[BUG]: Creating the two PipelineRuns makes one or both fail

[adelton]

Details

Describe the bug

When the user just pastes

oc create -f tekton/build-container-image-pipeline/aws-env-real.yaml
oc apply -k tekton/build-container-image-pipeline/
oc create -f tekton/build-container-image-pipeline/build-container-image-pipelinerun-bike-rentals.yaml
oc create -f tekton/build-container-image-pipeline/build-container-image-pipelinerun-tensorflow-housing.yaml

on a fresh namespace, there are high chances that

• build-container-image-bike-rentals-* PipelineRun will fail in git-clone-model-repo with

  + /ko-app/git-init -url=https://github.com/opendatahub-io/ai-edge.git -revision=main -refspec= -path=/workspace/output//model_dir/ -sslVerify=true -submodules=true -depth=1 -sparseCheckoutDirectories=
  {"level":"error","ts":1700162720.543732,"caller":"git/git.go:53","msg":"Error running git [remote add origin https://github.com/opendatahub-io/ai-edge.git]: exit status 128\nfatal: detected dubious ownership in repository at '/workspace/output/model_dir'\nTo add an exception for this directory, call:\n\n\tgit config --global --add safe.directory /workspace/output/model_dir\n","stacktrace":"github.com/tektoncd/pipeline/pkg/git.run\n\t/go/src/github.com/tektoncd/pipeline/pkg/git/git.go:53\ngithub.com/tektoncd/pipeline/pkg/git.Fetch\n\t/go/src/github.com/tektoncd/pipeline/pkg/git/git.go:109\nmain.main\n\t/go/src/github.com/tektoncd/pipeline/cmd/git-init/main.go:53\nruntime.main\n\t/usr/lib/golang/src/runtime/proc.go:250"}
  {"level":"fatal","ts":1700162720.54378,"caller":"git-init/main.go:54","msg":"Error fetching git repository: exit status 128","stacktrace":"main.main\n\t/go/src/github.com/tektoncd/pipeline/cmd/git-init/main.go:54\nruntime.main\n\t/usr/lib/golang/src/runtime/proc.go:250"}
  

• build-container-image-tensorflow-housing-* PipelineRun will fail in build-mlflow-container with

  error evaluating symlinks in build context path: lstat /workspace/source/model_dir/tensorflow-housing: no such file or directory
  

It's likely caused by a race condition over that new static model_dir part of the path introduced by #112 which the two pods are fighting for.

To Reproduce

Paste the commands all in one and then check the PipelineRuns in console.

Alternatively, do

oc create -f tekton/build-container-image-pipeline/aws-env-real.yaml
oc apply -k tekton/build-container-image-pipeline/
oc create -f tekton/build-container-image-pipeline/build-container-image-pipelinerun-tensorflow-housing.yaml

watch the first TaskRun (kserve-download-model) turn green in the console, then paste

oc create -f tekton/build-container-image-pipeline/build-container-image-pipelinerun-bike-rentals.yaml

and observe both the PipelineRuns in the console.

Expected behavior

Running the two PipelineRuns in parallel should still be possible the way it worked before, even if now one is git-based and the other one S3-based.

Screenshots (if applicable)

[LaVLaS]

@adelton The quick fix for this right now is to migrate your PipelineRun to use volumeClaimTemplate instead of a single pre-existing PVC for all Pipelines.

As we combine and refactor the pipelines (#177) to improve the workflow, we can add optional support for a collection of Pipelines to utilize a single PVC to archive data with support for purging older logs

[adelton]

@adelton The quick fix for this right now is to migrate your PipelineRun to use volumeClaimTemplate instead of a single pre-existing PVC for all Pipelines.

Could you do a PR so that we fix the repo content for everyone and folks don't need to do a one-off changes?