Scalix Server with replicated (not shared disk) data

[brockz]

Hello

is it possible to have the /var/opt/scalix stuff physicaly replicated and used on a second server (primary/primary state). I dont want t o use a shared wide disk !. I want to use the /var/opt/scalix stuff from server 1 on /var/opt/scalix server2 realtime replicated and the other way same. I readed der Scalix Active Passive HA Guide from DK . Actually i test a little bit with drbd 8(Primary/Primary Mode) with ocfs2 fs on it.
Ist there a better way to do this . Today i used Dirsync to acces on a second server the data from serv1 but i dont have the data on server 2 . We dont want to use a shared Disk we want a replicated one

[florian]

in principle, you can use filesystem-based replication instead of a shared disk.

various of our partners have implemented this using drbd and/or rsync and it works fine when done correctly.

However, note that it is recommended to use snapshot cycles using lvm snapshots to snap consistent states of the message store - similar to online backup.

also please note that we don't currently provide specific instructions and/or support for such configurations. If you need help, please contact Scalix professional services or a qualified Scalix reseller.

Cheers,
Florian.

[brockz]

is it possible to have lvm2 snapshoted message store sync
ed in two way

[florian]

what's the background of the question?

The server will only be able to be active on one side in the failover scenario, so there should be no changes on the "other" side, hence two-way sync is not needed.

Florian.

[brockz]

but i want an active-active scenario with repliucated message store without shared disk no failover so i can login same time on server 1 and server 2 and have same data physically on every server

[florian]

then you have to run 2 separate scalix instances with mutual failover, each with it's own message store, replicated in one direction and users manually distributed between the two instances. You cannot have two active instances accessing the same message store.

Florian.

[brockz]

After you last Post i to be unconcerned about my scalix szenario . I ve made a little picture with a 2 instanza topology with all that hints you give it to me. You think that can be realized ?

[florian]

as long as these are 2 separate instances and you decide which user account by default is homed on which instances.

are the two machines remote or local?

how many users in total?

also note that you will need enterprise edition for this as only enterprise edition is licensed for and supports multi-instance and clustering.

Florian.

[brockz]

we have enterprise lic 70 user . One Server is local and one is on a root server .
vpn rsa 1024 bit connected . On a SDSL 2 Mbit

[florian]

so why would you want a two-way sync in this case. i understand that you want a failover scneario, but i would probably make it one-way with active/passive setup.

high availability is usually improved through simplicity of a setup!

Florian.

[brockz]

because my boss want the message store physically on every server . Is there a better way to do that ?

[florian]

well, the active/passive case would do that.

in the active-active-case you would need to decide that some users would have their home on one server, some other users would have their home on the other server, so would need to split the user population between the two instances.

my understanding is that you want a failover server in the case one location/server goes down, but wouldn't that be active/passive failover clustering?

Florian.

[pinto]

Has anybody tried this with 2 servers (one failover for the other) but both with a commom NFS mounted storage location?

Thanks
Jaime

[florian]

We do not support putting the Scalix message store on NFS. This is based on the way Scalix accesses the disks (small, variable-szed, sync writes) and some NFS client stability issues with regards to proper locking behaviour in Linux 2.6.

Also, we do not support 2 Scalix instances running on separate nodes to access the same message store. It's instance needs to have it's own message store.

Cheers,
Florian.