Fix passing empty bytes in python tasks (#7045)

* ensure data_ won't be null_ptr when size == 0

* when data_sizes[i] == 0, we should Allocate an empty buffer

* work around for pyarrow.py_buffer

* fix comments

* add null ptr check

* add test for bytes

* lint

python/ray/_raylet.pyx

@@ -566,10 +566,13 @@ cdef write_serialized_object(
     from ray.serialization import Pickle5SerializedObject, RawSerializedObject
 
     if isinstance(serialized_object, RawSerializedObject):
-        buffer = Buffer.make(buf)
-        stream = pyarrow.FixedSizeBufferWriter(pyarrow.py_buffer(buffer))
-        stream.set_memcopy_threads(MEMCOPY_THREADS)
-        stream.write(pyarrow.py_buffer(serialized_object.value))
+        if buf.get() != NULL and buf.get().Size() > 0:
+            buffer = Buffer.make(buf)
+            # `Buffer` has a nullptr buffer underlying if size is 0,
+            # which will cause `pyarrow.py_buffer` crash
+            stream = pyarrow.FixedSizeBufferWriter(pyarrow.py_buffer(buffer))
+            stream.set_memcopy_threads(MEMCOPY_THREADS)
+            stream.write(pyarrow.py_buffer(serialized_object.value))
     elif isinstance(serialized_object, Pickle5SerializedObject):
         (<Pickle5Writer>serialized_object.writer).write_to(
             serialized_object.inband, buf, MEMCOPY_THREADS)

python/ray/tests/test_basic.py

@@ -43,6 +43,8 @@ def test_simple_serialization(ray_start_regular):
         0.9,
         1 << 62,
         1 << 999,
+        b"",
+        b"a",
         "a",
         string.printable,
         "\u262F",