concerning server list management, connection pooling

[Ofer Nave]

> -----Original Message-----
> From: Brad Fitzpatrick [mailto:brad at danga.com]
> Sent: Thursday, January 18, 2007 8:40 AM
> To: Ofer Nave
> Cc: memcached at lists.danga.com
> Subject: Re: concerning server list management, connection pooling
> 
> The code I wrote for consistent hashing can be used even without its
> integration into the client library (if you use the consistent hashing to
> build your servers => [...] list passed to the client constructor...).

You have code!  Link?

> But eventually it'll be integrated.  You might take a stab at that if you
> have time?

Time is always scarce, but it's important to me to have a solid
implementation of the consistent hashing algorithm, and I'd prefer to have
it integrated as upstream and seamlessly as possible so everyone benefits
(and everyone tests :).  Since I want it, I'm willing to do what I can to
make it happen.

> As for connection pooling:  what exactly do you mean?  The server does
> epoll/kqueue... just keep a ton of connections open from every
> mod_perl/php process on all machines.

Yes, well... I apologize for my ignorance.  I know that event based systems
scale connections effortlessly, but I assumed I'd still hit a limit with
dozens of web servers and dozens of memcaches, perhaps imposed by the
operating system.  I did more poking around in the list archives, and found
this:

"If you're using modern FreeBSD or modern Linux (Linux 2.6), you can have
tens of thousands of persistent connections per process without a sweat."
  -- http://lists.danga.com/pipermail/memcached/2005-May/001425.html

So, never mind - a persistent connections' life for me.

-ofer