Hello Brian,
On Do, 2021-07-08 at 13:42 -0700, Brian Bunker wrote:
> In a controller failover do not fail paths that are transitioning or
> an unexpected I/O error will return when accessing a multipath device.
>
> Consider this case, a two controller array with paths coming from a
> primary and a secondary controller. During any upgrade there will be a
> transition from a secondary to a primary state.
>
> [...]
> 4. It is not expected that the remaining 4 paths will also fail. This
> was not the case until the change which introduced BLK_STS_AGAIN into
> the SCSI ALUA device handler. With that change new I/O which reaches
> that handler on paths that are in ALUA state transitioning will result
> in those paths failing. Previous Linux versions, before that change,
> will not return an I/O error back to the client application.
> Similarly, this problem does not happen in other operating systems,
> e.g. ESXi, Windows, AIX, etc.
Please confirm that your kernel included ee8868c5c78f ("scsi:
scsi_dh_alua: Retry RTPG on a different path after failure").
That commit should cause the RTPG to be retried on other map members
which are not in failed state, thus avoiding this phenomenon.
> [...]
>
> 6. The error gets back to the user of the muitipath device
> unexpectedly:
> Thu Jul 8 13:33:59 2021: /opt/Purity/bin/bb/pureload I/O Error: io
> 43047 fd 36 op read offset 00000028ef7a7000 size 4096 errno 11
> rsize -1
>
> The earlier patch I made for this was not desirable, so I am proposing
> this much smaller patch which will similarly not allow the
> transitioning paths to result in immediate failure.
>
> Signed-off-by: Brian Bunker <brian@xxxxxxxxxxxxxxx>
> Acked-by: Krishna Kant <krishna.kant@xxxxxxxxxxxxxxx>
> Acked-by: Seamus Connor <sconnor@xxxxxxxxxxxxxxx>
>
> ____
> diff --git a/drivers/md/dm-mpath.c b/drivers/md/dm-mpath.c
> index bced42f082b0..d5d6be96068d 100644
> --- a/drivers/md/dm-mpath.c
> +++ b/drivers/md/dm-mpath.c
> @@ -1657,7 +1657,7 @@ static int multipath_end_io(struct dm_target
> *ti, struct request *clone,
> else
> r = DM_ENDIO_REQUEUE;
>
> - if (pgpath)
> + if (pgpath && (error != BLK_STS_AGAIN))
> fail_path(pgpath);
>
> if (!atomic_read(&m->nr_valid_paths) &&
>
I doubt that this is correct. If you look at the commit msg of
268940b80fa4 ("scsi: scsi_dh_alua: Return BLK_STS_AGAIN for ALUA
transitioning state"):
"When the ALUA state indicates transitioning we should not retry the command
immediately, but rather complete the command with BLK_STS_AGAIN to signal
the completion handler that it might be retried. This allows multipathing
to redirect the command to another path if possible, and avoid stalls
during lengthy transitioning times."
The purpose of that patch was to set the state of the transitioning
path to failed in order to make sure IO is retried on a different path.
Your patch would undermine this purpose.
Regards
Martin
