For a rails app on production is it a good practice, besides the autoscaler, to set up some puma workers? or is it better just to have more running pods?
For my experience, it is better to have an higher number of smaller (in terms or resource occupation) pods rather then a smaller number of bigger pods.
The reason why I came to this thinking are:
1) a smaller pod is quicker to spawn up and to be moved around by kube controller;
2) the failure of a pod instance is less impacting on the system overall performance (because there's an higher number of other replicas running);
3) a bigger pod could require the cluster autoscaler to spawn up a new node more frequently (it requires more resources to be available on a node in order to be scheduled).
That's my thought, I'd love to have other opinions though.