MediaFileManager: Missing file on Google App Engine

Hi All 🙂,

I have been going through the process of deploying a Streamlit application to https://radiotherapy.ai using Google App Engine. Once it was deployed it presented an interesting bug that I hadn’t seen before. The matplotlib plots were intermittently not showing up within the interface and the following error message was presented within the App Engine logs:

2021-05-26 05:10:15 default[20210526t145605]  2021-05-26 05:10:15.441 MediaFileManager: Missing file 5c5c1707408633221e9aab6c4c4f9437cdab6da0ed660ef129103e79.png

---

Here is what I saw

---

Here is what was able to make the app able to see if I dragged the slider to other values and then came back a few times:

---

I had a suspicion that this might be due to the new garbage collection process being undergone in 0.82.0:

https://github.com/streamlit/streamlit/blob/7ebd8bf4e54f88e37408d87080cbca521f2587e7/docs/changelog.md#L19-L25

So I downgraded to version 0.81.1, but unfortunately the issue was still present.

Then, wondering if maybe I wasn’t provisioning sufficient RAM I increased the instance from 5 GB up to 16 GB. However unfortunately this still didn’t rectify the issue.

I haven’t yet pointed my domain to the app engine version, so if you’d like to see this issue in action go to https://rtai-autocontouring.ts.r.appspot.com/.

---

I’d be more than happy to go through further debugging if you so desire.

---

I have copied in the code I am using to produce this below, of particular note, I am directly passing the fig object through to st.pyplot, so hopefully shouldn’t be running into the issue of using the global plt.figure() object.

# main.py

fig = _utilities.comparison_display(
    x, y, ct_image, gold_standard_masks, predicted_masks, zoom, windowing
)
st.pyplot(fig)

# _utilities.py

def comparison_display(
    x,
    y,
    ct_image,
    ground_truth_masks,
    predicted_masks=None,
    zoom=1,
    windowing=(None, None),
    figsize=(8, 6),
):
    fig, axs = plt.subplots(nrows=1, ncols=2, figsize=figsize)

    c = _individual_plots(
        axs[0], x, y, ct_image, ground_truth_masks, zoom=zoom, windowing=windowing
    )
    axs[0].set_title("Ground Truth")
    if not predicted_masks is None:
        _individual_plots(
            axs[1], x, y, ct_image, predicted_masks, zoom=zoom, windowing=windowing
        )
        axs[1].set_title("Model")

    label_to_handle_map = {}
    for ax in axs:
        handles, labels = ax.get_legend_handles_labels()
        for handle, label in zip(handles, labels):
            if not label in label_to_handle_map.keys():
                label_to_handle_map[label] = handle

    fig.colorbar(c, ax=axs[:], shrink=0.8, location="bottom")
    labels = list(label_to_handle_map.keys())
    handles = [label_to_handle_map[label] for label in labels]

    if len(labels) != 0:
        fig.legend(
            handles=handles,
            labels=labels,
            loc="upper center",
            bbox_to_anchor=(0.5, 1.05),
            ncol=len(labels),
        )

    return fig

Cheers 🙂, Simon

[charlielito]

I was having the same problem but for visualizing videos in a streamlit app but deployed to Google Cloud Run. I manage to get it working by setting Maximum requests per container to a high value (100) and #CPU=1

[kmcgrady]

Thanks @SimonBiggs! Really appreciate your help here!