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

RE: Software RAID and ext3 problem

I forgot to say that the /var partition is implemented as a RAID 1 array
(two identical partitions on two identical disks on different controllers -
on different Promise cards!).

I also follow HOWTOS - in this case the /var filesystem does have lots of
small files on it (the vast bulk of the files are news articles). So, I
think I'm justified in using 1Kb block size. The HOWTO doesn't say it's a
no-no... it just says that it influences performance. This is a home system,
so, I'm more interested in efficient use of the space rather than the high
performance offered by 4Kb.

I've been using this RAID setup for months (nearly six) under ext2. I've
only started to have problems since implementing ext3 on the filesystems
(one week ago - and I've had a problem every day since implementation, only
on the /var and /usr partitions). I've even copied the data elsewhere and
run mke2fs again to make sure that the fs was correct to start with. To me,
this rules out the IDE/controller being faulty or troublesome as I would
have had problems under ext2 if this was the case.

Now if somebody tells me that ext3, 1Kb blocksize and Software-RAID doesn't
work.... then I'll use 4Kb - but somehow, I think it should all work
together as it is.

To me this looks like some wierd timing thing, probably caused by load
(texpire causes load on the fs) and some interaction with software RAID and

Thanks for the advice and help,


-----Original Message-----
From: ext3-users-admin redhat com [mailto:ext3-users-admin redhat com]On
Behalf Of Matthew Sackman
Sent: 11 November 2001 20:01
To: ext3-users redhat com
Cc: john matthews-net org uk
Subject: Re: Software RAID and ext3 problem

Hi there,

My guess is it's either your IDE cables/controller or it's the 1024
block size - from the Software RAID howto:

The ext2fs blocksize severely influences the performance of the
filesystem. You should always use 4KBblock size on any filesystem
larger than a few hundred megabytes, unless you store a very large
number of very small files on it.

I've been using Software RAID 5 on ext3 for some time (even experimented
with it across enbd which was fun) without problems - but have always
followed the advice in the HOWTO to the letter.



> From: John Matthews <john matthews-net org uk>
> To: ext3-users redhat com
> Subject: Software RAID and ext3 problem
> Hi,
> I'm having a problem with ext3 on my system. I'm running 2.4.13 with the
> appropiate ext3 patch and a software raid array with paritiions as shown
> below:
> Filesystem            Size  Used Avail Use% Mounted on
> /dev/md5              939M  237M  654M  27% /
> /dev/md0               91M   22M   65M  25% /boot
> /dev/md6              277M  8.1M  254M   4% /tmp
> /dev/md7              1.8G  1.3G  595M  69% /usr
> /dev/md8              938M  761M  177M  82% /var
> /dev/md9              9.2G  2.6G  6.1G  30% /home
> /dev/md10              11G  2.1G  8.7G  19% /scratch
> /dev/md12              56G   43G   13G  77% /global
> The /usr and /var filesystems keep switching to ro mode following the
> detection of errors. This has been happening on a daily basis since
> switching to ext3 (usually the /var switches, /usr has done it once).
> I suppose my first question should be: do ext3 and software raid mix?
> If the answer to the above should be 'yes' then here are some more details
> about the error.
> The problem usually occurs during the expiration of news articles. The
> expiration is performed by the program texpire whcihc is part of the
> leafnode package. During the expiration, the following error occurs:
> Nov 10 07:39:17 nijinsky kernel: EXT3-fs warning (device md(9,8)):
> ext3_unlink: Deleting nonexistent file (70034), 0
> This alone doesn't seem to trip the fs into ro mode... leter in the syslog
> (virtually last entry before the fs goes readonly) we get:
> Nov 10 07:42:46 nijinsky kernel: attempt to access beyond end of device
> Nov 10 07:42:46 nijinsky kernel: 09:08: rw=2, want=128661505, limit=976640
> Nov 10 07:42:46 nijinsky kernel: attempt to access beyond end of device
> Nov 10 07:42:46 nijinsky kernel: 09:08: rw=2, want=91811841, limit=976640
> Then the fs goes readonly. Usually it seems to be the /var filesystem
> (which takes out quite a few things that get unhappy when they can't
> create log or lockfiles!).
> In case it makes a difference, I think the blocksize on the /var
> filesystem is 1024.
> Following the errors, it's normally a case of going into single user mode
> and doing a manual fsck on the disk....
> I've switched back to ext2 for my small filesystems but would like to use
> ext3.
> Any help appreciated....
> John.

Ext3-users mailing list
Ext3-users redhat com

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