Hello!
I was profiling some nodes in my project recently and noticed an issue in case of high incoming messages rate(~2-3K hz). In this case rcl_wait consumes significant amount of CPU. About 30-50% of rcl_wait execution time is rcutils_logging_logger_is_enabled_for:

It looks like most of overhead can be eliminated by removing rcutils_logging_severity_level_from_string and switching to integral log_level as mentioned in comment here and here