The prometheus-operator (Metrics) does not recognize when a pod is terminated, hence it keeps sending the metrics from terminated pods last received. In our receiving monitoring stack (ELK), these data lead to errors in the resulting metrics.

Open-shift cluster in the version mentioned above with activated prometheus. Run several Pods with software that cases metrics and configure them so that they are scraped by the prometheus, that is configured to send the metrics via remote-write to an external monitoring-stack (e.g.: elastic). Terminate some pods. 

1.
2.
3.

Prometheus keeps sending the data last received from terminated pods

Prometheus should not send metric-data from terminated pods and should not-send received metric-data more than once.