[ILUG] ReiserFS bugs

[Niall O Broin]

I was stress testing a new RAID subsystem with a 200GB Reiser FS running a
script called stress.sh that I found somewhere (takes a tree of data, by
default /usr/doc, puts an md5sum of every file in it into a file, then runs
parallel copies (by default 4, though I was running 8) of that tree on to
the FS to be stressed, puts an md5sum of every file in each copy into a
file, and diffs each of those files against the original md5sum file). It's
a reasonably stressful test - the load on an 850 MHz Athlon is around 10.
I've run this three times so far. Twice it was on a Reiser FS and each time
the machine hung - load went up to nearly 20 (before top stopped
refreshing), disk lights were all out after about 1 hour and 3 hours. 

On the advice of the RAID subsystem vendor, I retried the test on an ext2
filesystem and it ran overnight with no problems (well, that's a lie, there
was occasional output from the diff, but I think that if you're copying that
amount of data, an occasional bit error is not surprising).

This is a little distressing, to say the least, particularly since I have a
production box with a Reiser FS (although it's somewhat smaller than this one)

I just started another test on a Reiser FS and I got an error like the
following:

wait_buffer_until_released: nobody releases buffer (dev x7441, size 4096,
blocknr 5570560, count 3, state(UPTODATE, CLEAN, UNLOCKED)). Still waiting
(4800000)  JDIRTY !JWAIT

According to the RAID vendor

Solution : This is a bug in reiserfs which happens when the system uses
virtual memory.  Increasing RAM can clear the problem.  Changing  to  ext2
file system is  better.


although I had in fact turned swap off on the box before running this test,
so it appears that the problem is not dependent on VM. It also does not seem
to have been the cause of the earlier hangs because the only occurences of
this error in /var/log/messages are the two corresponding to this run of the
test which I just saw on the screen.

Have any of you using Reiser on big filesystems (Dave Rynne ?) noticed any
problems or glitches of any kind. I just had a look in the messages on my
production box and I don't see any instance of this error, nor have I
noticed anything untoward, but the existence of this problem has me worried.
A little googling returned a number of references to the problem, but no
suggestion as to its severity.



Regards,


Niall