The ideal test is probably discourse bench:
Setting it up is a bit tricky, I should at some point make a Dockerfile for it.