Need to support docker registries that store v1 and v2 images

[stuart-warren]

Some registries appear to store images that are only accessible via the same registry api version that they were pushed through.

for example:
docker2aci docker://gcr.io/google_containers/kubedns-amd64:1.7 # works!
docker2aci docker://gcr.io/google_containers/skydns:2015-10-13-8c72f8c # 404

Because the code currently assumes that a registry only supports one api version and v2 is checked first, if the registry supports v2 then any v1 images are not accessible.

[jonboulle]

@stuart-warren I think this is fixed via #214 - are you using master?

[stuart-warren]

@jonboulle I'm pretty sure I am yes.
Does docker2aci docker://gcr.io/google_containers/skydns:2015-10-13-8c72f8c work for you with master?

[jonboulle]

Nope, it does not :-(

[stuart-warren]

Pretty sure the issue is this if statement:
https://github.com/appc/docker2aci/blob/master/lib/internal/backend/repository/repository.go#L88

    if supportsV2 {
        return rb.getImageInfoV2(dockerURL)
    } else {
        URLSchema, supportsV1, err := rb.supportsRegistry(dockerURL.IndexURL, registryV1)
        if err != nil {
            return nil, nil, err
        }
        if !supportsV1 {
            return nil, nil, fmt.Errorf("registry doesn't support API v2 nor v1")
        }
        rb.schema = URLSchema + "://"
        return rb.getImageInfoV1(dockerURL)
    }

Will propose a fix

[lucab]

I guess this is a part of #216.

[lucab]

/cc @dgonyeo @s-urbaniak

[stuart-warren]

similar issues here:

docker2aci/lib/internal/backend/repository/repository.go

Line 104 in cd3f03b

if rb.hostsV2Support[dockerURL.IndexURL] {

[s-urbaniak]

It is a regression introduced in #220, as the above command works in 69aa7db.

[stuart-warren]

@s-urbaniak agreed 😞 , however docker2aci was not correctly implementing the spec for v2 registries.
In seems that dockerhub and quay.io are a bit more helpful in their implementations and redirect clients from /v2 to /v2/

[lucab]

This is an unfortunate situation, but I think that #220 goes in the right direction and we should continue here. Current docker client behavior is to re-try with v1 if a v2 operation fails. As much as I don't like automatic protocol downgrades, it looks like we should implement this in order not to break registries not doing v1<->v2 mirroring.

In order to untangle this from the bigger scope of #216, I'd suggest to assume that current docker2aci mode is auto (ie. registry-v2,registry-v1) and implement this behavior as "probe v2, try v2, downgrade-on-failure, probe v1, try v1, hardfail-on-failure".

@stuart-warren we are a bit constrained at the moment, so unfortunately I don't think this would be tackled in the upcoming rkt release cycle. Would you be interested in addressing just this out of #216?

[stuart-warren]

@lucab yeah I can give it a go.
Just to confirm, I'm to shuffle the code around to do the down grade automatically, but don't need to change the cli or any other tasks in that ticket?