ZFS Replication

[g---man]

Hi folks, need functionality confirmation from experts.

Here is the scenario: 2 software-identical freenas setups (both at 9.2.1.8), one is ZFS PUSH, second one is ZFS PULL.

Pull is on slow network (1.5Mb), push is on 30Mb. Data size is about 80GB, so first replication (obviously) takes time - 3 days already. PUSH shows 55% done.

Now I need to bounce PULL while initial replication still in progress. And guess what - PUSH resets to ZERO and starts from the beginning :( .

Am I missing something here or this is as-designed ?

 

[cyberjock]

I would say you missed nothing and this is "as expected". The data stream means nothing until it is complete. So your only options are to wait for it to finish to reboot PULL or wait for the replication to run again, from scratch.

Obviously, since you rebooted you're going to have to wait all over again.

Do note that, at the speed of your network, if your internet hiccups and disconnects you'll be right back where you are.

 

[g---man]

Thanks cyberjock.

So ZFS stream is transactional type (binary datastream?), "all or nothing" ? Not like rsync ?

If this is so I might need to rethink my offsite backup strategy, perhaps switching back to rsync.

 

[cyberjock]

ZFS replication is literally a binary datastream. Keep in mind that if you aren't changing a lot of data between replication tasks, ZFS replication offers superior options.

 

[g---man]

another related "issue" - first snapshot completed (took days), then some zero-size snapshots completed as well. Now it is on non-zero snapshot, and it is replicating as slow as first snapshot (~10% a day). However - it shows 116K in "Used" and 227K in "Refer".

Which makes me think "K" is not KB. What is it then? Number of files ?

 

[g---man]

Ignore, user error - 2 snapshots with same name, one refers to root filesystem, (size in K), second one to subfolder (size in G) - that explains long transfer time.