recent bug in dmsetup status of mirror

recent bug in dmsetup status of mirror

[Douglas McClendon]

Hello,

My name is Douglas McClendon, and this is my first post to this list.  I
cannot emphasize how insanely useful I find devicemapper.  Specifically,
in how snapshot is used for the Fedora LiveCDs, and how that enables the
ability to install/live-migrate the livecd filesystem to the host,
without rebooting.

Unfortunately, it appears that between the stock F7 kernel -
2.6.21-1.3194.fc7, and the updated F7 kernel 2.6.22.1-41.fc7, the
dmsetup status results on a mirror have regressed.  The userspace tools
are the same in both cases - device-mapper-1.02.17-7.fc7

With the 2.6.21 kernel, after creating a mirror, dmsetup status shows
the mirroring progress, e.g. [1234/56789].  With the 2.6.22 kernel, it
does not.  I don't have the exact output handy to cut and paste, if need
be I can do that, or file a bug.

I don't remember exactly, but I think it was buggy in that the dmsetup
status returned only information about 1 of the 2 devices, even after
the mirror process completed.  (the reason I know the process completed,
is because the target was a qemu sparse device, whose size I was
tracking with 'du -cms'.  And then upon completion, I reloaded the
table, which housed the system's rootfs with just the new device, and
things didn't fall over dead.  And then I wrote some files, rebooted,
and mounted the 2nd device, and the test files were there.  I.e.
everything was working correctly, just the dmsetup status output seemed
wrong).

Any ideas?

And while I'm here, I'll throw in another question-  Does anybody know
how to throttle the mirroring IO so that the process can take a longer
time, but leave the system more responsive during the process?  I know
that mdadm's raid1 supports write throttling.  I haven't yet figured out
a way to do that with dm.

Thanks,

-dmc
Douglas McClendon

Re: recent bug in dmsetup status of mirror

[Douglas McClendon]

Douglas McClendon wrote:

> 
> Unfortunately, it appears that between the stock F7 kernel -
> 2.6.21-1.3194.fc7, and the updated F7 kernel 2.6.22.1-41.fc7, the
> dmsetup status results on a mirror have regressed.  The userspace tools
> are the same in both cases - device-mapper-1.02.17-7.fc7
> 
> With the 2.6.21 kernel, after creating a mirror, dmsetup status shows
> the mirroring progress, e.g. [1234/56789].  With the 2.6.22 kernel, it
> does not.  I don't have the exact output handy to cut and paste, if need
> be I can do that, or file a bug.

just noticed it is fixed as of 2.6.22.4-65.fc7

I'm still curious whether write-throttling a mirror is possible. 
Anybody have any ideas?  Can the dm thread be ionice-d?  Can the 
destination device be throttled at the block layer independent of dm? 
Can the table just be reloaded rapidly, increasing the device size 
incrementally?  It sure does seem to make the system unusable.  I.e. 
when running the following test, it took 10 seconds to open another tab 
in a gnome-terminal, and then 30+ seconds to launch gedit.  (core duo 
laptop with 1G ram, and reasonably modern sata drive).

#!/bin/bash
dd if=/dev/zero of=./test10ga bs=1k count=1 seek=$(( 1024 * 1024 * 10))
dd if=/dev/zero of=./test10gb bs=1k count=1 seek=$(( 1024 * 1024 * 10))
loa=$( losetup -f )
losetup ${loa} ./test10ga
lob=$( losetup -f )
losetup ${lob} ./test10gb
echo "0 $( blockdev --getsize ${loa} ) mirror core 2 32 sync 2 ${loa} 0 
${lob} 0" | dmsetup create testmir
echo "testmir created"
dmsetup status testmir


thanks,

-dmc