zrep init with tags fails to send historical snapshots

[ajshea]

When using tags, for example, zrep -t zrep_xenon init zx/pluton xenon backup/zx-pluton the historical snapshots do not get sent to the target filesystem, even if ZREP_INC_FLAG=-I is set in environment or in /etc/default/zrep.

I cannot re-confirm the default behavior without a tag on this filesystem because it already has tags that can't be cleared due to other backup schemes in place.

I have used zrep to migrate filesystems when replacing drives (from a RAIDZ1 to a RAIDZ2 vdev) and between machines and my recollection was that all the historical snapshots came along when not using tags. Perhaps my memory is faulty?

Is it even possible to do this?

Perhaps a separate issue, but could be related -- intermediate snapshots don't seem to be destroyed/expired after they are destroyed on the master filesystem. Again, I had the strong impression that got destroyed on the slave filesystem when not using a tag, but I'm unable to confirm that at the moment.

[ppbrown]

You wrote:

I cannot re-confirm the default behavior without a tag on this filesystem because it already has tags that can't be cleared due to other backup schemes in place.

There arent many software makers that will touch a non-reproducible bug, I'm afraid.
I can look at this, but only if you test "zrep init" with and without the -t options, on some other filesystem.
Either on a seperate system with the same version of OS and ZFS, or if you just make a new filesystem on your current pair of systems.

[ajshea]

Just to clarify, is it possible to send all the historical ZFS snapshots from an existing filesystem to a new filesystem, or am I wasting time?

[ppbrown]

Its a bit.. iffy.

It may be practical, if and only if there are no child filesystems.
Then you can force use of "-R" in the init, and all snapshots will be sent in a somewhat reasonable way. But then ensure that you do NOT use -R on subsequent syncs.

Additionally.. I feel like this is probably a high risk operation when you are overlaying multiple replication chains with zrep and zrep tags.

consider this somewhat "unsupported" as far as zrep goes ;) but this is theoretically how you would do it.

[ajshea]

Thanks. That makes sense.

By "child filesystems" you mean something like "zx/fs1/child"? We only have "zx/fs1" "zx/fs2" but no third level.

In this particular case, we would be using two sets of zrep tags. zrep did not seem to like working without tags once tags had already been used on the filesystem -- it complained that it was already configured and I needed to "clear" the config, which I declined to do.

[ppbrown]

presuming "zx" is your pool name, then yes.

as far as your tags usage goes...
you have set an environment variable to use tags by default, and forgotten about it.
base zrep does not notice custom tags.

[ajshea]

Thank you very much.

Just to clarify, the backup script issues zrep -t zrep_tag_name sync zx/fs for each filesystem.

Unless I'm misunderstanding where it would be set, there is no ZREPTAG environment variable (its rather difficult to create persistent environment or O/S objects in XigmaNAS as it boots to RAMdisk, so nothing is saved unless its inserted in the config). I must have done something wonky when I tried to init new zrep syncs without tags.