chore(operator/scaling)!: Rework Scaler state to complex enum

crates/stackable-operator/CHANGELOG.md

@@ -9,6 +9,7 @@ All notable changes to this project will be documented in this file.
 - Add generic database connection mechanism ([#1163]).
 - Add `config_overrides` module with `KeyValueOverridesProvider` trait, enabling
   structured config file formats (e.g. JSON) in addition to key-value overrides ([#1177]).
+- Add `Scaler` CRD ([#1190], [#1195]).
 
 ### Changed
 
@@ -28,9 +29,11 @@ All notable changes to this project will be documented in this file.
 
 [#1163]: https://github.com/stackabletech/operator-rs/pull/1163
 [#1177]: https://github.com/stackabletech/operator-rs/pull/1177
+[#1190]: https://github.com/stackabletech/operator-rs/pull/1190
 [#1191]: https://github.com/stackabletech/operator-rs/pull/1191
 [#1192]: https://github.com/stackabletech/operator-rs/pull/1192
 [#1194]: https://github.com/stackabletech/operator-rs/pull/1194
+[#1195]: https://github.com/stackabletech/operator-rs/pull/1195
 
 ## [0.109.0] - 2026-04-07
 

crates/stackable-operator/crds/Scaler.yaml

@@ -65,38 +65,80 @@ spec:
                 type: string
               state:
                 description: The current state of the scaler state machine.
+                oneOf:
+                - required:
+                  - idle
+                - required:
+                  - preScaling
+                - required:
+                  - scaling
+                - required:
+                  - postScaling
+                - required:
+                  - failed
                 properties:
-                  details:
+                  failed:
+                    description: |-
+                      A hook returned an error.
+
+                      The scaler stays here until the user applies the [`Annotation::autoscaling_retry`] annotation
+                      to trigger a reset to [`ScalerState::Idle`].
                     properties:
                       failedIn:
-                        description: In which state the scaling operation failed.
+                        description: Which stage produced the error.
                         enum:
-                        - preScaling
-                        - scaling
-                        - postScaling
+                        - PreScaling
+                        - Scaling
+                        - PostScaling
                         type: string
-                      previous_replicas:
-                        maximum: 65535.0
-                        minimum: 0.0
-                        type: uint16
                       reason:
+                        description: Human-readable error message from the hook.
                         type: string
+                    required:
+                    - failedIn
+                    - reason
+                    type: object
+                  idle:
+                    description: No scaling operation is in progress.
+                    type: object
+                  postScaling:
+                    description: |-
+                      Running the `post_scale` hook (e.g. cluster rebalance).
+
+                      This stage additionally tracks the previous replica count to be able derive the direction
+                      of the scaling operation.
+                    properties:
+                      previousReplicas:
+                        format: uint16
+                        maximum: 65535.0
+                        minimum: 0.0
+                        type: integer
+                    required:
+                    - previousReplicas
+                    type: object
+                  preScaling:
+                    description: Running the `pre_scale` hook (e.g. data offload).
+                    type: object
+                  scaling:
+                    description: |-
+                      Waiting for the StatefulSet to converge to the new replica count.
+
+                      This stage additionally tracks the previous replica count to be able derive the direction
+                      of the scaling operation.
+                    properties:
+                      previousReplicas:
+                        format: uint16
+                        maximum: 65535.0
+                        minimum: 0.0
+                        type: integer
+                    required:
+                    - previousReplicas
                     type: object
-                  state:
-                    enum:
-                    - idle
-                    - preScaling
-                    - scaling
-                    - postScaling
-                    - failed
-                    type: string
-                required:
-                - state
                 type: object
             required:
+            - lastTransitionTime
             - replicas
             - state
-            - lastTransitionTime
             type: object
         required:
         - spec

crates/stackable-operator/src/crd/scaler/mod.rs

@@ -1,5 +1,3 @@
-use std::borrow::Cow;
-
 use k8s_openapi::apimachinery::pkg::apis::meta::v1::Time;
 use kube::CustomResource;
 use schemars::JsonSchema;
@@ -21,7 +19,7 @@ pub mod versioned {
         ),
         namespaced
     ))]
-    #[derive(Clone, Debug, PartialEq, CustomResource, Deserialize, Serialize, JsonSchema)]
+    #[derive(Clone, Debug, PartialEq, Eq, CustomResource, Deserialize, Serialize, JsonSchema)]
     pub struct ScalerSpec {
         /// Desired replica count.
         ///
@@ -40,7 +38,7 @@ pub mod versioned {
 }
 
 /// Status of a StackableScaler.
-#[derive(Clone, Debug, Deserialize, Serialize, JsonSchema)]
+#[derive(Clone, Debug, PartialEq, Eq, Deserialize, Serialize, JsonSchema)]
 #[serde(rename_all = "camelCase")]
 pub struct ScalerStatus {
     /// The current total number of replicas targeted by the managed StatefulSet.
@@ -59,25 +57,15 @@ pub struct ScalerStatus {
     pub last_transition_time: Time,
 }
 
-// We use `#[serde(tag)]` and `#[serde(content)]` here to circumvent Kubernetes restrictions in their
-// structural schema subset of OpenAPI schemas. They don't allow one variant to be typed as a string
-// and others to be typed as objects. We therefore encode the variant data in a separate details
-// key/object. With this, all variants can be encoded as strings, while the status can still contain
-// additional data in an extra field when needed.
-#[derive(Clone, Debug, Deserialize, Serialize, strum::Display)]
-#[serde(
-    tag = "state",
-    content = "details",
-    rename_all = "camelCase",
-    rename_all_fields = "camelCase"
-)]
+#[derive(Clone, Debug, PartialEq, Eq, Deserialize, Serialize, JsonSchema, strum::Display)]
+#[serde(rename_all = "camelCase", rename_all_fields = "camelCase")]
 #[strum(serialize_all = "camelCase")]
 pub enum ScalerState {
     /// No scaling operation is in progress.
-    Idle,
+    Idle {},
 
     /// Running the `pre_scale` hook (e.g. data offload).
-    PreScaling,
+    PreScaling {},
 
     /// Waiting for the StatefulSet to converge to the new replica count.
     ///
@@ -104,44 +92,9 @@ pub enum ScalerState {
     },
 }
 
-// We manually implement the JSON schema instead of deriving it, because kube's schema transformer
-// cannot handle the derived JsonSchema and proceeds to hit the following error: "Property "state"
-// has the schema ... but was already defined as ... in another subschema. The schemas for a
-// property used in multiple subschemas must be identical".
-impl JsonSchema for ScalerState {
-    fn schema_name() -> Cow<'static, str> {
-        "ScalerState".into()
-    }
-
-    fn json_schema(generator: &mut schemars::generate::SchemaGenerator) -> schemars::Schema {
-        schemars::json_schema!({
-            "type": "object",
-            "required": ["state"],
-            "properties": {
-                "state": {
-                    "type": "string",
-                    "enum": ["idle", "preScaling", "scaling", "postScaling", "failed"]
-                },
-                "details": {
-                    "type": "object",
-                    "properties": {
-                        "failedIn": generator.subschema_for::<FailedInState>(),
-                        "previous_replicas": {
-                            "type": "uint16",
-                            "minimum": u16::MIN,
-                            "maximum": u16::MAX
-                        },
-                        "reason": { "type": "string" }
-                    }
-                }
-            }
-        })
-    }
-}
-
 /// In which state the scaling operation failed.
-#[derive(Clone, Debug, Deserialize, Serialize, JsonSchema)]
-#[serde(rename_all = "camelCase")]
+#[derive(Clone, Debug, PartialEq, Eq, Deserialize, Serialize, JsonSchema)]
+#[serde(rename_all = "PascalCase")]
 pub enum FailedInState {
     /// The `pre_scale` hook returned an error.
     PreScaling,

crates/stackable-operator/src/lib.rs

@@ -31,6 +31,7 @@ pub mod product_config_utils;
 pub mod product_logging;
 pub mod role_utils;
 pub mod status;
+pub mod test_utils;
 pub mod utils;
 pub mod validation;
 

crates/stackable-operator/src/test_utils.rs

@@ -0,0 +1,17 @@
+/// Please use only in tests, as we have non-ideal error handling in case serde_yaml produced
+/// non-utf8 output.
+pub fn serialize_to_yaml_with_singleton_map<S>(input: &S) -> Result<String, serde_yaml::Error>
+where
+    S: serde::Serialize,
+{
+    use serde::ser::Error as _;
+
+    let mut buffer = Vec::new();
+    let mut serializer = serde_yaml::Serializer::new(&mut buffer);
+    serde_yaml::with::singleton_map_recursive::serialize(input, &mut serializer)?;
+    String::from_utf8(buffer).map_err(|err| {
+        serde_yaml::Error::custom(format!(
+            "For *some* reason, serde_yaml produced non-utf8 output: {err}"
+        ))
+    })
+}