Why does an intermediate snapshot have to be destroyed to rollback an earlier snapshot

[unkownerror1]

I use ZFS version 0.8.6.
I have 3 snapshots，xxx/fs2@1, xxx/fs2@2, xxx/fs2@3. when i rollback xxx/fs2@2， xxx/fs2@3 will be destroyed, why not keep xxx/fs2@3

[IvanVolosyuk]

Currently all the snapshots are organized in a sequence. What you suggest will make them organized in a tree: one parent snapshot can have multiple children snapshots. It makes management more complicated. You can actually keep the intermediate snapshot if you want by using 'zfs clone, zfs promote'. This will explicitly instruct zfs that you would like to have a tree of snapshots.

…

On Tue, Jul 19, 2022 at 1:01 PM unkownerror1 ***@***.***> wrote: I use ZFS version 0.8.6. I have 3 ***@***.***, ***@***.***, ***@***.*** when i rollback ***@***.***， ***@***.*** will be destroyed, why not keep ***@***.*** [image: image] <https://user-images.githubusercontent.com/109565529/179654234-f29d27de-ad87-473c-b398-6cbbc01c9f61.png> — Reply to this email directly, view it on GitHub <#13667>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABXQ6HKIKV3SI6JGRURK3MTVUYK7FANCNFSM536M2FXQ> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[unkownerror1]

I do not understand why following my suggestion would make them organized in a tree

[IvanVolosyuk]

If you keep @3 and rollback to @2 and then snapshot @4. This will make @3 and @4 children of @2.

…

On Tue, Jul 19, 2022 at 2:21 PM unkownerror1 ***@***.***> wrote: I do not understand why following my suggestion would make them organized in a tree — Reply to this email directly, view it on GitHub <#13667 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABXQ6HJWCTUR32YJFPNSYGDVUYUNFANCNFSM536M2FXQ> . You are receiving this because you commented.Message ID: ***@***.***>

[unkownerror1]

if clone filesystem ，Snapshots cannot be destroyed. What I need is a snapshot that can be destroyed, do you have any better suggestions

[Finix1979]

As I know, the original snapshot could not be destroyed if there is clone based on it even you modify whole data of the clone.

[IvanVolosyuk]

Well, the snapshot which used by two different filesystems (main + clone) shouldn't be deleted because it effectively branch point. Once you cut all the branches from it you can delete it. With 'zfs promote' you can change the ownership of the shared sequence of snapshots, e.g. move ownership to a clone and delete original. This way you are not locked on keeping the snapshot forever if that what you mean.

[rincebrain]

Currently, AIUI, the way ZFS decides whether something can be actually deleted is "are there any snapshots present between [creation of thing] and [deletion of thing]", which relies on there being one timeline of things with one creation and one deletion time.

If, given snapshots A,B,C,D, you rolled back to snapshot B, and then made changes, you would then have timeline A,B,C,D, and also timeline A,B,E, and the requirement for lifetimes of things to only have a single start and stop is broken and significantly more complicated to reason about.

(Conceptually, one could use something like #13392 to implement a clone-like operation where you end up with the equivalent of cp --reflink on an entire dataset, or do the same magic trick but on the thing you were planning to roll back and destroy to make an alternate timeline...but the devil would be in the details.)