Re: [PATCH RFC 9/9] btrfs: remove unused bgs on allocation failure

[Johannes Thumshirn <Johannes.Thumshirn@wdc.com>]

On 27.06.25 13:39, Christoph Hellwig wrote:
> On Fri, Jun 27, 2025 at 11:19:14AM +0200, Johannes Thumshirn wrote:
>> From: Johannes Thumshirn <johannes.thumshirn@wdc.com>
>>
>> In case find_free_extent() return ENOSPC, check if there are block-groups
>> in the filsystem which have been marked as 'unused' and if so, reclaim the
>> space occupied by these block-groups.
>>
>> Restart the search for free space to place the extent afterwards.
>>
>> In case the allocation is targeted for the data relocation root, skip this
>> step, as it can cause deadlocks between block group deletion and relocation.
> 
> Assuming an unused BG is one without space in it that just needs a zone
> reset or discard (a quick look at the code seems to confirm that, but
> with some extra caveats):  why don't you reclaim it ASAP once it becomes
> unused, at least modulo those space reservation caveats (which I don't
> understand from that quick look).
> 
> 

I've looked into it looks promising. Threw it into fstests and (up to 
now) nothing broke. So I'll run Damien's scripts on a ZNS drive and 
we'll see if it helps.