Scale Testing Cortex Writes with 200M Active Series

Thank you for so great and detailed article!

About the Avalanche memory issues, one of the options is to use some caching reverse-proxy in front of it. In one of VictoriaMetrics benchmarks the same trick was made for node-exporter. See https://valyala.medium.com/prometheus-vs-victoriametrics-ben...

> RAM usage was around 2.5TB across all of the Opstrace instance

Does it include memory used by Prometheis pods? If yes, what is the rough memory estimation if we exclude Avalanche+Prometheus from it?