The script will need to Talk to managed kafka to get a list of all topics within the whitelisted pattern. Talk to the clusters that are using it (dev/ephem) to get a list of namespaces If managed kafka has topics that exist for namespaces that don't on the clusters, remove them. It needs to happen in this order to prevent the destruction of topics that are new. i.e. if you first talk to the clusters and then talk to managed kafka, a namespace could have been provisioned in the time it has taken inbetween. Doing it this way lessens the risk, and will simply yield some extra topics that are missed until the next cleanup. It is also important to collect ALL namespaces from both / multiple clusters. Ideally we should have a managed kafka instance for each cluster and be careful about naming.