diff --git a/official/recommendation/constants.py b/official/recommendation/constants.py
index 33230e75e6c..d4faf5fd74a 100644
--- a/official/recommendation/constants.py
+++ b/official/recommendation/constants.py
@@ -58,6 +58,7 @@ def __init__(self, data_dir, cache_id=None):
 CYCLES_TO_BUFFER = 3  # The number of train cycles worth of data to "run ahead"
                       # of the main training loop.
 
+READY_FILE_TEMP = "ready.json.temp"
 READY_FILE = "ready.json"
 TRAIN_RECORD_TEMPLATE = "train_{}.tfrecords"
 
diff --git a/official/recommendation/data_async_generation.py b/official/recommendation/data_async_generation.py
index 30432bb5268..16ec7a792e1 100644
--- a/official/recommendation/data_async_generation.py
+++ b/official/recommendation/data_async_generation.py
@@ -282,11 +282,17 @@ def _construct_training_records(
       raise ValueError("Error detected: point counts do not match: {} vs. {}"
                        .format(num_pts, written_pts))
 
-  with tf.gfile.Open(os.path.join(record_dir, rconst.READY_FILE), "w") as f:
+  # We write to a temp file then atomically rename it to the final file, because
+  # writing directly to the final file can cause the main process to read a
+  # partially written JSON file.
+  ready_file_temp = os.path.join(record_dir, rconst.READY_FILE_TEMP)
+  with tf.gfile.Open(ready_file_temp, "w") as f:
     json.dump({
         "batch_size": train_batch_size,
         "batch_count": batch_count,
     }, f)
+  ready_file = os.path.join(record_dir, rconst.READY_FILE)
+  tf.gfile.Rename(ready_file_temp, ready_file)
 
   log_msg("Cycle {} complete. Total time: {:.1f} seconds"
           .format(train_cycle, timeit.default_timer() - st))