildir shouldn't have storage issues, my only concern there would be small file access. I'm setting up a test environment and I plan on comparing mail storage performance between the two storage types as well as a number of other factors. Importing gluster via nfs doesn't seem to have locking/caching issues accessing any of the files, at least not with 2 clients running postal. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20120410/7bf176e9/attachment.html>