Deleting first pipeline version of a pipeline and attempting pipeline run crashes ml-pipeline

[lyschoening]

What steps did you take:

There may be multiple bugs here.

1. I ran a pipeline on a specific pipeline version where there were multiple versions and the first pipeline version name had the same name as the pipeline, using the SDK with a code snippet looking roughly like this:

client = kfp.Client()
experiment = client.create_experiment(EXPERIMENT_NAME)
pipeline = client.get_pipeline(client.get_pipeline_id(PIPELINE_NAME))
client.run_pipeline(
            experiment_id=experiment.id,
            job_name='test run',
            params={},
            pipeline_id=pipeline.id,
            version_id=pipeline.default_version.id)

2. According to the pipeline UI the run was done with the correct specific pipeline version, however the Argo workflow used the template of the first pipeline version.
3. To debug, I deleted the first version of the pipeline via the UI.
4. I attempted to run the pipeline again, using the same code as before.

What happened:

When calling client.run_pipeline() the ml-pipeline pod appears to crash instantly. This happens every time. Via the SDK I get the following error message:

ApiException: (503)
Reason: Service Unavailable
HTTP response headers: HTTPHeaderDict({'Audit-Id': 'ea84b152-4123-429f-aa15-b5d8adff5b8d', 'Date': 'Wed, 19 Aug 2020 07:15:10 GMT', 'Transfer-Encoding': 'chunked'})
HTTP response body: Error: 'EOF'
Trying to reach: 'http://10.24.0.35:8888/apis/v1beta1/runs'

The pod fails with the following error:

panic: runtime error: invalid memory address or nil pointer dereference
[signal SIGSEGV: segmentation violation code=0x1 addr=0x138 pc=0x11b8766]

goroutine 110 [running]:
github.com/kubeflow/pipelines/backend/src/common/util.(*Workflow).VerifyParameters(0xc000907cf8, 0xc000553740, 0x0, 0xc000553740)
	backend/src/common/util/workflow.go:66 +0x96
github.com/kubeflow/pipelines/backend/src/apiserver/resource.(*ResourceManager).CreateRun(0xc00024c540, 0xc000914b60, 0xc0008e64e0, 0xc000326140, 0x2)
	backend/src/apiserver/resource/resource_manager.go:269 +0x227
github.com/kubeflow/pipelines/backend/src/apiserver/server.(*RunServer).CreateRun(0xc000904010, 0x1941ba0, 0xc0008e64e0, 0xc0008e6510, 0xc000904010, 0x1, 0x1)
	backend/src/apiserver/server/run_server.go:43 +0xc5
github.com/kubeflow/pipelines/backend/api/go_client._RunService_CreateRun_Handler.func1(0x1941ba0, 0xc0008e64e0, 0x1696ce0, 0xc0008e6510, 0x1, 0x0, 0xc000072700, 0x0)
	bazel-out/k8-opt/bin/backend/api/linux_amd64_stripped/go_client_go_proto%/github.com/kubeflow/pipelines/backend/api/go_client/run.pb.go:1269 +0x86
main.apiServerInterceptor(0x1941ba0, 0xc0008e64e0, 0x1696ce0, 0xc0008e6510, 0xc0002784c0, 0xc000278500, 0x15a5ec0, 0x25ad330, 0x171cca0, 0xc000188e00)
	backend/src/apiserver/interceptor.go:30 +0xf4
github.com/kubeflow/pipelines/backend/api/go_client._RunService_CreateRun_Handler(0x16c95e0, 0xc000904010, 0x1941ba0, 0xc0008e64e0, 0xc000185810, 0x18060e0, 0x0, 0x0, 0xc0008b4140, 0x91)
	bazel-out/k8-opt/bin/backend/api/linux_amd64_stripped/go_client_go_proto%/github.com/kubeflow/pipelines/backend/api/go_client/run.pb.go:1271 +0x158
google.golang.org/grpc.(*Server).processUnaryRPC(0xc00090c000, 0x194ea40, 0xc00090c180, 0xc000188e00, 0xc0008cc270, 0x2527f80, 0x0, 0x0, 0x0)
	external/org_golang_google_grpc/server.go:966 +0x4a2
google.golang.org/grpc.(*Server).handleStream(0xc00090c000, 0x194ea40, 0xc00090c180, 0xc000188e00, 0x0)
	external/org_golang_google_grpc/server.go:1245 +0xd61
google.golang.org/grpc.(*Server).serveStreams.func1.1(0xc00046e070, 0xc00090c000, 0x194ea40, 0xc00090c180, 0xc000188e00)
	external/org_golang_google_grpc/server.go:685 +0x9f
created by google.golang.org/grpc.(*Server).serveStreams.func1
	external/org_golang_google_grpc/server.go:683 +0xa1

What did you expect to happen:

1. Running a pipeline should use the pipeline definition of the default pipeline version.
2. Running a pipeline with a specific pipeline version should use the pipeline definition of that version.
3. Deleting an older pipeline version should not cause future pipeline runs to crash ml-pipeline.

Environment:

How did you deploy Kubeflow Pipelines (KFP)?

KFP version: 1.0.0

KFP SDK version: master branch

Anything else you would like to add:

[Miscellaneous information that will assist in solving the issue.]

/kind bug