Problem has improved, but is not gone

The high CPU issue that is causing midwest.social to slow way down and timeout on HTTP requests is still a mystery to me. But, it seems to do it less often and for less time now.

Here is CPU usage over 24 hours:

It's a bunch of SELECT statements in Postgres that all seem to fire at one time and take a while to complete as a batch. I've inspected the logs and haven't seen anything unusual. Just stuff federating and posts receiving upvotes.