Interference between status check of multiple Skaffold runs

[corneliusweig]

Skaffold now checks the deployment status. However, the checks for multiple runs can interfere.

For example, with this patch on master:

diff --git a/examples/react-reload/k8s/deployment.yaml b/examples/react-reload/k8s/deployment.yaml
index e4d64794..edc57ba2 100644
--- a/examples/react-reload/k8s/deployment.yaml
+++ b/examples/react-reload/k8s/deployment.yaml
@@ -12,7 +12,7 @@ spec:
 apiVersion: apps/v1
 kind: Deployment
 metadata:
-  name: node
+  name: node-react
 spec:
   selector:
     matchLabels:
diff --git a/pkg/skaffold/deploy/status_check.go b/pkg/skaffold/deploy/status_check.go
index 96b101e8..64317e89 100644
--- a/pkg/skaffold/deploy/status_check.go
+++ b/pkg/skaffold/deploy/status_check.go
@@ -96,6 +96,7 @@ func getDeployments(client kubernetes.Interface, ns string, l *DefaultLabeller)
 		} else {
 			deadline = *d.Spec.ProgressDeadlineSeconds
 		}
+		logrus.Debugf("Register deployment %s for status check", d.Name)
 		depMap[d.Name] = deadline
 	}
 

Skaffold prints the watched deployments.

When doing

cd examples/nodejs
skaffold run
cd ../../examples/react-reload
skaffold dev -v debug 2>&1 | grep Register

It shows that the second run also watches the first deployment:

time="2019-07-18T13:26:16+02:00" level=debug msg="Register deployment node for status check"
time="2019-07-18T13:26:16+02:00" level=debug msg="Register deployment node-react for status check"

Deleting the first deployment with

cd examples/nodejs
skaffold delete

will eventually (after the timeout) lead to a failure of the skaffold dev run.

---

A possible way out would be to add a new label with a UUID that is generated in every skaffold run. In other words deployments should be selected via

deps, err := client.AppsV1().Deployments(ns).List(metav1.ListOptions{
		LabelSelector: "skaffold-run-uuid=xyz-xyz",
	})

in pkg/skaffold/deploy/status_check.go:80,

cc @tejal29