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

CHANGELOG.md

@@ -29,13 +29,16 @@ 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 ([#814]).
+- The operator now watches all resources that it creates and early-exits the reconcile action when the
+  cluster is marked for deletion ([#821]).
 
 [#801]: https://github.com/stackabletech/hdfs-operator/pull/801
 [#806]: https://github.com/stackabletech/hdfs-operator/pull/806
 [#810]: https://github.com/stackabletech/hdfs-operator/pull/810
 [#811]: https://github.com/stackabletech/hdfs-operator/pull/811
 [#814]: https://github.com/stackabletech/hdfs-operator/pull/814
 [#819]: https://github.com/stackabletech/hdfs-operator/pull/819
+[#821]: https://github.com/stackabletech/hdfs-operator/pull/821
 
 ## [26.7.0] - 2026-07-21
 

Cargo.toml

@@ -17,6 +17,7 @@ built = { version = "0.8", features = ["chrono", "git2"] }
 clap = "4.6"
 const_format = "0.2"
 futures = { version = "0.3", features = ["compat"] }
+http = "1.3"
 indoc = "2.0"
 rstest = "0.26"
 semver = "1.0"

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

@@ -28,7 +28,8 @@ rules:
       - list
       - patch
       - watch
-  # serviceaccounts are applied via SSA and tracked for orphan cleanup.
+  # serviceaccounts are applied via SSA and tracked for orphan cleanup
+  # and watched by the controller.
   - apiGroups:
       - ""
     resources:
@@ -39,7 +40,9 @@ rules:
       - get
       - list
       - patch
-  # rolebindings are applied via SSA and tracked for orphan cleanup.
+      - watch
+  # rolebindings are applied via SSA and tracked for orphan cleanup
+  # and watched by the controller.
   - apiGroups:
       - rbac.authorization.k8s.io
     resources:
@@ -50,6 +53,7 @@ rules:
       - get
       - list
       - patch
+      - watch
   # statefulsets are applied via SSA, tracked for orphan cleanup.
   - apiGroups:
       - apps
@@ -62,7 +66,8 @@ rules:
       - list
       - patch
       - watch
-  # poddisruptionbudgets are applied via SSA and tracked for orphan cleanup.
+  # poddisruptionbudgets are applied via SSA and tracked for orphan cleanup
+  # and watched by the controller.
   - apiGroups:
       - policy
     resources:
@@ -73,6 +78,7 @@ 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/Cargo.toml

@@ -25,6 +25,7 @@ tracing-futures.workspace = true
 tracing.workspace = true
 
 [dev-dependencies]
+http.workspace = true
 rstest.workspace = true
 serde_yaml.workspace = true
 

rust/operator-binary/src/hdfs_controller.rs

@@ -84,6 +84,10 @@ pub async fn reconcile_hdfs(
 ) -> HdfsOperatorResult<Action> {
     tracing::info!("Starting reconcile");
 
+    if hdfs.meta().deletion_timestamp.is_some() {
+        return Ok(Action::await_change());
+    }
+
     let hdfs = hdfs
         .0
         .as_ref()
@@ -160,13 +164,25 @@ mod test {
     use std::str::FromStr;
 
     use stackable_operator::{
-        builder::pod::PodBuilder, commons::networking::DomainName, kube::api::ObjectMeta,
-        kvp::Labels, utils::cluster_info::KubernetesClusterInfo,
+        builder::pod::PodBuilder,
+        client::Client,
+        commons::networking::DomainName,
+        kube::{
+            Client as KubeClient, Config,
+            api::ObjectMeta,
+            runtime::{
+                controller::Action,
+                events::{Recorder, Reporter},
+            },
+        },
+        kvp::Labels,
+        utils::cluster_info::KubernetesClusterInfo,
         v2::types::operator::RoleGroupName,
     };
 
     use super::*;
     use crate::{
+        HDFS_FULL_CONTROLLER_NAME,
         controller::build::container::ContainerConfig,
         test_support::{deserialize_cluster, role_group_config, validate_cluster},
     };
@@ -261,4 +277,117 @@ spec:
             Some("group-value".to_string())
         );
     }
+
+    /// A [`Ctx`] whose client points at a closed port. Any API call made through it fails the
+    /// reconciliation, so an `Ok` result proves the reconciler returned before touching the
+    /// Kubernetes API.
+    fn unreachable_ctx() -> Arc<Ctx> {
+        let config = Config::new(
+            "http://127.0.0.1:1"
+                .parse::<http::Uri>()
+                .expect("valid static URI"),
+        );
+        let kube_client = KubeClient::try_from(config).expect("client from static config");
+
+        Arc::new(Ctx {
+            client: Client::new(
+                kube_client.clone(),
+                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: "hdfs-operator".to_owned(),
+                image_repository: "oci.stackable.tech/sdp".to_owned(),
+            },
+            event_recorder: Arc::new(Recorder::new(
+                kube_client,
+                Reporter {
+                    controller: HDFS_FULL_CONTROLLER_NAME.to_string(),
+                    instance: None,
+                },
+            )),
+        })
+    }
+
+    fn reconcile(hdfs: DeserializeGuard<v1alpha1::HdfsCluster>) -> Result<Action, Error> {
+        tokio::runtime::Builder::new_current_thread()
+            .enable_all()
+            .build()
+            .expect("current-thread tokio runtime")
+            .block_on(async { reconcile_hdfs(Arc::new(hdfs), unreachable_ctx()).await })
+    }
+
+    #[test]
+    fn reconcile_exits_early_for_deleted_cluster() {
+        let hdfs = serde_yaml::from_str(
+            r#"
+apiVersion: hdfs.stackable.tech/v1alpha1
+kind: HdfsCluster
+metadata:
+  name: hdfs
+  namespace: default
+  deletionTimestamp: "2026-08-14T12:00:00Z"
+spec:
+  image:
+    productVersion: 3.2.2
+"#,
+        )
+        .expect("valid cluster YAML");
+
+        let action = reconcile(hdfs).expect("a deleted cluster reconciles without any API call");
+
+        assert_eq!(action, Action::await_change());
+    }
+
+    #[test]
+    fn reconcile_exits_early_for_deleted_cluster_with_invalid_spec() {
+        let hdfs = serde_yaml::from_str(
+            r#"
+apiVersion: hdfs.stackable.tech/v1alpha1
+kind: HdfsCluster
+metadata:
+  name: hdfs
+  namespace: default
+  deletionTimestamp: "2026-08-14T12:00:00Z"
+spec: {}
+"#,
+        )
+        .expect("YAML parses; the invalid spec is captured inside the DeserializeGuard");
+
+        let action =
+            reconcile(hdfs).expect("a deleted cluster reconciles even when its spec is invalid");
+
+        assert_eq!(action, Action::await_change());
+    }
+
+    #[test]
+    fn reconcile_proceeds_for_live_cluster() {
+        let hdfs = serde_yaml::from_str(
+            r#"
+apiVersion: hdfs.stackable.tech/v1alpha1
+kind: HdfsCluster
+metadata:
+  name: hdfs
+  namespace: default
+spec:
+  image:
+    productVersion: 3.5.0
+  clusterConfig:
+    zookeeperConfigMapName: simple-znode
+"#,
+        )
+        .expect("valid cluster YAML");
+
+        let result = reconcile(hdfs);
+
+        assert!(
+            matches!(result, Err(Error::Dereference { .. })),
+            "a live cluster must reach the API but when dereferencing against the unreachable test server: {result:?}"
+        );
+    }
 }

rust/operator-binary/src/main.rs

@@ -17,7 +17,9 @@ use stackable_operator::{
     eos::EndOfSupportChecker,
     k8s_openapi::api::{
         apps::v1::StatefulSet,
-        core::v1::{ConfigMap, Service},
+        core::v1::{ConfigMap, Service, ServiceAccount},
+        policy::v1::PodDisruptionBudget,
+        rbac::v1::RoleBinding,
     },
     kube::{
         Api, CustomResourceExt, ResourceExt,
@@ -156,15 +158,27 @@ async fn main() -> anyhow::Result<()> {
             let hdfs_cluster_store = hdfs_controller.store();
             let hdfs_controller = hdfs_controller
                 .owns(
-                    watch_namespace.get_api::<DeserializeGuard<StatefulSet>>(&client),
+                    watch_namespace.get_api::<ConfigMap>(&client),
                     watcher::Config::default(),
                 )
                 .owns(
-                    watch_namespace.get_api::<DeserializeGuard<Service>>(&client),
+                    watch_namespace.get_api::<PodDisruptionBudget>(&client),
                     watcher::Config::default(),
                 )
                 .owns(
-                    watch_namespace.get_api::<DeserializeGuard<ConfigMap>>(&client),
+                    watch_namespace.get_api::<RoleBinding>(&client),
+                    watcher::Config::default(),
+                )
+                .owns(
+                    watch_namespace.get_api::<Service>(&client),
+                    watcher::Config::default(),
+                )
+                .owns(
+                    watch_namespace.get_api::<ServiceAccount>(&client),
+                    watcher::Config::default(),
+                )
+                .owns(
+                    watch_namespace.get_api::<StatefulSet>(&client),
                     watcher::Config::default(),
                 )
                 .watches(