Has anyone benchmarked the performance hit of CWPP agents on prod database hosts?

The marketing tech background is actually a benefit here, you're probably used to looking at real user impact instead of just box metrics. That's the right lens.

The one piece of advice I'd double down on is testing under actual replication load, not just a single instance. We saw weird network saturation on our read replicas during agent scans that didn't show up in primary benchmarks. It had to do with how the agent's traffic interacted with streaming replication buffers.

For Redis, the monitor vs block mode difference was negligible in our tests, but the kernel module's memory overhead made our Redis instances more prone to evictions under memory pressure. That's a subtle cost you won't see in a short demo.