GeezerVison

The World Wide Web, restored to 800 pixels of glory.

GeezerVison is showing github.com — perf: Replace MessageToDict with optimized custom dict builder by abhijeet-dhumal · Pull Request #6015 · feast-dev/feast

perf: Replace MessageToDict with optimized custom dict builder #6015

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged

franciscojavierarceo merged 12 commits into feast-dev:master from abhijeet-dhumal:fix/custom-json-serialization-6013

May 28, 2026

Uh oh!

There was an error while loading. Please reload this page.

Merged

perf: Replace MessageToDict with optimized custom dict builder #6015

Show file tree

Hide file tree

Changes from all commits

Commits

Show all changes

12 commits

Select commit Hold shift + click to select a range

b28ddf9

perf: Replace MessageToDict with optimized custom dict builder

abhijeet-dhumal Feb 25, 2026

40846ad

Fix test: Use GetOnlineFeaturesResponseMetadata instead of non-existe…

abhijeet-dhumal Mar 2, 2026

ad4882b

fix: Rename to convert_response_to_dict, fix timestamp float-rounding…

abhijeet-dhumal Mar 10, 2026

698d0a1

fix: Return raw bytes to match proto_json.patch() behavior

abhijeet-dhumal Mar 10, 2026

0c0637b

fix: serialize top-level status field in convert_response_to_dict

abhijeet-dhumal May 27, 2026

3ac1755

fix: replace deprecated ORJSONResponse with JSONResponse

abhijeet-dhumal May 27, 2026

5fff274

fix: handle set, list, and map Value types in _value_to_native

abhijeet-dhumal May 27, 2026

b2f2909

fix: warn on scalar_map_val; fix test_unix_timestamp_list_val; docume…

abhijeet-dhumal May 27, 2026

c559ac5

docs: document float_precision=18 trade-off; add double_val round-tri…

abhijeet-dhumal May 27, 2026

388f89b

fix: remove leftover orjson dead code causing mypy no-redef error

abhijeet-dhumal May 27, 2026

ecce3cf

fix: base64-encode bytes_val and bytes_list_val for JSON serialization

abhijeet-dhumal May 27, 2026

25e99b1

style: remove extra blank line in feature_server_utils.py

abhijeet-dhumal May 27, 2026

File filter

Filter by extension .py (3) All 1 file type selected

Viewed files

Clear filters

Conversations

Failed to load comments.

Uh oh!

There was an error while loading. Please reload this page.

Jump to

Jump to file

Failed to load files.

Uh oh!

There was an error while loading. Please reload this page.

Diff view

Unified

Split

Hide whitespace

Diff view

Unified

Split

Hide whitespace

There are no files selected for viewing

16 changes: 5 additions & 11 deletions 16 sdk/python/feast/feature_server.py

Show comments

View file

Open in desktop

Original file line number	Diff line number	Diff line change
Expand Up		@@ -40,7 +40,6 @@
		from fastapi.logger import logger
		from fastapi.responses import JSONResponse
		from fastapi.staticfiles import StaticFiles
		from google.protobuf.json_format import MessageToDict
		from pydantic import BaseModel, field_validator

		import feast
Expand All		@@ -52,6 +51,7 @@
		FeastError,
		)
		from feast.feast_object import FeastObject
		from feast.feature_server_utils import convert_response_to_dict
		from feast.feature_view_utils import get_feature_view_from_feature_store
		from feast.permissions.action import WRITE, AuthzedAction
		from feast.permissions.security_manager import assert_permissions
Expand Down Expand Up		@@ -392,12 +392,9 @@ async def get_online_features(request: GetOnlineFeaturesRequest) -> Any:
		)

		response_dict = await run_in_threadpool(
		MessageToDict,
		response.proto,
		preserving_proto_field_name=True,
		float_precision=18,
		convert_response_to_dict, response.proto
		)
		return response_dict
		return JSONResponse(content=response_dict)

		@app.post(
		"/retrieve-online-documents",
Expand Down Expand Up		@@ -434,12 +431,9 @@ async def retrieve_online_documents(
		)

		response_dict = await run_in_threadpool(
		MessageToDict,
		response.proto,
		preserving_proto_field_name=True,
		float_precision=18,
		convert_response_to_dict, response.proto
		)
		return response_dict
		return JSONResponse(content=response_dict)

		@app.post("/push", dependencies=[Depends(inject_user_details)])
		async def push(request: PushFeaturesRequest) -> Response:
Expand Down

141 changes: 141 additions & 0 deletions 141 sdk/python/feast/feature_server_utils.py

Show comments

View file

Open in desktop

Original file line number	Diff line number	Diff line change
		@@ -0,0 +1,141 @@
		"""Fast serialization utilities for Feature Server responses.

		Matches the output format of MessageToDict with proto_json.patch() applied.
		Values are serialized as native Python types (not wrapped dicts).
		"""

		import base64
		import logging
		from datetime import datetime, timezone
		from typing import Any, Dict, Optional

		from feast.protos.feast.serving.ServingService_pb2 import GetOnlineFeaturesResponse
		from feast.protos.feast.types.Value_pb2 import Value

		logger = logging.getLogger(__name__)

		# FieldStatus enum mapping (protos/feast/serving/ServingService.proto)
		_STATUS_NAMES: Dict[int, str] = {
		0: "INVALID",
		1: "PRESENT",
		2: "NULL_VALUE",
		3: "NOT_FOUND",
		4: "OUTSIDE_MAX_AGE",
		}
Comment thread devin-ai-integration[bot] marked this conversation as resolved. Resolved Uh oh! There was an error while loading. Please reload this page.


		def convert_response_to_dict(response: GetOnlineFeaturesResponse) -> Dict[str, Any]:
		"""Convert GetOnlineFeaturesResponse to a JSON-serializable dict.

		Matches the structure produced by MessageToDict(proto, preserving_proto_field_name=True)
		with proto_json.patch() applied, with one intentional difference:

		- double_val fields are returned as Python float objects (json.dumps uses Python 3.1+
		shortest round-trip form, ~15-17 sig digits) rather than 18 fixed significant digits
		(float_precision=18). Values are numerically identical; only the JSON string length
		may differ. This is safe for all ML feature types and avoids unnecessary precision
		overhead.
		"""
		result: Dict[str, Any] = {
		"results": [
		{
		"values": [_value_to_native(v) for v in feature_vector.values],
		"statuses": [
		_STATUS_NAMES.get(s, "INVALID") for s in feature_vector.statuses
		],
		**(
		{
		"event_timestamps": [
		_timestamp_to_str(ts)
		for ts in feature_vector.event_timestamps
		]
		}
		if feature_vector.event_timestamps
		else {}
		),
		}
		for feature_vector in response.results
		]
		}

		if response.HasField("metadata"):
		result["metadata"] = _metadata_to_dict(response.metadata)

		if response.status:
		result["status"] = response.status

		return result
Comment thread abhijeet-dhumal marked this conversation as resolved. Resolved Uh oh! There was an error while loading. Please reload this page.


		def _value_to_native(v: Value) -> Optional[Any]:
		"""Convert a Value proto to a JSON-serializable Python type.

		bytes_val and bytes_list_val are base64-encoded (RFC 4648) so that
		JSONResponse can serialize them without TypeError. This matches standard
		protobuf JSON encoding for bytes fields and is safe for all HTTP clients.
		"""
		which = v.WhichOneof("val")
		if which is None or which == "null_val":
		return None
		# bytes must be base64-encoded for JSON serialization
		elif which == "bytes_val":
		return base64.b64encode(v.bytes_val).decode("ascii")
		# RepeatedValue — nested Values that must be recursively converted
		elif which in ("list_val", "set_val"):
		return [_value_to_native(nested) for nested in getattr(v, which).val]
		# Map<string, Value> — recursively convert nested Values
		elif which in ("map_val", "struct_val"):
		return {k: _value_to_native(vv) for k, vv in getattr(v, which).val.items()}
		# MapList — list of Map<string, Value>
		elif which in ("map_list_val", "struct_list_val"):
		return [
		{k: _value_to_native(vv) for k, vv in m.val.items()}
		for m in getattr(v, which).val
		]
		# scalar_map_val has non-string keys; full conversion requires extra work and
		# this type is not returned by standard get_online_features paths today.
		elif which == "scalar_map_val":
		logger.warning(
		"scalar_map_val is not yet supported by convert_response_to_dict; value will be None"
		)
		return None
		# bytes_list_val / bytes_set_val — base64-encode each element
		elif which in ("bytes_list_val", "bytes_set_val"):
		return [base64.b64encode(b).decode("ascii") for b in getattr(v, which).val]
		# All other list/set types have scalar .val fields
		elif "_list_" in which or "_set_" in which:
		return list(getattr(v, which).val)
		else:
		return getattr(v, which)
Comment thread Comment on lines +77 to +109 devin-ai-integration Bot May 27, 2026 Copy link Copy Markdown Contributor There was a problem hiding this comment. Choose a reason for hiding this comment The reason will be displayed to describe this comment to others. Learn more. 🔴 `bytes_val` values cause ORJSONResponse to raise TypeError `_value_to_native` at `feature_server_utils.py:68` returns raw `bytes` for `bytes_val` fields. Even if `orjson` were installed, `orjson.dumps()` cannot serialize Python `bytes` objects and raises `TypeError: Type is not JSON serializable: bytes`. The old code path returned the dict to FastAPI which processed it through `jsonable_encoder` (capable of handling bytes). The switch to `ORJSONResponse` bypasses that encoder, causing a regression for any response containing bytes-typed feature values. Prompt for agents `In _value_to_native (feature_server_utils.py), when the value type is bytes_val, the raw bytes object is returned. orjson cannot serialize bytes. You need to either: (1) base64-encode bytes values before returning them (e.g. using base64.b64encode(val).decode('ascii')), or (2) not use ORJSONResponse and instead use a serializer that handles bytes. Note that bytes_list_val (handled by the _list_ branch) would also produce a list of bytes objects with the same issue.` Was this helpful? React with 👍 or 👎 to provide feedback. Sorry, something went wrong. Uh oh! There was an error while loading. Please reload this page. All reactions ntkathole May 27, 2026 Copy link Copy Markdown Member There was a problem hiding this comment. Choose a reason for hiding this comment The reason will be displayed to describe this comment to others. Learn more. @abhijeet-dhumal bytes_val is still problem since you are using JSONResponse Sorry, something went wrong. Uh oh! There was an error while loading. Please reload this page. All reactions abhijeet-dhumal May 27, 2026 Copy link Copy Markdown Contributor Author There was a problem hiding this comment. Choose a reason for hiding this comment The reason will be displayed to describe this comment to others. Learn more. Thanks for catching this, Nikhil! Now bytes_val and bytes_list_val now base64-encode the output before returning, so JSONResponse / json.dumps can serialize them without TypeError. Sorry, something went wrong. Uh oh! There was an error while loading. Please reload this page. All reactions abhijeet-dhumal May 27, 2026 Copy link Copy Markdown Contributor Author There was a problem hiding this comment. Choose a reason for hiding this comment The reason will be displayed to describe this comment to others. Learn more. Please review! Sorry, something went wrong. Uh oh! There was an error while loading. Please reload this page. All reactions


		def _timestamp_to_str(ts) -> str:
		"""Convert protobuf Timestamp to RFC 3339 format with Z suffix.

		Uses adaptive precision to match MessageToDict output:
		- No fractional seconds when nanos == 0
		- 3 digits (milliseconds) when nanos % 1_000_000 == 0
		- 6 digits (microseconds) when nanos % 1_000 == 0
		- 9 digits (nanoseconds) otherwise
		"""
		if ts.seconds == 0 and ts.nanos == 0:
		return "1970-01-01T00:00:00Z"
		dt = datetime.fromtimestamp(ts.seconds, tz=timezone.utc)
		base = dt.strftime("%Y-%m-%dT%H:%M:%S")
		nanos = ts.nanos
		if nanos == 0:
		return base + "Z"
		elif nanos % 1_000_000 == 0:
		return base + ".%03dZ" % (nanos // 1_000_000)
		elif nanos % 1_000 == 0:
		return base + ".%06dZ" % (nanos // 1_000)
		else:
		return base + ".%09dZ" % nanos


		def _metadata_to_dict(metadata) -> Dict[str, Any]:
		"""Convert FeatureResponseMeta to dict (matches proto_json.patch() format)."""
		result: Dict[str, Any] = {}
		if metadata.HasField("feature_names"):
		result["feature_names"] = list(metadata.feature_names.val)
		return result

Oops, something went wrong.

Uh oh!

There was an error while loading. Please reload this page.

Oops, something went wrong.

Uh oh!

There was an error while loading. Please reload this page.

Add this suggestion to a batch that can be applied as a single commit. This suggestion is invalid because no changes were made to the code. Suggestions cannot be applied while the pull request is closed. Suggestions cannot be applied while viewing a subset of changes. Only one suggestion per line can be applied in a batch. Add this suggestion to a batch that can be applied as a single commit. Applying suggestions on deleted lines is not supported. You must change the existing code in this line in order to create a valid suggestion. Outdated suggestions cannot be applied. This suggestion has been applied or marked resolved. Suggestions cannot be applied from pending reviews. Suggestions cannot be applied on multi-line comments. Suggestions cannot be applied while the pull request is queued to merge. Suggestion cannot be applied right now. Please check back later.

You are visitor 01402503 — GeezerVison does not store the pages you look up.
Best viewed with Netscape Navigator 4.7 or Internet Explorer 6 at 800×600. No cookies required.

GeezerVison

perf: Replace MessageToDict with optimized custom dict builder #6015

Uh oh!

perf: Replace MessageToDict with optimized custom dict builder #6015

Uh oh!

Uh oh!

Diff view

Diff view

There are no files selected for viewing

Uh oh!

Uh oh!

devin-ai-integration Bot May 27, 2026

Choose a reason for hiding this comment

Uh oh!

ntkathole May 27, 2026

Choose a reason for hiding this comment

Uh oh!

abhijeet-dhumal May 27, 2026

Choose a reason for hiding this comment

Uh oh!

abhijeet-dhumal May 27, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!