We push data into shared memory. Then we have clients that can read the memory and present it. This makes it possible to log millions of lines per second with a very limited cost.

This has the benefit of making logging more or less asynchronous. You still need to handle the logs coming out of this, of course.