From time to time you may wish to run two or more Kubernetes Operators locally for debugging purposes. If you try and run a second operator locally the second operator will shutdown you will hit an error like
WATCH_NAMESPACE=$(MY-OPERATOR-NAMESPACE) go run ./main.go
2022-01-07T15:16:11.786Z INFO controller-runtime.metrics metrics server is starting to listen {"addr": ":8383"}
2022-01-07T15:16:11.787Z ERROR controller-runtime.metrics metrics server failed to listen. You may want to disable the metrics server or use another port if it is due to conflicts {"error": "error listening on :8383: listen tcp :8383: bind: address already in use"}
So the metric port is already used.The Operator SDK generates ports for metrics and health automatically. These are set in the main.go
func main() {
var metricsAddr string
var enableLeaderElection bool
var probeAddr string
//Change the port numbers on the metrics-bind-address and health-probe-bind-address to a free port number
flag.StringVar(&metricsAddr, "metrics-bind-address", ":8383", "The address the metric endpoint binds to.")
flag.StringVar(&probeAddr, "health-probe-bind-address", ":8081", "The address the probe endpoint binds to.")
As you can see there are two port assignments here metrics-bind-address and health-probe-bind-address. These ports will be occupied on your machine by the first operator you run. You should change both to a free port numbers if you wish to run a second operator side by side locally, otherwise you will get a port conflict. You should now be able to debug both operators at once while running locally. Operators built with an older version of the Operator-sdk won't have the health-probe port.
NOTE: Sometimes the Operator gets stuck shutting down and you have to release the ports manually by killing the stuck process
# Get the process on the port metric port e.g. 8383
netstat -ltnp | grep -w ':8383'
tcp6 0 0 :::8383 :::* LISTEN 1480861/main
# kill the stuck process
sudo kill -9 1480861
Top comments (0)