Timeout pushing image for a Web API container app

[ciwchris]

When trying to deploy a Web API as a Azure Container App I receive a timeout error pushing the image.

error CONTAINER1013: Failed to push to the output registry: The request was canceled due to the configured HttpClient.Timeout of 100 seconds elapsing.

If I trim the project, set PublishTrimmed to true, then I can get it to deploy most of the time. But trimming the app breaks the app, in it's current form, and trimming isn't officially supported for this type of app.

Are there other workarounds for pushing the image without timing out?

[davidfowl]

cc @baronfel

[baronfel]

We don't currently have a way to set this timeout with the container tooling. I've logged dotnet/sdk-container-builds#543 to track adding some kind of way to work around this. In the interim as a workaround I'd suggest pushing to a local docker daemon and then using docker push to have Docker push the image to your destination registry.

[ciwchris]

Thank you for getting it routed to the right place.

Okay, thank you. This project is just a learning experiment, so yeah, for now I push the image myself.

[tmds]

If I trim the project, set PublishTrimmed to true, then I can get it to deploy most of the time. But trimming the app breaks the app, in it's current form, and dotnet/aspnetcore#44997 (comment).

It sounds like the timeout is hit because what gets uploaded is "large", that is: size / 100 s > bandwidth.

Are there other workarounds for pushing the image without timing out?

You could try to force the upload to chunked by setting this envvar SDK_CONTAINER_DEBUG_REGISTRY_FORCE_CHUNKED_UPLOAD=true.

The chunks will only be 64kB which might cause a slow upload.
You can increase the size by setting SDK_CONTAINER_REGISTRY_CHUNKED_UPLOAD_SIZE_BYTES.

[tmds]

cc @ellismg

[baronfel]

It would also help if azd created binlogs for these publishes - we log trace-level diagnostics about the push to the binlog, and that would help us know if our intended chunk-based upload fallback is happening or not. We intended that if the single atomic push fails for any reason (including timeouts) the SDK should fallback to uploading chunks in the way that @tmds has described.