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

Re: xfs, reiserfs, ext2 and sync directory updates

On Fri, 20 Jul 2001 03:19, Bulent Murtezaoglu wrote:
> >>>>> "CS" == Craig Sanders <cas@taz.net.au> writes:
> (quoting Russel Coker, all snipped sorry if snipped too much)
>     RC> I haven't bothered investigating this in depth.  If one of my
>     RC> servers crashes the possibility that one message may be lost is
>     RC> really the least of my concerns (email client programs crash
>     RC> far more often and lose much more mail).
>     CS> me too.
> Hmm.  I beg to differ.  I expect people I buy infrastructure from to
> care about not losing mail.  If the daemon 250's the DATA in the SMTP
> conversation it should guard it with more than its life.  In an ideal
> world.

Linux running ReiserFS and Postfix is far less likely to lose mail than 99% 
of all servers on the net.

Causes of mail loss I've seen include failings of Sun Cluster manager (which 
trashes file systems and presumably loses files).  Crashes of commercial mail 
server software also tends to lose lots of mail.  Then there's the issue of 
stupid mail servers that are based on databases (a company I know of was 
considering "rm -rf" on a directory containing 500,000 mail boxes because 
they thought it would take 48 hours to rebuild the indexes).

> Data preservation is a good thing to shoot for.  I don't mean to offend
> either of you and I think I do understand where you're coming from,
> it just seemed an odd thing to have "me too" kind of agreements on.

The thing is, what we are discussing is not something that's guaranteed to 
lose mail.  It's a rather unlikely corner case that would be difficult to 
generate if you tried.  Due to the nature of how it works it's unlikely to 
lose more than one or two emails in a reboot.

If you are running a small mail server (say 1000 accounts) then you will 
probably never see such a condition.  If you are running a large mail server 
then losing one message out of 1,000,000 accounts per crash is of no great 
concern compared to all the other problems a crash incurrs (down-time, 
complaining customers, costs of people who fix the machine and investigate 
it, costs of further down-time for more tests, etc).

http://www.coker.com.au/bonnie++/     Bonnie++ hard drive benchmark
http://www.coker.com.au/postal/       Postal SMTP/POP benchmark
http://www.coker.com.au/projects.html Projects I am working on
http://www.coker.com.au/~russell/     My home page

Reply to: