To avoid restarting FLPs every time a metric is (re)configured, FLP could watch them directly using the kube api

Or alternatively, it could use fs.notify to get mounted configmap updates and reload its pipeline, but this has a bigger sync delay afaik