Describe the feature would like to see added to OpenZFS
When resuming an interrupted send of multiple snapshots, zfs send -t should attempt to send not only the resumed snapshot, but all following snapshots that were originally requested, if possible.
Consider pool1/dataset has ten snapshots named @snapshot1, @snapshot2, ...@snapshot10, and the following command is used:
During the transfer of @snapshot3 the send is interrupted, and the command is resumed as follows:
zfs send -t "$(zfs get -Ho value receive_resume_token pool2/dataset)" | zfs receive -sd pool2
The send should now produce a stream to finish @snapshot3, before continuing to send @snapshot4 etc. until it reaches @snapshot10 as originally requested.
In the event that @snapshot10 no longer exists, an error will be produced after @snapshot3 has finished sending, since there is no way to continue the original request.
How will this feature improve OpenZFS?
It will make it easier to resume interrupted sends involving multiple snapshots as a result of zfs send -R or zfs send -I @snap or similar.
Additional context
The behaviour of this feature with regards to zfs send -R is a bit complex – the current behaviour effectively ignores the -R flag already, so the most basic proposal would be for the resume to be limited only to the specific dataset that was being resumed (all snapshots will be sent, but child/sibling datasets will be skipped).
However, ideally the resume should resume the full replication stream if that is at all possible – in the worst case this will resume the interrupted snapshot, finish that and then fail, resulting in the same as current behaviour, but in an ideal case it would also complete the replication stream by sending all other snapshots for the current dataset, then moving onto the remaining datasets that were not started (if any).
Describe the feature would like to see added to OpenZFS
When resuming an interrupted send of multiple snapshots,
zfs send -tshould attempt to send not only the resumed snapshot, but all following snapshots that were originally requested, if possible.Consider
pool1/datasethas ten snapshots named@snapshot1,@snapshot2, ...@snapshot10, and the following command is used:During the transfer of
@snapshot3the send is interrupted, and the command is resumed as follows:The send should now produce a stream to finish
@snapshot3, before continuing to send@snapshot4etc. until it reaches@snapshot10as originally requested.In the event that
@snapshot10no longer exists, an error will be produced after@snapshot3has finished sending, since there is no way to continue the original request.How will this feature improve OpenZFS?
It will make it easier to resume interrupted sends involving multiple snapshots as a result of
zfs send -Rorzfs send -I @snapor similar.Additional context
The behaviour of this feature with regards to
zfs send -Ris a bit complex – the current behaviour effectively ignores the-Rflag already, so the most basic proposal would be for the resume to be limited only to the specific dataset that was being resumed (all snapshots will be sent, but child/sibling datasets will be skipped).However, ideally the resume should resume the full replication stream if that is at all possible – in the worst case this will resume the interrupted snapshot, finish that and then fail, resulting in the same as current behaviour, but in an ideal case it would also complete the replication stream by sending all other snapshots for the current dataset, then moving onto the remaining datasets that were not started (if any).