Improve image build waiting mechanism

[pditommaso]

Wave can build container on-fly when the container request provides a dockerfile.

https://github.com/seqeralabs/wave/blob/2b24fb330bab6220870141140f62164ed7404e4e/wave-api/src/main/groovy/io/seqera/wave/api/SubmitContainerTokenRequest.groovy#L30-L30

However, the build process can take some minutes. Therefore subsequence image requests need to wait for the build completion before starting to pull the corresponding layers.

Now it's uses a very simple blicking mechanism in the controller handler.

https://github.com/seqeralabs/wave/blob/2b24fb330bab6220870141140f62164ed7404e4e/src/main/groovy/io/seqera/wave/controller/RegistryProxyController.groovy#L62-L63

But this can be very inefficient. Likely a better approach consistent using reactive response.

Also the build response status should be taken into account an return an error when the build fails

[jorgeaguileraseqera]

Seems Docker inspects the Retry-After header if a 50x error is returned but I'll confirm this error with pull is solved:

https://forums.docker.com/t/force-docker-client-to-retry-pulling-layers/31947

by other hand, I think the best approach is to create a background task that performs the build when the dockerfile is provided and store the result in some storage. When the docker client request again we can check if it's present in the storage and serve it

[pditommaso]

by other hand, I think the best approach is to create a background task that performs the build when the dockerfile is provided and store the result in some storage

this is how it works! look here

https://github.com/seqeralabs/wave/blob/26a9fe70b3cb2806b8e96dcb5b5a69a265292b0f/src/main/groovy/io/seqera/wave/service/builder/ContainerBuildServiceImpl.groovy#L183-L183

[jorgeaguileraseqera]

I prefer to use the background task feature of Micronaut (https://guides.micronaut.io/latest/micronaut-scheduled-gradle-groovy.html#scheduling-a-job-manually)

[pditommaso]

Dont' think impact much, it's just another way to execute the task in a separate thread.