Add Client Pool

[bufferx]

Usage

from toredis.pool import ClientPool

from tornado.gen import engine as gen_engine
from tornado.gen import Task as gen_Task

from tornado.ioloop import IOLoop

@gen_engine
def info():
    redis_pool = ClientPool(pool_size=20, host='127.0.0.1', port=6379)
    response = yield gen_Task(redis_pool.client.info)

    print response
    IOLoop.instance().stop()

info()
IOLoop.instance().start()

[mrjoes]

I was thinking about connection pool before, but I don't see how it'll help and here's why:

1. Redis is single threaded. It does not matter if there's one connection or multiple connections, it will execute commands one by one, sequentially. toredis is asynchronous redis client, you can send commands as fast as you want and whenever redis sends result back, your callbacks will be triggered;
2. Pool adds complexity - some commands should lock connection in the pool and some should not.

Do you see any benefit to have a connection pool?

[bufferx]

@mrjoes thanks for comment.

In my testing, connection pool will improve the performance when transfer large-byte data.
Statistical data as below:

We can see that 20-50% increase in TPS indicator, especially when dealing with large-byte data over the network, so I think the reason is

1. Both of Tornado and Redis is single threaded. the entire data flow will be perform one by one in a unified channel, it can not process the next GET command generated by another HTTP REQUEST timely when transfer large-byte data which cause delay including network-delay, redis-self delay and tornado-self read delay.

2. Multiple Connection make the processing of the entire data flow much more simultaneously, therefore reduce waiting for delay, and improve the indicator of transaction per second.

TEST DEMO: toredis_pool_test.py