Is the performance benefit of a logback AsyncAppender diluted when in a k8s container that simply writes to stdout?
Stdout/err from the container is, usually, connected to a file. It's possibly via some number of pipes rather than a direct thing but somewhere under all of it is probably still a disk that needs to keep up with all the writes. That said, the pipe buffers in your CRI plugin can probably eat most spikes and level them out so "diluted" might be fair, but that depends on the specifics of your CRI and how it's configured.