Parallel callback execution with ApproximateTimeSynchronizer

[karl-schulz]

Currently, the ApproximateTimeSynchronizer only allows the execution of one callback simultaneously, as the call to self.signalMessage(*msgs) is inside of the lock.

For parallel execution oif callbacks (e.g. using a MultithreadedExecutor) it would be desirable to have the callback call outside of this lock. Parallel threads could then acquire the lock again and start comparing timestamps. This would allow a higher throughput for callbacks that take some time to process, in our case a deep learning model.

Additionally, the usage of a context manager for the lock and more inline comments would probably make the code more readable.

To sum it up, handle inside the lock:

• Compare timestamps
• Clear outdated messages from the buffer
• Find matches, remember them somewhere and remove them from the buffer

Afterwards, outside of the lock

• Call the callback of the matched messages

What do you think about this? We will implement this change internally and report if we can parallel execution to work.

[karl-schulz]

I just realized this is not the Python repo... sorry! Where can I find it?

[rawnam]

Did you find a solution to this problem? Please let me know your thoughts on this.