Re: unable to remove device due to enospc

[Ross Vandegrift <ross@kallisti.us>]

On Sat, Jan 15, 2022 at 05:39:12PM -0700, Chris Murphy wrote:
> On Tue, Jan 11, 2022 at 12:41 AM Ross Vandegrift <ross@kallisti.us> wrote:
> >
> > Unallocated:
> >    /dev/mapper/backup      7.24TiB
> >    /dev/mapper/backup_a    2.55TiB
> >    /dev/mapper/backup_b    2.55TiB
> 
> You might be running into this bug I mentioned today in another
> thread. It's an overcommit related bug where the initial overcommit is
> based on a single device's unallocated space, in this case the 7T
> device - but then later logic results in ENOSPC because there aren't
> two devices that can handle that amount of overcommit. If you  can
> remove /dev/mapper/backup, leaving just _a and _b with equal
> unallocated, then try to reproduce. If you can't, you've likely hit
> the bug I'm thinking of. If you still can, then you're hitting
> something different.

I know I can't remove /dev/mapper/backup - I'm trying to do that anyhow, it
fails with enospc.

> However, since these unallocated values are orders of magnitude bigger
> than discussed in the other thread, and orders of magnitude more space
> than is needed to successfully create new chunks on multiple devices -
> I'm not sure. Also I'm not sure without going through the change log
> for 5.15 and 5.16 if this is fixed in one of those so the easiest
> thing to do is try and reproduce it with 5.16. If ENOSPC happens, then
> try to remove the device with 7T unallocated.

I'll build 5.16 and see if it helps.

Thanks,
Ross