phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2026-01-02 04:44:31 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

fix: Call pandas.read_* in a seperate thread

Browse source 
These block on io reads which in turn block the
server. Move them to their own thread.

Closes: #1697
Signed-off-by: Derek Higgins <derekh@redhat.com>
This commit is contained in:
Derek Higgins 2025-03-19 14:58:32 +00:00
parent 7c0448456e
commit e48e36a136
2 changed files with 11 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

10
llama_stack/providers/utils/datasetio/url_utils.py
View file

@ -4,6 +4,7 @@
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
import asyncio
import base64
import io
from urllib.parse import unquote
@ -13,12 +14,15 @@ import pandas
from llama_stack.providers.utils.memory.vector_store import parse_data_url
def get_dataframe_from_uri(uri: str):
async def get_dataframe_from_uri(uri: str):
df = None
if uri.endswith(".csv"):
df = pandas.read_csv(uri)
# Moving to its own thread to avoid io from blocking the eventloop
# This isn't ideal as it moves more then just the IO to a new thread
# but it is as close as we can easly get
df = await asyncio.to_thread(pandas.read_csv, uri)
elif uri.endswith(".xlsx"):
df = pandas.read_excel(uri)
df = await asyncio.to_thread(pandas.read_excel, uri)
elif uri.startswith("data:"):
parts = parse_data_url(uri)
data = parts["data"]