Autoscaling in kubernetes for collector and enrich pubsub


I have deployed in GCP kubernetes collector and enrich pubsub with vertical and horizontal autoscaling and I’ve encountered the following problem:

  • some pods were evicted from the nodes because the node remained out of resources eg. insufficient memory or cpu

I want to avoid this from happening by setting requests and limits on pods. Is there somewhere documented in the snowplow docs where I can find minimal and optimal upper limits values for CPU and memory usage for collector and enrich pubsub pods to run?