There is a need for a low-impact info logging that is capable of filtering out unwanted message. Currently it is not feasible to enable log_info in production or high load situations as the logging mechanism is a bottleneck that serializes all threads behind one global mutex. Even if the logging is made lock-free, the sheer volume of messages becomes the next, and often the only, problem. This means that there needs to be a mechanism with which log messages can be selectively chosen for logging.

This could be built on top of the session_trace feature that already collects a trace log in memory before writing it to the log when the session is closed under abnormal conditions. The filtering can be done either when the session is about to close or when some other predefined filtering condition is met (e.g. in-memory log is about to overflow).