Stuck in processing forever when using queue()

[jamie0725]

Describe the bug

I have a demo that segments an image with a large model, and for some reason, I have to run it on CPU, making the generation lasts > 60s.

I know I need to use .queue() to keep the connection alive in this situation. But, after enabling the queue, the progress bar is stuck in "processing" forever, even after my function already returns the generated image (as shown in the image below).

Can anyone help me fix the problem I encountered, thanks!

Have you searched existing issues? 🔎

• I have searched and found no existing issues

Reproduction

My code is as below, I use fastapi to serve the demo:

def model_segment_image(np_image, mask_history):
    LogUtil.info("model image segmentation: {}".format(np_image.shape))
    time.sleep(120) # <- minimum reproducible code, as long as it takes longer than 60s to return
    segment_img_path = "some image.png"
    return segment_img_path, mask_history

def get_blocks():
    with gr.Blocks() as demo:
        with gr.Tab():
            with gr.Row():
                mask_history = gr.State([])
                segment_img_input = gr.Image(label="Source Image").style(height=500)
                segment_img_output = gr.Image(label="Segmented Image").style(height=500)
            segment_button = gr.Button("Segment")
            segment_button.click(
                model_segment_image,
                inputs=[segment_img_input, mask_history],
                outputs=[segment_img_output, mask_history],
                queue=True,
            )
    demo.queue()
    return demo

app = FastAPI()
app.include_router(whoami_router.router)
app = gr.mount_gradio_app(app, image_io, path="/v1/image")

Screenshot

As can be seen from the container log above, for some reason the reset API is invoked immediately after 60s despite that the queue is enabled.

Logs

No response

System Info

Google GKE Container

Severity

Blocking usage of gradio

[abidlabs]

Is this using the latest version of gradio?

[jamie0725]

Is this using the latest version of gradio?

we use the gradio version 3.34.0

[abidlabs]

Could you try upgrading to 3.49?

[jamie0725]

Could you try upgrading to 3.49?

Sure, I will try now.

[fivelittlec]

@abidlabs However, when we try to upgrade gradio version to 3.49.0, we encounter Pandas dependency problem that Pandas requires version '3.1.2' or newer of 'jinja2'. The version of jinja2 we currently installed is 3.0.1, if we upgrade to 3.1.2, we would face another error that module 'jinja2' has no attribute 'contextfunction'. Can you give us some advices? Thank you!

[freddyaboulton]

What version of pandas do you have installed? I have 1.5.2 and everything is working for me.

[jamie0725]

Seems like updating gradio solves the problem, so I'm closing the issue for now, thanks~