SQLite mailbox data replication between different datacenters

lyudmil-todorov-EPRO

Dear Grommunio,
One question:
In case exchange database availability groups-like deployment is needed in primary/disaster recovery datacenter configuration, there must be mailbox/SQLite instance for every user on grommunio servers in the second datacenter. Is that correct? These must be kept replicated from mailboxes in the primary center and must be activated in case of primary datacenter outage. How are data synched in that case? Which are the approaches, what software is used ?

Thank you

jengelh

This is outside the scope of grommunio. But FWIW, you apply the same cluster concepts/mechanisms just as you would for most other generic software: globally available storage, and a STONITH mechanism to limit the use of the data files to a single host. One conceivable implemetation is building something with corosync/pacemaker + rbd.
Or you use a virtualization layer which is capable of the same (vMotion or something).

mwilliams

jengelh

(To me) known installations with similar setups use:

CephFS
GFS2
NetApp MetroCluster

There is also a known BeeGFS installation which is not in production yet.

For grommunio in clustering, there are only 3 requirements:

a POSIX-compatible filesystem is required (on mounting side)
realistic latency (<10ms)
a proper fencing (like STONITH) to provide cluster-in-recess integrity

Scale-out and cluster handling is properly handled by mysql multiserver meta-information (and properly defined OCF resources for failover handling). For these kind of setups it is recommended to get in touch with our sales team and/or professional services.

lyudmil-todorov-EPRO

Thank you very much! The suggested ideas will be considered.

lyudmil-todorov-EPRO

Of course, for production deployment, we will get in touch with the sales team and/or professional services- now we have POC license. But due to geographical distance "realistic latency (<10ms)" is not possible. Thus, one last question: Does grommunio have a solution when the latency between datacenters is above 10ms (12 - 14 ms in our case) ?

mwilliams

lyudmil-todorov-EPRO

Realistic latency from grommunio to storage. If you have replicas on 2 locations, both locations will have a local dataset mirror with reasonable latency, if not, something is going south. For network, there is no sensible limit. Installations spread over 5 continents perfectly doable if you are able to eliminate multi-master contention (storage). The larger you get, the more it makes sense to think about share-nothing mailbox clusters (logically tied together only with replicated mysql metadata). This way you can "delegate" mailboxes. In comparison, with large exchange scenarios you also don't replicate dags (database availability groups) everywhere.