[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [dm-devel] [ANNOUNCE] multipath-tools-0.2.4



On jeu, 2004-07-01 at 18:10, Andy wrote:
> On Mon, Jun 28, 2004 at 06:58:54PM +0200, christophe varoqui free fr wrote:
> > Hello,
> > 
> > here goes multipath-tools-0.2.4
> > 
> > Changelog for this release :
> >  
> > 2004-06-28 multipath-tools-0.2.4
> >         * [multipathd] break free from system's libsysfs for now
> >           as it is not that common these days
> >
> Multipathd does not compile, tar does not like the ../libsysfs stuff.  I
> just changed it to use the system libsysfs until it gets fixed.
> 
Can you be more specific, as it works for me.
Anyway, thanks for reporting.

> I'm using the 2.6.7 kernel with the multipath patches only (could that be a
> problem?).
> 
Depends on the quality of your manual patching :)
I would say it's quite risky at best. Better to hand-patch a 2.6.7-mm1
in my opnion.

> It's been a while since I've had a change to work with the dm multipathing
> stuff.  While Multipathing work, it did not fair well when I pulled one of
> the fibre cables out (basically it didn't work at all).  Does multipathd need
> to be running (see below).
> 
Sure : the daemon is in charge of detecting changes on multipaths and
engage reconfiguration on such event.
Event are caught by path-checkers threads (failed paths -> good) and
device-mapper event listeners (good -> failed).

Without the daemon, all you get is the device mapper stopping to route
IO on failed paths ... until there's no paths left as it doesn't do
revalidation.

> Multipathd does not run, it creates the pid file then exits (unless it's a
> hidden process somehow).
> 
Not good. Need more details.
1) strace
2) compile with DEBUG=3 and send daemon.log
3) version crashing, and last version that did not.

> How can I get the dm multipath stuff to work properly with devices that
> already have a partition and fs on them? i.e I have sdb (and others that
> point to the same device, I'll just use sdb to keep it simple) with 1
> partition sdb1, I have a xfs filesystem on sdb1.  When I look in udev I see
> dm-0, if I fdisk dm-0 it has a dm-0p1 partition in it.  I would have to
> mount dm-0p1 right, or does that not work? but that device does not exist in the udev directory. 
> Do I have to create my own dm table entry using the sdb1 entry?
> 
Yes, that's the idea behind dmpartx.
Try dmpartx -a /dev/dm-0 and see if it create the good partition
mapping. I ought to push that into hotplug scripts or udev.

It should cover :
* partitioned devmaps (multipaths or logical volumes)
* partitioned loops
* partitioned MD raids
* partitioned special bdev like cciss and such

regards,
cvaroqui


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]