Re: [PATCH] btrfs: balance RAID1/RAID10 mirror selection

[Wang Yugui <wangyugui@e16-tech.com>]

Hi, Louis Jencka
Cc: Anand Jain

Maybe we still need BTRFS_READ_POLICY_PID because of readahead.

There are readahead inside OS and readahead inside some disk.

For most SSD/SAS and SSD/SATA, there seems readahead inside the disk.
But for some SSD/NVMe,  there seems NO readahead inside the disk.

BTRFS_READ_POLICY_PID cooperates readahead better in some case.

Best Regards
Wang Yugui (wangyugui@e16-tech.com)
2020/10/27

> Hi, Louis Jencka
> 
> Can we move 'atomic_t rr_counter' into 'struct btrfs_fs_info' to
> support multiple mounted btrfs filesystem?
> 
> Although 'readmirror feature (read_policy sysfs)'  is talked about, 
> round-robin is a replacement for BTRFS_READ_POLICY_PID in most case, 
> we no longer need BTRFS_READ_POLICY_PID ?
> 
> Best Regards
> Wang Yugui (wangyugui@e16-tech.com)
> 2020/10/23
> 
> > Balance RAID1/RAID10 mirror selection via plain round-robin scheduling. This should roughly double throughput for large reads.
> > 
> > Signed-off-by: Louis Jencka <louis@waffle.tech>
> > ---
> >  fs/btrfs/volumes.c | 6 +++++-
> >  1 file changed, 5 insertions(+), 1 deletion(-)
> > 
> > diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
> > index 58b9c419a..45c581d46 100644
> > --- a/fs/btrfs/volumes.c
> > +++ b/fs/btrfs/volumes.c
> > @@ -333,6 +333,9 @@ struct list_head * __attribute_const__ btrfs_get_fs_uuids(void)
> >  	return &fs_uuids;
> >  }
> >  
> > +/* Used for round-robin balancing RAID1/RAID10 reads. */
> > +atomic_t rr_counter = ATOMIC_INIT(0);
> > +
> >  /*
> >   * alloc_fs_devices - allocate struct btrfs_fs_devices
> >   * @fsid:		if not NULL, copy the UUID to fs_devices::fsid
> > @@ -5482,7 +5485,8 @@ static int find_live_mirror(struct btrfs_fs_info *fs_info,
> >  	else
> >  		num_stripes = map->num_stripes;
> >  
> > -	preferred_mirror = first + current->pid % num_stripes;
> > +	preferred_mirror = first +
> > +	    (unsigned)atomic_inc_return(&rr_counter) % num_stripes;
> >  
> >  	if (dev_replace_is_ongoing &&
> >  	    fs_info->dev_replace.cont_reading_from_srcdev_mode ==
>