Re: [squid-users] Re: False disk full errors

From: Toens Bueker <toens.bueker@dont-contact.us>
Date: Wed, 7 Oct 1998 18:40:21 +0200

On Tue, Sep 22, 1998 at 07:56:50AM +0100, Chris Tilbury
wrote:

> Daryl Collins (daryl@beaker.htb.com.au) has mailed me to tell me that he's
> having the same problems with Solaris 2.6 and squid beta versions since
> 1.2b19.
>
> From what Daryl has discovered, it looks like a problem with the allocation
> of space within large filesystems and fragmentation, under Solaris 2.6 (he's
> not using disksuite, and has 20x4.2Gb disks).
>
> I've done a little digging and found a Sun bug report (#4141030) to the
> effect that "newfs" was broken in Solaris 2.6 and that it passes incorrect
> parameters to mkfs when it generates filesystems. In essence, newfs will
> create the filesystem with too few cylinder groups:
>
> root@bluebell [scripts]# newfs -Nv /dev/md/rdsk/d32
> mkfs -F ufs -o N /dev/md/rdsk/d32 24914320 80 19 8192 1024 256 1 90 8192 t 0 -1 8 16
> /dev/md/rdsk/d32: 24914320 sectors in 16391 cylinders of 19 tracks, 80 sectors
> 12165.2MB in 261 cyl groups (63 c/g, 46.76MB/g, 5888 i/g)
>
> instead of
>
> root@bluebell [scripts]# newfs -Nv -c 16 /dev/md/rdsk/d32
> mkfs -F ufs -o N /dev/md/rdsk/d32 24914320 80 19 8192 1024 16 1 90 8192 t 0 -1 8 16
> /dev/md/rdsk/d32: 24914320 sectors in 16391 cylinders of 19 tracks, 80 sectors
> 12165.2MB in 1025 cyl groups (16 c/g, 11.88MB/g, 1472 i/g)
>
> (the latter is what a working 2.5.1 would have done). Daryl mentions he's
> seen the problem under 2.5.1, but it occurs less frequently and at higher
> disk capacities. This sounds quite fishy, therefore.
>
> Alas, the only workaround for this bug is to recreate the filesystem. I've
> done this and now all I can do is wait for another week to go by to see what
> happens when I hit the same cache storage levels (we couldn't afford the
> cache downtime needed to backup and restore the cache filesystem).

Ok. We ran into the same problem lately. Obviously there
is a bug in solaris 2.6. Sun said you could overcome the
problem by setting a bigger blocksize for newfs - but
blocksizes bigger than 4k are not supported for sun4u
architectures :-(
You can go with only 50% disk-usage (i guess Sun opted for
that to sell more disks) - or leave Solaris and go with
Linux.

By
Töns

-- 
_o)
 /\\ pgp fingerprint: 9B AC A5 CB C8 CC FC DC  25 B5 26 9A 5D 28 C0 3D
_\_V 
Received on Wed Oct 07 1998 - 09:41:49 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:42:21 MST