Re: [PATCH RFC PATCH 3/3] btrfs: zoned: kick cleaner kthread if low on space

[Boris Burkov]

On Thu, Mar 28, 2024 at 02:56:33PM +0100, Johannes Thumshirn wrote:
> From: Johannes Thumshirn <johannes.thumshirn@xxxxxxx>
> 
> Kick the cleaner kthread on chunk allocation if we're slowly running out
> of usable space on a zoned filesystem.

I'm really excited about this and think it probably makes sense on
not-zoned as well.

Have you found that this is a fast enough to help real allocations that
are in trouble? I'd be worried that it's a lot faster than waiting 30
seconds but still not urgent enough to save the day for someone trying
to allocate right now.

Not a blocking request for this patch, which I think is a definite
improvement, but we could do something like add a pass to find_free_extent
(before allocating a fresh BG?!?) that blocks on reclaim running if we
have kicked it (or errors out in a way that signals to the outer loop
that we can wait for reclaim, if needed for locking or whatever)

> 
> Signed-off-by: Johannes Thumshirn <johannes.thumshirn@xxxxxxx>
> ---
>  fs/btrfs/zoned.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/fs/btrfs/zoned.c b/fs/btrfs/zoned.c
> index fb8707f4cab5..25c1a17873db 100644
> --- a/fs/btrfs/zoned.c
> +++ b/fs/btrfs/zoned.c
> @@ -1040,6 +1040,7 @@ int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror)
>  u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start,
>  				 u64 hole_end, u64 num_bytes)
>  {
> +	struct btrfs_fs_info *fs_info = device->fs_info;
>  	struct btrfs_zoned_device_info *zinfo = device->zone_info;
>  	const u8 shift = zinfo->zone_size_shift;
>  	u64 nzones = num_bytes >> shift;
> @@ -1051,6 +1052,11 @@ u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start,
>  	ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size));
>  	ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size));
>  
> +	if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags) &&
> +	    btrfs_zoned_should_reclaim(fs_info)) {
> +		wake_up_process(fs_info->cleaner_kthread);
> +	}
> +
>  	while (pos < hole_end) {
>  		begin = pos >> shift;
>  		end = begin + nzones;
> 
> -- 
> 2.35.3
>