Decode and skip transactional control records in consumer (#2499)

Author: dpkp

kafka/consumer/fetcher.py

@@ -457,10 +457,20 @@ def _unpack_message_set(self, tp, records):
             batch = records.next_batch()
             while batch is not None:
 
-                # LegacyRecordBatch cannot access either base_offset or last_offset_delta
+                # Try DefaultsRecordBatch / message log format v2
+                # base_offset, last_offset_delta, and control batches
                 try:
                     self._subscriptions.assignment[tp].last_offset_from_message_batch = batch.base_offset + \
                                                                                         batch.last_offset_delta
+                    # Control batches have a single record indicating whether a transaction
+                    # was aborted or committed.
+                    # When isolation_level is READ_COMMITTED (currently unsupported)
+                    # we should also skip all messages from aborted transactions
+                    # For now we only support READ_UNCOMMITTED and so we ignore the
+                    # abort/commit signal.
+                    if batch.is_control_batch:
+                        batch = records.next_batch()
+                        continue
                 except AttributeError:
                     pass
 
@@ -677,7 +687,7 @@ def _create_fetch_requests(self):
                 if next_offset_from_batch_header > self._subscriptions.assignment[partition].position:
                     log.debug(
                         "Advance position for partition %s from %s to %s (last message batch location plus one)"
-                        " to correct for deleted compacted messages",
+                        " to correct for deleted compacted messages and/or transactional control records",
                         partition, self._subscriptions.assignment[partition].position, next_offset_from_batch_header)
                     self._subscriptions.assignment[partition].position = next_offset_from_batch_header
 

kafka/record/default_records.py

@@ -269,8 +269,12 @@ def _read_msg(
                 "payload, but instead read {}".format(length, pos - start_pos))
         self._pos = pos
 
-        return DefaultRecord(
-            offset, timestamp, self.timestamp_type, key, value, headers)
+        if self.is_control_batch:
+            return ControlRecord(
+                offset, timestamp, self.timestamp_type, key, value, headers)
+        else:
+            return DefaultRecord(
+                offset, timestamp, self.timestamp_type, key, value, headers)
 
     def __iter__(self):
         self._maybe_uncompress()
@@ -362,6 +366,45 @@ def __repr__(self):
         )
 
 
+class ControlRecord(DefaultRecord):
+    __slots__ = ("_offset", "_timestamp", "_timestamp_type", "_key", "_value",
+                 "_headers", "_version", "_type")
+
+    KEY_STRUCT = struct.Struct(
+        ">h"  # Current Version => Int16
+        "h"  # Type => Int16 (0 indicates an abort marker, 1 indicates a commit)
+    )
+
+    def __init__(self, offset, timestamp, timestamp_type, key, value, headers):
+        super(ControlRecord, self).__init__(offset, timestamp, timestamp_type, key, value, headers)
+        (self._version, self._type) = self.KEY_STRUCT.unpack(self._key)
+
+    # see https://kafka.apache.org/documentation/#controlbatch
+    @property
+    def version(self):
+        return self._version
+
+    @property
+    def type(self):
+        return self._type
+
+    @property
+    def abort(self):
+        return self._type == 0
+
+    @property
+    def commit(self):
+        return self._type == 1
+
+    def __repr__(self):
+        return (
+            "ControlRecord(offset={!r}, timestamp={!r}, timestamp_type={!r},"
+            " version={!r}, type={!r} <{!s}>)".format(
+                self._offset, self._timestamp, self._timestamp_type,
+                self._version, self._type, "abort" if self.abort else "commit")
+        )
+
+
 class DefaultRecordBatchBuilder(DefaultRecordBase, ABCRecordBatchBuilder):
 
     # excluding key, value and headers:

test/record/test_records.py

@@ -60,6 +60,15 @@
     b'\x00\xff\xff\xff\xff\x00\x00\x00\x03123'
 ]
 
+# Single record control batch (abort)
+control_batch_data_v2 = [
+    b'\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00R\x00\x00\x00\x00'
+    b'\x02e\x97\xff\xd0\x00\x20\x00\x00\x00\x00\x00\x00\x00\x00\x00'
+    b'\x98\x96\x7f\x00\x00\x00\x00\x00\x98\x96'
+    b'\x7f\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff'
+    b'\x00\x00\x00\x01@\x00\x00\x00\x08\x00\x00\x00\x00,opaque-control-message\x00'
+]
+
 
 def test_memory_records_v2():
     data_bytes = b"".join(record_batch_data_v2) + b"\x00" * 4
@@ -230,3 +239,18 @@ def test_memory_records_builder_full(magic, compression_type):
         key=None, timestamp=None, value=b"M")
     assert metadata is None
     assert builder.next_offset() == 1
+
+
+def test_control_record_v2():
+    data_bytes = b"".join(control_batch_data_v2)
+    records = MemoryRecords(data_bytes)
+
+    assert records.has_next() is True
+    batch = records.next_batch()
+    assert batch.is_control_batch is True
+    recs = list(batch)
+    assert len(recs) == 1
+    assert recs[0].version == 0
+    assert recs[0].type == 0
+    assert recs[0].abort is True
+    assert recs[0].commit is False