>btw. this can be fixed with readiness probes. and applications should implement them.

What rate would you set on readiness probes for high throughput services (100k requests per second) to keep +99% uptime? :)