When using the Python logger (i.e. not a console logger via PYCBC_LOG_LEVEL env variable) the SDK can sometimes crash as the C++ logger is flushing out its remaining log messages. The problem is that we need to the Python process to build logging messages and we can end up trying to make a C-Python call after the Python process is no longer available. We should look into calling the C++ logger's shutdown method as some point prior to the Python process shutting down. To do that, we could hook into Python's atexit() or add some logic around when the Python SDK's logger sink goes out of scope.

Also, maybe look into the cluster shutdown path to see about shutting down the logger as well as that might allow us to pick up the remaining log messages that might otherwise be truncated if shutting down the logger prior to trying to cluster the cluster connection.