More test results

More test results

[Dipl.-Ing. Michael Niederle]

Hi!

I reinstalled my pen drive and continued testing just doing my regular work. (I
always work on a pen drive based system.)

I did *not* start the gc-daemon and worked without any problems for several
hours. Free space was reduced by several GBs - but I started with plenty of free
space left.

During the night I run the gc to claim back unused space. In the morning
the gc seemed to have crashed again. The PC still worked but essential parts of
the system were frozen - so no clean shutdown was possible.

This time the system was severely damaged. The next reboot failed with an
error message (failed mount). (Sorry, I was in a hurry and did not memoize the
exact reading of the message.)

Another issue: Sometimes the gc stops after some time leaving the following
message in the kernel log. (The message seems to stem from another part of the
kernel, but I'm sure it is connected to the gc-daemon.)

vmap() failed while attempting to map about 256 MB of memory

(Sorry, once more - I had stored the exact message on the crashed file system
and had to repeat it from memory.)

I'm using an unmodified 2.6.30 kernel with builtin nilfs (no module).

Greetings, Michael

Re: More test results

[Ryusuke Konishi]

Hi Michael,
On Mon, 15 Jun 2009 12:43:59 +0200, "Dipl.-Ing. Michael Niederle" wrote:
> Hi!
>
> I reinstalled my pen drive and continued testing just doing my
> regular work. (I always work on a pen drive based system.)  I did
> *not* start the gc-daemon and worked without any problems for
> several hours. Free space was reduced by several GBs - but I started
> with plenty of free space left.
>
> During the night I run the gc to claim back unused space. In the
> morning the gc seemed to have crashed again. The PC still worked but
> essential parts of the system were frozen - so no clean shutdown was
> possible.
>
> This time the system was severely damaged. The next reboot failed
> with an error message (failed mount). (Sorry, I was in a hurry and
> did not memoize the exact reading of the message.)

Hmm, did you apply any change to accelerate GC?
Do you think how can I rate up the reproducibility of your problem?

Sounds some effort is needed to clean up quality problems of the
garbage collection.

> Another issue: Sometimes the gc stops after some time leaving the
> following message in the kernel log. (The message seems to stem from
> another part of the kernel, but I'm sure it is connected to the
> gc-daemon.)
>
> vmap() failed while attempting to map about 256 MB of memory

This message means that an ioctl routine of nilfs failed to allocate
virtual memory in kernel.  Setting nsegments_per_clean to a high value
makes this likely.

> (Sorry, once more - I had stored the exact message on the crashed
> file system and had to repeat it from memory.)
>
> I'm using an unmodified 2.6.30 kernel with builtin nilfs (no module).
> 
> Greetings, Michael

Thanks,
Ryusuke Konishi