2
SmarterMail HA Environments
Question asked by Robbie Wright - April 1, 2016 at 9:55 AM
Unanswered
Still trying to fit SmarterMail into an actual enterprise environment and digging into some more high availability options. Having multiple instances load balanced for SMTP/IMAP and webmail is what we'd like to do, but the file store is obviously the hold up. Has anybody put their SM datastore on an NFS share between two servers? Both servers could access the data, I'm curious what happens with file locking, permissions issues, etc. Or if you can run to edge smtp instances, both accepting inbound mail, writing to the spool, and then to the file store on the share.
 
Amazon is getting ready to release their "Elastic File System" which is basically glorified NFS between instances. We run HyperV internally and are getting ready to test putting the datastore on an NFS share between two load balances VM's, but I wanted to know if anyone else has tested a setup like this.

1 Reply

Reply to Thread
1
Scarab Replied
April 1, 2016 at 12:08 PM
Robbie,
 
I have tried similar using Microsoft Windows Server 2012R2 iSCSI Initiator and File & Storage Services to access the SmarterMail store located on a SAN. I never got to load testing because the environment became unstable due to iSCSI contention whenever I introduced a third server (seemed to work fine with two servers accessing the same iSCSI Target but would crash on all three simultaneously the moment I connected a third to the same Target). I never could figure out whether it was an issue with Microsoft's implementation of iSCSI (as the Target LUN was an FreeBSD based SAN not a Microsoft iSCSI Target Service) or contention caused by running Windows Server 2012R2 on ESXi rather than HyperV. I regrettably didn't have enough free resources to try it outside of a Virtual environment (or on a HyperV environment) or with the Target LUN located on another Microsoft Windows Server running iSCSI Target Service to isolate the actual cause of the contention.

Never tried NFS though. In theory it should work better in a virtual environment as there shouldn't be any kind of conflict between Windows and the Virtual Host fighting for management control of the resource (as it is at the File System level rather than Block level), and if you only have Windows Servers accessing the NFS Share then there shouldn't be an issue with byte range locking. In theory the only sticky-widget would be latency while one server is waiting for a lock on a file to be released before being able to write to it. Running a medium to high volume of mail traffic could certainly cause potential high-latency problems.

I'm too am curious if anyone has succeeded at doing this.

Reply to Thread