After some benchmarking on the tegola deployment on k8s it doesn't look like we have any sort of capacity problems. Here is our setup:

• We built a karotherian module to be able to mirror traffic to different backends
• We gradually increased from 0% to 100% of the production traffic mirrored from eqiad kartotherian to codfw tegola

We didn't find any latency/io/load issue to any component:

• Postgres (maps2005..2010)
• Swift (thanos)
• Kubernetes

Some findings regarding resource usage:

• We removed the dependency to Cassandra and we now rely on an existing Swift cluster (thanos).
• Based on the load testing i don' think we can reduce resources on Postgres nodes.
• The new maps cluster consists of less machines compared to the cluster we decommisioned.

Given that we already have some docs around infrastructure for the new architecture I would consider this ticket resolved.