bhetherman/open-webui
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

perf: optional orjson JSON codec behind ENABLE_ORJSON (#27583)

Browse Source 
Swap the JSON encoder/decoder used across the backend from stdlib json to
orjson when ENABLE_ORJSON is set — HTTP request bodies, JSONResponse
bodies, upstream provider responses, SSE chunks, and socket.io/Redis
payloads.

The flag defaults to off, in which case the app uses stdlib json and
engineio's codec verbatim, so default behaviour is unchanged.

- json_codec exports JSONCodec (stdlib json or the orjson codec) and
  SOCKETIO_JSON (engineio's codec or the orjson codec); call sites import
  JSONCodec and stay implementation-agnostic
- apply_orjson_http_json() is a no-op when the flag is off, leaving
  starlette's Request.json / JSONResponse.render untouched
- the orjson codec falls back to the stdlib for inputs orjson rejects
  (non-str dict keys, ints beyond 64 bits, NaN literals)
- orjson is imported only when the flag is on
- FastAPI(default_response_class=...) is deliberately not used: an
  explicit default disables the Pydantic direct-to-bytes fast path for
  response_model routes
This commit is contained in:
Classic298
2026-07-27 03:45:37 -04:00
committed by GitHub
parent 602004dd5f
commit 72fdf238a8
13 changed files with 176 additions and 51 deletions
Download Patch File Download Diff File Expand all files Collapse all files
backend/open_webui/utils/middleware.py
+23 -22
View File
@@ -94,6 +94,7 @@ from open_webui.utils.filter import (
process_filter_functions,
)
from open_webui.utils.json_codec import JSONCodec
from open_webui.utils.mcp.client import MCPClient
from open_webui.utils.memory import add_memory_context, review_memory_after_turn
from open_webui.utils.misc import (
@@ -299,7 +300,7 @@ def get_citation_source_from_tool_result(
try:
try:
tool_result = json.loads(tool_result)
tool_result = JSONCodec.loads(tool_result)
except (json.JSONDecodeError, TypeError):
pass # keep tool_result as-is (e.g. fetch_url returns plain text)
if isinstance(tool_result, dict) and 'error' in tool_result:
@@ -1000,7 +1001,7 @@ async def process_tool_result(
text = item.get('text', '')
if isinstance(text, str):
try:
text = json.loads(text)
text = JSONCodec.loads(text)
except json.JSONDecodeError:
pass
tool_response.append(text)
@@ -1028,7 +1029,7 @@ async def process_tool_result(
text = resource.get('text', '')
if isinstance(text, str) and text:
try:
text = json.loads(text)
text = JSONCodec.loads(text)
except json.JSONDecodeError:
pass
tool_response.append(text)
@@ -1103,7 +1104,7 @@ async def terminal_event_handler(
parsed = tool_result
if isinstance(parsed, str):
try:
parsed = json.loads(parsed)
parsed = JSONCodec.loads(parsed)
except (json.JSONDecodeError, TypeError):
pass
if isinstance(parsed, dict) and parsed.get('exists') is False:
@@ -1141,7 +1142,7 @@ async def chat_completion_tools_handler(
content = None
if hasattr(response, 'body_iterator'):
async for chunk in response.body_iterator:
data = json.loads(chunk.decode('utf-8', 'replace'))
data = JSONCodec.loads(chunk.decode('utf-8', 'replace'))
content = data['choices'][0]['message']['content']
# Cleanup any remaining background tasks if necessary
@@ -1221,7 +1222,7 @@ async def chat_completion_tools_handler(
if not content:
raise Exception('No JSON object found in the response')
result = json.loads(content)
result = JSONCodec.loads(content)
async def tool_call_handler(tool_call):
nonlocal skip_files
@@ -1390,7 +1391,7 @@ async def chat_web_search_handler(request: Request, form_data: dict, extra_param
# user message as the search query.
if isinstance(res, JSONResponse):
try:
error_body = json.loads(res.body)
error_body = JSONCodec.loads(res.body)
detail = error_body.get('detail', 'Query generation failed')
except Exception:
detail = 'Query generation failed'
@@ -1406,7 +1407,7 @@ async def chat_web_search_handler(request: Request, form_data: dict, extra_param
raise Exception('No JSON object found in the response')
response = response[bracket_start:bracket_end]
queries = json.loads(response)
queries = JSONCodec.loads(response)
queries = queries.get('queries', [])
except Exception as e:
queries = [response]
@@ -1735,7 +1736,7 @@ async def chat_image_generation_handler(request: Request, form_data: dict, extra
# Handle JSONResponse from error paths
if isinstance(res, JSONResponse):
try:
error_body = json.loads(res.body)
error_body = JSONCodec.loads(res.body)
detail = error_body.get('detail', 'Image prompt generation failed')
except Exception:
detail = 'Image prompt generation failed'
@@ -1751,7 +1752,7 @@ async def chat_image_generation_handler(request: Request, form_data: dict, extra
raise Exception('No JSON object found in the response')
response = response[bracket_start:bracket_end]
response = json.loads(response)
response = JSONCodec.loads(response)
prompt = response.get('prompt', [])
except Exception as e:
prompt = user_message
@@ -1855,7 +1856,7 @@ async def chat_completion_files_handler(
raise Exception('No JSON object found in the response')
queries_response = queries_response[bracket_start:bracket_end]
queries_response = json.loads(queries_response)
queries_response = JSONCodec.loads(queries_response)
except Exception as e:
queries_response = {'queries': [queries_response]}
@@ -1966,7 +1967,7 @@ def apply_params_to_form_data(form_data, model):
if isinstance(value, str):
try:
# Attempt to parse the string as JSON
custom_params[key] = json.loads(value)
custom_params[key] = JSONCodec.loads(value)
except json.JSONDecodeError:
# If it fails, keep the original string
pass
@@ -1988,7 +1989,7 @@ def apply_params_to_form_data(form_data, model):
logit_bias = convert_logit_bias_input_to_json(params['logit_bias'])
if logit_bias:
form_data['logit_bias'] = json.loads(logit_bias)
form_data['logit_bias'] = JSONCodec.loads(logit_bias)
except Exception as e:
log.exception(f'Error parsing logit_bias: {e}')
@@ -3031,7 +3032,7 @@ def get_response_data(response):
if isinstance(response, JSONResponse):
if isinstance(response.body, bytes):
try:
response_data = json.loads(response.body.decode('utf-8', 'replace'))
response_data = JSONCodec.loads(response.body.decode('utf-8', 'replace'))
except json.JSONDecodeError:
response_data = {'error': {'detail': 'Invalid JSON response'}}
else:
@@ -3090,7 +3091,7 @@ def update_assistant_message_from_stream(assistant_message, raw):
continue
try:
data = json.loads(part)
data = JSONCodec.loads(part)
except Exception:
continue
@@ -3278,7 +3279,7 @@ async def background_tasks_handler(ctx):
]
try:
follow_ups = json.loads(follow_ups_string).get('follow_ups', [])
follow_ups = JSONCodec.loads(follow_ups_string).get('follow_ups', [])
await event_emitter(
{
'type': 'chat:message:follow_ups',
@@ -3336,7 +3337,7 @@ async def background_tasks_handler(ctx):
title_string = title_string[title_string.find('{') : title_string.rfind('}') + 1]
try:
title = json.loads(title_string).get('title', user_message)
title = JSONCodec.loads(title_string).get('title', user_message)
except Exception as e:
title = ''
@@ -3388,7 +3389,7 @@ async def background_tasks_handler(ctx):
tags_string = tags_string[tags_string.find('{') : tags_string.rfind('}') + 1]
try:
tags = json.loads(tags_string).get('tags', [])
tags = JSONCodec.loads(tags_string).get('tags', [])
await Chats.update_chat_tags_by_id(metadata['chat_id'], tags, user)
await event_emitter(
@@ -4212,7 +4213,7 @@ async def streaming_chat_response_handler(response, ctx):
# (without SSE `data:` prefix). Try to normalize these into standard
# error events so frontend and DB paths still receive them.
try:
raw_obj = json.loads(data)
raw_obj = JSONCodec.loads(data)
raw_error = raw_obj.get('error') if isinstance(raw_obj, dict) else None
if raw_error:
if save_to_chat:
@@ -4235,7 +4236,7 @@ async def streaming_chat_response_handler(response, ctx):
data = data[5:].strip()
try:
data = json.loads(data)
data = JSONCodec.loads(data)
if filter_functions:
data, _ = await process_filter_functions(
@@ -4937,7 +4938,7 @@ async def streaming_chat_response_handler(response, ctx):
params = {}
if tool_args and tool_args.strip():
try:
params = json.loads(tool_args)
params = JSONCodec.loads(tool_args)
except Exception:
try:
params = ast.literal_eval(tool_args)
@@ -5613,7 +5614,7 @@ async def streaming_chat_response_handler(response, ctx):
)
if event:
yield wrap_item(json.dumps(event))
yield wrap_item(JSONCodec.dumps(event))
async for data in original_generator:
data, _ = await process_filter_functions(