Technical and Product News and Insights from Rackspace
With the release of Sitecore® version 10, Kubernetes® is now a supported option. This opens up some great benefits, such as better resource utilization, version-controlled images, and overall operations ease. While building out an automated pipeline, I came across an error during the Kubernetes job step. Let’s dive in and see the fix I used to get successful deployments.
Following the install document, you deploy the ingress controller, secrets, and then the external dependencies. These
dependencies include Redis®, MSSQL®, and Solr™. Looking at the following helm install command, notice it
nodeSelector to target a Linux® node for the Nginx® ingress. While Sitecore runs on Windows® containers,
this usage informs us that we need to have Linux and Windows nodes for a full Sitecore deployment.
helm repo add stable https://kubernetes-charts.storage.googleapis.com/ helm install nginx-ingress stable/nginx-ingress --set controller.replicaCount=1 --set controller.nodeSelector."beta\.kubernetes\.io/os"=linux --set defaultBackend.nodeSelector."beta\.kubernetes\.io/os"=linux --set-string controller.config.proxy-body-size=10m --set controller.service.externalTrafficPolicy=Local kubectl apply -k ./secrets/ kubectl apply -f ./external/ kubectl apply -f ./init/
Each step of my deploy went smoothly until I hit the init YAML step, which runs jobs within Kubernetes. While monitoring
my pods, I noticed ErrImagePull for the solr initialization job. I ran a
kubectl describe pod solr-init-m5dgj
command to view more information, which yielded the following output:
QoS Class: BestEffort Node-Selectors: <none> Tolerations: node.kubernetes.io/not-ready:NoExecute for 300s node.kubernetes.io/unreachable:NoExecute for 300s Events: Type Reason Age From Message ---- ------ ---- ---- ------- Normal Scheduled <unknown> default-scheduler Successfully assigned default/solr-init-m6dgj to aks-nodepool1-19665653-vmss000001 Normal Pulling 3m5s (x4 over 4m33s) kubelet, aks-nodepool1-19665653-vmss000001 Pulling image "scr.sitecore.com/sxp/sitecore-xm1-solr-init:10.0.0-ltsc2019" Warning Failed 3m5s (x4 over 4m32s) kubelet, aks-nodepool1-19665653-vmss000001 Failed to pull image "scr.sitecore.com/sxp/sitecore-xm1-solr-init:10.0.0-ltsc2019": rpc error: code = Unknown desc = Error response from daemon: Get https://scr.sitecore.com/v2/: x509: certificate signed by unknown authority Warning Failed 3m5s (x4 over 4m32s) kubelet, aks-nodepool1-19665653-vmss000001 Error: ErrImagePull Warning Failed 2m53s (x6 over 4m32s) kubelet, aks-nodepool1-19665653-vmss000001 Error: ImagePullBackOff Normal BackOff 2m41s (x7 over 4m32s) kubelet, aks-nodepool1-19665653-vmss000001 Back-off pulling image "scr.sitecore.com/sxp/sitecore-xm1-solr-init:10.0.0-ltsc2019"
Having worked with Kubernetes for a while, I instantly spotted the error. While the message gave an
error, look at the job’s assigned node. aks-nodepool1-19665653-vmss000001 is a Linux node, and the image it pulled
ltsc2019, which is an indicator for Windows. Because I used AKS, the Windows VMSS nodes are named with a structure
of aksnpwin000000, aksnpwin000001, and so on. A Windows image cannot run off a Linux node.
The question comes up: Why did this randomly work in previous deploys? Well, the Kubernetes scheduler handles the placement
of where to run the job, but we can target specific nodes based on the operating system. Auditing the Sitecore deployment YAML
files, I noticed they were indeed targeting Windows nodes, but the init jobs were missing this entry. I verified this by
looking at the preceding describe pod output, which has
Node-Selectors: <none> set to
To fix this, open up each init job YAML file and add a
nodeSelector entry targeting the operating system,
spec: template: spec: nodeSelector: kubernetes.io/os: windows imagePullSecrets: - name: sitecore-docker-registry
After editing the YAML file, delete the failed job and apply the new init YAML files.
Describing the pod should show a status of ContainerCreating and a Node-Selectors set to Windows, as shown in the following image:
I have informed Sitecore about the missing node selector for the Kubernetes jobs, but in the meantime, it is an easy fix to edit the YAML files. For more information about windows containers, reference the kubernetes documentation.