From de71fd8d58d76f4076aae36b05dc7be539751802 Mon Sep 17 00:00:00 2001
From: Han Zhang <623606860@qq.com>
Date: Wed, 24 Nov 2021 15:10:42 +0800
Subject: [PATCH] [Fix] Avoid infinite GPU waiting in dist training (#6501)

* [#6495] fix infinite GPU waiting in dist training

* print log_vars keys in assertion msg

* linting issue
---
 mmdet/models/detectors/base.py | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/mmdet/models/detectors/base.py b/mmdet/models/detectors/base.py
index 762c957c9f8..76fbda1b450 100644
--- a/mmdet/models/detectors/base.py
+++ b/mmdet/models/detectors/base.py
@@ -381,6 +381,16 @@ def _parse_losses(self, losses):
         loss = sum(_value for _key, _value in log_vars.items()
                    if 'loss' in _key)
 
+        # If the loss_vars has different length, GPUs will wait infinitely
+        if dist.is_available() and dist.is_initialized():
+            log_var_length = torch.tensor(len(log_vars), device=loss.device)
+            dist.all_reduce(log_var_length)
+            message = (f'rank {dist.get_rank()}' +
+                       f' len(log_vars): {len(log_vars)}' + ' keys: ' +
+                       ','.join(log_vars.keys()))
+            assert log_var_length == len(log_vars) * dist.get_world_size(), \
+                'loss log variables are different across GPUs!\n' + message
+
         log_vars['loss'] = loss
         for loss_name, loss_value in log_vars.items():
             # reduce loss when distributed training