Are people really running software that is making 1k Redis calls/message?
As for email, if email was broken down and nicely distributed over the same period I likely would have no issues. However, there are times where there is much more scanning going on, and a good 4 hours of peak time. This is where issues can crop up. If rspamd scan can't get a return fast enough, that test will be skipped. It still works, but I'd rather have all tests go through. So lets then take a single email and where redis is used
*Ratelimit checking
* Replies plugin
* Some multi map are stored in redis
*mx check module
* neural module (a bit like bayes but automatic and short term)
* fuzzy module used from spam traps
* history module - for redis admin, stats
So a single email is doing more than a single redis query for each email. The mail server (for smtp), which is not rspamd, is also using redis on its own.
Now imagine a spike of emails, perhaps a large spamming activity that is being stopped. Needing to scan 100 emails in a second isn't impossible at peak. The scanning is also for incoming and outgoing email.
Even at 100 ops/message you should be able to do > 1k messages/sec without breaking a sweat.