Move to immutable backups
At the moment, we have 10 HealthShare instance servers (5 x mirrored pairs), where we implement an External Backup approach, using the freeze/thaw commands against whichever server of the pair is the backup mirror member, to complete a VM level backup. These backups are stored to a disk within our control, to purge as required. This approach allows us to deliver a zero downtime backup approach.
Our future backup solution will be storing immutable backups, with the main concern being that the 10 HealthShare instance servers are quite sizable and that our current approach of backing up the current backup mirror member at the time of the backup could generate 10 immutable VM backups and the size this would consume, when the ideal solution would be to only be capturing one side of each pair consistently.
With this in mind, I don't know whether anybody has any experience in a similar situation as to what would be an optimal solution? I have considered the following options and potential concerns:
1) To only ever backup one side of each pair, with a manual effort to keep one server as the preferred primary - the concerns being if it's the current primary that did get backed up, the freeze/thaw will impact system use, or if there was a VM issue for any period of time, we might have no valid backup
2) Implement online backups of the primary member to a centralised drive location and backup the drive, rather than the entire VM - the concerns being the time to complete such large DB backups, the time to restore and in a DR situation, the time to implement a new VM environment to restore the databases
3) Look at implementing a Concurrent External Backup approach - similar concerns to approach 2
Any thoughts are experiences would be gratefully received.
Regards,
Mark
Comments
With a mirror you can use iris qlist to detect the primary and not run there. It is a great idea to snapshot the failover backup/DR async as there can not be a user impact that way.
I would look to SAN devices that offer deduped and compressed and immutable snapshots. Many storage vendors can do this now. The WRC might have reference designs and you should ask them.
Thank you Alexander for your input and advice; through some conversations we believe the way forward will be snapshotting at the storage array level, to eliminate the quiesce impact of VM snapshotting, which again would remove the check on active/standby nodes.