Node-wide rate limits

[dlex]

Problem

Currently, most of our rate limits are shard local. This is sufficient for "protective" rate limits that control a workload's impact on shard-local resources, but insufficient for protecting shared node-wide resources, such as disk.

Who is this for and what problem do they have today?

A customer who wants to protect Redpanda node resources like network bandwith and disk from overloading by thoughput unbound clients on both ingress and egress sides.

Solution

New parameters are defined to limit ingress and egress rate limits. Both these limits are node wide, and they are distributed across node shards in the way that the connections may use them efficiently.

Additional notes

This issue sets ground for #6453.