Push mapper data to reducers after execution

Author: hekaisheng

mars/dataframe/merge/merge.py

@@ -808,7 +808,7 @@ def merge(
     method: str = "auto",
     auto_merge: str = "both",
     auto_merge_threshold: int = 8,
-    bloom_filter: Union[bool, str] = "auto",
+    bloom_filter: Union[bool, str] = False,
     bloom_filter_options: Dict[str, Any] = None,
 ) -> DataFrame:
     """

mars/services/storage/api/oscar.py

@@ -13,7 +13,7 @@
 # limitations under the License.
 
 import sys
-from typing import Any, List, Type, TypeVar, Union
+from typing import Any, List, Tuple, Type, TypeVar, Union
 
 from .... import oscar as mo
 from ....lib.aio import alru_cache
@@ -163,7 +163,7 @@ async def batch_delete(self, args_list, kwargs_list):
     @mo.extensible
     async def fetch(
         self,
-        data_key: str,
+        data_key: Union[str, Tuple],
         level: StorageLevel = None,
         band_name: str = None,
         remote_address: str = None,

mars/services/subtask/worker/processor.py

@@ -26,6 +26,8 @@
     Fetch,
     FetchShuffle,
     execute,
+    MapReduceOperand,
+    OperandStage,
 )
 from ....metrics import Metrics
 from ....optimization.physical import optimize
@@ -424,6 +426,28 @@ async def set_chunks_meta():
         # set result data size
         self.result.data_size = result_data_size
 
+    async def push_mapper_data(self, chunk_graph):
+        # TODO: use task api to get reducer bands
+        reducer_idx_to_band = dict()
+        if not reducer_idx_to_band:
+            return
+        storage_api_to_fetch_tasks = defaultdict(list)
+        for result_chunk in chunk_graph.result_chunks:
+            key = result_chunk.key
+            reducer_idx = key[1]
+            if isinstance(key, tuple):
+                # mapper key is a tuple
+                address, band_name = reducer_idx_to_band[reducer_idx]
+                storage_api = StorageAPI(address, self._session_id, band_name)
+                fetch_task = storage_api.fetch.delay(
+                    key, band_name=self._band[1], remote_address=self._band[0]
+                )
+                storage_api_to_fetch_tasks[storage_api].append(fetch_task)
+        batch_tasks = []
+        for storage_api, tasks in storage_api_to_fetch_tasks.items():
+            batch_tasks.append(asyncio.create_task(storage_api.fetch.batch(*tasks)))
+        await asyncio.gather(*batch_tasks)
+
     async def done(self):
         if self.result.status == SubtaskStatus.running:
             self.result.status = SubtaskStatus.succeeded
@@ -495,6 +519,8 @@ async def run(self):
                 await self._unpin_data(input_keys)
 
         await self.done()
+        # after done, we push mapper data to reducers in advance.
+        await self.push_mapper_data(chunk_graph)
         if self.result.status == SubtaskStatus.succeeded:
             cost_time_secs = (
                 self.result.execution_end_time - self.result.execution_start_time