Socket pool

[bakins]

A little rough. Adds an options table to connect() with a pool option per issue 133.

[agentzh]

Thanks! I'll look into your patch tomorrow :)

[agentzh]

Some questions/suggestions:

1. Could you please add some test cases for this new feature (including the cases for both TCP sockets and unix sockets)?
2. It seems that your patch disallow option tables for unix domain sockets?
3. I hope that the "pool" option is the whole key for distinguishing connection pools, that is, we should not prepend "host:port" or "path" to the "pool" option value such that the user can choose to make several different backends to share a single pool if she wishes.

Thanks!
-agentzh

[bakins]

Sure. I hacked that together in between meetings, so I know it's pretty rough.

on number 3 -- good point, I didn't do it that way but that makes more sense. We'd just need to document the behavior.

I may start from scratch as it's pretty ugly as is.

[bakins]

I probably need to squash down the commits. Added tests.

[agentzh]

No worries, I'll commit a single diff rather than your original commits :)

[bakins]

I'm a commit early commit often sort of guy ;)

Tests passed. I need better tests, but unix sockets are working now. Took more hackery than I wanted. I'm sure the code can be cleaned up.

[agentzh]

i've committed a modified version of your patch to the "socket-pool" branch:

1d8d157

Please have a look ;) Feedback welcome!

Thanks!
-agentzh

[bakins]

Looks good. I'll run some tests tomorrow.

You handled the table/key much better than I did.