quay.io/operator-framework/olm for v0.18.0 cannot start container "util" due to missing /bin/cp

[zhaminglei]

What did you do?
Installed OLM v0.18.0
created a catalogsource with an index built by me
created a subscription to the catalogsource

What did you expect to see?
The flow is able to parse my index and moving forward, generate Installplan etc..

What did you see instead? Under which circumstances?
Job failed due to following issue:

Successfully assigned blah/6d587b34c86ea704dce32c7abc361c7cf3a078db7d3e86673e7406d3f88nwzv to minikube
  Normal   Pulling    22s                kubelet            Pulling image "quay.io/operator-framework/olm@sha256:e9de77ac5c08643202ad42a0311d15c98ffbfd8a1dc2eab4e9f2dcaeed7110ac"
  Normal   Pulled     22s                kubelet            Successfully pulled image "quay.io/operator-framework/olm@sha256:e9de77ac5c08643202ad42a0311d15c98ffbfd8a1dc2eab4e9f2dcaeed7110ac" in 747.878744ms
  Warning  BackOff    17s (x2 over 18s)  kubelet            Back-off restarting failed container
  Normal   Created    3s (x3 over 21s)   kubelet            Created container util
  Normal   Pulled     3s (x2 over 19s)   kubelet            Container image "quay.io/operator-framework/olm@sha256:e9de77ac5c08643202ad42a0311d15c98ffbfd8a1dc2eab4e9f2dcaeed7110ac" already present on machine
  Warning  Failed     2s (x3 over 21s)   kubelet            Error: failed to start container "util": Error response from daemon: OCI runtime create failed: container_linux.go:367: starting container process caused: exec: "/bin/cp": stat /bin/cp: no such file or directory: unknown

Environment

• operator-lifecycle-manager version:
  v0.18.0

• Kubernetes version information:

kubectl version
Client Version: version.Info{Major:"1", Minor:"21", GitVersion:"v1.21.0", GitCommit:"cb303e613a121a29364f75cc67d3d580833a7479", GitTreeState:"clean", BuildDate:"2021-04-08T21:16:14Z", GoVersion:"go1.16.3", Compiler:"gc", Platform:"darwin/amd64"}
Server Version: version.Info{Major:"1", Minor:"20", GitVersion:"v1.20.2", GitCommit:"faecb196815e248d3ecfb03c680a4507229c2a56", GitTreeState:"clean", BuildDate:"2021-01-13T13:20:00Z", GoVersion:"go1.15.5", Compiler:"gc", Platform:"linux/amd64"}

• Kubernetes cluster kind:

minikube with virtualbox as driver.

Possible Solution
Please verify the file in the images and executable

Additional context

v0.17.0 is working fine for me
Another issue opened in operator-sdk: operator-framework/operator-sdk#4833

[benluddy]

Thanks for reporting this. I marked v0.18.0 as "pre-release", which appears to make v0.17.0 be considered the latest release again. I hope that will prevent other operator-sdk users from hitting this.

The base image gcr.io/distroless/static:debug introduced in 14df116 does have cp via busybox, but it's not at the path that the bundle unpacker expects. A symlink should fix it, or maybe cpb could be extended to copy itself?

This also highlights a testing gap: the E2E test jobs build an image locally using e2e.Dockerfile. That could probably be unified with upstream.Dockerfile, and there should also be an end-to-end run for tagged releases that installs OLM using the release payload.

I'll leave this issue here to track a v0.19.0 roll-forward release.

[njhale]

The patch merged: #2133

[pedjak]

operatorhub.io instructions are still advertising the usage of OLM 0.18 version

[kevinrizza]

Looks like this should be resolved with the release of https://github.com/operator-framework/operator-lifecycle-manager/releases/tag/v0.18.1