refactor: additional kuttl checks and fixture consolidation

Author: sweb

rust/operator-binary/src/test_support.rs

@@ -66,7 +66,7 @@ pub fn cluster_info() -> KubernetesClusterInfo {
     }
 }
 
-fn operator_environment() -> OperatorEnvironmentOptions {
+pub(crate) fn operator_environment() -> OperatorEnvironmentOptions {
     OperatorEnvironmentOptions {
         operator_namespace: "stackable-operators".to_owned(),
         operator_service_name: "zookeeper-operator".to_owned(),

rust/operator-binary/src/zk_controller/build/resource/statefulset.rs

@@ -754,9 +754,10 @@ mod tests {
     #[test]
     fn vector_agent_adds_vector_container_to_statefulset() {
         // Enabling the Vector agent wires a `vector` sidecar onto the StatefulSet, mounting the
-        // `config` (for `vector.yaml`) and `log` volumes. Its env vars come from the upstream
-        // `vector_container` helper, so this pins only the seam this operator owns: that the sidecar
-        // is added when the agent is enabled.
+        // `config` (for `vector.yaml`) and `log` volumes. The env var format comes from the
+        // upstream `vector_container` helper, but the identity values (cluster/role/role-group) and
+        // the aggregator ConfigMap reference are wiring this operator supplies, so those are pinned
+        // here (they were previously only checked by the kuttl smoke `14-assert` heredoc).
         let sts = build_sts(
             r#"
             apiVersion: zookeeper.stackable.tech/v1alpha1
@@ -807,6 +808,31 @@ mod tests {
             mount_names.contains(&LOG_VOLUME_NAME.as_ref()),
             "vector container missing `log` volume mount: {mount_names:?}"
         );
+
+        // Identity wiring this operator passes into the upstream `vector_container` helper.
+        let env = vector.env.as_ref().expect("vector container env");
+        let env_value = |name: &str| {
+            env.iter()
+                .find(|e| e.name == name)
+                .unwrap_or_else(|| panic!("vector env {name} missing"))
+                .value
+                .as_deref()
+        };
+        assert_eq!(env_value("CLUSTER_NAME"), Some("simple-zookeeper"));
+        assert_eq!(env_value("ROLE_NAME"), Some("server"));
+        assert_eq!(env_value("ROLE_GROUP_NAME"), Some("default"));
+
+        // The aggregator address resolves from the discovery ConfigMap named in `clusterConfig`.
+        let aggregator = env
+            .iter()
+            .find(|e| e.name == "VECTOR_AGGREGATOR_ADDRESS")
+            .expect("VECTOR_AGGREGATOR_ADDRESS env var")
+            .value_from
+            .as_ref()
+            .and_then(|source| source.config_map_key_ref.as_ref())
+            .expect("VECTOR_AGGREGATOR_ADDRESS resolved from a ConfigMap key");
+        assert_eq!(aggregator.name, "vector-aggregator-discovery");
+        assert_eq!(aggregator.key, "ADDRESS");
     }
 
     #[test]

rust/operator-binary/src/znode_controller/validate.rs

@@ -208,12 +208,10 @@ pub fn validate(
 /// (product version, `zookeeper_security`, identity) matches production.
 #[cfg(test)]
 pub(crate) mod test_support {
-    use stackable_operator::cli::OperatorEnvironmentOptions;
-
     use super::{ValidatedZnode, validate};
     use crate::{
         crd::{authentication::DereferencedAuthenticationClasses, v1alpha1},
-        test_support::minimal_zk,
+        test_support::{minimal_zk, operator_environment},
         znode_controller::dereference::DereferencedObjects,
     };
 
@@ -233,14 +231,6 @@ pub(crate) mod test_support {
         znode
     }
 
-    fn operator_environment() -> OperatorEnvironmentOptions {
-        OperatorEnvironmentOptions {
-            operator_namespace: "stackable-operators".to_owned(),
-            operator_service_name: "zookeeper-operator".to_owned(),
-            image_repository: "oci.example.org".to_owned(),
-        }
-    }
-
     /// Runs the real znode validate step against a minimal znode fixture and the given
     /// `ZookeeperCluster` YAML (which controls product version and `zookeeper_security`).
     pub fn validated_znode(znode: &v1alpha1::ZookeeperZnode, zk_yaml: &str) -> ValidatedZnode {

tests/templates/kuttl/smoke/13-assert.yaml.j2

@@ -7,7 +7,11 @@
 #
 #   * the StatefulSets actually reconciled to a running state (status.readyReplicas)
 #   * the data PVCs were provisioned and bound (status.phase: Bound)
-#   * the PDB this operator builds is actually applied (bare existence)
+#   * the PDB this operator builds is applied and matched the running server pods
+#     (status.expectedPods)
+#   * the metrics Services this operator builds actually reconciled (bare existence
+#     per role group): readiness does not depend on them, so a unit test alone would
+#     not catch a reconcile that stopped emitting them
 #   * the resources this operator does NOT build itself exist at runtime:
 #       - the cluster-internal `test-zk-server` Service (created by the
 #         listener-operator from the Listener object)
@@ -33,12 +37,26 @@ status:
   readyReplicas: 1
   replicas: 1
 ---
-# Built by this operator; its shape is unit-tested in pdb.rs. Kept here as a bare
-# existence check so the smoke test still proves the PDB is actually applied.
+# Built by this operator; its shape is unit-tested in pdb.rs. Kept here to prove the
+# PDB is applied and that k8s matched it against the 3 running server pods.
 apiVersion: policy/v1
 kind: PodDisruptionBudget
 metadata:
   name: test-zk-server
+status:
+  expectedPods: 3
+---
+# Metrics Services built by this operator (shape unit-tested in service.rs); one bare
+# existence check per role group so a reconcile that stopped emitting them fails here.
+apiVersion: v1
+kind: Service
+metadata:
+  name: test-zk-server-primary-metrics
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: test-zk-server-secondary-metrics
 ---
 # Created at runtime by the listener-operator, not by this operator's build code.
 apiVersion: v1