diff --git a/c/src/ml-api-service-training-offloading.c b/c/src/ml-api-service-training-offloading.c
index 13722432..c8a8c1bb 100644
--- a/c/src/ml-api-service-training-offloading.c
+++ b/c/src/ml-api-service-training-offloading.c
@@ -513,6 +513,7 @@ _training_offloading_services_request (ml_service_s * mls)
 
 error:
   g_free (service_name);
+  g_free (pipeline);
   g_free (transfer_data);
   g_free (contents);
   g_list_free (list);
@@ -569,6 +570,11 @@ _training_offloading_check_received_data (ml_training_services_s * training_s)
 
   g_return_val_if_fail (training_s != NULL, FALSE);
 
+  if (training_s->received_thread) {
+    g_thread_join (training_s->received_thread);
+    training_s->received_thread = NULL;
+  }
+
   training_s->received_thread = g_thread_new ("check_received_file",
       _check_received_data_thread, training_s);
 
@@ -709,6 +715,7 @@ _ml_service_training_offloading_prepare_receiver (ml_service_s * mls,
 
   pipe = json_object_get_object_member (pipeline_obj, "pipeline");
   if (json_object_has_member (pipe, "description")) {
+    g_clear_pointer (&training_s->receiver_pipe, g_free);
     training_s->receiver_pipe =
         g_strdup (_ml_service_get_json_string_member (pipe, "description"));
   } else {
@@ -809,6 +816,7 @@ _ml_service_training_offloading_process_received_data (ml_service_s * mls,
 
   if (training_s->type == ML_TRAINING_OFFLOADING_TYPE_RECEIVER) {
     if (service_type == ML_SERVICE_OFFLOADING_TYPE_PIPELINE_RAW) {
+      g_clear_pointer (&training_s->receiver_pipe_json_str, g_free);
       training_s->receiver_pipe_json_str = g_strdup (data);
       _ml_logd ("Received JSON string pipeline:%s",
           training_s->receiver_pipe_json_str);
@@ -821,6 +829,7 @@ _ml_service_training_offloading_process_received_data (ml_service_s * mls,
         _ml_error_report_return (ret,
             "Failed to get name while processing the ml-offloading service.");
       }
+      g_clear_pointer (&training_s->trained_model_path, g_free);
       training_s->trained_model_path =
           g_build_path (G_DIR_SEPARATOR_S, dir_path, name, NULL);
       _ml_logd ("Reply: name:%s, received trained_model:%s", name,