The collector was OOMKilled when receiving massive events

[dxsup]

As we all know, the collector consumes events from the probe via the UNIX socket domain using ZeroMQ. I have found that if the probe sends too many events and the collector is unable to keep up with the rate of incoming messages, the collector will use more and more memory until it is OOMKilled. And even if I stopped the test load on this node which means the collector was able to keep up with the events, the memory usage would not be deallocated. I know there is an option ZMQ_HWM which should have limited the memory allocated, but after I set this option, the problem was still there.

This seems like what ZeroMQ expects to do, so we should adapt the usage of ZeroMQ to prevent this happening. See more at zeromq/libzmq#4218

[dxsup]

This issue is invalid after merging #264 because ZeroMQ has been removed.