Add DDA rollback functionality

api/datadoghq/v2alpha1/datadogagent_types.go

@@ -2349,6 +2349,43 @@ type RemoteConfigConfiguration struct {
 	Features *DatadogFeatures `json:"features,omitempty"`
 }
 
+// ExperimentPhase is the lifecycle phase of a Fleet Automation experiment.
+// +kubebuilder:validation:Enum=running;stopped;rollback;timeout;promoted;aborted
+type ExperimentPhase string
+
+const (
+	// ExperimentPhaseRunning is set by RC when an experiment starts (startExperiment).
+	ExperimentPhaseRunning ExperimentPhase = "running"
+	// ExperimentPhaseStopped is set by RC to request a rollback (stopExperiment).
+	ExperimentPhaseStopped ExperimentPhase = "stopped"
+	// ExperimentPhaseRollback is set by the operator after processing a stopped signal and restoring the previous spec.
+	ExperimentPhaseRollback ExperimentPhase = "rollback"
+	// ExperimentPhaseTimeout is set by the operator when the experiment exceeds the timeout and is auto-rolled back.
+	ExperimentPhaseTimeout ExperimentPhase = "timeout"
+	// ExperimentPhasePromoted is set by RC when an experiment succeeds (promoteExperiment).
+	ExperimentPhasePromoted ExperimentPhase = "promoted"
+	// ExperimentPhaseAborted is set by the operator when a manual spec change is detected during a running experiment.
+	ExperimentPhaseAborted ExperimentPhase = "aborted"
+)
+
+// ExperimentStatus defines the state of a Fleet Automation experiment.
+// +k8s:openapi-gen=true
+type ExperimentStatus struct {
+	// Phase is the current state of the experiment.
+	// +optional
+	Phase ExperimentPhase `json:"phase,omitempty"`
+	// ID is the unique experiment ID sent by Fleet Automation.
+	// +optional
+	ID string `json:"id,omitempty"`
+	// Generation is the DDA metadata.generation recorded when the experiment started.
+	// Used to detect manual spec changes while the experiment is running: if the
+	// current DDA generation differs from this value, the operator aborts the experiment.
+	//
+	// This value must be recorded after the DDA is patched for a startExperiment signal.
+	// +optional
+	Generation int64 `json:"generation,omitempty"`
+}
+
 // DatadogAgentStatus defines the observed state of DatadogAgent.
 // +k8s:openapi-gen=true
 type DatadogAgentStatus struct {
@@ -2376,6 +2413,9 @@ type DatadogAgentStatus struct {
 	// RemoteConfigConfiguration stores the configuration received from RemoteConfig.
 	// +optional
 	RemoteConfigConfiguration *RemoteConfigConfiguration `json:"remoteConfigConfiguration,omitempty"`
+	// Experiment tracks the state of an active or recent Fleet Automation experiment.
+	// +optional
+	Experiment *ExperimentStatus `json:"experiment,omitempty"`
 }
 
 // DatadogAgent defines Agent configuration, see reference https://github.com/DataDog/datadog-operator/blob/main/docs/configuration.v2alpha1.md
@@ -2387,6 +2427,7 @@ type DatadogAgentStatus struct {
 // +kubebuilder:printcolumn:name="cluster-agent",type="string",JSONPath=".status.clusterAgent.status"
 // +kubebuilder:printcolumn:name="cluster-checks-runner",type="string",JSONPath=".status.clusterChecksRunner.status"
 // +kubebuilder:printcolumn:name="age",type="date",JSONPath=".metadata.creationTimestamp"
+// +kubebuilder:printcolumn:name="experiment-phase",type="string",JSONPath=".status.experiment.phase",priority=1
 // +k8s:openapi-gen=true
 // +genclient
 type DatadogAgent struct {

config/crd/bases/v1/datadoghq.com_datadogagents.yaml

@@ -29,6 +29,10 @@ spec:
         - jsonPath: .metadata.creationTimestamp
           name: age
           type: date
+        - jsonPath: .status.experiment.phase
+          name: experiment-phase
+          priority: 1
+          type: string
       name: v2alpha1
       schema:
         openAPIV3Schema:
@@ -8470,6 +8474,32 @@ spec:
                   x-kubernetes-list-map-keys:
                     - type
                   x-kubernetes-list-type: map
+                experiment:
+                  description: Experiment tracks the state of an active or recent Fleet Automation experiment.
+                  properties:
+                    generation:
+                      description: |-
+                        Generation is the DDA metadata.generation recorded when the experiment started.
+                        Used to detect manual spec changes while the experiment is running: if the
+                        current DDA generation differs from this value, the operator aborts the experiment.
+
+                        This value must be recorded after the DDA is patched for a startExperiment signal.
+                      format: int64
+                      type: integer
+                    id:
+                      description: ID is the unique experiment ID sent by Fleet Automation.
+                      type: string
+                    phase:
+                      description: Phase is the current state of the experiment.
+                      enum:
+                        - running
+                        - stopped
+                        - rollback
+                        - timeout
+                        - promoted
+                        - aborted
+                      type: string
+                  type: object
                 otelAgentGateway:
                   description: The actual state of the OTel Agent Gateway as a deployment.
                   properties:

config/crd/bases/v1/datadoghq.com_datadogagents_v2alpha1.json

@@ -8175,6 +8175,34 @@
           ],
           "x-kubernetes-list-type": "map"
         },
+        "experiment": {
+          "additionalProperties": false,
+          "description": "Experiment tracks the state of an active or recent Fleet Automation experiment.",
+          "properties": {
+            "generation": {
+              "description": "Generation is the DDA metadata.generation recorded when the experiment started.\nUsed to detect manual spec changes while the experiment is running: if the\ncurrent DDA generation differs from this value, the operator aborts the experiment.\n\nThis value must be recorded after the DDA is patched for a startExperiment signal.",
+              "format": "int64",
+              "type": "integer"
+            },
+            "id": {
+              "description": "ID is the unique experiment ID sent by Fleet Automation.",
+              "type": "string"
+            },
+            "phase": {
+              "description": "Phase is the current state of the experiment.",
+              "enum": [
+                "running",
+                "stopped",
+                "rollback",
+                "timeout",
+                "promoted",
+                "aborted"
+              ],
+              "type": "string"
+            }
+          },
+          "type": "object"
+        },
         "otelAgentGateway": {
           "additionalProperties": false,
           "description": "The actual state of the OTel Agent Gateway as a deployment.",

internal/controller/datadogagent/controller.go

@@ -76,6 +76,8 @@ type ReconcilerOptions struct {
 	DatadogAgentProfileEnabled  bool
 	DatadogAgentInternalEnabled bool
 	CreateControllerRevisions   bool
+	// ExperimentTimeout overrides ExperimentDefaultTimeout. Zero means use the default.
+	ExperimentTimeout time.Duration
 }
 
 // Reconciler is the internal reconciler for Datadog Agent