[Dataset] Validate sort key in Sort LogicalOperator (ray-project#34282

)

As a followup of ray-project#32133, we should validate key with block.py:_validate_key_fn(), in generate_sort_fn() before doing sort.

Signed-off-by: Scott Lee <[email protected]>

python/ray/data/_internal/plan.py

@@ -16,10 +16,10 @@
 )
 
 import ray
+from ray.data._internal.util import unify_block_metadata_schema
 from ray.data.block import BlockMetadata
 from ray.data._internal.util import capitalize
 from ray.types import ObjectRef
-from ray.data._internal.arrow_ops.transform_pyarrow import unify_schemas
 from ray.data._internal.block_list import BlockList
 from ray.data._internal.compute import (
     UDF,
@@ -419,29 +419,10 @@ def _get_unified_blocks_schema(
             blocks.ensure_metadata_for_first_block()
 
         metadata = blocks.get_metadata(fetch_if_missing=False)
-        # Some blocks could be empty, in which case we cannot get their schema.
-        # TODO(ekl) validate schema is the same across different blocks.
 
-        # First check if there are blocks with computed schemas, then unify
-        # valid schemas from all such blocks.
-        schemas_to_unify = []
-        for m in metadata:
-            if m.schema is not None and (m.num_rows is None or m.num_rows > 0):
-                schemas_to_unify.append(m.schema)
-        if schemas_to_unify:
-            # Check valid pyarrow installation before attempting schema unification
-            try:
-                import pyarrow as pa
-            except ImportError:
-                pa = None
-            # If the result contains PyArrow schemas, unify them
-            if pa is not None and any(
-                isinstance(s, pa.Schema) for s in schemas_to_unify
-            ):
-                return unify_schemas(schemas_to_unify)
-            # Otherwise, if the resulting schemas are simple types (e.g. int),
-            # return the first schema.
-            return schemas_to_unify[0]
+        unified_schema = unify_block_metadata_schema(metadata)
+        if unified_schema is not None:
+            return unified_schema
         if not fetch_if_missing:
             return None
         # Synchronously fetch the schema.

python/ray/data/_internal/planner/sort.py

@@ -14,6 +14,8 @@
 )
 from ray.data._internal.planner.exchange.sort_task_spec import SortKeyT, SortTaskSpec
 from ray.data._internal.stats import StatsDict
+from ray.data._internal.util import unify_block_metadata_schema
+from ray.data.block import _validate_key_fn
 from ray.data.context import DataContext
 
 
@@ -22,7 +24,6 @@ def generate_sort_fn(
     descending: bool,
 ) -> AllToAllTransformFn:
     """Generate function to sort blocks by the specified key column or key function."""
-    # TODO: validate key with block._validate_key_fn.
 
     def fn(
         key: SortKeyT,
@@ -31,11 +32,15 @@ def fn(
         ctx: TaskContext,
     ) -> Tuple[List[RefBundle], StatsDict]:
         blocks = []
+        metadata = []
         for ref_bundle in refs:
-            for block, _ in ref_bundle.blocks:
+            for block, block_metadata in ref_bundle.blocks:
                 blocks.append(block)
+                metadata.append(block_metadata)
         if len(blocks) == 0:
             return (blocks, {})
+        unified_schema = unify_block_metadata_schema(metadata)
+        _validate_key_fn(unified_schema, key)
 
         if isinstance(key, str):
             key = [(key, "descending" if descending else "ascending")]

python/ray/data/_internal/stage_impl.py

@@ -328,13 +328,14 @@ def do_sort(
                 block_list.clear()
             else:
                 blocks = block_list
+            schema = ds.schema(fetch_if_missing=True)
             if isinstance(key, list):
                 if not key:
                     raise ValueError("`key` must be a list of non-zero length")
                 for subkey in key:
-                    _validate_key_fn(ds, subkey)
+                    _validate_key_fn(schema, subkey)
             else:
-                _validate_key_fn(ds, key)
+                _validate_key_fn(schema, key)
             return sort_impl(blocks, clear_input_blocks, key, descending, ctx)
 
         super().__init__(

python/ray/data/_internal/util.py

@@ -9,6 +9,7 @@
 
 import ray
 from ray.air.constants import TENSOR_COLUMN_NAME
+from ray.data._internal.arrow_ops.transform_pyarrow import unify_schemas
 from ray.data.context import DataContext
 from ray._private.utils import _get_pyarrow_version
 
@@ -462,3 +463,33 @@ def get_table_block_metadata(
     return BlockAccessor.for_block(table).get_metadata(
         input_files=None, exec_stats=stats.build()
     )
+
+
+def unify_block_metadata_schema(
+    metadata: List["BlockMetadata"],
+) -> Optional[Union[type, "pyarrow.lib.Schema"]]:
+    """For the input list of BlockMetadata, return a unified schema of the
+    corresponding blocks. If the metadata have no valid schema, returns None.
+    """
+    # Some blocks could be empty, in which case we cannot get their schema.
+    # TODO(ekl) validate schema is the same across different blocks.
+
+    # First check if there are blocks with computed schemas, then unify
+    # valid schemas from all such blocks.
+    schemas_to_unify = []
+    for m in metadata:
+        if m.schema is not None and (m.num_rows is None or m.num_rows > 0):
+            schemas_to_unify.append(m.schema)
+    if schemas_to_unify:
+        # Check valid pyarrow installation before attempting schema unification
+        try:
+            import pyarrow as pa
+        except ImportError:
+            pa = None
+        # If the result contains PyArrow schemas, unify them
+        if pa is not None and any(isinstance(s, pa.Schema) for s in schemas_to_unify):
+            return unify_schemas(schemas_to_unify)
+        # Otherwise, if the resulting schemas are simple types (e.g. int),
+        # return the first schema.
+        return schemas_to_unify[0]
+    return None

python/ray/data/aggregate.py

@@ -91,7 +91,7 @@ def _set_key_fn(self, on: KeyFn):
         self._key_fn = on
 
     def _validate(self, ds: "Datastream") -> None:
-        _validate_key_fn(ds, self._key_fn)
+        _validate_key_fn(ds.schema(fetch_if_missing=True), self._key_fn)
 
 
 @PublicAPI

python/ray/data/block.py

@@ -40,7 +40,6 @@
     import pandas
     import pyarrow
 
-    from ray.data import Datastream
     from ray.data._internal.block_builder import BlockBuilder
     from ray.data.aggregate import AggregateFn
 
@@ -58,9 +57,11 @@
 KeyFn = Union[None, str, Callable[[T], Any]]
 
 
-def _validate_key_fn(ds: "Datastream", key: KeyFn) -> None:
-    """Check the key function is valid on the given datastream."""
-    schema = ds.schema(fetch_if_missing=True)
+def _validate_key_fn(
+    schema: Optional[Union[type, "pyarrow.lib.Schema"]],
+    key: KeyFn,
+) -> None:
+    """Check the key function is valid on the given schema."""
     if schema is None:
         # Datastream is empty/cleared, validation not possible.
         return

python/ray/data/dataset.py

@@ -1718,7 +1718,7 @@ def groupby(self, key: Optional[KeyFn]) -> "GroupedData[T]":
         # Always allow None since groupby interprets that as grouping all
         # records into a single global group.
         if key is not None:
-            _validate_key_fn(self, key)
+            _validate_key_fn(self.schema(fetch_if_missing=True), key)
 
         return GroupedData(self, key)
 

python/ray/data/tests/test_execution_optimizer.py

@@ -643,6 +643,53 @@ def test_sort_e2e(
     # assert [d["one"] for d in r2] == list(reversed(range(100)))
 
 
+def test_sort_validate_keys(
+    ray_start_regular_shared,
+    enable_optimizer,
+):
+    ds = ray.data.range(10)
+    assert ds.sort().take_all() == list(range(10))
+
+    invalid_col_name = "invalid_column"
+    with pytest.raises(
+        ValueError,
+        match=f"String key '{invalid_col_name}' requires datastream format to be "
+        "'arrow' or 'pandas', was 'simple'",
+    ):
+        ds.sort(invalid_col_name).take_all()
+
+    ds_named = ray.data.from_items(
+        [
+            {"col1": 1, "col2": 2},
+            {"col1": 3, "col2": 4},
+            {"col1": 5, "col2": 6},
+            {"col1": 7, "col2": 8},
+        ]
+    )
+
+    ds_sorted_col1 = ds_named.sort("col1", descending=True)
+    r1 = ds_sorted_col1.select_columns(["col1"]).take_all()
+    r2 = ds_sorted_col1.select_columns(["col2"]).take_all()
+    assert [d["col1"] for d in r1] == [7, 5, 3, 1]
+    assert [d["col2"] for d in r2] == [8, 6, 4, 2]
+
+    with pytest.raises(
+        ValueError,
+        match=f"The column '{invalid_col_name}' does not exist in the schema",
+    ):
+        ds_named.sort(invalid_col_name).take_all()
+
+    def dummy_sort_fn(x):
+        return x
+
+    with pytest.raises(
+        ValueError,
+        match=f"Callable key '{dummy_sort_fn}' requires datastream format to be "
+        "'simple'",
+    ):
+        ds_named.sort(dummy_sort_fn).take_all()
+
+
 def test_aggregate_operator(ray_start_regular_shared, enable_optimizer):
     planner = Planner()
     read_op = Read(ParquetDatasource())