[CoreEngine] save the relationship between endpoint and run id.

Author: fedml-alex

python/fedml/computing/scheduler/master/deploy_job_launcher.py

@@ -3,7 +3,6 @@
 from fedml.computing.scheduler.model_scheduler import device_client_constants
 from fedml.computing.scheduler.model_scheduler.device_model_cards import FedMLModelCards
 from fedml.computing.scheduler.scheduler_entry.constants import Constants
-from fedml.computing.scheduler.scheduler_core.compute_cache_manager import ComputeCacheManager
 
 
 class FedMLDeployJobLauncher:
@@ -90,11 +89,7 @@ def check_model_device_ready_and_deploy(self, request_json, run_id, master_devic
         serving_devices.extend(device_slave_ids)
 
         # Start to deploy the model
-        endpoint_id = FedMLDeployJobLauncher.deploy_model(serving_devices, request_json, run_id=run_id)
+        FedMLDeployJobLauncher.deploy_model(serving_devices, request_json, run_id=run_id)
 
-        # Save the relationship between run id and endpoint
-        ComputeCacheManager.get_instance().set_redis_params()
-        ComputeCacheManager.get_instance().get_gpu_cache().set_endpoint_run_id_map(
-            endpoint_id, run_id)
 
 

python/fedml/computing/scheduler/slave/base_slave_protocol_manager.py

@@ -257,6 +257,12 @@ def callback_start_train(self, topic, payload):
                 model_master_device_id=model_master_device_id,
                 model_slave_device_id=model_slave_device_id)
         else:
+            # Save the relationship between run id and endpoint
+            ComputeCacheManager.get_instance().set_redis_params()
+            ComputeCacheManager.get_instance().get_gpu_cache().set_endpoint_run_id_map(
+                endpoint_id, run_id)
+
+            # Report the run status with finished status and return
             self.generate_status_report(run_id, edge_id, server_agent_id=server_agent_id).report_client_id_status(
                 edge_id, GeneralConstants.MSG_MLOPS_CLIENT_STATUS_FINISHED, run_id=run_id)
             return