On 08-Apr-01 at 14:53:22 Philip Hazel wrote:
> Forget about Exim for the moment, and think of the requirement. The only
> possibly way to do this "properly" is to have the two systems accessing
> the same file. It's no good having one system tell the other what it has
> done - by that time the other system may have received a message and
> sent a second autoreply (and be trying to tell the first system...) This
> is likely to be worse after a period when one system is down - there
> will be lots of stuff stacked up on the other, which will take some time
> to get through.
>
Possible, but from looking at the logs not only do we have few vacation
users, but they receive few emails. Obviously this will increase as Summer
approaches, but is also dependant on who the recipient is - I receive a lot
of mail each day.
> How can you get the two systems (not really "master" and "slave" if they
> can both do the entire job) to use the same file? Well, one way is to
[snipped]
Sorry, but I think the idea of another machine is not going to go too far.
We had enough trouble just getting a second mailhub! We are a sizeable
University now and it took some convincing that the staff/students had been
'lucky' so far that our (then) single mailhub had not suffered as serious
fault. If/when it did then there would be no email for the users until we
had configured some other system (which we do not really have) to handle the
mail. It was appreciated that this was not really good enough now - hence we
were allowed to buy a second machine. To ask for a third will probably just
get a 'no' regardless of any reason specified.
> I think the general point I am making is that you can take precautions
> against a machine failing by making regular copies of the files, but you
> have to use just ONE live file at the actual point of doing the job. I
> cannot see any way round this.
>
Okay, I think we have beaten this topic to death now :-) I am not going to
get a 100% perfect solution - I wasn't looking/asking for that - but you
have given me ideas which we shall use.
My current, and probably final thoughts, on this is that we can actually get
both system to do exactly the same thing. The master/slave terminology was
because we had a couple of other 'functions' which needed co-ordination -
however, these have been resolved and vacation processing was the last.
Both systems will send the original message to the recipient - no problem.
Both systems will perform the vacation processing and send back to the
sender an 'I'm on holiday' message if required. Both systems will send a copy
of the original message to the 'other' system, with an added X- header. Both
systems, upon seeing the X- header, will update their once/log files, and
set the 'to=' option to something (probably I'll put the original sender in
the X- header) appended with the 'plymouth.ac.uk' address. By doing this I
can actually get rid of the transport filter and the 'return_message'
option. All plymouth addresses with the X- header will then be blackholed.
So, and this is still to be configured and tested, hopefully the log file
will contain the original sender, albeit with the plymouth domain tacked
onto the end. Your scenario above about backlogged mail when one system is
down will be true. However, I think that is something we will have to test
(?!) to see what amount of vacation mail we do get. If possible we may try
and get all the system-to-system stuff handled first, and then open the mail
service up to the Internet. Of course, how we do that I haven't thought
through yet...
If I can I'd like to get a test version going this week, but with Easter
looming it may be a push. We'll see :-)
Many thanks for the thoughts and suggestions.
Regards,
John.
------------------------------------------------------------------------
John Horne, University of Plymouth, UK Tel: +44 (0)1752 233914
E-mail: jhorne@???
PGP key available from public key servers
