The Toolforge Kubernetes cluster has a suprisingly large number of pods that are always in CrashLoopBackOff state. No-one seems to notice or care about said failing pods and the pods likely can't do anything useful if they are constantly crashing, but they still consume resources on the cluster.
I propose adding a cronjob of some sort to regularly look for pods that haven't been able to start for a while and just removes that pod (and any deployments that control it). The script should send a warning in advance for the tool maintainers to fix the problem and a notification after killing it.