What happens to replications scheduled during power off? What if replication interrupted?

[exp]

I have a node which I only manually power up for replication (I am aware that this is not the intended usage). I would like to manually kick off the replication when I power up that node.

1. What happens if replication job is scheduled but the target node is not reachable?
2. What happens if running replication is interrupted by shutting down a node?

I am trying to decide if it's better to set replication to say, every 5 minutes or to once a year (=basically never) and I handle kicking off replication entirely by myself.

 

[esi_y]

What happens if replication job is scheduled but the target node is not reachable?

It will fail, your last replicated state would be from when it was last successful.

What happens if running replication is interrupted by shutting down a node?

I am sure you will have an error message in the log, but it would appear on the target system as if never even initiated.

 

[exp]

thanks! And what happens if it’s scheduled while already running? I assume it also just fails and retries on the NEXT scheduled time?

Is it then a better idea to schedule all replication jobs every 5 minutes such that once the node turns on one run will work out?

Or is it better to disable automatical scheduling and running it completely by myself via script?

 

[esi_y]

thanks! And what happens if it’s scheduled while already running? I assume it also just fails and retries on the NEXT scheduled time?

I would need to go check in the code, what exactly is happening there, but this is common enough I'd like to believe they figured it out e.g. you can schedule it every minute and create such a big difference on slow link that when next one is scheduled, the first one has not even finished. But if you worry about consistency, that's not going to be a problem. You could be even doing multiple sends with different snapshots concurrently that's not a problem for ZFS.

Is it then a better idea to schedule all replication jobs every 5 minutes such that once the node turns on one run will work out?

Other than plenty of error messages which you might want to filter out, I do not see why it would not work.

Or is it better to disable automatical scheduling and running it completely by myself via script?

They even had own completely separate pve-zsync [1], you may want to check what works better for your use case. I remember myself I was so confused by these that when I needed simple backup, I did my own zfs send | mbuffer ...

[1] https://pve.proxmox.com/wiki/PVE-zsync