Stream manager not stopped gracefully

[glrf]

Note: I looked into the stream manager running on a local cluster, but as far as I can tell other parts of Heron and other kinds of clusters may also be affected.

When stopping a topology (heron kill) on a local cluster, the local scheduler sends a SIGTERM signal to all heron executors. The executor handles the signal and in turn sends a SIGTERM to all heron components running in its container.

The stream manager however does not register a handler to handle this SIGTERM, which means it will be killed without calling its destructors or any cleanup code. This seems unintuitive and can lead to unexpected behavior .

In my case, I tried to use the stream managers destructor to export additional latency metrics I collected and noticed that the destructor is never called.

I suggest handling SIGTERM signals in the stream manager and exit gracefully. The eventlib used in the event loop can handle the signal and there is already a function in place that stops the event loop (EventLoopImpl::loopExit()), so the fix is quite simple and as far as I can tell should not have any side effects.

@antiguru

[kramasamy]

@Glorfischi - it will be great if you could send a quick PR.

[antiguru]

See #2950 for the corresponding pull request.