Multiple Vmware Esxi Virtual Zimbra machine + Openfiler

[lovelord]

Hi all,

in your opinion is possible to use multiple Esxi v.machines to create a sort of high availability service with an Openfiler iSCSi Storage?

Example:

           
  
                incoming mail from xxx@xxx.xxx
                               |
                               |
                       zimbra.mydomain.com
                       /       |        \
                 zbox1       zbox2      zbox3
                       \       |        /
                         \     |      /
                           OpenFiler
                             iSCSi 
                            STORAGE

into specific:

zbox1 = zbox2 = zbox3

so if I've to turn off zbox1 to do maintenance I can still use zbox2 and zbox3 to receive messages, and if while zbox1 is turned off for maintenance and zbox2 "explode" (that quite enough on Murphi's Rule, cough cough... ) still zbox3 is receiveing mail, just in time to turn on zbox1 ...

is this clear? probably not, just ask me more if you need. Thanks 4 support.

PS:

Already read: Need settings for Zimbra OpenLDAP

Already read: NAS/iSCSI

Already read: UNIX and Windows Accounts in Zimbra LDAP and Zimbra Admin UI - Zimbra :: Wiki

[phoenix]

You can certainly do what you're describing but it's not a HA solution, you're only using each instance of Zimbra as a backup server should the others go down. You'd only need to setup your DNS records for multiple backup servers and modify postfix (on the backup servers) to relay back to the primary server when it comes on-line. You caould also consider a multi-server setup, have a look at the documentation here: Documentation for the open source Zimbra Collaboration Suite; email, contacts, and group calendaring The other alternative is to use a Cluster for HA, don't forget that if you're running VM on a single ESXi server you still have a single point of failure in the ESXi server.

[lovelord]

Originally Posted by phoenix

You can certainly do what you're describing but it's not a HA solution, you're only using each instance of Zimbra as a backup server should the others go down. You'd only need to setup your DNS records for multiple backup servers and modify postfix (on the backup servers) to relay back to the primary server when it comes on-line. You caould also consider a multi-server setup, have a look at the documentation here: Documentation for the open source Zimbra Collaboration Suite; email, contacts, and group calendaring The other alternative is to use a Cluster for HA, don't forget that if you're running VM on a single ESXi server you still have a single point of failure in the ESXi server.

Thanks Phoenix,

started reading Multiple Server Documentation on this lunch time.
So what I've to do (if I've correctly understand) is 3 Virtual Machines (LDAP Auth Server only, Mailbox Server only, MTA server only) + 1 Virtual Machine for Openfiler/FreeNas system (or a phisical server, or Hardware Nas) where my Mailbox Server have to point for it's stores (I've to mount external NAS volume I suppose).

This is correct , cannot speak about HA solutions, but it should be most performant then an all-in-one solution. If Esxi server crash we are stopped. If a virtual machine crash we can perform a restore of that specific snapshot and for datas in NAS we can provide surely a raid protection.

There is any way , with Zimbra O.S.E. , to perform a sort of "service continuity" in case one mailserver or component crash?

[lovelord]

Ok, just an UPDATE:

I've created 3 VMware machines running Zimbra separate services:

1- Ldap Auth
2- Mbox Store
3- MTA server

everything is running fine, just i can't understand:

1- how to configure domain -> external LDAP auth -> Filters/Base search (??)

I've followed step by step http://files.zimbra.com/website/docs/Zimbra%20OS%20Multi-Server%20Install.pdf 'till LDAP replication (unusefull to replicate something still to test)

2- I've configured another VMware machine Ubuntu 8.04LTS based to simulate an external NAS; told my storage Zimbra to mount a /mnt/nas on its starts, and everything is goin' very well. My Zimbra storage see /mnt/nas/zimbra/index and /mnt/nas/zimbra/store , user permissions are right and I can configure under my admin UI on store server an

Index2 -> /mnt/nas/zimbra/index
Message2 -> /mnt/nas/zimbra/store

I simulate an accident on index1 or store1 and switch index2 and store2 as active on zimbra store, restarted zimbra servers but if I send a message and tracing it I obtain:

....
Apr 28 18:43:12 mta sshd[7323]: Accepted publickey for zimbra from xxx.xxx.xxx.xxx port 52171 ssh2
Apr 28 18:43:12 mta sshd[7325]: Received disconnect from xxx.xxx.xxx.xxx: 11: Closed due to user request.
Apr 28 18:43:17 mta sshd[7384]: Accepted publickey for zimbra from xxx.xxx.xxx.xxx port 52173 ssh2
Apr 28 18:43:17 mta postfix/qmgr[5378]: 5CFD820620: from=<aaaaa@bbbbbb.cc>, size=7035, nrcpt=1 (queue active)
Apr 28 18:43:17 mta sshd[7386]: Received disconnect from xxx.xxx.xxx.xxx: 11: Closed due to user request.
Apr 28 18:43:17 mta postfix/lmtp[7142]: 5CFD820620: to=<aaaaa@bbbbbb.cc>, relay=zbox.zimbra.xx[xxx.xxx.xxx.xxx]:7025, delay=32, delays=32/0/0/0.07, dsn=4.0.0, status=deferred (host zbox.zimbra.xx[xxx.xxx.xxx.xxx] said: 451 4.0.0 Temporary message delivery failure try again (in reply to end of DATA command))

NB: If i keep open samba share on original "fake" NAS I've shared on Zimbra storage as volume, i can see in store folder that "incoming" and "X" (where X is a number) folders created, and inside them message delivered .... Index folder: empty.

there is something I've to do, like reindex or I've missed some around?

Thanks 4 Support!

[Klug]

If you use a virtualisation software that handles High Availability (VMware with the correct options, XenServer with Essentials, VirtualIron, etc) you'll get HA.

You'll get HA through the virtualisation layer (means if your host crashed, VMs will be moved to another host).

[lovelord]

Originally Posted by Klug

If you use a virtualisation software that handles High Availability (VMware with the correct options, XenServer with Essentials, VirtualIron, etc) you'll get HA.

You'll get HA through the virtualisation layer (means if your host crashed, VMs will be moved to another host).

Thats right, but we are not so big to afford a multiple Esx server license with Vmotion and similar services so... we cannot afford nothing comparable to a very HA solution. What I'm trying to reach is something not closed on itself, but still a little reliable. (We are using a free Esxi license at the moment).

[lovelord]

UPDATE:

Doing a

zimbra@zbox:~$ cat /opt/zimbra/log/mailbox.log |grep nas
java.io.IOException: Unable to rename /mnt/nas/zimbra/store/incoming/1241001684562-0.msg to /mnt/nas/zimbra/store/0/3/msg/0/340-800.msg

I've understand why I've got

[...] said: 451 4.0.0 Temporary message delivery failure try again (in reply to end of DATA command)

it was simply a right permissions on mounted external nas volume. When zbox machine starts and mount in sambafs external nas volume with (/etc/fstab)

//xxx.xxx.xxx.xxx/nas /mnt/nas smbfs user,defaults,credentials=/root/.smbpasswd,uid=zimbra,gid=zimbra 0 0

it gives to local mounted volume on zbox right credentials zimbra:zimbra, but these are different from local ones because gid and uid of zimbra on zbox are different from xxx.xxx.xxx.xxx zimbra gid and uid. So i've done something different remounting external nas volume with right stuff and restart zbox machine.

Now when I switch index1/message1 (local machine) to index2/message2 (nas storage) message are correctly delivered.

Something I'm still wondering ... when I log in into a client , how can I see all my stored old messages if i switched from store1 to store2 ? In theory I should see messages arrived from switching on and turning back to store1 messages arrived from switching back ... Am I mistaking?

There is a way to keep synched store1(local) with store2(nas) , I've to schedule a cronjob with rsync probably?

Last but not the least: Why my index2(nas) is still empty even if I've switched from index1(local) to index2(nas)? Is zbox machine still searching message indexes in index1(local)?

Thanks all.

Andrea.

[lovelord]

updated last post

[bdial]

maybe you want drbd?

[SOLVED] Zimbra on DRBD

[lovelord]

thx for reply bdial,

DRBD is a clustering solution , this will surely work, but I'm trying to reach this goal w/o a clustering solution, due to the fact that we haven't got any NAS solution at the moment. Maybe we'll got one in next months, but still not schedulable.

Following my "schema" and fixing some permissions here and there, now everything is goin' well while testing.
On storage server I've scheduled an incremental backup system and Bacula on an external storage does its work.

Now I'm asking if , over bacula solution, I can implement an Rsync cronjob to sync my messages

rsync -avEK --delete /opt/zimbra/store/ /external/store/

to save messages in real time, but probably this is not enough, and a suppletive rsync is needed to save in real time indexes for messages location, something like

rsync -avEK --delete /opt/zimbra/db/data/ /external/db/data/

Am I mistaking?