Re: tcp_outgoing_address accounting.

On Wed, 15 May 2002, Henrik Nordström wrote:

> You completely lost the ACL part here, didn't you (#if 0 segment)? The ACL
> part is quite important to many people...

Yeah, thats where it gains most of its *hack* status, i think i totally
butchered it ;) If you have a look this is what i was thinking of
attempting...

addr = aclMapAddr(Config.accessList.outgoing_address, &ch);
if (addr.s_addr == INADDR_ANY)
        return adaptiveLoadBalance(Config.accessList.outgoing_address);

But we will match on the last tcp_outgoing_address x.x.x.x, so that won't
work.

> What would be great is if this could be done using an generic ACL mechanism
> to select 1 of N in a group of things.. This way the same mechanism could
> apply to
>
> tcp_outgoing_address
> tcp_outgoing_tos
> cache_peer_access
>
> and any other ACL driven selection mechanims.
>
> This would put the implementaion mainly in acl.c..

I'll have to dig deeper into that..

> And then we have the question on persistence.. Most people would really like
> each user to stick to one of the IP's to work around stupid websites using
> source IP in authentication/session management..

That would be fixed if i can get the ACL checking re-incorporated.

> And then there is the question on supporting secondary alternatives when the
> first fails.. which makes me thing that perhaps this should be done in a two
> stage process
> a) Collect possible the alternatives by ACL driven processing
> b) Load distribution, by "Select 1 of N", or by "Reorder alternatives
> according to load".

I'm going to rethink the whole thing and try and undo my breakage first
then graft in the feature afterwards.

> Obviously some brain work needed here to find the correct approach.

Unfortunately for me a resource in short supply ;)

Regards,
        Zwane Mwaikambo

-- 
http://function.linuxpower.ca