Syncoid: sync incremental - hold selected snapshots during sync

[NicholasRush]

Hi there,

i have tested sanoid so much and it is a great tool, but there is a problem when it comes to exclude or include snapshots. The problem is, that the incremental snapshots on the source could be delete through a snapshotengine during the sync. The best would be to hold all these snaps during the sync. If the sync is interrupted, it should check if theres a previous hold tag, finish the sync and remove all holds of the prevoious selected snapshots.

[phreaker0]

@NicholasRush if a snapshot is being used in an ongoing send it can't be destroyed AFAIK and zfs will return an "busy" error.

[jimsalterjrs]

@NicholasRush if a snapshot is being used in an ongoing send it can't be destroyed AFAIK and zfs will return an "busy" error.

Can confirm; I've tested this extensively while making certain syncoid didn't have unexpected failure modes. If the snapshot is necessary for replication to continue, the destroy operation fails "snapshot busy". If the snapshot isn't necessary for replication to continue, the destroy operation succeeds.

[BryanJacobs]

@NicholasRush if a snapshot is being used in an ongoing send it can't be destroyed AFAIK and zfs will return an "busy" error.

Can confirm; I've tested this extensively while making certain syncoid didn't have unexpected failure modes. If the snapshot is necessary for replication to continue, the destroy operation fails "snapshot busy". If the snapshot isn't necessary for replication to continue, the destroy operation succeeds.

These comments are technically correct, but also kind of miss the point/problem here.

Let's say I'm running syncoid and sanoid. I'm sending snapshots of a 40TB dataset. The delta between two monthly snapshots, June and July, is 10TB.

The syncoid send process gets 8TB of the way through that 10TB, then... it fails. The network connection was interrupted.

Sanoid will happily jump in and delete the June snapshot even though I ran syncoid with --use-holds. Now my 8TB of progress is gone.

I'd love it if I had an option to make syncoid hold EACH base snapshot, and leave the holds present after it exits, and only remove them when the sync actually finishes.