Rollout Status

An alternative to kubectl rollout status.

It is intended as a building block for your CI/CD pipeline. Unlike kubectl it does not rely only on progressDeadlineSeconds and fails fast when appropriate, such as when image does not exist or environment properties are misconfigured.

Usage

Docker Image is available on GitHub packages.

docker run -v "$HOME/.kube:/root/.kube:ro" ghcr.io/socialgouv/rollout-status:latest -namespace "$NAMESPACE" -selector "$SELECTOR"

Since the program is intended for a CI/CD pipeline, you will likely copy the binary to your own linux-amd64 image:

# ...

COPY --from=ghcr.io/socialgouv/rollout-status:latest /rollout-status /opt/rollout-status 

Overview of handled states

Listed by error.code:

• not-found: Deployment or the ReplicaSet was not found in given -namespace. This is may be a pipeline setup error or indicate other deployment issues that prevented your manifests from being saved to Kubernetes.
• invalid-config: Includes ErrImagePull, ImagePullBackOff, init:ErrImagePull, init:ImagePullBackOff. Fails the deployment immediately as we expect the image to be built and published before we attempt application deployment. Also incldues CreateContainerConfigError and init:CreateContainerConfigError which happen when invalid envs are defined (such as mounting non-existent ConfigMap) etc.
• resource-limits-exceeded: Aborts deployment when ReplicaSet is unable to create pods. This may happen when available resources are limited with ResourceQuota or LimitRange.
• process-crashing: Application is scheduled and runs, but exits. This encompases RunContainerError, CrashLoopBackOff as well as init container errors init:RunContainerError and init:CrashLoopBackOff and termination Error.
• scheduling: Waits for Pending pods for (currently hardcoded) deadline. If not scheduled, deployment fails.
• not-progressing: Waits until .spec.progressDeadlineSeconds. If Pods do not become ready, deployment fails.

Transient blocking states. Multiple states such as waiting for scheduling, PodInitializing, ContainerCreating as well as waiting for pods to become ready block rollout-status. The final result is reported only after a pod fails or all pods transition to ready state and succeed.

Example

go run github.com/SocialGouv/rollout-status/cmd -selector app=crashloop-backoff

{
  "success": false,
  "error": {
    "code": "process-crashing",
    "message": "Container \"main\" is in \"CrashLoopBackOff\": back-off 5m0s restarting failed container=main pod=crashloop-backoff-7fd845849c-cfvqd_default(617f6364-5bd4-4e69-b19f-fcf3ce4c171a)",
    "type": "rollout",
    "log": "panic: invalid configuration: [unable to read client-key /Users/mikulas/.minikube/profiles/minikube/client.key for minikube due to open /Users/mikulas/.minikube/profiles/minikube/client.key: no such file or directory, unable to read certificate-authority /Users/mikulas/.minikube/ca.crt for minikube due to open /Users/mikulas/.minikube/ca.crt: no such file or directory]\n\ngoroutine 1 [running]:\nmain.makeClientset(0xc000041440, 0x12, 0x0)\n\t/src/cmd/main.go:57 +0xfb\nmain.main()\n\t/src/cmd/main.go:34 +0x21c\n"
  }
}

Output

See schema.json for details. The program blocks until rollout succeeds (as success=true) or is not progressing and effectively failed (success=false). In the latter case an error object is returned. It contains all details that should be neccessary to remediate the issue. Besides error.mesage your pipeline should also switch on error.code and print help specific to your pipeline.