Describe the feature would like to see added to OpenZFS
zfs holds should allow the same % notation like zfs destroy, so zfs holds pool/dataset@% should show holds for all snapshots just like zfs destroy pool/dataset@% destroys all snapshots.
How will this feature improve OpenZFS?
When tags are used with replication the most common pattern is create a tag from the last replication snapshot and remove the tag from the previous one. But when resuming a replication with receive_resume_token there's no way to tell which both of those snapshots were, so you don't know what to run zfs release on. To find out which one you have to pass a list of all the snapshots to zfs holds, but how many you can pass is limited by the shell. Or you can run zfs holds for every snapshot which is inefficient and slow especially if it's over ssh.
And it will be useful when zfs release fails for any reason like due to a bad ssh connection. Since zfs hold and zfs release can't be combined into an atomic operation there are times when you'll have a leftover tag that you don't know about, and replication will fail in the future when it tries to destroy a snapshot with a tag on it. By passing a range with % you can efficiently see the holds on all snapshots you're interested in and do clean up if necessary.
Additional context
It will make zfs holds more consistent with zfs destroy.
dawidmachon
commented
10 months ago
Describe the feature would like to see added to OpenZFS
zfs holdsshould allow the same%notation likezfs destroy, sozfs holds pool/dataset@%should show holds for all snapshots just likezfs destroy pool/dataset@%destroys all snapshots.How will this feature improve OpenZFS?
When tags are used with replication the most common pattern is create a tag from the last replication snapshot and remove the tag from the previous one. But when resuming a replication with
receive_resume_tokenthere's no way to tell which both of those snapshots were, so you don't know what to runzfs releaseon. To find out which one you have to pass a list of all the snapshots tozfs holds, but how many you can pass is limited by the shell. Or you can runzfs holdsfor every snapshot which is inefficient and slow especially if it's over ssh.And it will be useful when
zfs releasefails for any reason like due to a bad ssh connection. Sincezfs holdandzfs releasecan't be combined into an atomic operation there are times when you'll have a leftover tag that you don't know about, and replication will fail in the future when it tries to destroy a snapshot with a tag on it. By passing a range with%you can efficiently see the holds on all snapshots you're interested in and do clean up if necessary.Additional context
It will make
zfs holdsmore consistent withzfs destroy.