Re: [PATCH v1 5/5] events: Unmount all lost mounts at task fail

[Baurzhan Ismagulov <ibr@radix50.net>]

On Thu, Jul 08, 2021 at 06:28:38PM +0200, Jan Kiszka wrote:
> >> No looping, so everything you miss here will continue to be cleaned up
> >> by build_completed() - so what's the point of these changes?
...
> I do get the idea of a perfect world where a single attempt to umount
> works and cleanly pairs with the regular umount when the task succeeds.
> The problem is that I cannot see that this has been validate to truly
> catch the corner cases that the lazy, late-triggered scheme addresses so
> far. If it does not handle that, these umounts are pointless. If it
> does, we can drop the build_completed() handler. Having both indicates
> no progress, just more complexity.

Thanks Jan for the review and the use cases,

I agree that some patches could be refined and better documented.

We'll work on the topics you've raised and respond. I'm afraid we'd need your
help regarding reproducing and analysis of the old issues, because in general,
many issues so far have been rarely reproducible across different build hosts.

With kind regards,
Baurzhan.