A short resume of the original posting:
|> In our site, with 20 hosts, we have setup a mail-server, where all
|> machines mount /var/spool/mail, so the mailbox is in common for everybody.
|> Mail works realy nice...
|>
|> BUT, When for other reasons the MAIL server is down, all machines suffer.
|> Then the response is very slow in those machines (even few minutes till you
|> get response from shell (csh or tcsh)), all this happens even if
|> /var/spool/mail is the only NFS mounted directory.
|>
|> Here is the escenario:
|> 1) at the MAIL SERVER:(Called ariadna)
|> - /etc/exports:
|> ...
|> /var/spool/mail
|> ...
|>
|> 2) At the suffering machines:
|> - /etc/fstab:
|> ...
|> ariadna:/var/spool/mail /var/spool/mail nfs soft,rw,bg,noquota 0 0
|> ...
|>
The answers:
1) If you don't add the `noquota' to the /etc/fstab on the client side, you
will get a very slow response, (When the NFS server is down), specially
during the `login' time, that is, because the machine is trying to check
quotas..
Also you may work-around by linking the `quota check' program to /bin/true.
2) Other reason is that the shell, checks the mailbox (/var/spool/mail/$USER)
each time you go back to command mode. This can be fixed by setting or
changing some environment variables. (biff, mail, MAIL, MAILPATH, MAILCHECK)
you may do a setenv MAIL to /dev/null, (or unnseting the mail variable),
and in place, try using a kind of daemon that checks for in-comming mail..
(like xbiff in X-windows)... so if the daemon hangs, who cares?..
This is nice handled by mailtool(1).
3) If you know that the mail-server will be down, you may do some changes
to prevent other machines do behave slow, by moving the mailhost to other
machine.
To be ready for this, you need:
- Create an alias for the mailhost in the NIS maps:
From: To:
.... ....
128.0.2.116 vivaldi 128.0.2.116 vivaldi mailhost
128.0.2.133 ariadna mailhost 128.0.2.133 ariadna
.... ....
- Move /var/spool/mail in the mail-server to /var/spool/mail.dir
- Create a symbolic link from /var/spool/mail -> /var/spool/mail.dir
- In the /etc/fstab file of each client that mounts /var/spool/mail:
....
mailhost:/var/spool/mail /var/spool/mail nfs rw,bg,noquota 0 0
....
This mounts mailhost:/var/spool/mail to /var/spool/mail.dir,
then, if the server goes down, you can at least scramble to change
the symbolic links, so that they do NOT refer to the hung NFS
mount /var/spool/mail.dir
- Check the `R' option in the /etc/sendmail.cf file, must be:
....
ORmailhost
....
If you change this, don't forget to re-build the /etc/sendmail.fc file!!
(Assuming NIS, and the Master NIS server is not the mailhost machine!!)
So yo use the NIS maps to change the location of the mail-server
(This is the way I have procceded!! Ugly but works)
4) Some told me to reduce the `timeout' option, (in the /etc/fstab file),
the default is 7 (tenths of sec.) and the reentry to 1.. but it caused
/var/spool/mail mounting point to timeout each time, when the the
mail-server (or the network) is overloaded..
5) We still use `soft' option (in the /etc/fstab file), because with `hard'
NFS mounts, the machine gets stuck. (We know that with soft-links we may
loose some info..)
I didn't want to summarize till the results were applied on our site.
I would like to thanks to: (In no special order..)
jaf@jupiter.Sun.CSD.unb.ca (J. Anthony Fitzgerald)
robm@ncsa.uiuc.edu (Rob McCool)
cks@hawkwind.utcs.toronto.edu (Chris Siebenmann)
wolfgang@wsrcc.com (Wolfgang S. Rupprecht)
hermes.chpc.utexas.edu!jones (Bill Jones)
rothstei@mcs.kent.edu (Michael Rothstein)
scott@netcom (Scott Corcoran)
lwv26@cas.org (Larry W. Virden)
mjs@tropel.gca.com (Michael J Shon)
I appologise for all those I may have forgotten, or their replies got after
this posting.
--- Ricky Marek
-- Ricky Marek (System Manager) Fibronics Ltd., Matam Industrial Park, Haifa 31905, ISRAEL phones: +972-4-313690 Fax: +972-4-550550 e-mail: ricky@FibHaifa.com or ricky@fibronics.UUCP
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:39 CDT