Docker's logic for resolving credentials from .dockercfg accepts two kinds of matches:
1. an exact match between the dockercfg entry and the image prefix
2. a hostname match between the dockercfg entry and the image prefix
This change implements the latter, which permits the docker client to take .dockercfg entries of the form:
https://quay.io/v1/
and use them for images of the form:
quay.io/foo/bar
even though they are not a prefix-match.
Break up the monolithic volumes code in kubelet into very small individual
modules with a well-defined interface. Move them all into their own packages
and beef up testing along the way.
This is a partial rollback of commit 6e6f465a36 ("Fix a crash for
kubelet when without EtcdClient") in which we used the `reflect` module
to inspect that the pointer stored inside the interface was `nil`, but
as pointed out by @lavalamp, the correct solution is to make the
function return the interface type, in which case a `return nil` will
return the interface nil and not a nil pointer that turns into a non-nil
value when coerced into an interface.
For more details, see http://golang.org/doc/faq#nil_error and the
discussion in PR #3356.
Tested by installing a kubelet built from head with this patch into a
containervm instance and confirming it did not crash on standalone.go.
Confirmed that by only removing the `reflect.IsNil()` comparison but not
changing the return type of `EtcdClientOrDie()` did indeed cause that
same crash, so changing the return type does indeed fix the issue.
Signed-off-by: Filipe Brandenburger <filbranden@google.com>
Added a kubelet config source for watching pods on apiserver.
The pods are converted to boundpods for merging with other
config sources.
The preferred way to create a kubelet is now to pass an apiserver
client but not an etcd client. Changed cmd/integration to use
apiserver to talk to kubelets. And cmd/kubernetes.
Unit, integration, and e2e tests pass, except for a failure of the pd
e2e test which was unrelated.
