Datadog monitoring and alert for k8s

kubernetes_state.deployment.replicas_available should not be too far off from kubernetes_state.deployment.replicas_desired

Keep a timeseries on # of running pods by node or replica set, and correlate it with resource metrics

For cpu, memory usage, prefer standard metrics from docker, rather than from k8s. Similar princple applies to time-series data - but later on, you can break the data down by pod, and then filter by k8s labels (K8s labels are already applied to Docker metrics)

Focus on monitoring sum of requests on hosts, instead of simple CPU and memory usage

Group/filter your docker metrics by k8s labels instead of hosts