feast-dev
diff --git a/‎docs/reference/type-system.md‎
Lines changed: 47 additions & 0 deletions b/‎docs/reference/type-system.md‎
Lines changed: 47 additions & 0 deletions
diff --git a/‎protos/feast/types/Value.proto‎
Lines changed: 4 additions & 0 deletions b/‎protos/feast/types/Value.proto‎
Lines changed: 4 additions & 0 deletions
diff --git a/‎sdk/python/feast/field.py‎
Lines changed: 31 additions & 7 deletions b/‎sdk/python/feast/field.py‎
Lines changed: 31 additions & 7 deletions
diff --git a/‎sdk/python/feast/infra/offline_stores/offline_utils.py‎
Lines changed: 39 additions & 9 deletions b/‎sdk/python/feast/infra/offline_stores/offline_utils.py‎
Lines changed: 39 additions & 9 deletions
diff --git a/‎sdk/python/feast/infra/online_stores/remote.py‎
Lines changed: 15 additions & 1 deletion b/‎sdk/python/feast/infra/online_stores/remote.py‎
Lines changed: 15 additions & 1 deletion
diff --git a/‎sdk/python/feast/proto_json.py‎
Lines changed: 19 additions & 0 deletions b/‎sdk/python/feast/proto_json.py‎
Lines changed: 19 additions & 0 deletions
@@ -86,6 +86,25 @@ All primitive types (except `Map` and `Json`) have corresponding set types for s
 - Set types are best suited for **online serving** use cases where feature values are written as Python sets and retrieved via `get_online_features`.
 {% endhint %}
 
+### Nested Collection Types
+
+Feast supports arbitrarily nested collections using a recursive `VALUE_LIST` / `VALUE_SET` design. The outer container determines the proto enum (`VALUE_LIST` for `Array(…)`, `VALUE_SET` for `Set(…)`), while the full inner type structure is persisted via a mandatory `feast:nested_inner_type` Field tag.
+
+| Feast Type | Python Type | ValueType | Description |
+|------------|-------------|-----------|-------------|
+| `Array(Array(T))` | `List[List[T]]` | `VALUE_LIST` | List of lists |
+| `Array(Set(T))` | `List[List[T]]` | `VALUE_LIST` | List of sets |
+| `Set(Array(T))` | `List[List[T]]` | `VALUE_SET` | Set of lists |
+| `Set(Set(T))` | `List[List[T]]` | `VALUE_SET` | Set of sets |
+| `Array(Array(Array(T)))` | `List[List[List[T]]]` | `VALUE_LIST` | 3-level nesting |
+
+Where `T` is any supported primitive type (Int32, Int64, Float32, Float64, String, Bytes, Bool, UnixTimestamp) or another nested collection type.
+
+**Notes:**
+- Nesting depth is **unlimited**. `Array(Array(Array(T)))`, `Set(Array(Set(T)))`, etc. are all supported.
+- Inner type information is preserved via Field tags (`feast:nested_inner_type`) and restored during deserialization. This tag is mandatory for nested collection types.
+- Empty inner collections (`[]`) are stored as empty proto values and round-trip as `None`. For example, `[[1, 2], [], [3]]` becomes `[[1, 2], None, [3]]` after a write-read cycle.
+
 ### Map Types
 
 Map types allow storing dictionary-like data structures:
@@ -233,6 +252,10 @@ user_features = FeatureView(
         Field(name="metadata", dtype=Map),
         Field(name="activity_log", dtype=Array(Map)),
 
+        # Nested collection types
+        Field(name="weekly_scores", dtype=Array(Array(Float64))),
+        Field(name="unique_tags_per_category", dtype=Array(Set(String))),
+
         # JSON type
         Field(name="raw_event", dtype=Json),
 
@@ -290,6 +313,30 @@ related_sessions = [uuid.uuid4(), uuid.uuid4(), uuid.uuid4()]
 unique_devices = {uuid.uuid4(), uuid.uuid4()}
 ```
 
+### Nested Collection Type Usage Examples
+
+Nested collections allow storing multi-dimensional data with unlimited depth:
+
+```python
+# List of lists — e.g., weekly score history per user
+weekly_scores = [[85.0, 90.5, 78.0], [92.0, 88.5], [95.0, 91.0, 87.5]]
+
+# List of sets — e.g., unique tags assigned per category
+unique_tags_per_category = [["python", "ml"], ["rust", "systems"], ["python", "web"]]
+
+# 3-level nesting — e.g., multi-dimensional matrices
+Field(name="tensor", dtype=Array(Array(Array(Float64))))
+
+# Mixed nesting
+Field(name="grouped_tags", dtype=Array(Set(Array(String))))
+```
+
+**Limitation:** Empty inner collections round-trip as `None`:
+```python
+# Input:  [[1, 2], [], [3]]
+# Output: [[1, 2], None, [3]]  (empty [] becomes None after write-read cycle)
+```
+
 ### Map Type Usage Examples
 
 Maps can store complex nested data structures:
 
@@ -63,6 +63,8 @@ message ValueType {
     TIME_UUID_LIST = 39;
     UUID_SET = 40;
     TIME_UUID_SET = 41;
+    VALUE_LIST = 42;
+    VALUE_SET = 43;
   }
 }
 
@@ -108,6 +110,8 @@ message Value {
     StringList time_uuid_list_val = 39;
     StringSet uuid_set_val = 40;
     StringSet time_uuid_set_val = 41;
+    RepeatedValue list_val = 42;
+    RepeatedValue set_val = 43;
   }
 }
 
 
@@ -23,6 +23,7 @@
 from feast.value_type import ValueType
 
 STRUCT_SCHEMA_TAG = "feast:struct_schema"
+NESTED_COLLECTION_INNER_TYPE_TAG = "feast:nested_inner_type"
 
 
 @typechecked
@@ -118,7 +119,7 @@ def __str__(self):
 
     def to_proto(self) -> FieldProto:
         """Converts a Field object to its protobuf representation."""
-        from feast.types import Array
+        from feast.types import Array, Set
 
         value_type = self.dtype.to_value_type()
         vector_search_metric = self.vector_search_metric or ""
@@ -128,6 +129,11 @@ def to_proto(self) -> FieldProto:
             tags[STRUCT_SCHEMA_TAG] = _serialize_struct_schema(self.dtype)
         elif isinstance(self.dtype, Array) and isinstance(self.dtype.base_type, Struct):
             tags[STRUCT_SCHEMA_TAG] = _serialize_struct_schema(self.dtype.base_type)
+        # Persist nested collection type info in tags
+        if isinstance(self.dtype, (Array, Set)) and isinstance(
+            self.dtype.base_type, (Array, Set)
+        ):
+            tags[NESTED_COLLECTION_INNER_TYPE_TAG] = _feast_type_to_str(self.dtype)
         return FieldProto(
             name=self.name,
             value_type=value_type.value,
@@ -155,17 +161,24 @@ def from_proto(cls, field_proto: FieldProto):
         # Reconstruct Struct type from persisted schema in tags
         from feast.types import Array
 
+        internal_tags = {STRUCT_SCHEMA_TAG, NESTED_COLLECTION_INNER_TYPE_TAG}
         dtype: FeastType
         if value_type == ValueType.STRUCT and STRUCT_SCHEMA_TAG in tags:
             dtype = _deserialize_struct_schema(tags[STRUCT_SCHEMA_TAG])
-            user_tags = {k: v for k, v in tags.items() if k != STRUCT_SCHEMA_TAG}
+            user_tags = {k: v for k, v in tags.items() if k not in internal_tags}
         elif value_type == ValueType.STRUCT_LIST and STRUCT_SCHEMA_TAG in tags:
             inner_struct = _deserialize_struct_schema(tags[STRUCT_SCHEMA_TAG])
             dtype = Array(inner_struct)
-            user_tags = {k: v for k, v in tags.items() if k != STRUCT_SCHEMA_TAG}
+            user_tags = {k: v for k, v in tags.items() if k not in internal_tags}
+        elif (
+            value_type in (ValueType.VALUE_LIST, ValueType.VALUE_SET)
+            and NESTED_COLLECTION_INNER_TYPE_TAG in tags
+        ):
+            dtype = _str_to_feast_type(tags[NESTED_COLLECTION_INNER_TYPE_TAG])
+            user_tags = {k: v for k, v in tags.items() if k not in internal_tags}
         else:
             dtype = from_value_type(value_type=value_type)
-            user_tags = tags
+            user_tags = {k: v for k, v in tags.items() if k not in internal_tags}
 
         return cls(
             name=field_proto.name,
@@ -198,6 +211,7 @@ def _feast_type_to_str(feast_type: FeastType) -> str:
     from feast.types import (
         Array,
         PrimitiveFeastType,
+        Set,
     )
 
     if isinstance(feast_type, PrimitiveFeastType):
@@ -209,6 +223,8 @@ def _feast_type_to_str(feast_type: FeastType) -> str:
         return json.dumps({"__struct__": nested})
     elif isinstance(feast_type, Array):
         return f"Array({_feast_type_to_str(feast_type.base_type)})"
+    elif isinstance(feast_type, Set):
+        return f"Set({_feast_type_to_str(feast_type.base_type)})"
     else:
         return str(feast_type)
 
@@ -218,6 +234,7 @@ def _str_to_feast_type(type_str: str) -> FeastType:
     from feast.types import (
         Array,
         PrimitiveFeastType,
+        Set,
     )
 
     # Check if it's an Array type
@@ -226,6 +243,12 @@ def _str_to_feast_type(type_str: str) -> FeastType:
         base_type = _str_to_feast_type(inner)
         return Array(base_type)
 
+    # Check if it's a Set type
+    if type_str.startswith("Set(") and type_str.endswith(")"):
+        inner = type_str[4:-1]
+        base_type = _str_to_feast_type(inner)
+        return Set(base_type)
+
     # Check if it's a nested Struct (JSON encoded)
     if type_str.startswith("{"):
         try:
@@ -243,9 +266,10 @@ def _str_to_feast_type(type_str: str) -> FeastType:
     try:
         return PrimitiveFeastType[type_str]
     except KeyError:
-        from feast.types import String
-
-        return String
+        raise ValueError(
+            f"Unknown FeastType: {type_str!r}. "
+            f"Valid primitive types: {[t.name for t in PrimitiveFeastType]}"
+        )
 
 
 def _serialize_struct_schema(struct_type: Struct) -> str:
 
@@ -1,3 +1,4 @@
+import logging
 import uuid
 from dataclasses import asdict, dataclass
 from datetime import datetime, timedelta, timezone
@@ -21,6 +22,7 @@
 from feast.repo_config import RepoConfig
 from feast.type_map import feast_value_type_to_pa
 from feast.utils import _get_requested_feature_views_to_features_dict, to_naive_utc
+from feast.value_type import ValueType
 
 DEFAULT_ENTITY_DF_EVENT_TIMESTAMP_COL = "event_timestamp"
 
@@ -241,6 +243,37 @@ def get_offline_store_from_config(offline_store_config: Any) -> OfflineStore:
     return offline_store_class()
 
 
+_PA_BASIC_TYPES = {
+    "int32": pa.int32(),
+    "int64": pa.int64(),
+    "double": pa.float64(),
+    "float": pa.float32(),
+    "string": pa.string(),
+    "binary": pa.binary(),
+    "bool": pa.bool_(),
+    "large_string": pa.large_string(),
+    "null": pa.null(),
+}
+
+
+def _parse_pa_type_str(pa_type_str: str) -> pa.DataType:
+    """Parse a PyArrow type string to preserve inner element types for nested lists."""
+    pa_type_str = pa_type_str.strip()
+    if pa_type_str.startswith("list<item: ") and pa_type_str.endswith(">"):
+        inner = pa_type_str[len("list<item: ") : -1]
+        return pa.list_(_parse_pa_type_str(inner))
+    if pa_type_str in _PA_BASIC_TYPES:
+        return _PA_BASIC_TYPES[pa_type_str]
+    if pa_type_str.startswith("timestamp"):
+        return pa.timestamp("us")
+    logger = logging.getLogger(__name__)
+    logger.warning(
+        "Unrecognized PyArrow type string '%s', falling back to pa.string()",
+        pa_type_str,
+    )
+    return pa.string()
+
+
 def get_pyarrow_schema_from_batch_source(
     config: RepoConfig, batch_source: DataSource, timestamp_unit: str = "us"
 ) -> Tuple[pa.Schema, List[str]]:
@@ -250,15 +283,12 @@ def get_pyarrow_schema_from_batch_source(
     pa_schema = []
     column_names = []
     for column_name, column_type in column_names_and_types:
-        pa_schema.append(
-            (
-                column_name,
-                feast_value_type_to_pa(
-                    batch_source.source_datatype_to_feast_value_type()(column_type),
-                    timestamp_unit=timestamp_unit,
-                ),
-            )
-        )
+        value_type = batch_source.source_datatype_to_feast_value_type()(column_type)
+        if value_type in (ValueType.VALUE_LIST, ValueType.VALUE_SET):
+            pa_type = _parse_pa_type_str(column_type)
+        else:
+            pa_type = feast_value_type_to_pa(value_type, timestamp_unit=timestamp_unit)
+        pa_schema.append((column_name, pa_type))
         column_names.append(column_name)
 
     return pa.schema(pa_schema), column_names
 
@@ -106,6 +106,11 @@ def _proto_value_to_transport_value(proto_value: ValueProto) -> Any:
         if val_attr == "json_list_val":
             return list(getattr(proto_value, val_attr).val)
 
+        # Nested collection types use feast_value_type_to_python_type
+        # which handles recursive conversion of RepeatedValue protos.
+        if val_attr in ("list_val", "set_val"):
+            return feast_value_type_to_python_type(proto_value)
+
         # Map/Struct types are converted to Python dicts by
         # feast_value_type_to_python_type.  Serialise them to JSON strings
         # so the server-side DataFrame gets VARCHAR columns instead of
@@ -204,6 +209,12 @@ def online_read(
             logger.debug("Able to retrieve the online features from feature server.")
             response_json = json.loads(response.text)
             event_ts = self._get_event_ts(response_json)
+            # Build feature name -> ValueType mapping so we can reconstruct
+            # complex types (nested collections, sets, etc.) that cannot be
+            # inferred from raw JSON values alone.
+            feature_type_map: Dict[str, ValueType] = {
+                f.name: f.dtype.to_value_type() for f in table.features
+            }
             # Iterating over results and converting the API results in column format to row format.
             result_tuples: List[
                 Tuple[Optional[datetime], Optional[Dict[str, ValueProto]]]
@@ -223,13 +234,16 @@ def online_read(
                             ]
                             == "PRESENT"
                         ):
+                            feature_value_type = feature_type_map.get(
+                                feature_name, ValueType.UNKNOWN
+                            )
                             message = python_values_to_proto_values(
                                 [
                                     response_json["results"][index]["values"][
                                         feature_value_index
                                     ]
                                 ],
-                                ValueType.UNKNOWN,
+                                feature_value_type,
                             )
                             feature_values_dict[feature_name] = message[0]
                         else:
 
@@ -63,6 +63,12 @@ def to_json_object(printer: _Printer, message: ProtoMessage) -> JsonObject:
         # to JSON. The parse back result will be different from original message.
         if which is None or which == "null_val":
             return None
+        elif which in ("list_val", "set_val"):
+            # Nested collection: RepeatedValue containing Values
+            repeated = getattr(message, which)
+            value = [
+                printer._MessageToJsonObject(inner_val) for inner_val in repeated.val
+            ]
         elif "_list_" in which:
             value = list(getattr(message, which).val)
         else:
@@ -86,6 +92,19 @@ def from_json_object(
             if len(value) == 0:
                 # Clear will mark the struct as modified so it will be created even if there are no values
                 message.int64_list_val.Clear()
+            elif isinstance(value[0], list) or (
+                value[0] is None and any(isinstance(v, list) for v in value)
+            ):
+                # Nested collection (list of lists).
+                # Check any() to handle cases where the first element is None
+                # (empty inner collections round-trip through proto as None).
+                # Default to list_val since JSON transport loses the
+                # outer/inner set distinction.
+                rv = RepeatedValue()
+                for inner in value:
+                    inner_val = rv.val.add()
+                    from_json_object(parser, inner, inner_val)
+                message.list_val.CopyFrom(rv)
             elif isinstance(value[0], bool):
                 message.bool_list_val.val.extend(value)
             elif isinstance(value[0], str):
Original file line number	Diff line number	Diff line change
`@@ -63,6 +63,8 @@ message ValueType {`
`63`	`63`	`TIME_UUID_LIST = 39;`
`64`	`64`	`UUID_SET = 40;`
`65`	`65`	`TIME_UUID_SET = 41;`
	`66`	`+ VALUE_LIST = 42;`
	`67`	`+ VALUE_SET = 43;`
`66`	`68`	`}`
`67`	`69`	`}`
`68`	`70`
`@@ -108,6 +110,8 @@ message Value {`
`108`	`110`	`StringList time_uuid_list_val = 39;`
`109`	`111`	`StringSet uuid_set_val = 40;`
`110`	`112`	`StringSet time_uuid_set_val = 41;`
	`113`	`+ RepeatedValue list_val = 42;`
	`114`	`+ RepeatedValue set_val = 43;`
`111`	`115`	`}`
`112`	`116`	`}`
`113`	`117`