[SDK] Document in-place replica scaling in create examples

Show how Inference/CServe/Dynamo updates that only change min/max
replicas reuse the selected revision without a rolling update.

Author: michaelshin

examples/sdk/create_cserve.py

@@ -51,6 +51,16 @@ def main():
         print("Deployment details: ", deployment)
 
         """
+        ### Scale replicas in place (no new revision / no rolling update)
+        # Changing only min_replicas and/or max_replicas on the same Create*
+        # request updates the selected revision in place; revision_number stays put.
+        # Any other field change still creates a new revision.
+        qwen_config.min_replicas = 2
+        qwen_config.max_replicas = 4
+        cclient.update_cserve(deployment.id, qwen_config)
+        scaled = cclient.get_cserve(deployment.id)
+        print(f"Scaled to {scaled.min_replicas}-{scaled.max_replicas}; revision {scaled.revision_number}")
+
         ### Pause the deployment
         cclient.pause(deployment.id)
 

examples/sdk/create_dynamo.py

@@ -58,9 +58,15 @@ def main():
         # client.resume(deployment.id)
         # client.delete(deployment.id)
         #
-        # To update a Dynamo deployment, construct a new
-        # CreateDynamoDeploymentRequest and call:
-        # client.update_dynamo(deployment.id, updated_request)
+        # Scale replicas in place (no new revision / no rolling update):
+        # Changing only min_replicas and/or max_replicas on the same Create*
+        # request updates the selected revision in place; revision_number stays put.
+        # Any other field change still creates a new revision.
+        # request.min_replicas = 2
+        # request.max_replicas = 4
+        # client.update_dynamo(deployment.id, request)
+        # scaled = client.get_dynamo(deployment.id)
+        # print(f"Scaled to {scaled.min_replicas}-{scaled.max_replicas}; revision {scaled.revision_number}")
 
     print(f"Created Dynamo deployment {deployment.id}: {deployment.endpoint_url}")
     print(f"Model: {deployment.model}")

examples/sdk/create_inference.py

@@ -38,6 +38,16 @@ def main():
         print("Deployment details: ", deployment)
 
         '''
+        ### Scale replicas in place (no new revision / no rolling update)
+        # Changing only min_replicas and/or max_replicas on the same Create*
+        # request updates the selected revision in place; revision_number stays put.
+        # Any other field change still creates a new revision.
+        request.min_replicas = 2
+        request.max_replicas = 5
+        cclient.update_inference(deployment.id, request)
+        scaled = cclient.get_inference(deployment.id)
+        print(f"Scaled to {scaled.min_replicas}-{scaled.max_replicas}; revision {scaled.revision_number}")
+
         ### Pause the deployment
         cclient.pause(deployment.id)
 

examples/sdk/create_inference_vllm.py

@@ -37,6 +37,16 @@ def main():
         print("Deployment details: ", deployment)
 
         '''
+        ### Scale replicas in place (no new revision / no rolling update)
+        # Changing only min_replicas and/or max_replicas on the same Create*
+        # request updates the selected revision in place; revision_number stays put.
+        # Any other field change still creates a new revision.
+        request.min_replicas = 1
+        request.max_replicas = 3
+        cclient.update_inference(deployment.id, request)
+        scaled = cclient.get_inference(deployment.id)
+        print(f"Scaled to {scaled.min_replicas}-{scaled.max_replicas}; revision {scaled.revision_number}")
+
         ### Pause the deployment
         cclient.pause(deployment.id)