Describe the bug
In a Kubernetes (k8s) environment, the unload event of Gradio Blocks is triggered with a long delay (e.g., ~30 minutes) after the client disconnects or the interface is closed. This occurs consistently with the latest versions of Gradio 4, 5, and 6. The issue does not appear when running the same application locally (outside k8s).
Temporarily, setting the heartbeat_interval to a smaller value (e.g., 1 second) resolves the problem – the unload event then fires as expected within a short time.
Code:
from datetime import datetime
import gradio as gr
def on_load():
print(f"load: {datetime.now()}")
def on_unload():
print(f"unload: {datetime.now()}")
with gr.Blocks() as demo:
demo.load(on_load)
demo.unload(on_unload)
if __name__ == "__main__":
demo.launch(server_name="0.0.0.0", server_port=9527)Output (the browser tab is closed immediately after the page is loaded):
load: 2026-04-30 15:10:20.650028
unload: 2026-04-30 15:26:35.641765
Setting GRADIO_IS_E2E_TEST environment variable to change heartbeat interval from 15s to 0.25s can fix this problem. Setting heartbeat interval to 1s can also resolve the issue.
Have you searched existing issues? 🔎
Reproduction
from datetime import datetime
import gradio as gr
def on_load():
print(f"load: {datetime.now()}")
def on_unload():
print(f"unload: {datetime.now()}")
with gr.Blocks() as demo:
demo.load(on_load)
demo.unload(on_unload)
if __name__ == "__main__":
demo.launch(server_name="0.0.0.0", server_port=9527)Screenshot
No response
Logs
System Info
Gradio Environment Information:
------------------------------
Operating System: Linux
gradio version: 5.49.1
gradio_client version: 1.13.3
------------------------------------------------
gradio dependencies in your environment:
aiofiles: 24.1.0
anyio: 4.13.0
audioop-lts is not installed.
brotli: 1.2.0
fastapi: 0.136.1
ffmpy: 1.0.0
gradio-client: 1.13.3
groovy: 0.1.2
httpx: 0.28.1
huggingface-hub: 1.11.0
jinja2: 3.1.6
markupsafe: 3.0.3
numpy: 2.4.4
orjson: 3.11.8
packaging: 26.1
pandas: 2.3.3
pillow: 11.3.0
pydantic: 2.11.10
pydub: 0.25.1
python-multipart: 0.0.26
pyyaml: 6.0.3
ruff: 0.15.11
safehttpx: 0.1.7
semantic-version: 2.10.0
starlette: 0.52.1
tomlkit: 0.13.3
typer: 0.24.2
typing-extensions: 4.15.0
uvicorn: 0.46.0
mcp is not installed.
pydantic: 2.11.10
authlib is not installed.
itsdangerous is not installed.
gradio_client dependencies in your environment:
fsspec: 2026.3.0
httpx: 0.28.1
huggingface-hub: 1.11.0
packaging: 26.1
typing-extensions: 4.15.0
websockets: 15.0.1
Severity
Blocking usage of gradio
Describe the bug
In a Kubernetes (k8s) environment, the
unloadevent of Gradio Blocks is triggered with a long delay (e.g., ~30 minutes) after the client disconnects or the interface is closed. This occurs consistently with the latest versions of Gradio 4, 5, and 6. The issue does not appear when running the same application locally (outside k8s).Temporarily, setting the heartbeat_interval to a smaller value (e.g., 1 second) resolves the problem – the unload event then fires as expected within a short time.
Code:
Output (the browser tab is closed immediately after the page is loaded):
Setting
GRADIO_IS_E2E_TESTenvironment variable to change heartbeat interval from 15s to 0.25s can fix this problem. Setting heartbeat interval to 1s can also resolve the issue.Have you searched existing issues? 🔎
Reproduction
Screenshot
No response
Logs
System Info
Severity
Blocking usage of gradio