Bug#843953: multipath-tools: LVM on internal disks + multipath-toosl = no multipaths

[Vincent McIntyre]

On Fri, Nov 11, 2016 at 04:06:44PM +0530, Ritesh Raj Sarraf wrote:
> > Related notes:
> > 
> >    On some reboots the system log shows multipath timing out. Below is 'sdd'.
> >    The timeout occurs 33 seconds after the disk was attached.
> >    I was unable to determine the cause of this or reproduce consistently.
> > 
> >    Nov 11 11:11:53  kernel: sd 1:0:0:4: [sdd] 25769805824 512-byte logical
> > blocks: (13.1 TB/12.0 TiB)
> >    Nov 11 11:11:53  kernel: sd 1:0:0:4: [sdd] Write Protect is off
> >    Nov 11 11:11:53  kernel: sd 1:0:0:4: [sdd] Mode Sense: 97 00 10 08
> >    Nov 11 11:11:53  kernel: sd 1:0:0:4: [sdd] Write cache: enabled, read
> > cache: enabled, supports DPO and FUA
> >    Nov 11 11:11:53  kernel:  sdd: sdd1
> >    Nov 11 11:11:53  kernel: sd 1:0:0:4: [sdd] Attached SCSI disk
> >    ...
> >    Nov 11 11:12:25  systemd-udevd[346]: timeout '/sbin/multipath -v0 /dev/sdd'
> >    Nov 11 11:12:26  systemd-udevd[346]: timeout: killing '/sbin/multipath -v0
> > /dev/sdd' [459]
> >    Nov 11 11:12:26  systemd-udevd[346]: '/sbin/multipath -v0 /dev/sdd' [459]
> > terminated by signal 9 (Killed)
> > 
> 
> This is mostly the locking issue which you mentioned in the other bug.

This still occurs, but the system is able to recover from it.
I'll send you the log out of band.

> > Requests:
> > 
> > Can we please have sg3-utils v1.42 added to a stable point release?
> > Also multipath-tools needs to depend on sg3-utils-udev.
> > 
> 
> So as I understand, for now, you've already picked the fixed
> versions from testing for your setup.
 
Yes, that's the best way to summarise it.

> For this issue, multipath-tools doesn't really have any code
> change. The change is required in sg3-utils. For the stable
> release, I'm not sure what to do. 
> 
> 1. New releases are not allowed in stable
> 2. Backports could cover this case, but I really can't commit
>    right now, on when I can get this done.
 
I understand. Probably backports, since it is hard to know what could
happen on systems that work already without the package.

> > It seems a shame to not include the shared lock patch as it avoids
> > a known deadlock and the system still works fine with it included.
> > 
> 
> Indeed. But, as I mentioned in other bug report, it was submitted
> upstream very recently only. And unless something is committed
> upstream, I don't pick it as a fix for Debian Stable.

Oh. I thought the link I gave was to the upstream repository...hm.
That's the repo Christoph's page links to.
I can see the commit in the master branch, what am I missing?

> I'd suggest, you pick the contents of sg3-utils-udev, for now.
> There's nothing other than the udev rules, in that package.
 
I've pretty much done that. I'll stick with the newer sg3-utils too.

> PS: You may also want to make plans for Stretch now. There are
> many more changes in multipath in the Stretch version. Having
> a test setup and reporting bugs in the development phase helps
> much more.

I am planning on this. I had some 'adventures' with 0.6.3 and the
4.7 backport kernel before trying the path I've described.

You may as well close this bug as wontfix, the workaround is just
install sg3-utils-udev.

Vince