Default values and warnings for DDP env variables

Author: Alexander Reshytko

pytorch_lightning/core/lightning.py

@@ -927,6 +927,18 @@ def init_ddp_connection(
         if is_slurm_managing_tasks:
             self._init_slurm_connection()
 
+        if 'MASTER_ADDR' not in os.environ:
+            log.warning("MASTER_ADDR environment variable is not defined. Set as localhost")
+            os.environ['MASTER_ADDR'] = '127.0.0.2'
+
+        if 'MASTER_PORT' not in os.environ:
+            log.warning("MASTER_PORT environment variable is not defined. Set as 12910")
+            os.environ['MASTER_PORT'] = '12910'
+
+        if 'WORLD_SIZE' in os.environ and os.environ['WORLD_SIZE'] != world_size:
+            log.warning("WORLD_SIZE environment variable is not equal to the computed "
+                        "world size. Ignored.")
+
         torch_distrib.init_process_group('nccl', rank=proc_rank, world_size=world_size)
 
     def configure_apex(

pytorch_lightning/trainer/distrib_data_parallel.py

@@ -281,6 +281,7 @@ def ddp_train(self, gpu_idx, model):
             node_id = os.environ['SLURM_NODEID'] if self.is_slurm_managing_tasks else os.environ['RANK']
             self.node_rank = int(node_id)
         except Exception:
+            log.warning("SLURM_NODEID or RANK environment variable is not defined. Set as 0.")
             self.node_rank = 0
 
         # show progressbar only on progress_rank 0