Implementing thread based PrefetcherIterDataPipe

ghstack-source-id: 635a985ea38c220345a1b7c08d5220e7a24c15c1
Pull Request resolved: #770

test/test_iterdatapipe.py

@@ -258,6 +258,16 @@ def odd_even_bug(i: int) -> int:
         result_dp = source_dp.zip_with_map(map_dp, odd_even)
         self.assertEqual(len(source_dp), len(result_dp))
 
+    def test_prefetcher_iterdatapipe(self) -> None:
+        source_dp = IterableWrapper(range(50000))
+        prefetched_dp = source_dp.prefetch(10)
+        # check if early termination resets child thread properly
+        for _, _ in zip(range(100), prefetched_dp):
+            pass
+        expected = list(source_dp)
+        actual = list(prefetched_dp)
+        self.assertEqual(expected, actual)
+
     def test_repeater_iterdatapipe(self) -> None:
         import itertools
 

torchdata/datapipes/iter/__init__.py

@@ -106,6 +106,7 @@
     LineReaderIterDataPipe as LineReader,
 )
 from torchdata.datapipes.iter.util.prefetch import FullSyncIterDataPipe as FullSync
+from torchdata.datapipes.iter.util.prefetcher import PrefetcherIterDataPipe as Prefetcher
 from torchdata.datapipes.iter.util.randomsplitter import RandomSplitterIterDataPipe as RandomSplitter
 from torchdata.datapipes.iter.util.rararchiveloader import RarArchiveLoaderIterDataPipe as RarArchiveLoader
 from torchdata.datapipes.iter.util.rows2columnar import Rows2ColumnarIterDataPipe as Rows2Columnar
@@ -190,6 +191,7 @@
     "OnlineReader",
     "ParagraphAggregator",
     "ParquetDataFrameLoader",
+    "Prefetcher",
     "RandomSplitter",
     "RarArchiveLoader",
     "Repeater",

torchdata/datapipes/iter/util/prefetcher.py

@@ -0,0 +1,100 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+import threading
+import time
+
+from typing import Optional
+
+from torchdata.dataloader2 import communication
+
+from torchdata.datapipes import functional_datapipe
+from torchdata.datapipes.iter import IterDataPipe
+
+PRODUCER_SLEEP_INTERVAL = 0.0001  # Interval between buffer fullfilment checks
+CONSUMER_SLEEP_INTERVAL = 0.0001  # Interval between checking items availablitity in buffer
+
+
+class _PrefetchData:
+    def __init__(self, source_datapipe, buffer_size):
+        self.run_prefetcher = True
+        # TODO: Potential optimization is changing buffer from list to dequeue
+        self.prefetch_buffer = []
+        self.buffer_size = buffer_size
+        self.source_datapipe = source_datapipe
+
+
+@functional_datapipe("prefetch")
+class PrefetcherIterDataPipe(IterDataPipe):
+    def __init__(self, source_datapipe, buffer_size=10):
+        self.source_datapipe = source_datapipe
+        self.buffer_size = buffer_size
+        self.thread: Optional[threading.Thread] = None
+
+    @staticmethod
+    def thread_worker(prefetch_data):
+        itr = iter(prefetch_data.source_datapipe)
+        stop_iteration = False
+        while prefetch_data.run_prefetcher:
+            if len(prefetch_data.prefetch_buffer) < prefetch_data.buffer_size and not stop_iteration:
+                try:
+                    item = next(itr)
+                    prefetch_data.prefetch_buffer.append(item)
+                except StopIteration:
+                    stop_iteration = True
+                except communication.iter.InvalidStateResetRequired:
+                    stop_iteration = True
+                except communication.iter.TerminateRequired:
+                    prefetch_data.run_prefetcher = False
+            elif stop_iteration and len(prefetch_data.prefetch_buffer) == 0:
+                prefetch_data.run_prefetcher = False
+            else:  # Buffer is full, waiting for main thread to consume items
+                # TODO: Calculate sleep interval based on previous consumption speed
+                time.sleep(PRODUCER_SLEEP_INTERVAL)
+
+    def __iter__(self):
+        self.reset()
+        if self.buffer_size < 1:
+            yield from self.source_datapipe
+        else:
+            try:
+                prefetch_data = _PrefetchData(self.source_datapipe, self.buffer_size)
+                self.prefetch_data = prefetch_data
+                self.thread = threading.Thread(
+                    target=PrefetcherIterDataPipe.thread_worker, args=(prefetch_data,), daemon=True
+                )
+                self.thread.start()
+                while prefetch_data.run_prefetcher:
+                    if len(prefetch_data.prefetch_buffer) > 0:
+                        yield prefetch_data.prefetch_buffer[0]
+                        prefetch_data.prefetch_buffer = prefetch_data.prefetch_buffer[1:]
+                    else:
+                        # TODO: Calculate sleep interval based on previous availability speed
+                        time.sleep(CONSUMER_SLEEP_INTERVAL)
+            finally:
+                prefetch_data.run_prefetcher = False
+                if self.thread is not None:
+                    self.thread.join()
+                    self.thread = None
+
+    # def __getstate__(self):
+    #     """
+    #     Getting state in threading enviroment requires next operations:
+    #         1) Stopping of the producer thread.
+    #         2) Saving buffer.
+    #         3) Adding lazy restart of producer thread when __next__ is called again
+    #           (this will guarantee that you only change state of the source_datapipe
+    #            after entire state of the graph is saved).
+    #     """
+    #     pass
+
+    def reset(self):
+        if self.thread is not None:
+            self.prefetch_data.run_prefetcher = False
+            self.thread.join()
+
+    def reset_iterator(self):
+        self.reset()