good dm-mpath test setup?

good dm-mpath test setup?

[Christoph Hellwig]

I've been looking into support blk-mq/scsi-mq in dm-mpath, but I'm
already running into issues with mainline.  I created a couple luns
(without second path) using iet, and then using the iscsi initiator
to access them.  But it seems like no I/O is possible through dm-mpath
for this simple setup as the kpartx processes spawned by udev already
hang.  Is dm-mpath (default config, no multipath.conf) support to work
for a single path setup, or is this expected behavior?

Does anyone have a good default iet or maybe lio config for dm-mpath
testing?

Re: good dm-mpath test setup?

[Hannes Reinecke]

[-- Attachment #1: Type: text/plain, Size: 1346 bytes --]

On 06/05/2014 01:50 PM, Christoph Hellwig wrote:
> I've been looking into support blk-mq/scsi-mq in dm-mpath, but I'm
> already running into issues with mainline.  I created a couple luns
> (without second path) using iet, and then using the iscsi initiator
> to access them.  But it seems like no I/O is possible through dm-mpath
> for this simple setup as the kpartx processes spawned by udev already
> hang.  Is dm-mpath (default config, no multipath.conf) support to work
> for a single path setup, or is this expected behavior?
>
Hmm. This shouldn't happen.

But be aware, that multipath doesn't play well together with systemd 
currently.
I've done plenty of fixes to get it up and running for SLES12; once 
they've stabilized I'll be sending them mainline.

But for a stable testbed I would recommend against systemd currently.

> Does anyone have a good default iet or maybe lio config for dm-mpath
> testing?
>
The easiest bit would be set set up lio in loopback mode, and map 
two luns pointing to the same device.

I've attached a small script I'm using for a quick test here.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		      zSeries & Storage
hare@suse.de			      +49 911 74053 688
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: J. Hawn, J. Guild, F. Imendörffer, HRB 16746 (AG Nürnberg)

[-- Attachment #2: setup_loopback.sh --]
[-- Type: application/x-shellscript, Size: 1240 bytes --]

[-- Attachment #3: Type: text/plain, Size: 0 bytes --]

Re: good dm-mpath test setup?

[Hannes Reinecke]

On 06/05/2014 01:50 PM, Christoph Hellwig wrote:
> I've been looking into support blk-mq/scsi-mq in dm-mpath, but I'm
> already running into issues with mainline.  I created a couple luns
> (without second path) using iet, and then using the iscsi initiator
> to access them.  But it seems like no I/O is possible through dm-mpath
> for this simple setup as the kpartx processes spawned by udev already
> hang.  Is dm-mpath (default config, no multipath.conf) support to work
> for a single path setup, or is this expected behavior?
>
On a complete different footing, I've whipped together some patches 
to simplify the request/bio cloning when using multipath;
basically do _not_ clone the bios, but just modify the rq_end_io 
handler.

You mentioned something at LSF that the current way of handling is 
quite insane, and looking at it I tend to agree.

So if you'd be interested I'll be posting them.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		      zSeries & Storage
hare@suse.de			      +49 911 74053 688
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: J. Hawn, J. Guild, F. Imendörffer, HRB 16746 (AG Nürnberg)

Re: good dm-mpath test setup?

[Christoph Hellwig]

On Thu, Jun 05, 2014 at 02:17:10PM +0200, Hannes Reinecke wrote:
> On a complete different footing, I've whipped together some patches
> to simplify the request/bio cloning when using multipath;
> basically do _not_ clone the bios, but just modify the rq_end_io
> handler.
> 
> You mentioned something at LSF that the current way of handling is
> quite insane, and looking at it I tend to agree.
> 
> So if you'd be interested I'll be posting them.

Please do - that's something mostly orthogonal to my patches, but with
the bio cloning removed and my conversion to blk-mq dm-mpath will become
really lean and simple.

Re: good dm-mpath test setup?

[Christoph Hellwig]

On Thu, Jun 05, 2014 at 02:14:17PM +0200, Hannes Reinecke wrote:
> Hmm. This shouldn't happen.
> 
> But be aware, that multipath doesn't play well together with systemd
> currently.
> I've done plenty of fixes to get it up and running for SLES12; once
> they've stabilized I'll be sending them mainline.
> 
> But for a stable testbed I would recommend against systemd currently.

I'm not going anywhere near that stinking pile if I can avoid it.
This is a Debian Wheezy setup.

> >Does anyone have a good default iet or maybe lio config for dm-mpath
> >testing?
> >
> The easiest bit would be set set up lio in loopback mode, and map
> two luns pointing to the same device.
> 
> I've attached a small script I'm using for a quick test here.

Thanks!