[Data] Disable multithreaded reads if preserve_order is enabled (#4…

…6135)

Multithreaded reads can cause non-deterministic ordering of output rows.

Signed-off-by: Balaji Veeramani <balaji@anyscale.com>

python/ray/data/datasource/file_based_datasource.py

@@ -233,6 +233,11 @@ def create_read_task_fn(read_paths, num_threads):
             def read_task_fn():
                 nonlocal num_threads, read_paths
 
+                # TODO: We should refactor the code so that we can get the results in
+                # order even when using multiple threads.
+                if ctx.execution_options.preserve_order:
+                    num_threads = 0
+
                 if num_threads > 0:
                     if len(read_paths) < num_threads:
                         num_threads = len(read_paths)