This removes the need for the core context code to import
`github.com/docker/cli/cli/context/kubernetes` which in turn reduces the
transitive import tree in this file to not pull in all of Kubernetes.
Note that this means that any calling code which is interested in the
kubernetes endpoint must import `github.com/docker/cli/cli/context/kubernetes`
itself somewhere in order to trigger the dynamic registration. In practice
anything which is interested in Kubernetes must import that package (e.g.
`./cli/command/context.list` does for the `EndpointFromContext` function) to do
anything useful, so this restriction is not too onerous.
As a special case a small amount of Kubernetes related logic remains in
`ResolveDefaultContext` to handle error handling when the stack orchestrator
includes Kubernetes. In order to avoid a circular import loop this hardcodes
the kube endpoint name.
Similarly to avoid an import loop the existing `TestDefaultContextInitializer`
cannot continue to unit test for the Kubernetes case, so that aspect of the
test is carved off into a very similar test in the kubernetes context package.
Lastly, note that the kubernetes endpoint is now modifiable via
`WithContextEndpointType`.
Signed-off-by: Ian Campbell <ijc@docker.com>
These are needed by any dynamically registered (via
`RegisterDefaultStoreEndpoints`) endpoint type to write a useful/sensible unit
test.
Signed-off-by: Ian Campbell <ijc@docker.com>
Previously an endpoint registered using `RegisterDefaultStoreEndpoints` would
not be taken into consideration by `resolveDefaultContext` and so could not
provide any details.
Resolve this by passing a `store.Config` to `resolveDefaultContext` and using
it to iterate over all registered endpoints. Any endpoint can ensure that their
type implements the new `EndpointDefaultResolver` in order to provide a default.
The Docker and Kubernetes endpoints are special cased, shortly the Kubernetes
one will be refactored to be dynamically registered.
Signed-off-by: Ian Campbell <ijc@docker.com>
I'm about to refactor the code which includes the Kubernetes support in a way
which relies on something vendoring `./cli/context/kubernetes/` in order to
trigger the inclusion of support for the Kubernetes endpoint in the final
binary.
In practice anything which is interested in Kubernetes must import that package
(e.g. `./cli/command/context.list` does for the `EndpointFromContext`
function). However if it was somehow possible to build without that import then
the `KUBERNETES ENDPOINT` column would be mysteriously empty.
Out of an abundance of caution add a specific check on the final binary.
Signed-off-by: Ian Campbell <ijc@docker.com>
This reverts commit 59defcb34d which caused #1892
since the timeout applied not only to the dial phase but to everything, so it
would kill `docker logs -f ...` if the container was not chatty enough.
Signed-off-by: Ian Campbell <ijc@docker.com>
The comment on `github.com/docker/cli/kubernetes.NewKubernetesConfig` said:
// Deprecated: Use github.com/docker/compose-on-kubernetes/api.NewKubernetesConfig instead
By making this switch in `github.com/docker/cli/context/kubernetes/load.go` we
break a vendoring chain:
`github.com/docker/cli/cli/command`
→ `vendor/github.com/docker/cli/cli/context/kubernetes/load.go`
→ `vendor/github.com/docker/cli/kubernetes`
→ `github.com/docker/compose-on-kubernetes/api/compose/...`
This means that projects which just want `github.com/docker/cli/cli/command`
(which is itself pulled in transitively by
`github.com/docker/cli/cli-plugins/plugin`) which do not themselves need the
compose-on-kubernetes API avoid a huge pile of transitive dependencies.
On one of my private projects the diff on the vendor dir is:
280 files changed, 21 insertions(+), 211346 deletions(-)
and includes dropping:
* `github.com/docker/compose-on-kubernetes/api/compose/{clone,impersonation}`
* `github.com/docker/compose-on-kubernetes/api/compose/{v1alpha3,v1beta1,v1beta2,v1beta3}`
* `github.com/google/btree`
* `github.com/googleapis/gnostic`
* `github.com/gregjones/httpcache`
* `github.com/peterbourgon/diskv`
* `k8s.io/api/*` (_lots_ of subpackages)
* `k8s.io/client-go/{discovery,kubernetes/scheme}`
and I've gone from:
$ du -sh vendor/k8s.io/
8.1M vendor/k8s.io/
to:
$ du -sh vendor/k8s.io/
2.1M vendor/k8s.io/
(overall I went from 36M → 29M of vendor dir for this particular project)
The change to `cli/command/system/version.go` is just for consistency and
allows us to drop the now unused alias.
Signed-off-by: Ian Campbell <ijc@docker.com>
Refactor `RunCreate` slightly so that all three paths always produce the same
output, namely the name of the new context of `stdout` (for scripting) and the
success log message on `stderr`.
Validate by extending the existing unit tests to always check the output is as
expected.
Signed-off-by: Ian Campbell <ijc@docker.com>
This partially mitigates #1739 ("Docker commands take 1 minute to timeout if
context endpoint is unreachable") and is a simpler alternative to #1747 (which
completely defers the client connection until an actual call is attempted).
Note that the previous 60s delay was the culmination of two separate 30s
timeouts since the ping is tried twice. This with this patch the overall
timeout is 20s. https://github.com/moby/moby/pull/39206 will remove the second
ping and once that propagates to this tree the timeout will be 10s.
Signed-off-by: Ian Campbell <ijc@docker.com>
Ian Campbell
Sebastiaan van Stijn
Silvin Lubecki
Vincent Demeester
Sebastiaan van Stijn
Tibor Vass
Tibor Vass
Harald Albers