Sending snapshots doesn't use the latest snapshot created by znapzend, but rather the latest available

[cserem]

Hello!

I have a system with 2 pools: live_pool archive_pool

There is a znapzend schedule for live_pool to send snapshots to the archive_pool. But I also have another system pulling backups from live_pool, quite rapidly. Often znapzend uses this other snapshot for replication instead of its own.

Example: Znapzend wakes up and creates live_pool@2021-09-25-090000 In the background this other task is creating live_pool@whatever-created-just-after-2021-09-25-090000 The same Znapzend job, when transferring live_pool to archive_pool it is then using this snapshot instead of it's own: zfs send -i live/somedata@2021-09-25-084500 live/somedata@whatever-created-just-after-2021-09-25-090000 '|zfs recv -F archive/somedata

If znapzend is running with '-I', this is not a problem, since the snapshot created by znapzend would also be transferred (just have to have a cleanup job for the "alien" snapshots, generated by the other task.)

But this is a problem if it is running with '-i' since the SRC and DST might come out of sync in respect of snapshots managed by znapzend.

Would there be a way to force znapzend to use the snapshot created by itself for replication?

Thanks

[jimklimov]

Check the pattern your znapzendzetup defined to identify snapshots created by znapzend? On my systems it is quite an explicit znapzend-<timestamp> but by default IIRC it could be just <timestamp> so perhaps your other tool's snapshots also match?

[jimklimov]

From my side, I did post a number of PRs in the past to optionally allow a sort of opposite effect: to let znapzend replicate not only increments between its own snapshots (that matched the pattern) but also others, e.g. manually named points that would not be cleared away, such as before/after some significant reconfiguration or upgrade (the latter is usually a series of snapshots named by beadm and pkg on Solaris and descendants).

These were supposed to be non-default options, so now I worry if something misfired from my contributions? :)

[cserem]

Thank you for picking this up! In my case znapzend's snapshots are only named after creation (no "znapzend-" prefix like in your case): Example: 2021-11-07-121500

The other machine makes: syncoid_backup_2021-11-07:12:38:39-GMT01:00

..even if znapzend doesn't use exact pattern match (eg: ^<timestamp>$ vs <timestamp> ) it shouldn't match the other machine's snapshots because of the different hour:minute:seconds pattern. (That is why I am guessing it uses the most recent one.)

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.