Redis memory leak.

[robertnishihara]

To reproduce the problem, do the following

import ray
import numpy as np

ray.init()

@ray.remote
def f(x):
    return 1

x = np.zeros(1000)

for _ in range(100000): f.remote(x)

Now the non-primary Redis shard should be using a decent amount of memory. Check how much it is using with ps aux | grep "redis-server ".

Attach a Redis client to it with

redis-cli -p <PORT>

You can get the port from the output of ray.init() or from ps aux | grep "redis-server ".

In the Redis client, do FLUSHALL to flush all of the keys.

Check how much memory the Redis server is using. It's still using the same amount.

[robertnishihara]

cc @TWAnthony

[robertnishihara]

Fixed by #1797.