Resilver progress should be displayed accumulated across reboots

[spotlesscoder]

Describe the feature would like to see added to OpenZFS

The total percentage of resilver progress and the estimated total remaining resilver duration should be displayed when running zpool status
Thus, the counters of resilver need to be persisted on disk in the metadata somehow.

How will this feature improve OpenZFS?

It will help admins give an estimation when the pool is completely safe again

Additional context

• none

[amotin]

Resilver does save the position across reboots. But due to design of sequential scrub/resilver it can be saved only at certain checkpoints, that might appear spaced quite far in time, depending on the pool size and amount of RAM in the system, that result in process reverting quite far back, possibly even to the beginning.

[spotlesscoder]

Ah, interesting, I didn't know that.

While my issue is primarily about the progress indicator calculation (which I guess is just a consequence of the behaviour @amotin described) - I wonder why it even is implemented like this

I thought that every 5 seconds ZFS writes all pending changes to the disk.

Will the progress also revert if the system is rebooted cleanly (/sbin/reboot)?

PS: @scineram can you please elaborate why you think my proposal is not a good idea?

[amotin]

I thought that every 5 seconds ZFS writes all pending changes to the disk.

It does. And usually much more often even. But the problem is that once we start scrubbing/resilvering in sorted order instead of natural, we can't easily save the position as a single number. We'd have to save all the list, which is excessively expensive.

[spotlesscoder]

You're referring to what is called Sequential Resilver in this presentation, is that correct?
https://docs.google.com/presentation/d/1vLsgQ1MaHlifw40C9R2sPsSiHiQpxglxMbK2SMthu0Q/edit?slide=id.p1#slide=id.p1

No. That presentation is about different thing implemented much later. I am talking about this: https://www.youtube.com/watch?v=SZFwv8BdBj4 .

[n0xena]

in reference to my #16602 a way to pause/abort a current resilver would be nice
in my case I encountered a multi-failure with one bad drive slowing the controller to a crawl
in the end I had to force-powerdown the system as a clean shutdown also got hung by the non-interruptable resilver
given by the mentioned checkpoints a force override would also be neat as in my case even get to the next checkpoint could had taken days
I'm aware this might be an edge case but any option to handle such without everything just locking up (zfs/zpool command, zfs driver, the kernel itself) would be highly appreciate as a system locking itself up on an failure zfs is meant to at least mitigate against is somewhat crontadicting