Add support to properly run out of cluster #5109
Conversation
We must be able to find the current namespace. If not is specified/detected we cannot keep going on When running out of cluster, we have no image thus we use the default image in that case, when requested
| @@ -295,7 +295,8 @@ func getOperatorImage(ctx context.Context, c ctrl.Reader) (string, error) { | |||
| ns := platform.GetOperatorNamespace() | |||
| name := platform.GetOperatorPodName() | |||
| if ns == "" || name == "" { | |||
| return "", nil | |||
| // We are most likely running out of cluster. Let's take a chance and use the default value | |||
There was a problem hiding this comment.
I think the previous assumption was correct. I mean, if we're running out of cluster, then, the software binary is not distributed in a container image. If we enable this, then, we would trick the operator to think that it is running from a container, for example, enabling the possibility to run pipeline in a pod (which would instead spin off the kamel builder from a distribution that is different from the running application). IMO, we must return as empty, as also claimed in the comment.
There was a problem hiding this comment.
But currently if you remove this, we cannot run out of cluster as this function is used to set the global platform.Operator which is eventually used, at least, in pkg/controller/integrationkit/build.go
This will result in a build resource that has not ToolImage thus it cannot be run at all.
This is actually one of the reason for the other PR #5110 which would allow to workaround this by explicitly specify a ToolImage regardless if we are running in a pod or not.
There was a problem hiding this comment.
That's the point. The local execution should not allow to run a pipeline but only execute the code that can find "locally". It's because the local operator is running a given binary (typically built from source), whilst the builder would be run in the Pod with potentially a different software (as it is packaged in a container from somewhere else). I think it's wise to keep this design as it to avoid consistency problems.
|
This PR has been automatically marked as stale due to 90 days of inactivity. |
This PR intends to solve some issue encountered while running the operator out of cluster.
Current namespace requirement
We must be able to find the current namespace. If not is specified/detected we cannot keep going on as we will eventually require the namespace to be defined to create the integration plaform objects.
The proposed solution is to not allow both NAMESPACE and WATCH_NAMESPACE to be empty as it creates an issue.
Current operator image detection
The other issue, it that when running out of cluster, we inherently cannot, when requested, detect the current image as we are not in a pod.
The proposed solution is to simply return the default operator image and never return an empty image as it is used later in the operator execution.
Release Note