feat: rolloutrun status support SkipTolerations

config/crd/rollout/rollout.kusionstack.io_rolloutruns.yaml

@@ -2264,32 +2264,31 @@ spec:
                       - targets
                       type: object
                     type: array
-                  toleration:
-                    description: Toleration is the toleration policy of the canary strategy
-                    properties:
-                      initialDelaySeconds:
-                        description: Number of seconds after the toleration check has started before the task are initiated.
-                        format: int32
-                        type: integer
-                      taskFailureThreshold:
-                        anyOf:
-                        - type: integer
-                        - type: string
-                        description: |-
-                          FailureThreshold indicates how many failed pods can be tolerated before marking the rollout task as success
-                          If not set, the default value is 0, which means no failed pods can be tolerated
-                          This is a task level threshold.
-                        x-kubernetes-int-or-string: true
-                      workloadTotalFailureThreshold:
-                        anyOf:
-                        - type: integer
-                        - type: string
-                        description: |-
-                          WorkloadFailureThreshold indicates how many failed pods can be tolerated in all upgraded pods of one workload.
-                          The default value is 0, which means no failed pods can be tolerated.
-                          This is a workload level threshold.
-                        x-kubernetes-int-or-string: true
-                    type: object
+                  tolerations:
+                    description: |-
+                      Tolerations records the accumulated toleration from skipped batches per workload.
+                      When a batch is skipped, the gap between expected and actual replicas for each workload
+                      is accumulated into this field, allowing subsequent batches to tolerate the deficit.
+                    items:
+                      description: RolloutRunTolerationTarget records the toleration value accumulated from skipped batches for a specific workload.
+                      properties:
+                        cluster:
+                          description: Cluster indicates the name of cluster
+                          type: string
+                        name:
+                          description: Name is the resource name
+                          type: string
+                        toleration:
+                          description: |-
+                            Toleration is the accumulated toleration value from skipped batches.
+                            It represents how many replicas the workload is allowed to be short of.
+                          format: int32
+                          type: integer
+                      required:
+                      - name
+                      - toleration
+                      type: object
+                    type: array
                 type: object
               canary:
                 description: Canary defines the canary strategy

rollout/v1alpha1/rolloutrun_types.go

@@ -76,9 +76,11 @@ type RolloutRunBatchStrategy struct {
 	// Batches define the order of phases to execute release in batch release
 	Batches []RolloutRunStep `json:"batches,omitempty"`
 
-	// Toleration is the toleration policy of the canary strategy
+	// Tolerations records the accumulated toleration from skipped batches per workload.
+	// When a batch is skipped, the gap between expected and actual replicas for each workload
+	// is accumulated into this field, allowing subsequent batches to tolerate the deficit.
 	// +optional
-	Toleration *TolerationStrategy `json:"toleration,omitempty"`
+	Tolerations []RolloutRunTolerationTarget `json:"tolerations,omitempty"`
 }
 
 type RolloutRunStep struct {
@@ -151,6 +153,15 @@ type RolloutRunStatus struct {
 	TargetStatuses []RolloutWorkloadStatus `json:"targetStatuses,omitempty"`
 }
 
+// RolloutRunTolerationTarget records the toleration value accumulated from skipped batches for a specific workload.
+type RolloutRunTolerationTarget struct {
+	CrossClusterObjectNameReference `json:",inline"`
+
+	// Toleration is the accumulated toleration value from skipped batches.
+	// It represents how many replicas the workload is allowed to be short of.
+	Toleration int32 `json:"toleration"`
+}
+
 type RolloutRunBatchStatus struct {
 	// RolloutBatchStatus contains status of current batch
 	RolloutBatchStatus `json:",inline"`