Merge pull request #56864 from juanvallejo/jvallejo/add-selector-kubectl-drain

Automatic merge from submit-queue (batch tested with PRs 55751, 57337, 56406, 56864, 57347). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.

Add pod-selector kubectl drain

**Release note**:
```release-note
Added the ability to select pods in a chosen node to be drained, based on given pod label-selector
```

This patch adds the ability to select pods in a chosen node to be drained, based on given pod label-selector. Related downstream issue: https://github.com/openshift/origin/issues/17554

Further, it removes explicit, specific, pod-controller check. The `drain` command currently fails if a pod has a controller of a `kind` [not explicitly handled in the command itself](https://github.com/kubernetes/kubernetes/blob/master/pkg/kubectl/cmd/drain.go#L331). This causes `drain` to be unusable if a node contains pods managed by third-party, or "unknown" controllers.

Based on [this comment](https://github.com/kubernetes/kubernetes/blob/master/pkg/kubectl/cmd/drain.go#L353), the expectation was to fail if a pod's controller was not found for whatever reason. I believe that the `drain` command should not care about the existence of a pod controller. It should only care whether a pod has one, and act according to that controller kind. This solves a downstream bug: https://github.com/openshift/origin/issues/17563

cc @fabianofranz @deads2k @kubernetes/sig-cli-misc

hack/make-rules/test-cmd-util.sh

@@ -4300,6 +4300,51 @@ run_cluster_management_tests() {
 
   kube::test::get_object_assert nodes "{{range.items}}{{$id_field}}:{{end}}" '127.0.0.1:'
 
+  # create test pods we can work with
+  kubectl create -f - "${kube_flags[@]}" << __EOF__
+{
+  "kind": "Pod",
+  "apiVersion": "v1",
+  "metadata": {
+    "name": "test-pod-1",
+    "labels": {
+      "e": "f"
+    }
+  },
+  "spec": {
+    "containers": [
+      {
+        "name": "container-1",
+        "resources": {},
+        "image": "test-image"
+      }
+    ]
+  }
+}
+__EOF__
+
+  kubectl create -f - "${kube_flags[@]}" << __EOF__
+{
+  "kind": "Pod",
+  "apiVersion": "v1",
+  "metadata": {
+    "name": "test-pod-2",
+    "labels": {
+      "c": "d"
+    }
+  },
+  "spec": {
+    "containers": [
+      {
+        "name": "container-1",
+        "resources": {},
+        "image": "test-image"
+      }
+    ]
+  }
+}
+__EOF__
+
   ### kubectl cordon update with --dry-run does not mark node unschedulable
   # Pre-condition: node is schedulable
   kube::test::get_object_assert "nodes 127.0.0.1" "{{.spec.unschedulable}}" '<no value>'
@@ -4314,6 +4359,20 @@ run_cluster_management_tests() {
   kube::test::get_object_assert nodes "{{range.items}}{{$id_field}}:{{end}}" '127.0.0.1:'
   kube::test::get_object_assert "nodes 127.0.0.1" "{{.spec.unschedulable}}" '<no value>'
 
+  ### kubectl drain with --pod-selector only evicts pods that match the given selector
+  # Pre-condition: node is schedulable
+  kube::test::get_object_assert "nodes 127.0.0.1" "{{.spec.unschedulable}}" '<no value>'
+  # Pre-condition: test-pod-1 and test-pod-2 exist
+  kube::test::get_object_assert "pods" "{{range .items}}{{.metadata.name}},{{end}}" 'test-pod-1,test-pod-2,'
+  kubectl drain "127.0.0.1" --pod-selector 'e in (f)'
+  # only "test-pod-1" should have been matched and deleted - test-pod-2 should still exist
+  kube::test::get_object_assert "pods/test-pod-2" "{{.metadata.name}}" 'test-pod-2'
+  # delete pod no longer in use
+  kubectl delete pod/test-pod-2
+  # Post-condition: node is schedulable
+  kubectl uncordon "127.0.0.1"
+  kube::test::get_object_assert "nodes 127.0.0.1" "{{.spec.unschedulable}}" '<no value>'
+
   ### kubectl uncordon update with --dry-run is a no-op
   # Pre-condition: node is already schedulable
   kube::test::get_object_assert "nodes 127.0.0.1" "{{.spec.unschedulable}}" '<no value>'