Hi David,
On Fri, 27 Mar 2009 15:20:05 +0900 (JST), Ryusuke Konishi wrote:
> Hi,
> On Fri, 27 Mar 2009 06:55:56 +0100, David Arendt <[email protected]> wrote:
> > Hi,
> >
> > one thing I forgot to mention, in /etc/nilfs_cleanerd.conf I changed
> > n_segments_per clean to 20 in order to clean faster when running the
> > cleaner manually. Could this have any influence ?
>
> Yes, maybe. It raises memory pressure then may induce unusual path of
> execution like cache invalidation. It may even increase the chance of
> revealing underlying problems in relocation of on-disk blocks.
>
> Decreasing cleaning_interval is safer in general. We'll try the
> condition.
>
> Regards,
> Ryusuke
I examined the case of nsegments_per_clean = 20 and met an
inconsistent state as follows:
# lssu -a
SEGNUM DATE TIME STAT NBLOCKS
...
7418 2009-03-27 18:41:33 -d- 2048
7419 2009-03-27 18:41:48 -d- 2048
7420 2009-03-27 18:42:08 -d- 2048
7421 2009-03-27 18:42:28 -d- 2048
7422 2009-03-27 18:42:48 --- 2048
7423 2009-03-27 18:43:03 --- 2048
7424 2009-03-27 18:43:23 -d- 2048
7425 2009-03-27 18:43:33 ad- 1166
7426 ---------- --:--:-- ad- 0
7427 ---------- --:--:-- --- 0
...
Here, the segment 7422 and 7423 are in-use but not dirty.
This is crucial because these segments will be reallocated and
overridden later. I suspect there is a bug of error handling
somewhere, and it evaporates the dirty flag and causes the crash.
If you have a (not broken) nilfs partition made under heavy stress,
could you try ``lssu -a'' likewise ?
I'll dig into this from now.
Regards,
Ryusuke Konishi
_______________________________________________
users mailing list
[email protected]
https://www.nilfs.org/mailman/listinfo/users
