Always flush or close handlers before shutting down the process; otherwise
buffered records may be lost.
Monitor logger.get_dropped() and the warnings emitted by each handler to
detect back pressure early. Increase handler capacities or switch to
blocking/timeout policies when drops are unacceptable.
File-based handlers count flush_interval in records. If you need
time-based flushing, add a periodic handler.flush() in your application.
Blocking overflow policies affect the thread that calls logger.log(). Use
them only when you are comfortable with logging back pressure slowing the
producer because that is what will happen.
Socket handlers serialize to MessagePack with a one-megabyte default frame
limit. Large payloads are silently dropped; consider truncating or chunking
messages before logging.
reset_manager() is a destructive operation intended for tests. Do not call
it while other threads might still hold references to existing loggers.