example to reproduce issue with hitting limit of pool

[ikondrat]

What has changed and why?

• Fetch coco-10k dataset (you can get it from PR)
• Run make start-coco-10k and set the limit of images per row at least 7
• Scroll
• It would loading stop on some step (check my video)

In the logs you can see

INFO:     ::1:57872 - "GET /api/datasets/6b2ec36a-2498-48b4-91c2-9fbf4eff6857/metadata/info HTTP/1.1" 200 OK
ERROR:asyncio:socket.accept() out of system resource
socket: <asyncio.TransportSocket fd=19, family=AddressFamily.AF_INET6, type=SocketKind.SOCK_STREAM, proto=6, laddr=('::1', 8001, 0, 0)>
Traceback (most recent call last):
  File "/Users/kondrat/.local/share/uv/python/cpython-3.8.20-macos-aarch64-none/lib/python3.8/asyncio/selector_events.py", line 164, in _accept_connection
  File "/Users/kondrat/.local/share/uv/python/cpython-3.8.20-macos-aarch64-none/lib/python3.8/socket.py", line 292, in accept
OSError: [Errno 24] Too many open files

or

sqlalchemy.exc.TimeoutError: QueuePool limit of size 5 overflow 10 reached,

I think it depends on where it hits the limit

UI wise it looks like:

• hitting the limit of pool connections of sql alchemy - will look like non-listing all of the items in the Grid
• Hitting the TCP connections limit (256 default for macos) looks like freesing interface because fastapi starts to timeout requests

[ikondrat]

Screen.Recording.2025-11-27.at.14.45.56.mov