Additional snapshots on replicated datasets

[skek]

Hi All,

I have a freenas box running 11.0-U1 (soon to move to 11.2-U3) that is replicating datasets to a backup system on 11.2-U3. This works.

I want to keep or make snapshots on the remote/destination system longer than the source. I do not wish to keep this data around on the primary for capacity reasons.

If I do not "delete stale snapshots on remote system" I end up with many tens of thousands of snapshots on the backup platform that need to be managed independently. If I do delete stale snapshots things are nicely in sync but I lose the above goal.

After the initial sync of a dataset, can I safely snapshot the remote (receive) datasets on the backup server? Can I use the auto-snapshot capabilities in the GUI to create and delete snapshots without impacting those "managed" by the source? (ie, if I use daily/weekly on the source, and monthly on the remote?) or should I do this manually (ie, using a cron and naming the snaps so as to not conflict with the auto system).

Is there a better way, like not deleting stale snapshots in replication settings and instead using a script to prune out most stale ones but leave a monthly for example?

Is this whole idea a bad one?

Thoughts appreciated.

 

[sretalla]

Is there a better way, like not deleting stale snapshots in replication settings and instead using a script to prune out most stale ones but leave a monthly for example?

I think you have it there... it's just that simple. I think there's even a script in @Chris Moore or @Spearfoot signature for snapshot pruning. I'm sure a little forum searching will turn it up.

 

[sretalla]

And here it is...

https://www.ixsystems.com/community...napshots-similar-to-apples-timemachine.10304/

Props to @fracai for the script.

 

[blanchet]

I want to keep or make snapshots on the remote/destination system longer than the source

You can use zfs hold for this purpose. See the Solaris ZFS documentation for more details.

 

[skek]

Thanks @sretalla and @blanchet

It looks like the script by @fracai is the solution to my problem and thanks for pointing me right to it. I confess I had seen that but hadn't quite appreciated it was the reasonable solution to my problem. I've had a play with the script and I will do some testing with it and see how it goes.

I believe that if I keep a small number of snapshots on the primary system and then use the prune script to keep just some of these for a long time on the replication destination this will work, especially if I ensure the snapshots that exist on the primary also exist on the replication. I'm not positive how important that is (except for the latest one) but time will tell. I don't want to have to resend whole datasets if I stuff it up.

 

[blanchet]

You may be interested also by this thread about dataset replication