As a quick workaround for flush not doing what you want, you can use use_queues=False when constructing the log handler to make each log message trigger a log delivery API call.

I agree in general that the semantics of flush() as specified by https://docs.python.org/3/library/logging.html#logging.Handler.flush mean it should not trigger shutdown by default.

@kislyuk do you know if this is still the case? It looks like it's not since flush looks like it just puts a sentinel on the queue and will force the queues to send when they see it but they continue monitoring the queue after that:

and