[svlug] Now THAT'S wierd!!!

[Nate Campi]

On Tue, Oct 09, 2001 at 10:23:58PM -0700, J C Lawrence wrote:
> On Tue, 9 Oct 2001 17:25:11 -0700 
> Nate Campi <nate at wired.com> wrote:
> 
> > See recent discussions on the postfix mailing list for reasons not
> > to use certain journaling filesystems on busy mailservers.
> 
> Quoting the last message in that thread:
> 
>   On Mon, Oct 08, David Chin wrote:
>   > So, am I correct in assuming that the upshot is: "don't use
>   > ReiserFS with= =20 > Postfix because ReiserFS is buggy (and
>   > still buggy in Linux 2.4.10)"? Ca= n=20
> 
>   That's simply not true.
>   Better would be:
>   There are still buggy versions of reiserfs out there.
>   Please go to one of the mailinglist archives of postfix and search for
>   reiserfs.
>   Not long ago we detected a bug in reiserfs with setrlimit, which Chris
>   Mason solved ASAP and I also posted a Bugfix.
>   Maybe that is the very same source of the problem.
> 
> Kanga.Nu processes ~.75M - 1M messages a day via Postfix atop
> ReiserFS.  While its proves nothing/little, I've seen no problems
> there (I also keep my kernel patches up to date).

I use reiserfs as well, with never the smallest problem. However, 
that the thread closed with the fact that the setrlimit problem was 
not even the problem that started that thread, but a second, unrelated 
bug.

I couldn't get that archive search tool to properly gather the whole
thread into one page :(

My busiest mailserver sends out over a million messages in a mass
mailing every day for Wired News (opt-in mailing list), but it's
FreeBSD. I have a Linux box built and ready to take over, but its ext2,
same as anything else production around here. I now wish I had had
reiser on it to see how it works out.

I've been using reiser on all my personal systems since the stable
kernel release that it went into the kernel (2.4.1 IIRC), and have been
quite happy. I just don't use in worst-case scenario conditions like
Wietse details in the thread.

I just got done with a bunch of BIND 8/9 and djbdns testing, I guess
it's time for linux filesystem testing. I never really trust other
peoples results until I try it out myself anyways.
-- 
Nate Campi, UNIX Ops WiReD SF, Terra Lycos DNS, (415) 276-8678  

Nothing so needs reforming as other people's habits.
		-- Mark Twain, "Pudd'nhead Wilson's Calendar"