Thanks for the hard numbers. That also makes sense. It should be easy to rationalize that SIGKILL recovery is slower than SIGTERM shutdown for such a database

Thanks for the hard numbers. That also makes sense. It should be easy to rationalize that SIGKILL recovery is slower than SIGTERM shutdown for such a database. If your in-memory cache is empty, it won't be providing any speedups, right? Thus you'll need to go back to disk for everything.