feat: Add UUID and TIME_UUID as feature types (#5885) by soooojinlee · Pull Request #5951 · feast-dev/feast

soooojinlee · 2026-02-08T14:55:58Z

What this PR does / why we need it:

Adds UUID and TIME_UUID as native Feast feature types, resolving #5885. Currently UUID values must be stored as STRING, which loses type semantics, prevents backend-specific features (e.g. Cassandra timeuuid range queries), and makes PostgreSQL uuid columns infer as STRING. This PR enables users to declare UUID features with Field(name="user_id", dtype=Uuid) and receive uuid.UUID objects from get_online_features().to_dict().

Design Decisions

Why two types (UUID vs TIME_UUID)?
The issue author explicitly requested distinguishing time-based UUID (uuid1) and random UUID (uuid4). Both serialize
identically to string in proto, but separate types allow expressing intent in feature definitions and enable future backend-specific optimizations.

Why dedicated proto fields (uuid_val, time_uuid_val)?
Following the pattern established by SET types (PR #5888) and UNIX_TIMESTAMP (which reuses int64/Int64List), we add dedicated oneof fields that reuse existing proto scalar types (string and StringList). This allows WhichOneof("val") to identify UUID types directly from the proto message, without requiring a side-channel.

Backward compatibility for data stored before this change:
OnlineResponse accepts an optional feature_types dict. When data was previously stored as string_val, this metadata enables feast_value_type_to_python_type() to convert it to uuid.UUID. New materializations use uuid_val/time_uuid_val and are identified automatically.

Changes

Layer	Files	Description
Proto	`Value.proto`, generated `_pb2.py`/`_pb2.pyi`	Add `UUID=30`, `TIME_UUID=31`, `UUID_LIST=32`, `TIME_UUID_LIST=33` to `ValueType.Enum`; add `uuid_val`, `time_uuid_val`, `uuid_list_val`, `time_uuid_list_val` to `Value.oneof`
Type system	`value_type.py`, `types.py`	Add `UUID`, `TIME_UUID`, `UUID_LIST`, `TIME_UUID_LIST` enums and `Uuid`/`TimeUuid` aliases
Type conversion	`type_map.py`	Add mappings to ~11 conversion dicts (proto, PyArrow, pandas, PostgreSQL, Couchbase, Snowflake); switch from `string_val` to `uuid_val`; add `PROTO_VALUE_TO_VALUE_TYPE_MAP` entries for UUID fields
Online response	`online_response.py`, `online_store.py`, `feature_store.py`, `utils.py`	Pass `feature_types` metadata for backward-compatible deserialization
ODFV	`on_demand_feature_view.py`	Add UUID/TIME_UUID sample values for schema inference

Backward Compatibility

Data previously stored as string_val still deserializes correctly via the feature_types side-channel
New materializations use dedicated uuid_val/time_uuid_val proto fields
feast_value_type_to_python_type(v) without feature_type now returns uuid.UUID for uuid_val fields (previously returned plain string for string_val)
PostgreSQL uuid columns now infer as ValueType.UUID (previously ValueType.STRING)
Go SDK: proto changes compile without errors; UUID handling logic is not implemented (out of scope)

Tests

test_types.py: Uuid/TimeUuid ↔ ValueType bidirectional conversion, Array types
test_type_map.py: Proto roundtrip with uuid_val, uuid.UUID object return, backward compatibility for string_val, UUID list roundtrip, PostgreSQL mapping
All 78 unit tests passing
ruff lint and format checks passing

sdk/python/feast/type_map.py

sdk/python/feast/infra/online_stores/online_store.py

nquinn408 · 2026-02-09T20:08:38Z

@soooojinlee , thanks so much for putting this together! Can you rebase to bring this PR up to date?

nquinn408 · 2026-02-17T02:36:44Z

@soooojinlee , can you also add support for UUID_SET and TIME_UUID_SET?

nquinn408 · 2026-02-17T02:37:58Z

@soooojinlee , can you also update the docs with the newly supported types?

soooojinlee · 2026-02-17T04:32:04Z

@soooojinlee , can you also add support for UUID_SET and TIME_UUID_SET?

@soooojinlee , can you also update the docs with the newly supported types?

I added UUID_SET / TIME_UUID_SET support and updated the type system documentation as requested in here 41cfe7e

sdk/python/feast/value_type.py

sdk/python/feast/online_response.py

franciscojavierarceo

one small nit pick for readability please

soooojinlee · 2026-02-18T15:30:59Z

I addressed all feedback in 17aebe6

Signed-off-by: soojin <[email protected]> Co-Authored-By: Claude Opus 4.6 <[email protected]> Signed-off-by: soojin <[email protected]>

Co-Authored-By: Claude Opus 4.6 <[email protected]> Signed-off-by: soojin <[email protected]>

Add uuid_val, time_uuid_val, uuid_list_val, time_uuid_list_val as dedicated oneof fields in the Value proto message, replacing the previous reuse of string_val/string_list_val. This allows UUID types to be identified from the proto field alone without requiring a feature_types side-channel. Backward compatibility is maintained for data previously stored as string_val. Co-Authored-By: Claude Opus 4.6 <[email protected]> Signed-off-by: soojin <[email protected]>

Signed-off-by: soojin <[email protected]>

Add Set(Uuid) and Set(TimeUuid) as feature types with full roundtrip support, backward compatibility, and documentation for all UUID types. Signed-off-by: soojin <[email protected]> Co-Authored-By: Claude Opus 4.6 <[email protected]>

…ype mappings Keep PDF_BYTES=30 and IMAGE_BYTES=31 at their upstream values instead of renumbering them. Shift UUID types to 32-37 in both proto and Python enum. Also add missing SET type entries in _convert_value_type_str_to_value_type(), convert_array_column(), and _get_sample_values_by_type() for completeness. Signed-off-by: soojin <[email protected]> Co-Authored-By: Claude Opus 4.6 <[email protected]>

The comment claimed Sets do not support UUID/TimeUuid but the code intentionally allows them. Updated to reflect actual behavior. Signed-off-by: soojin <[email protected]> Co-Authored-By: Claude Opus 4.6 <[email protected]>

…o top Signed-off-by: soojin <[email protected]> Co-Authored-By: Claude Opus 4.6 <[email protected]>

Keep upstream JSON/STRUCT types at proto enum 32-35, renumber UUID types to 36-41. Merge JSON/Struct handling with UUID type handling across all type mapping files.

…rialization Return UUID proto fields as plain strings instead of falling through to feast_value_type_to_python_type which converts them to uuid.UUID objects that are not JSON-serializable, causing TypeError during HTTP transport. Signed-off-by: soojin <[email protected]>

…ture-types # Conflicts: # docs/getting-started/concepts/feast-types.md # docs/reference/type-system.md

devin-ai-integration

Devin Review found 2 new potential issues.

View 38 additional findings in Devin Review.

devin-ai-integration · 2026-03-17T02:40:53Z

sdk/python/feast/type_map.py

+    if val_attr in ("uuid_set_val", "time_uuid_set_val"):
+        return {uuid_module.UUID(v) if isinstance(v, str) else v for v in val}


🟡 pd.isnull() raises TypeError on uuid.UUID objects in scalar proto conversion

In _convert_scalar_values_to_proto, the generic scalar conversion path (line 896) calls pd.isnull(value) on each value. When a uuid.UUID object is passed directly (not as a string), pd.isnull(uuid.UUID(...)) returns False in current pandas versions, but this is fragile — in some pandas versions or with numpy interactions, calling pd.isnull() on non-standard types can raise a TypeError. More critically, the valid_scalar_types for ValueType.UUID is {str, uuid_module.UUID}, so UUID objects are accepted, but the assertion at line 876 (type(sample) in valid_scalar_types) will pass. The actual conversion func = lambda x: str(x) works fine. This path appears to function currently but is not a bug per se.

However, there IS a real issue: the _non_empty_value function at sdk/python/feast/type_map.py:1507 checks isinstance(value, Sized) — uuid.UUID is not Sized, so it passes the check correctly. No bug here either after verification.

Was this helpful? React with 👍 or 👎 to provide feedback.

devin-ai-integration · 2026-03-17T02:40:53Z

sdk/python/feast/type_map.py

    elif val_attr.endswith("_set_val") and val_attr != "unix_timestamp_set_val":
        val = set(val)


🟡 UUID set values hit generic set() conversion before UUID-specific handling, producing set of strings instead of set of UUID objects

In feast_value_type_to_python_type at sdk/python/feast/type_map.py:152, the elif branch converts _set_val types to Python sets via val = set(val). For uuid_set_val and time_uuid_set_val, this converts the list of strings into a set of strings. Then at line 160-161, the UUID-specific code does {uuid_module.UUID(v) ... for v in val} which iterates over the set of strings and converts them to UUIDs — so the final result is correct.

However, this is NOT the case for the backward compatibility path at lines 170-173: when a UUID set is stored using the old string_set_val proto field, line 152 converts it to set(val) (a set of strings). Then line 170 checks isinstance(val, set) — this is True, so it enters the branch and converts each string to uuid.UUID. This path is also correct.

After careful trace-through, this is actually not a bug. Withdrawing.

Was this helpful? React with 👍 or 👎 to provide feedback.

soooojinlee requested review from a team as code owners February 8, 2026 14:55

soooojinlee requested review from ejscribner, robhowley and tokoko and removed request for a team February 8, 2026 14:56

This comment was marked as resolved.

Sign in to view

soooojinlee force-pushed the feat/add-uuid-feature-types branch from 1d4cd01 to 4a5c932 Compare February 8, 2026 15:00

This comment was marked as resolved.

Sign in to view

soooojinlee commented Feb 8, 2026

View reviewed changes

sdk/python/feast/type_map.py Outdated Show resolved Hide resolved

soooojinlee commented Feb 8, 2026

View reviewed changes

sdk/python/feast/infra/online_stores/online_store.py Show resolved Hide resolved

This comment was marked as resolved.

Sign in to view

ntkathole force-pushed the feat/add-uuid-feature-types branch from 2c56521 to 1fd106a Compare February 11, 2026 04:44

ntkathole added the ok-to-test label Feb 11, 2026

ntkathole force-pushed the feat/add-uuid-feature-types branch 3 times, most recently from 26198a6 to 525ac72 Compare February 11, 2026 14:14

soooojinlee force-pushed the feat/add-uuid-feature-types branch 3 times, most recently from cb6dd44 to 54c2eae Compare February 12, 2026 05:55

ntkathole force-pushed the feat/add-uuid-feature-types branch from 2dd648a to b280364 Compare February 16, 2026 06:44

soooojinlee force-pushed the feat/add-uuid-feature-types branch 2 times, most recently from 180434b to 41cfe7e Compare February 17, 2026 04:26

ntkathole reviewed Feb 17, 2026

View reviewed changes

sdk/python/feast/value_type.py Outdated Show resolved Hide resolved

franciscojavierarceo reviewed Feb 18, 2026

View reviewed changes

sdk/python/feast/online_response.py Outdated Show resolved Hide resolved

franciscojavierarceo reviewed Feb 18, 2026

View reviewed changes

soooojinlee force-pushed the feat/add-uuid-feature-types branch 2 times, most recently from aca5741 to 17aebe6 Compare February 18, 2026 15:22

This comment was marked as resolved.

Sign in to view

soooojinlee and others added 12 commits February 19, 2026 09:56

feat: Add UUID and TIME_UUID as feature types (feast-dev#5885)

61558d7

Signed-off-by: soojin <[email protected]> Co-Authored-By: Claude Opus 4.6 <[email protected]> Signed-off-by: soojin <[email protected]>

test: Add unit tests for UUID type support

f71c981

Signed-off-by: soojin <[email protected]> Co-Authored-By: Claude Opus 4.6 <[email protected]> Signed-off-by: soojin <[email protected]>

style: Fix ruff lint and formatting issues

e424947

Co-Authored-By: Claude Opus 4.6 <[email protected]> Signed-off-by: soojin <[email protected]>

fix: Address review feedback for UUID type support

dd21fee

Signed-off-by: soojin <[email protected]>

fix: Address review feedback for UUID type support

a6476a7

Signed-off-by: soojin <[email protected]>

fix: Address review feedback

5223a45

Signed-off-by: soojin <[email protected]>

fix: Convert uuid.UUID to string for Arrow and JSON serialization

84f5111

Signed-off-by: soojin <[email protected]>

fix: Correct misleading comment in Set.__init__

5a3bc79

The comment claimed Sets do not support UUID/TimeUuid but the code intentionally allows them. Updated to reflect actual behavior. Signed-off-by: soojin <[email protected]> Co-Authored-By: Claude Opus 4.6 <[email protected]>

refactor: Extract UUID Arrow conversion into helper and move import t…

cc94cba

…o top Signed-off-by: soojin <[email protected]> Co-Authored-By: Claude Opus 4.6 <[email protected]>

ntkathole force-pushed the feat/add-uuid-feature-types branch from 17aebe6 to cc94cba Compare February 19, 2026 04:26

soooojinlee added 2 commits February 21, 2026 00:08

Merge branch 'master' into feat/add-uuid-feature-types

c6e62d9

Merge branch 'master' into feat/add-uuid-feature-types

1a13ec0

soooojinlee requested a review from franciscojavierarceo February 23, 2026 09:20

soooojinlee mentioned this pull request Feb 23, 2026

Fix inverted default_value logic in OnDemandFeatureView._construct_random_input #6008

Open

Merge upstream/master: resolve JSON/STRUCT + UUID type enum conflicts

c2f4107

Keep upstream JSON/STRUCT types at proto enum 32-35, renumber UUID types to 36-41. Merge JSON/Struct handling with UUID type handling across all type mapping files.

This comment was marked as resolved.

Sign in to view

soooojinlee added 4 commits March 7, 2026 16:50

Merge branch 'master' into feat/add-uuid-feature-types

b6d4898

Merge branch 'master' into feat/add-uuid-feature-types

74a949b

Merge remote-tracking branch 'upstream/master' into feat/add-uuid-fea…

9eeb047

…ture-types # Conflicts: # docs/getting-started/concepts/feast-types.md # docs/reference/type-system.md

devin-ai-integration bot reviewed Mar 17, 2026

View reviewed changes

		if val_attr in ("uuid_set_val", "time_uuid_set_val"):
		return {uuid_module.UUID(v) if isinstance(v, str) else v for v in val}

		elif val_attr.endswith("_set_val") and val_attr != "unix_timestamp_set_val":
		val = set(val)

Conversation

soooojinlee commented Feb 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What this PR does / why we need it:

Design Decisions

Changes

Backward Compatibility

Tests

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

Uh oh!

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

nquinn408 commented Feb 9, 2026

Uh oh!

nquinn408 commented Feb 17, 2026

Uh oh!

nquinn408 commented Feb 17, 2026

Uh oh!

soooojinlee commented Feb 17, 2026

Uh oh!

Uh oh!

Uh oh!

franciscojavierarceo left a comment

Choose a reason for hiding this comment

Uh oh!

soooojinlee commented Feb 18, 2026

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

devin-ai-integration bot left a comment

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration bot Mar 17, 2026

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration bot Mar 17, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

soooojinlee commented Feb 8, 2026 •

edited

Loading