Merge pull request #54 from jimmidyson/restclient-tpr-check

Fixes error handling when waiting for TPRs

pkg/alertmanager/operator.go

@@ -112,6 +112,7 @@ func (c *Operator) Run(stopc <-chan struct{}) error {
 		if err != nil {
 			return err
 		}
+		c.logger.Log("msg", "TPR API endpoints ready")
 	case <-stopc:
 		return nil
 	}

pkg/k8sutil/k8sutil.go

@@ -19,6 +19,7 @@ import (
 	"net/http"
 	"time"
 
+	"k8s.io/client-go/1.5/pkg/api/errors"
 	"k8s.io/client-go/1.5/pkg/api/v1"
 	"k8s.io/client-go/1.5/pkg/util/wait"
 	"k8s.io/client-go/1.5/rest"
@@ -31,19 +32,23 @@ func WaitForTPRReady(restClient *rest.RESTClient, tprGroup, tprVersion, tprName
 		res := restClient.Get().AbsPath("apis", tprGroup, tprVersion, tprName).Do()
 		err := res.Error()
 		if err != nil {
+			// RESTClient returns *errors.StatusError for any status codes < 200 or > 206
+			// and http.Client.Do errors are returned directly.
+			if se, ok := err.(*errors.StatusError); ok {
+				if se.Status().Code == http.StatusNotFound {
+					return false, nil
+				}
+			}
 			return false, err
 		}
 
 		var statusCode int
 		res.StatusCode(&statusCode)
-		switch statusCode {
-		case http.StatusOK:
-			return true, nil
-		case http.StatusNotFound: // not set up yet. wait.
-			return false, nil
-		default:
+		if statusCode != http.StatusOK {
 			return false, fmt.Errorf("invalid status code: %d", statusCode)
 		}
+
+		return true, nil
 	})
 }
 

pkg/prometheus/operator.go

@@ -164,14 +164,30 @@ func New(conf Config, logger log.Logger) (*Operator, error) {
 func (c *Operator) Run(stopc <-chan struct{}) error {
 	defer c.queue.ShutDown()
 
-	v, err := c.kclient.Discovery().ServerVersion()
-	if err != nil {
-		return fmt.Errorf("communicating with server failed: %s", err)
-	}
-	c.logger.Log("msg", "connection established", "cluster-version", v)
+	errChan := make(chan error)
+	go func() {
+		v, err := c.kclient.Discovery().ServerVersion()
+		if err != nil {
+			errChan <- fmt.Errorf("communicating with server failed: %s", err)
+			return
+		}
+		c.logger.Log("msg", "connection established", "cluster-version", v)
 
-	if err := c.createTPRs(); err != nil {
-		return err
+		if err := c.createTPRs(); err != nil {
+			errChan <- err
+			return
+		}
+		errChan <- nil
+	}()
+
+	select {
+	case err := <-errChan:
+		if err != nil {
+			return err
+		}
+		c.logger.Log("msg", "TPR API endpoints ready")
+	case <-stopc:
+		return nil
 	}
 
 	go c.worker()