fix: Add missing owns/watches and test early-exit reconcile

CHANGELOG.md

@@ -31,13 +31,16 @@ All notable changes to this project will be documented in this file.
 - Fix a longstanding problem of including empty `categories`, `shortNames` and `additionalPrinterColumns` in the CRDs,
   which could cause problems with GitOps tools (e.g. ArgoCD) reporting a diff in the custom resources.
   See [our internal issue](https://github.com/stackabletech/hdfs-operator/issues/626) and [the fix](https://github.com/kube-rs/kube/pull/2042) for details ([#860]).
+- The operator now watches all resources that it creates and early-exits the reconcile action when the
+  cluster is marked for deletion ([#867]).
 
 [#841]: https://github.com/stackabletech/druid-operator/pull/841
 [#846]: https://github.com/stackabletech/druid-operator/pull/846
 [#855]: https://github.com/stackabletech/druid-operator/pull/855
 [#856]: https://github.com/stackabletech/druid-operator/pull/856
 [#860]: https://github.com/stackabletech/druid-operator/pull/860
 [#865]: https://github.com/stackabletech/druid-operator/pull/865
+[#867]: https://github.com/stackabletech/druid-operator/pull/867
 
 ## [26.7.0] - 2026-07-21
 

deploy/helm/druid-operator/templates/clusterrole-operator.yaml

@@ -13,46 +13,28 @@ rules:
       - nodes/proxy
     verbs:
       - get
-  # Manage core workload resources created per DruidCluster.
-  # All resources are applied via Server-Side Apply (create + patch) and tracked for
-  # orphan cleanup (list + delete).
+  # Manage core workload resources created per DruidCluster: ConfigMaps and Services,
+  # the shared internal authentication Secret (cookie passphrase and internal client
+  # password; no orphan cleanup — Kubernetes GC via owner reference instead), and the
+  # ServiceAccount providing workload pod identity. All are applied via Server-Side
+  # Apply (create + patch), tracked for orphan cleanup where applicable (list +
+  # delete) and watched by the controller (`.owns()` lists before it watches).
   - apiGroups:
       - ""
     resources:
       - configmaps
-      - services
-    verbs:
-      - create
-      - delete
-      - get
-      - list
-      - patch
-      - watch
-  # Shared internal authentication secret (cookie passphrase and internal client password).
-  # Orphan cleanup not needed (instead, Kubernetes GC via owner reference).
-  - apiGroups:
-      - ""
-    resources:
       - secrets
-    verbs:
-      - create
-      - delete
-      - get
-      - patch
-  # ServiceAccount created per DruidCluster for workload pod identity.
-  # Applied via SSA and tracked for orphan cleanup. Not watched by the controller.
-  - apiGroups:
-      - ""
-    resources:
       - serviceaccounts
+      - services
     verbs:
       - create
       - delete
       - get
       - list
       - patch
+      - watch
   # RoleBinding created per DruidCluster to bind the product ClusterRole to the workload
-  # ServiceAccount. Applied via SSA and tracked for orphan cleanup. Not watched by the controller.
+  # ServiceAccount. Applied via SSA and tracked for orphan cleanup and watched by the controller.
   - apiGroups:
       - rbac.authorization.k8s.io
     resources:
@@ -63,6 +45,7 @@ rules:
       - get
       - list
       - patch
+      - watch
   # Required to bind the product ClusterRole to the per-cluster ServiceAccount.
   - apiGroups:
       - rbac.authorization.k8s.io
@@ -85,8 +68,7 @@ rules:
       - list
       - patch
       - watch
-  # PodDisruptionBudget created per role. Applied via SSA and tracked for orphan cleanup.
-  # Not watched by the controller.
+  # PodDisruptionBudget created per role. Applied via SSA and tracked for orphan cleanup and watched by the controller.
   - apiGroups:
       - policy
     resources:
@@ -97,6 +79,7 @@ rules:
       - get
       - list
       - patch
+      - watch
   # Required for maintaining the CRDs within the operator (including the conversion webhook info).
   # Also for the startup condition check before the controller can run.
   - apiGroups:

rust/operator-binary/src/controller.rs

@@ -17,6 +17,7 @@ use stackable_operator::{
         rbac::v1::RoleBinding,
     },
     kube::{
+        Resource,
         core::{DeserializeGuard, error_boundary},
         runtime::controller::Action,
     },
@@ -115,6 +116,11 @@ pub async fn reconcile_druid(
     ctx: Arc<Ctx>,
 ) -> Result<Action> {
     tracing::info!("Starting reconcile");
+
+    if druid.meta().deletion_timestamp.is_some() {
+        return Ok(Action::await_change());
+    }
+
     let druid = druid
         .0
         .as_ref()
@@ -173,9 +179,15 @@ mod test {
     use std::str::FromStr;
 
     use rstest::*;
-    use stackable_operator::v2::types::operator::RoleGroupName;
+    use stackable_operator::{
+        client::Client,
+        commons::networking::DomainName,
+        kube::{Client as KubeClient, Config, runtime::controller::Action},
+        utils::cluster_info::KubernetesClusterInfo,
+        v2::types::operator::RoleGroupName,
+    };
 
-    use super::{CONTROLLER_NAME, OPERATOR_NAME, PRODUCT_NAME};
+    use super::{CONTROLLER_NAME, OPERATOR_NAME, PRODUCT_NAME, *};
     use crate::{
         controller::build::{
             properties::ConfigFileName, resource::config_map::build_rolegroup_config_map,
@@ -254,4 +266,54 @@ mod test {
             "role group {tested_rolegroup_name}"
         );
     }
+
+    /// The client points at a closed port, so any API call would fail the reconciliation: an `Ok`
+    /// proves that a cluster being deleted returns before the reconciler touches the Kubernetes
+    /// API, and because the spec is invalid, before the [`DeserializeGuard`] is unwrapped.
+    #[test]
+    fn reconcile_exits_early_for_deleted_cluster() {
+        let druid = serde_yaml::from_str(
+            r#"
+apiVersion: druid.stackable.tech/v1alpha1
+kind: DruidCluster
+metadata:
+  name: druid
+  namespace: default
+  deletionTimestamp: "2026-08-14T12:00:00Z"
+spec: {}
+"#,
+        )
+        .expect("YAML parses; the invalid spec is captured inside the DeserializeGuard");
+
+        let action = tokio::runtime::Builder::new_current_thread()
+            .enable_all()
+            .build()
+            .expect("current-thread tokio runtime")
+            .block_on(async {
+                let ctx = Arc::new(Ctx {
+                    client: Client::new(
+                        KubeClient::try_from(Config::new(
+                            "http://127.0.0.1:1".parse().expect("valid static URI"),
+                        ))
+                        .expect("client from static config"),
+                        None,
+                        "default".to_owned(),
+                        KubernetesClusterInfo {
+                            cluster_domain: DomainName::from_str("cluster.local")
+                                .expect("valid cluster domain"),
+                        },
+                    ),
+                    operator_environment: OperatorEnvironmentOptions {
+                        operator_namespace: "stackable-operators".to_owned(),
+                        operator_service_name: "druid-operator".to_owned(),
+                        image_repository: "oci.stackable.tech/sdp".to_owned(),
+                    },
+                });
+
+                reconcile_druid(Arc::new(druid), ctx).await
+            })
+            .expect("a deleted cluster reconciles without any API call");
+
+        assert_eq!(action, Action::await_change());
+    }
 }

rust/operator-binary/src/main.rs

@@ -15,7 +15,9 @@ use stackable_operator::{
     eos::EndOfSupportChecker,
     k8s_openapi::api::{
         apps::v1::StatefulSet,
-        core::v1::{ConfigMap, Service},
+        core::v1::{ConfigMap, Secret, Service, ServiceAccount},
+        policy::v1::PodDisruptionBudget,
+        rbac::v1::RoleBinding,
     },
     kube::{
         CustomResourceExt, ResourceExt,
@@ -125,19 +127,35 @@ async fn main() -> anyhow::Result<()> {
             let config_map_store = druid_controller.store();
             let druid_controller = druid_controller
                 .owns(
-                    watch_namespace.get_api::<Service>(&client),
+                    watch_namespace.get_api::<DeserializeGuard<ConfigMap>>(&client),
+                    watcher::Config::default(),
+                )
+                .owns(
+                    watch_namespace.get_api::<DeserializeGuard<Listener>>(&client),
+                    watcher::Config::default(),
+                )
+                .owns(
+                    watch_namespace.get_api::<DeserializeGuard<PodDisruptionBudget>>(&client),
+                    watcher::Config::default(),
+                )
+                .owns(
+                    watch_namespace.get_api::<DeserializeGuard<RoleBinding>>(&client),
+                    watcher::Config::default(),
+                )
+                .owns(
+                    watch_namespace.get_api::<DeserializeGuard<Secret>>(&client),
                     watcher::Config::default(),
                 )
                 .owns(
-                    watch_namespace.get_api::<StatefulSet>(&client),
+                    watch_namespace.get_api::<DeserializeGuard<Service>>(&client),
                     watcher::Config::default(),
                 )
                 .owns(
-                    watch_namespace.get_api::<ConfigMap>(&client),
+                    watch_namespace.get_api::<DeserializeGuard<ServiceAccount>>(&client),
                     watcher::Config::default(),
                 )
                 .owns(
-                    watch_namespace.get_api::<Listener>(&client),
+                    watch_namespace.get_api::<DeserializeGuard<StatefulSet>>(&client),
                     watcher::Config::default(),
                 )
                 .watches(