Don't delete resources unless field managers match

provider/pkg/await/await.go

@@ -48,6 +48,7 @@ import (
 	"k8s.io/apimachinery/pkg/apis/meta/v1/unstructured"
 	"k8s.io/apimachinery/pkg/runtime/schema"
 	"k8s.io/apimachinery/pkg/types"
+	"k8s.io/apimachinery/pkg/util/sets"
 	"k8s.io/apimachinery/pkg/watch"
 	"k8s.io/apiserver/pkg/storage/names"
 	"k8s.io/client-go/dynamic"
@@ -834,6 +835,30 @@ func Deletion(c DeleteConfig) error {
 		PropagationPolicy: &deletePolicy,
 	}
 
+	live, err := client.Get(c.Context, c.Name, metav1.GetOptions{})
+	if err != nil {
+		return nilIfGVKDeleted(err)
+	}
+
+	actualSSAManagers := sets.Set[string]{}
+	for _, f := range live.GetManagedFields() {
+		// Ignore fields not managed by pulumi SSA.
+		if !strings.HasPrefix(f.Manager, "pulumi-kubernetes-") {
+			continue
+		}
+		actualSSAManagers = actualSSAManagers.Insert(f.Manager)
+	}
+	if actualSSAManagers.Len() > 0 {
+		fmt.Println("OVERLAP")
+	}
+	if c.ServerSideApply && !actualSSAManagers.Has(c.FieldManager) {
+		// Didn't find our expected manager on the object. Assume it was
+		// upserted by another manager and refuse to delete it. For the sake of
+		// the program's state, report that it has been deleted since we are no
+		// longer managing it.
+		return nil
+	}
+
 	err = client.Delete(c.Context, c.Name, deleteOpts)
 	if err != nil {
 		return nilIfGVKDeleted(err)

provider/pkg/await/await_test.go

@@ -787,6 +787,34 @@ func TestDeletion(t *testing.T) {
 		}
 	}
 
+	field := []byte(`{
+			"f:metadata": {
+				"f:generateName": {},
+				"f:labels": {},
+				"f:ownerReferences": {}
+			}
+		}`)
+
+	original := validPodUnstructured.DeepCopy()
+	original.SetManagedFields([]metav1.ManagedFieldsEntry{{
+		Manager:    "pulumi-kubernetes-123",
+		Operation:  "Update",
+		FieldsType: "FieldsV1",
+		FieldsV1: &metav1.FieldsV1{
+			Raw: field,
+		},
+	}})
+
+	upserted := validPodUnstructured.DeepCopy()
+	upserted.SetManagedFields([]metav1.ManagedFieldsEntry{{
+		Manager:    "pulumi-kubernetes-XYZ",
+		Operation:  "Update",
+		FieldsType: "FieldsV1",
+		FieldsV1: &metav1.FieldsV1{
+			Raw: field,
+		},
+	}})
+
 	tests := []struct {
 		name      string
 		client    client
@@ -909,6 +937,19 @@ func TestDeletion(t *testing.T) {
 			condition: awaitNoop,
 			expect:    []expectF{succeeded()},
 		},
+		{
+			name: "Field manager mismatch",
+			args: args{
+				resType:         tokens.Type("kubernetes:core/v1:Pod"),
+				name:            "foo",
+				objects:         []runtime.Object{original},
+				inputs:          original,
+				outputs:         original,
+				serverSideApply: true,
+			},
+			condition: awaitNoop,
+			expect:    []expectF{succeeded()},
+		},
 	}
 
 	for _, tt := range tests {

provider/pkg/await/condition/deleted.go

@@ -119,6 +119,7 @@ func (dc *Deleted) Satisfied() (bool, error) {
 	}
 
 	uns := dc.Object()
+
 	r, _ := status.Compute(uns)
 	if r.Message != "" {
 		dc.logger.LogStatus(diag.Info, r.Message)

provider/pkg/await/condition/deleted_test.go

@@ -191,4 +191,46 @@ func TestDeleted(t *testing.T) {
 		assert.False(t, done)
 		assert.Contains(t, buf.String(), "boom")
 	})
+
+	t.Run("fieldManager mismatch", func(t *testing.T) {
+		field := []byte(`{
+			"f:metadata": {
+				"f:generateName": {},
+				"f:labels": {},
+				"f:ownerReferences": {}
+			}
+		}`)
+
+		original := pod.DeepCopy()
+		original.SetManagedFields([]metav1.ManagedFieldsEntry{{
+			Manager:    "pulumi-kubernetes-123",
+			Operation:  "Update",
+			FieldsType: "FieldsV1",
+			FieldsV1: &metav1.FieldsV1{
+				Raw: field,
+			},
+		}})
+
+		upserted := pod.DeepCopy()
+		upserted.SetManagedFields([]metav1.ManagedFieldsEntry{{
+			Manager:    "pulumi-kubernetes-XYZ",
+			Operation:  "Update",
+			FieldsType: "FieldsV1",
+			FieldsV1: &metav1.FieldsV1{
+				Raw: field,
+			},
+		}})
+
+		ctx := context.Background()
+		getter := get404{}
+
+		cond, err := NewDeleted(ctx, Static(nil), getter, stdout, pod)
+		assert.NoError(t, err)
+
+		cond.Range(nil)
+
+		done, err := cond.Satisfied()
+		assert.NoError(t, err)
+		assert.True(t, done)
+	})
 }

tests/sdk/java/delete_test.go

@@ -0,0 +1,67 @@
+package provider
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	"github.com/pulumi/providertest/pulumitest"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestDeleteDueToRename(t *testing.T) {
+	t.Parallel()
+	ctx := context.Background()
+	test := pulumitest.NewPulumiTest(t, "delete/rename")
+	t.Cleanup(func() {
+		test.Destroy()
+	})
+
+	test.Up()
+
+	// Change our Pod's resource name
+	test.UpdateSource("testdata/delete/rename/step2")
+	test.Up()
+
+	// Renaming the namespace should not have deleted it. Perform a refresh and
+	// make sure our pod is still running -- if it's not, Pulumi will have
+	// deleted it from our state.
+	refresh, err := test.CurrentStack().Refresh(ctx)
+	assert.NoError(t, err)
+	assert.NotContains(t, refresh.StdOut, "deleted", refresh.StdOut)
+}
+
+func TestDeletePatchResource(t *testing.T) {
+	t.Parallel()
+	ctx := context.Background()
+	test := pulumitest.NewPulumiTest(t, "testdata/delete/patch")
+	t.Cleanup(func() {
+		test.Destroy()
+	})
+
+	test.Up()
+
+	time.Sleep(60 * time.Second)
+
+	outputs, err := test.CurrentStack().Outputs(ctx)
+	require.NoError(t, err)
+
+	// The ConfigMap should have 3 managed fields -- one for the URN annotation
+	// and 2 for the patches.
+	mf, ok := outputs["managedFields"]
+	require.True(t, ok)
+	assert.Len(t, mf.Value, 3)
+
+	// Delete a patch.
+	test.UpdateSource("testdata/delete/patch/step2")
+	test.Up()
+
+	// One ConfigMapPatch should still be applied, plus a manager for our URN
+	// annotation.
+	outputs, err = test.CurrentStack().Outputs(ctx)
+	require.NoError(t, err)
+	mf, ok = outputs["managedFields"]
+	require.True(t, ok)
+	assert.Len(t, mf.Value, 2)
+}

tests/sdk/java/testdata/delete/patch/Pulumi.yaml

@@ -0,0 +1,37 @@
+name: delete-patch-resource
+runtime: yaml
+description: |
+  Deleting a logical patch resource should not delete the underlying physical
+  resource.
+
+outputs:
+  managedFields: ${patch2.metadata.managedFields}
+
+resources:
+  configmap:
+    type: kubernetes:core/v1:ConfigMap
+    properties:
+      metadata:
+        name: patched-configmap
+
+  patch1:
+    type: kubernetes:core/v1:ConfigMapPatch
+    properties:
+      metadata:
+        name: patched-configmap
+      data:
+        foo: bar
+    options:
+      dependsOn:
+        - ${configmap}
+
+  patch2:
+    type: kubernetes:core/v1:ConfigMapPatch
+    properties:
+      metadata:
+        name: patched-configmap
+      data:
+        boo: baz
+    options:
+      dependsOn:
+        - ${patch1}

tests/sdk/java/testdata/delete/patch/step2/Pulumi.yaml

@@ -0,0 +1,40 @@
+name: delete-patch-resource
+runtime: yaml
+description: |
+  Deleting a logical patch resource should not delete the underlying physical
+  resource.
+
+outputs:
+  managedFields: ${patch1.metadata.managedFields}
+
+resources:
+  configmap:
+    type: kubernetes:core/v1:ConfigMap
+    properties:
+      metadata:
+        name: patched-configmap
+
+  patch1:
+    type: kubernetes:core/v1:ConfigMapPatch
+    properties:
+      metadata:
+        name: patched-configmap
+      data:
+        foo: bar
+    options:
+      dependsOn:
+        - ${configmap}
+
+  # Delete patch2 - the underlying ConfigMap should not be deleted, and patch1
+  # should still be applied.
+  #
+  # patch2:
+  #   type: kubernetes:core/v1:ConfigMapPatch
+  #   properties:
+  #     metadata:
+  #       name: patched-configmap
+  #     data:
+  #       boo: baz
+  #   options:
+  #     dependsOn:
+  #       - ${patch1}

tests/sdk/java/testdata/delete/rename/Pulumi.yaml

@@ -0,0 +1,16 @@
+name: delete-with-rename
+runtime: yaml
+description: |
+  Changing a resource's name, but leaving .metadata untouched, should not
+  result in a deletion from the cluster.
+
+resources:
+  pod:
+    type: kubernetes:core/v1:Pod
+    properties:
+      spec:
+        containers:
+          - image: nginx:1.14.2
+            name: nginx
+            ports:
+              - containerPort: 80

tests/sdk/java/testdata/delete/rename/step2/Pulumi.yaml

@@ -0,0 +1,18 @@
+name: delete-with-rename
+runtime: yaml
+description: |
+  Changing a resource's name, but leaving .metadata untouched, should not
+  result in a deletion from the cluster.
+
+resources:
+  # Change the resource's name from "pod" to "mypod" but leave everything
+  # else the same.
+  mypod:
+    type: kubernetes:core/v1:Pod
+    properties:
+      spec:
+        containers:
+          - image: nginx:1.14.2
+            name: nginx
+            ports:
+              - containerPort: 80