Broken TCP connection and endless waiting for events from the dark

[nazarkhanov]

Found a bug which occurres when the connection is broken: the worker waits until the data is sent from the socket. This is my fault. I have no idea how to solve this problem without losing the data that was not sent before closing the socket.

A possible hotfix would be to set a constant timeout to wait for the socket write buffer to clear.

[cbeaujoin-stellar]

Use beat transport, it has an acknowledgment feature

[eht16]

I implemented a timeout, configurable via constants and with a default of 30 seconds.

@nazarkhanov
Do you think 30 seconds are a reasonable default?

Could you maybe test the changes?
I tried to reproduce this by rate-limiting outgoing packets to Logstash via iptables. This seemed to work but is a rather artificial case.