----- Original Message -----
From: "Jason R. Mastaler" <[EMAIL PROTECTED]>
To: <[EMAIL PROTECTED]>
Sent: Tuesday, March 11, 2003 6:56 PM
Subject: Re: cdb whitelist


> Jesse Guardiani <[EMAIL PROTECTED]> writes:
>
> > Does anyone know what will happen if someone receives mail while I'm
> > rebuilding a flat file global whitelist?
>
> Rebuild the whitelist to a temp file first, and when that's finished,
> overwrite the original with the temp file. On a POSIX OS, rename()
> operations are guaranteed to be atomic.

Thanks! That will help in the short run. I'll try to implement this tommorrow so that 
I can get this server up and running.


>
> > Is it going to really really slow down my system's performance above
> > about 100 domains?
>
> If you are using wildcards, probably, yes. If there is any way you can
> use a flat file w/out wildcards (just individual addresses) combined
> with the -autocdb option you'll be much better off. Even a .cdb with
> 100,000 addresses shouldn't cause TMDA to break a sweat.

Now, I've moved this to the workers list, because I'd like to talk about the "long 
run" here. A long term solution.

Currently, I only have 50 domains. Most of which DO NOT use TMDA, and most of which 
have a relatively small number of users. My
server is NOT a powerhouse, but I'm currently running about 98% idle CPU most of the 
time.

However, within about two weeks, I'm going to be offering TMDA to all of my current 
virtual domain customers, as well as migrate
another large domain to the server and offer TMDA to them too. I'm a little worried 
that even just with 50 domains TMDA may put a
rather high load on my server if I require each TMDA customer's tmda-filter program to 
look through a 50 domain wildcarded flat file
for every incoming and outgoing email.


Now, as I understand it, CDBs are hash tables, correct? And you can get extremely good 
retrieval speed as long as you know exactly
what you're looking for, correct? It's like a hit or miss thing.

For the current wildcarding system, using a CDB is out of the question, because you 
wouldn't have a clue what you're looking for.
It's a wildcard! It could be any number of things.

But my case is rather specialized. I JUST want to see if the incoming or outgoing 
email's DOMAIN is listed in the CDB. This can be
accomplished without wildcarding, but I don't think it can be accomplished with 
current TMDA filter functionality. I think it may
require a special filter type, an option to a current filter, or some more overhead to 
the current wildcarding code. (I vote for a
new filter type, personally, and I'd like to see the -autocdb and -autodbm 
functionality added for good measure.)

I would think that this functionality will be widely desired. After all, even the main 
TMDA page mentions that less than 6% of
Jason's contacts had to confirm. Why should my trusted customers have to confirm, or I 
to them? I think the proposed functionality
will make TMDA a better software package as a whole, and I think it fits the TMDA way 
of doing things.

So, Jason, would you be opposed to the creation of a new CDB filter type (or similar 
functionality) that simply looks up the domain
part of an email in a CDB?

Or can you or anyone else think of a better way to do this?


>
> > Does anyone know of a better way to do this?
>
> Another way would be what I mentioned before, which is to figure out a
> way in the MTA to circumvent TMDA if the message was generated by a
> trusted domain.

I've thought about this. With qmail, it would be extremely difficult. TMDA is invoked 
on the qmail-local side of the queue. At that
point, it's just an email. No way to know if it's trusted or not without extensive 
patching. I think a domain cdb is a better bet.

Thanks for reading!

Jesse

> _____________________________________________
> tmda-users mailing list ([EMAIL PROTECTED])
> http://tmda.net/lists/listinfo/tmda-users
>

_________________________________________________
tmda-workers mailing list ([EMAIL PROTECTED])
http://tmda.net/lists/listinfo/tmda-workers

Reply via email to