Builds are freezing on ‘preparing build cache for export’·

[nihil-admirari]

E.g. https://github.com/yt-dlp/FFmpeg-Builds/actions/runs/17149050620 and multiple builds before

2025-08-22T08:23:44.3931579Z #161 exporting cache to registry
2025-08-22T08:23:44.3932129Z #161 preparing build cache for export
2025-08-22T13:36:23.6663281Z ##[error]The operation was canceled.

Only win32-gpl and win64-gpl-shared target-variants are freezing, others never do.

Most likely related to moby/buildkit#6008. Proposed workaround: downgrade Docker to 0.22.

[BtbN]

You'll need to delete the :cache image for the affected jobs and re-run them.

[nihil-admirari]

It helped. Thanks.

[BtbN]

I kinda wanna leave this open, since it's happening locally to me occasionally, and right now in mass on CI, much worse than in the past.
Not sure if there is any reasonable way to fix it, other than deleting the :cache tags whenever it happens. But it's annoying as hell and inflates build times a ton.

[nihil-admirari]

Now it fails with

2025-08-30T13:55:55.8811166Z #8 importing cache manifest from ghcr.io/yt-dlp/ffmpeg-builds/win32-gpl-7.1:cache
2025-08-30T13:55:56.0945667Z #8 inferred cache manifest type: application/vnd.oci.image.manifest.v1+json done
2025-08-30T13:55:57.0978700Z ERROR: failed to build: failed to receive status: rpc error: code = Unavailable desc = error reading from server: EOF

or

2025-08-30T14:32:22.0282463Z #223 exporting cache to registry
2025-08-30T14:32:22.0282761Z #223 preparing build cache for export
2025-08-30T14:32:26.8106506Z ERROR: failed to build: failed to receive status: rpc error: code = Unavailable desc = error reading from server: EOF

[BtbN]

Yeah, see latest commit.
I guess an immediate error beats hour-long spinning though, so I kept it.

[gamer191]

For anyone wondering, the specific crash is being tracked in moby/buildkit#6129 PR comments (this repo uses that PR because vanilla Docker gets stuck in endless loops)

[gamer191]

@BtbN that PR got a few commits today which may or may not fix it. Might be worth updating https://github.com/BtbN/buildkit

[BtbN]

The image the builds are using are currently using my own PR that attempts to fix the issue, since the other one was failing way too frequently for no apparent reason.

[gamer191]

Huh? Is the image not the same as the code at https://github.com/BtbN/buildkit

[BtbN]

It's the code at https://github.com/BtbN/buildkit/tree/loop_detection_fix
It's not a CI built image, I couldn't get that to work, I just push it manually. I don't intend to keep using a self-built image forever.
So far my impression of the fix is good, haven't noticed any hour long cache-preparations or fails.

Edit: I do intend to switch it back over to the other PR branch soon though, to test it more after its recent changes.