Faulty Remote Resources are accepted by Remote Resolution

[Chitrang Patel]

Expected Behavior

apiVersion: tekton.dev/v1
kind: PipelineRun
metadata:
  name: ref-0001
spec:
  pipelineSpec:
    tasks:
    - name: task1
      taskSpec:
        steps:
        - name: step1
          image: ubuntu
          script: |
            echo "Hello World from a TaskSpec!"
    - name: task2
      taskRef:
        resolver: git
        params:
        - name: url
          value: <git-repo>
        - name: revision
          value: main
        - name: pathInRepo
          value: bad-task.yaml
---
# where bad-task looks like:
apiVersion: tekton.dev/v1
kind: Task
metadata:
  name: bad-task
spec:
  steps:
    - image: ubuntu
       foo: bar

Notice that the field foo is not supposed to exist in the above Task.

At runtime, a validation error is expected when the remotely referenced resource is inspected.
In fact, if we kubectl apply the bad yaml to the cluster, we do get a webhook validation error.

Actual Behavior

The remote resolution discards the bad field foo and runs the rest of it so that error is not caught and the Task runs as if there was nothing wrong.

Issue

We use the universal deserializer to unmarshal the bytes to the pipeline struct.

pipeline/pkg/remote/resolution/resolver.go

Line 100 in 13f45bf

 obj, _, err := scheme.Codecs.UniversalDeserializer().Decode(data, nil, nil) 

The CodecFactory that is used here does not have serializer.EnableStrict as an argument. Hence, the unmarshilling process zaps the bad field.

Solution

The following diff in the resolver code does the magic:

+       serializer "k8s.io/apimachinery/pkg/runtime/serializer"


@@ -97,7 +98,8 @@ func ResolvedRequest(resolved resolutioncommon.ResolvedResource, err error) (run
-       obj, _, err := scheme.Codecs.UniversalDeserializer().Decode(data, nil, nil)
+       codecs := serializer.NewCodecFactory(scheme.Scheme, serializer.EnableStrict)
+       obj, _, err := codecs.UniversalDeserializer().Decode(data, nil, nil)

[Chitrang Patel]

@tektoncd/core-maintainers PTAL here 🙏 I think this is a bug that we should fix for consistent behavior and catch bad resources during remote resolution.