This page is where I put commands and script-fu that I tend to use most often for OpenShift 3 (Kubernetes/Docker). Originally written to aid my poor memory, but I share it here in case you find it useful.

OpenShift is an awesome platform for developing and deploying apps in containers. To try it out, you can:



Version Release date Notes
4.1 June 2019  
4.0 April/May 2019 Preview release only
3.11 October 2018  
3.10 July 2018  
3.9 March 2018 Includes features and fixes from Kubernetes 1.8 and 1.9. There was no 3.8 release.
3.7 November 2017  
3.6 August 2017  
3.5 April 2017  

Getting started

Using minishift (for OpenShift 3.x only)

Up and running with the upstream minishift, on Fedora:

sudo dnf install libvirt qemu-kvm
sudo usermod -a -G libvirt $(whoami)
newgrp libvirt
sudo curl -L -o /usr/local/bin/docker-machine-driver-kvm
sudo chmod +x /usr/local/bin/docker-machine-driver-kvm

curl -OL${MINISHIFT_VERSION}/minishift-${MINISHIFT_VERSION}-linux-amd64.tgz
tar -xvf minishift-${MINISHIFT_VERSION}-linux-amd64.tgz

mv minishift-${MINISHIFT_VERSION}-linux-amd64/minishift ~/bin

sudo systemctl start virtlogd
sudo systemctl enable virtlogd
sudo systemctl start libvirtd
sudo systemctl enable libvirtd

minishift oc-env

Container Development Kit - Using a Red Hat developer subscription

Upstream minishift doesn’t support a RHEL-based VM and productised OpenShift out of the box.

To use a RHEL-based VM and productised OpenShift, install the Container Development Kit (CDK), first download the CDK from the Red Hat Developers site (ensure third-party cookies are enabled and you have a valid Developer subscription) - this is a large download as it includes a RHEL7 ISO.

Then set up the CDK as below:

mv ~/Downloads/cdk-3.9.0-1-minishift-linux-amd64 ~/bin/minishift
chmod u+x ~/bin/minishift
minishift setup-cdk
export MINISHIFT_USERNAME=your-rh-developers-username

minishift start    # first run may take 5-10 mins to start up
minishift console

Using oc cluster up

Mac: Start a simple local all-in-one OpenShift cluster with a configured registry, router, image streams, and default templates:

brew install openshift-cli
oc cluster up

Fedora: Start an all-in-one cluster with oc cluster up:

$ sudo dnf install -y docker origin-clients

# (Optional) Add yourself into the `docker` user group to avoid needing sudo
$ sudo groupadd docker && sudo gpasswd -a ${USER} docker && sudo systemctl restart docker
$ sudo systemctl start docker

# Then in `/etc/containers/registries.conf`, add to `registries.insecure`

$ newgrp docker
$ oc cluster up

Single node cluster in AWS

Start a cluster in AWS - first spin up a RHEL machine, then (example instructions from Wildfly-Camel docs):

ssh -i ~/.ssh/id_rsa_ec2.pem ec2-user@

curl -fsSL | sh
sudo usermod -aG docker ec2-user
sudo systemctl enable docker.service
sudo systemctl start docker
docker run hello-world

curl -L
tar xzf openshift-origin-server-v3.10.0-dd10d17-linux-64bit.tar.gz
oc cluster up

And updated on 03/09/2019. First start a RHEL instance in AWS (RHEL 8.x seems to be OK), allow incoming traffic to ports 22, 80 and 443, then:

sudo yum install -y podman podman-docker
curl -OL
tar xzvf openshift-origin-server-v3.11.0-0cbc58b-linux-64bit.tar.gz

mkdir ~/.local/bin
mv openshift-origin-server-v3.11.0-0cbc58b-linux-64bit/* ~/.local/bin/
oc cluster up --public-hostname=${EXTERNAL_HOSTNAME}

sudo dnf clean all
sudo subscription-manager register
sudo subscription-manager attach
sudo subscription-manager repos --enable rhel-7-server-devtools-rpms

sudo rm -r /var/cache/dnf

Installing on AWS using Installer-Provisioned Infrastructure (IPI)

Follow a process like this:

  1. Create a dedicated public hosted zone in Route 53. The zone must be authoritative for the domain.

  2. Create a bastion host in AWS. This is preferable, because the installer will spit out

Installing Helm

sudo curl -L -o /usr/local/bin/helm
sudo chmod +x /usr/local/bin/helm


Container-Native Storage comprises two technologies:

  • Red Hat Gluster Storage - containerized distributed storage. Each Red Hat Gluster Storage volume is composed of a collection of bricks, where each brick is the combination of a node and an export directory.
  • Heketi - for Red Hat Gluster Storage volume life cycle management

Images and image streams

The image field inside a Pod spec in a Deployment Config object:

  • When deploying from an image stream:

    • Set the dc.spec.template.spec.containers.image field to the name of the image. This will cause OpenShift to resolve the full spec to the image from the image stream.
    • Also, under triggers, add an ImageChange trigger which references the ImageStreamTag, and set automatic: true
  • When the OpenShift console is used to deploy an image (using Add to Project -> Deploy Image) and an imagestream is selected, the deploymentconfig that ultimately gets created contains the full image spec of the image, e.g. "image": ""

  • If triggers[0].imageChangeParams.automatic is false, OpenShift will not automatically rollout the DC when it is created.

To build your own custom layers on top of “gold” vendor-distributed images from an external registry:

  1. Create an Image Stream for the gold image, e.g. RHEL
  2. Create a BuildConfig which has an ImageChangeTrigger on the external image and uses the external image as the base/from
  3. Define a schedule for importing/updating the external imagestream, because it is uses an external registry which won’t be automatically updated. e.g. use oc import-image (this imports the metadata for the Image Stream)


Networking inside a pod:

  • Pods take the contents of their host node’s resolv.conf DNS configuration file.
  • OpenShift appends search and nameserver details, and dnsIP, which is set in /etc/origin/node/node-config.yml. If not set, the Kubernetes service IP is used.

Communicating from OpenShift pods to external services:

  • By default, traffic from pods in OpenShift to external destinations will have the source IP of whatever node the pod is currently running on.
  • If the nodes have NAT configured, then the address will be different and dependent on the NAT configuration.
  • Alternatively, you can configure OpenShift to add an Egress IP against a particular namespace (netnamespace)

Accessing services in another namespace:

  • If a service myapp is defined in namespace goats then it can be accessed using the DNS name myapp.goats.svc.cluster.local - this is achieved using the ovs-subnet SDN plugin (default).
  • Namespaces can be optionally isolated using the ovs-multitenant SDN (Software-Defined Networking) plugin which means pods in different namespaces cannot send/receive packets to/from pods in another namespace.


The default router is HAProxy. Plugins are available to use other routers if required.


Authenticating to the cluster

In OpenShift you don’t define users - that responsibility is delegated to the identity provider which has been configured, e.g.:

  • Htpasswd (authentication via an htpasswd file which should exist on all of the master nodes)
  • LDAP
  • Keystone
  • etc.

Permissions are granted by adding a user to a group, and then defining a role binding for the group, which grants role(s) to the group.

The kubeadmin user is created on installation and the password is written to the install log. To delete the kubeadmin user:

oc delete secret kubeadmin -n kube-system

Security inside containers and SCCs

UID inside a container

OpenShift runs containers using an arbitrarily assigned user ID, by default:

UID ranges

To check the user ID range for a Pod, look at the automatically generated annotation on the namespace. Then, you can go inside a Pod and observe that the user will have an ID from the range. In this case, 1004580000:

# Check the UID range for this Project
$ oc get namespace $(oc project -q) -o yaml | grep sa.scc.uid-range 1004580000/10000

# Get a terminal inside a Pod, and check what UID the container is running as
$ oc rsh some-pod-in-the-namespace-0
sh-4.2$ id
uid=1004580000 gid=0(root) groups=0(root),1004580000

Can Pods within the same Project run as the same UID?

Yes. Here’s an example. 3 pods in the same Project, which all have the same UID:

$ oc exec myapp-6876b7c677-hc9mk -- id
uid=1004580000(1004580000) gid=0(root) groups=0(root),1004580000
$ oc exec amq-broker-5d899dc74c-966nm -- id
uid=1004580000 gid=0(root) groups=0(root),1004580000
$ oc exec eap-app-amq-1-vrk5m -- id
uid=1004580000 gid=0(root) groups=0(root),1004580000

What group is the container user in?

The user inside the container is also always a member of the root group (0):

“For an image to support running as an arbitrary user, directories and files that may be written to by processes in the image should be owned by the root group and be read/writable by that group.

Running containers as root

Pods can also run as root if required:

  • For images that expect to run as root, run them under a named service account, and then give that service account the anyuid SCC
  • e.g. oc adm policy add-scc-to-user anyuid system:serviceaccount:myproject:mysvcacct


Applying YAML

To apply some arbitrary YAML to the cluster (multi-line):

oc create -f - <<API

For example - to create a ServiceAccount:

oc create -f - <<API
apiVersion: v1
kind: ServiceAccount
name: metrics-deployer
- name: metrics-deployer

Working with objects

Copy a bunch of objects from one namespace to another, removing all of the metadata we don’t want:

oc get dc,svc,is,route -l app=kylie-fan-club -o json | jq 'del( .items[].status, \
  .items[], \
  .items[].metadata.namespace, \
  .items[].metadata.uid, \
  .items[].metadata.selfLink, \
  .items[].metadata.resourceVersion, \
  .items[].metadata.creationTimestamp, \
  .items[].metadata.generation), \
  .items[].spec.clusterIP' | oc apply -n mynewnamespace -f -


Get a pod name using awk and head:

oc get pods | grep your-pod-base-name | awk '{ print $1 }' | head -1

Get the pod name(s) from a deployment (e.g. matching a given label):

oc get pod -l application=sso --output=jsonpath='{}'


Start a build and follow (tail) the log onscreen:

oc start-build your-build-name --follow

Add a trigger to a build, on completion of another build (e.g. if the build pushes to the ImageStreamTag my-build:latest):

oc set trigger bc/my-build-after --from-image=my-build:latest

Add a source secret to an existing build using oc patch (e.g. when cloning from a Git repository that requires credentials, or a certificate):

oc patch bc/your-build-name -p '{"spec":{"source":{"sourceSecret":{"name":"builder-secret-name"}}}}'


Create (“expose”) a Route from a Service:

oc expose service your-service-name

Get the hostname (host) of the Route using the Template output format:

oc get route your-route -o template –template=’{{}}’

Images and image streams

Create an empty image stream:

oc create is your-image-stream-name

To import an image from an external registry (create an image stream from an external image) use oc import-image .... If the image stream doesn’t already exist you’ll need to add --confirm. Some examples:

oc import-image fuse-java-openshift:1.7 --confirm
oc import-image jaeger-agent --confirm -n mynamespace
# and in the old public RH registry...
oc import-image amq-interconnect-1.3-openshift:latest -n openshift --confirm
oc import-image jboss-amq-62:1.3 -n openshift --confirm

Import image stream definitions, and then import a specific image tag:

oc replace --force  -f \
oc import-image amq-broker-72-openshift:1.2

To add a new tag to an existing image stream, use import-image and specify the source (from) manually. This is useful when you have an image stream that is pointing to images in an external repository, e.g. the Red Hat Container Registry:

oc import-image fuse7-java-openshift:1.3 -n openshift --as=system:admin

Grant permissions for a build to pull an image from another project:

oc policy add-role-to-user system:image-puller system:serviceaccount:yourbuildproject:builder -n namespace-to-pull-from


Create a ConfigMap containing a bunch of arbitrary XML (e.g. an AMQ Broker config file):

apiVersion: v1
kind: ConfigMap
  name: my-broker-configmap
  broker.xml: |
    <?xml version="1.0" encoding="UTF-8" standalone="no"?>
    <configuration xmlns="urn:activemq"
        xsi:schemaLocation="urn:activemq /schema/artemis-configuration.xsd">


Secrets also have different types:


Create a new source secret for a build (where the source is located in a Git repository that requires authentication):

oc create secret generic gitlab-secret \
    --from-literal=username=MYUSERNAME \
    --from-literal=password=MYPASSWORD \

Get the username value from a secret using the template option:

oc get secret -n my-project mongodb --template="{{ .data.username }}"

Applying a secret:

oc create -f - <<API
apiVersion: v1
kind: Secret
  name: my-test-secret
type: Opaque
  truststorepassword: changeit
  keystorepassword: changeit


Create resources from a Template, and set a parameter:

oc process mytemplate -p BARLOW_TYPE=deirdre | oc create -f -


See the dedicated page on OpenShift Operators.


Basic StatefulSet

oc create -f - <<API
apiVersion: apps/v1beta1
kind: StatefulSet
  generation: 1
    test: tom
  name: my-statefulset
  podManagementPolicy: Parallel
  replicas: 1
      app: my-statefulset
  serviceName: spring-boot-camel-xa-headless
      creationTimestamp: null
        app: my-statefulset
        - env:
            - name: KUBERNETES_NAMESPACE
                  apiVersion: v1
                  fieldPath: metadata.namespace
          image: this-image-does-not-exist-eggs
          imagePullPolicy: IfNotPresent
          name: app-container
            - containerPort: 8080
              name: http
              protocol: TCP
            - containerPort: 9779
              name: prometheus
              protocol: TCP
            - containerPort: 8778
              name: jolokia
              protocol: TCP
              cpu: '1'
              memory: 256Mi
              cpu: 200m
              memory: 256Mi
            privileged: false
          terminationMessagePath: /dev/termination-log
          terminationMessagePolicy: File
      dnsPolicy: ClusterFirst
      restartPolicy: Always
      terminationGracePeriodSeconds: 180
    type: RollingUpdate

External registry

Pull an image from a private AWS ECR (Elastic Container Registry) repository

# This will last for 12 hours only.
ECR_PASSWORD=$(aws ecr get-login-password)

oc create secret docker-registry ecr-secret --docker-username AWS --docker-password ${ECR_PASSWORD} --docker-server ${REGISTRY_HOST}

oc secrets link default ecr-secret --for=pull

Working with Red Hat images

Pulling images

Red Hat images on now require authentication. The simplest way to do this is to create a Secret in the openshift namespace to authenticate against the Red Hat registry:

oc create secret docker-registry rh-secret --docker-username=${RH_USERNAME} --docker-password=${RH_PASSWORD} -n openshift --as system:admin

Then try to reimport the ImageStreamTag:

oc import-image fuse7-java-openshift:1.2 -n openshift --as=system:admin

Or, if needing authentication so that Jenkins can pull images from the Red Hat Container Registry:

oc create secret docker-registry rh-secret .....
oc secrets link deployer rh-secret

Using a Red Hat base image where authentication is required, in a build:

oc create secret docker-registry rh-secret ...
oc secrets link builder rh-secret
oc new-build

Inspecting S2I images

Get the location of the S2I scripts for an image:

podman inspect | grep s2i.scripts-url

Cat the run script:

podman run --rm cat /usr/local/s2i/run

Red Hat Middleware for OpenShift

JBoss Fuse, AMQ, EAP. All the lush and beautiful things.

Install the JBoss middleware image streams:

oc create -f -n openshift

Install the AMQ image streams:

# 7.2 (1.2) GA
oc create -f

Install one of the JBoss middleware templates to allow you to create AMQ, EAP, etc instances, from the web console or CLI:

oc create -f -n openshift

Create a new app using one of the templates, and download dependencies from a local Maven mirror, instead of Maven Central:

oc new-app jboss-webserver30-tomcat8-openshift~ -e MAVEN_MIRROR_URL=

Jolokia agent in Java images

To list all available Jolokia operations on a Java container which has the Jolokia agent installed:

OCTOKEN=$(oc whoami -t)
curl -k -H "Authorization: Bearer $OCTOKEN" https://$OCMASTER:8443/api/v1/namespaces/$OCPROJECT/pods/https:$OCPOD:8778/proxy/jolokia/list

View heap memory usage (watch this to monitor the heap in realtime without needing Hawtio):

$ curl -k -H "Authorization: Bearer $OCTOKEN" https://$OCMASTER:8443/api/v1/namespaces/$OCPROJECT/pods/https:$OCPOD:8778/proxy/jolokia/read/java.lang:type=Memory/HeapMemoryUsage

View cache size for any local Infinispan caches:

$ CACHENAME=mycachename
$ curl -k -H "Authorization: Bearer $OCTOKEN" https://$OCMASTER:8443/api/v1/namespaces/$OCPROJECT/pods/https:$OCPOD:8778/proxy/jolokia/read/org.infinispan:type\=Cache,name\=%22$CACHENAME%28local%29%22,manager\=%22DefaultCacheManager%22,component\=Statistics/numberOfEntries

View Artemis broker info:

curl http://quarkus:quarkus@\=\"${ARTEMIS_POD}\"

Internal Docker registry

Verify that the registry is up and running in the default project:

oc get all -n default

Find the IP address of the Docker registry (by showing all services in the default project):

oc get svc -n default

Redeploy the internal Docker registry:

oc deploy docker-registry --retry


Get information about rolebindings

See information about all ClusterRoleBindings:

oc describe clusterrolebinding.rbac

Looking at a ClusterRoleBinding named admin (although it can be named anything):

$ oc describe clusterrolebinding.rbac/admin
Name:         admin
Labels:       <none>
Annotations:  <none>
  Kind:  ClusterRole
  Name:  admin
  Kind  Name                 Namespace
  ----  ----                 ---------

Means that I have the admin ClusterRole. Which is:

$ oc describe clusterrole/admin
Name:         admin
Annotations: true
  Resources                                                  Non-Resource URLs  Resource Names                                Verbs
  ---------                                                  -----------------  --------------                                -----                                   []                 []                                            [* create update patch delete get list watch]


  builds/details                                             []                 []                                            [update]                          []                 []                                            [update]

Get role or clusterrole for a specific user:

$ oc get clusterrolebindings.authorization \,,user:.userNames.* | \
  awk -v user=JEFF_MILLS '$3 == user {print}'
edit-21           edit     JEFF_MILLS

Which shows that the user only has the clusterrolebinding edit-21 assigned to them, which is:

$ oc describe clusterrolebinding edit-21
Name:         edit-21
Labels:       <none>
Annotations:  <none>
  Kind:  ClusterRole
  Name:  edit
  Kind  Name                     Namespace
  ----  ----                     ---------

This will give you the clusterrole that has been assigned to the user. You can figure out what permissions that role has:

$ oc describe clusterrole edit

Create a group and add users

To grant permissions to individual users, you can add them to a group, and then define a role binding for that group. To create the group and add members:

oc adm groups new my-new-group
oc adm groups add-users my-new-group dave barry

Give Jenkins permissions to do stuff outside its namespace

Grant the admin cluster role to the jenkins service account user and the ocp-devs group, and allow images to be pulled from another namespace.

Bind user groups to roles for each of the new project - i.e. give “ocp-devs” edit permissions in the projects:

oc create rolebinding ocp-devs_admin_role --clusterrole=admin --group=ocp-devs -n myapp-dev
oc create rolebinding ocp-devs_admin_role --clusterrole=admin --group=ocp-devs -n myapp-test
oc create rolebinding ocp-devs_admin_role --clusterrole=admin --group=ocp-devs -n myapp-perftest

Allow Jenkins to operate outside of the project he’s created in:

oc create rolebinding serviceaccount-labs-ci-cd-jenkins-edit-role --clusterrole=admin --serviceaccount=labs-ci-cd:jenkins -n myapp-dev
oc create rolebinding serviceaccount-labs-ci-cd-jenkins-edit-role --clusterrole=admin --serviceaccount=labs-ci-cd:jenkins -n myapp-test
oc create rolebinding serviceaccount-labs-ci-cd-jenkins-edit-role --clusterrole=admin --serviceaccount=labs-ci-cd:jenkins -n myapp-perftest

Give Jenkins the image-puller role, so he can fetch Docker images stored in another namespace:

oc create rolebinding serviceaccount-labs-ci-cd-default-edit-role --clusterrole=system:image-puller --serviceaccount=labs-ci-cd:default -n myapp-dev
oc create rolebinding serviceaccount-labs-ci-cd-default-edit-role --clusterrole=system:image-puller --serviceaccount=labs-ci-cd:default -n myapp-test
oc create rolebinding serviceaccount-labs-ci-cd-default-edit-role --clusterrole=system:image-puller --serviceaccount=labs-ci-cd:default -n myapp-perftest

Create a rolebinding in YAML

Role bindings in YAML: to define a role binding which grants the admin Role to the group junior-devs in the project development:

kind: RoleBinding
  name: my-admin-role
  namespace: development
- junior-devs
  name: admin
- kind: Group
  name: junior-devs
userNames: null

View information about a role

Grant a user sudoer / system:admin

Create a clusterrolebinding, to allow a basic user to impersonate a privileged user (e.g. when running a local cluster using oc cluster up and granting system:admin impersonation):

oc create clusterrolebinding willsmith-sudo --clusterrole=sudoer --user=will-smith

Allow a container to run as root

Need to run a container as root? Or you’re running a RH container as a random user, which seems to have no write access to the filesystem inside the container? Then create a service account which allows a pod to run as root, then configure the DeploymentConfig to use that service account:

oc create serviceaccount useroot
oc adm policy add-scc-to-user anyuid -z useroot
oc patch dc/appthatneedsroot --patch '{"spec":{"template":{"spec":{"serviceAccountName": "useroot"}}}}'

The add-scc-to-user is actually a shortcut to add the user to the special ClusterRoleBinding object called system:openshift:scc:anyuid, which grants the anyuid role to the ServiceAccount:

$ oc get clusterrolebinding/system:openshift:scc:anyuid -o yaml
kind: ClusterRoleBinding
  name: system:openshift:scc:anyuid
  kind: ClusterRole
  name: system:openshift:scc:anyuid
- kind: ServiceAccount
  name: default
  namespace: toms-temp
- kind: ServiceAccount
  name: default
  namespace: my-dev
- kind: ServiceAccount
  name: default
  namespace: argocd

Set up an htpasswd provider and create a user

Create an htpasswd file and define a user:

$ touch htpasswd
$ htpasswd -Bb htpasswd tom letmein

Now set up an identity provider of type HTPasswd. This also creates a new Log in with option called “Local Password” when you log into the web console:

oc [--user=admin] create secret generic htpasswd \
    --from-file=htpasswd -n openshift-config

oc replace -f - <<API
kind: OAuth
  name: cluster
  - name: Local Password
    mappingMethod: claim
    type: HTPasswd
        name: htpasswd

oc adm groups new mylocaladmins

oc adm groups add-users mylocaladmins tom bernie cliff

oc get groups

oc adm policy add-cluster-role-to-group cluster-admin mylocaladmins

NB: This might take a few minutes to apply to the cluster. So if your new login doesn’t work immediately, try again in a few minutes.

Create a custom resource and grant permissions

Once you’ve created a CustomResourceDefinition (e.g., you can grant users with the ClusterRoles admin or edit permissions to create/delete instances of the CRD with this YAML below.

Use the special aggregate-to-admin and aggregate-to-edit labels to make sure these permissions are added into the admin and edit roles:

kind: ClusterRole
  - metadata:
      name: aggregate-pizzas-admin-edit
      labels: "true" "true"
      - apiGroups: [""]
        resources: ["pizzas"]
        verbs: ["get", "list", "watch", "create",
                "update", "patch", "delete", "deletecollection"]
  - metadata:
      name: aggregate-pizzas-view
        # Add these permissions to the "view" default role. "true" "true"
      - apiGroups: [""]
        resources: ["pizzas"]
        verbs: ["get", "list", "watch"]

oc patch-fu

Fun uses of oc patch.

Modify a BuildConfig to not use cached layers in a Docker build

Don’t use cached layers in a Docker build:

oc patch bc/myapp -p '{"spec":{"strategy":{"dockerStrategy":{"noCache":true}}}}'

Modify the timeout of a Knative Service to 20 seconds:

oc patch ksvc greeter -n knativetutorial --type=json -p='[{"op": "replace", "path": "/spec/template/spec/timeoutSeconds", "value":20}]'

Add an environment variable populated from a ConfigMap to a DeploymentConfig

This uses the /- trick to add an additional element to the env array in the JSON, which populates an environment variable from a key in a ConfigMap - in this case, updating the AMQ BROKER_XML environment variable:

oc patch dc/london-amq --type=json -p='[{"op": "add", "path": "/spec/template/spec/containers/0/env/-", "value": { "name": "BROKER_XML", "valueFrom": { "configMapKeyRef": { "name": "london-broker", "key": "broker.xml" }}}}]'

Thanks to

OpenShift API

The OpenShift API is documented using Swagger at (when using oc-cluster up):



Get the version of OpenShift:

oc get clusterversion

Get a shell onto an OpenShift node / SSH into a node

You can get a shell onto an OpenShift node (e.g. a worker machine) using the oc debug command.

First find the node where a Pod is running (if you want to debug a specific container):

oc get pod prometheus-k8s-0 --template="{{ .spec.nodeName }}"

Then oc debug that Node:

oc debug node/

Once on the node, use chroot to access the node’s binaries, then you can look at which containers are running, with crictl:

chroot /host
crictl ps


Get pod usage statistics for a project (memory)

You need cluster-reader for this:

oc adm top pods -n myproject


Deleting/pruning old pods using an initContainer:

- command: ["oc", "delete", "jobs", "-l", "foo=bar"]
  image: openshift3/ose
  imagePullPolicy: IfNotPresent
  name: mypod-cleanup

Demos & wonderful little things

Sample app - Basic Apache HTTP Server web site

Deploy a basic demonstration web site running on Apache HTTPD:

oc new-app centos/httpd-24-centos7~
oc expose svc/httpd-ex

StatefulSet - Python simple web server, with a 30s delay on startup

This uses the python image from Docker Hub to run a simple HTTP server and serve some basic content. With a delay on startup using sleep:

apiVersion: v1
kind: List
- apiVersion: apps/v1beta1
  kind: StatefulSet
      test: test
    name: sleepy-python
    replicas: 1
        app: sleepy-python
        creationTimestamp: null
          app: sleepy-python
          - command: ["/bin/sh", "-c"]
            args: ["echo Sleeping... && sleep 30 && echo Starting web server... && echo HELLO >> /tmp/index.html && python -m http.server 8000 --directory /tmp/"]
            image: python:3.8
            imagePullPolicy: IfNotPresent
            name: python
              - containerPort: 8000
                name: http
                protocol: TCP
                path: /
                port: 8000
              initialDelaySeconds: 10
                path: /
                port: 8000
              initialDelaySeconds: 30


The oc command line tool

(Windows) Tell OC to use your local C: drive as the location for its .kube config file. Useful in environments where your home drive is set to a network drive, or your .kube config file is otherwise inaccessible:

set KUBECONFIG=C:\Users\username\.kube\config

Or use these:

set HOMEPATH=C:\Users\username

minishift/Container Development Kit

Delete the Minishift VM because something’s gone wrong:

$ minishift delete [--clear-cache]
$ rm -rf ~/.kube       # optional, but sometimes necessary

Start the CDK when inside an evil corporate network/behind a proxy (or just try and avoid running behind a proxy if possible):

$ minishift setup-cdk
$ minishift start --http-proxy http://<YOURPROXY>:<PORT> --https-proxy https://<YOURPROXY>:<PORT>

To see full logs when starting up, add:

$ minishift [...] --show-libmachine-logs --v=10 --alsologtostderr

“Could not set oc CLI context for ‘minishift’ profile: Error during setting ‘minishift’ as active profile: The specified path to the kube config ‘/Users/tdonohue/.minishift/machines/minishift_kubeconfig’ does not exist” means something went wrong with your minishift config:

$ minishift delete && rm -rf ~/.minishift

Build fails when building from an image stream in the Red Hat Container Registry - 401 Unauthorized when trying to pull the image (e.g. the Fuse base image, etc.):

  • Try deploying the image manually (Web Console → Add to Project → Deploy Image → point to the image stream tag). This should work (deployer seems to have permissions to deploy the image, but perhaps builder does not)

The all-in-one cluster (oc cluster up)

The all-in-one cluster is a local OpenShift cluster on a single machine, incorporating a registry, router, image streams and default templates. All of these run as (Docker) containers.

See all OpenShift infrastructure containers (e.g. registry, router, etc) running on your workstation:

docker ps

Open a terminal in the origin container (where the all-in-one OpenShift server is located):

docker exec -it origin bash

View logs from the origin container:

docker logs origin

View the master-config file in the origin container:

docker exec -it origin cat /var/lib/origin/openshift.local.config/master/master-config.yaml

Edit the master-config file, when using the oc-cluster wrapper utility:

vim ~/.oc/profiles/[profile-name]/config/master/master-config.yml

List the kube utils that are available in the origin container:

# docker exec origin ls /usr/bin | grep kube

OpenShift 4.x on AWS

Delete a Machine that’s causing problems

If there are problems with a Node (e.g. status of NotReady), you can delete the underlying Machine (a Machine represents an underlying virtual machine, or EC2 Instance in the case of AWS):

  • Check which nodes have problems, using oc get nodes. Look for a status of NotReady.
  • oc get machines -n openshift-machine-api -o wide will show all Machines and which Node each corresponds to.
  • oc delete machine xxx -n openshift-machine-api to delete a specific Machine. The machine-api Operator will create a new machine in AWS to replace it.
  • For further info, oc get machineset -n openshift-machine-api will show all of the configured MachineSets and their replica counts (there is usually 1 MachineSet for each AWS availability zone)
  • Once things are back up, check the status of the cluster operators - oc get co

Debugging an AWS cluster where the API is down

If the API is down, then you probably won’t be able to use oc at all. Sad. This calls for desperate measures.

  • See if you can access the cluster from the original machine which installed OpenShift. The installer leaves a kubeconfig file, somewhere after the installation. Run export KUBECONFIG=/path/tokubeconfig, then oc get pods or something similar.
  • oc get nodes to get the list of nodes. Look for a status of NotReady.
  • Get node logs using oc adm node-logs -u crio

Other problems and solutions

Q. My computer starts burning up and/or running out of RAM. Also, Java containers are hanging on startup.

  • Increase the RAM available to Docker for Mac (this will require a Docker restart)
  • docker stop any non-essential containers that you may be running outside OpenShift
  • Check docker stats to see CPU usage of the origin container; docker restart origin if necessary

Q. The Router does not seem to start properly when running OpenShift locally.

$ oc get pods -n default | grep router
router-1-deploy                 0/1       Error       0          11m

Possibly you might have a port conflict. Check the events in the default namespace:

$ oc get events -n default
14m       14m       1         router-1-uc7mo   Pod                 Warning   FailedSync   kubelet, localhost   Error syncing pod, skipping: failed to "StartContainer" for "POD" with RunContainerError: "runContainer: Error response from daemon: {\"message\":\"driver failed programming external connectivity on endpoint k8s_POD.ec88479e_router-1-uc7mo_default_7b212800-f797-11e7-a5c3-fe66a2ce528b_5b61cfa1 (a039fa984569e6ab4d2f4cae417b454578190c52ec7a824c5e8e8ea29adbe90f): Error starting userland proxy: Bind for unexpected error (Failure EADDRINUSE)\"}"

Apps that might cause this: Laravel Valet. Terminate the offending app and restart the router:

$ oc deploy router -n default --retry

Q. Multiple pods are running even though I only want one!

Maybe you’ve got multiple pods with different index numbers running:

$ oc get pods
NAME                          READY     STATUS              RESTARTS   AGE
ocp-job-manager-5-oo9yc       0/1       ContainerCreating   0          1s
ocp-job-manager-6-cqqvv       0/1       ContainerCreating   0          1s

This is probably because you’re trying to roll out more than one ReplicationController:

$ oc get rc
NAME                DESIRED   CURRENT   READY     AGE
ocp-job-manager-4   0         0         0         14m
ocp-job-manager-5   1         1         0         9m
ocp-job-manager-6   1         1         0         33s

Just scale down the one you don’t want:

$ oc scale rc/ocp-job-manager-5 --replicas=0

Q. I’ve pushed a new image to a tag but my pods still seem to be using the old image.

Check whether your pod is using the cached image that is already present on the node:

$ oc get events | grep 'already present'
6m        6m        1         funbobby-nknku84e-debug   Pod       spec.containers{int0032}   Normal    Pulled       kubelet,   Container image "172.x.x.x:5000/mynamespace/funbobby:1.0.0-SNAPSHOT" already present on machine

If so, update your template or deploymentconfig and set ImagePullPolicy to Always.

Q. ImagePullBackOff. What?

  • You’ve specified an image in your DeploymentConfig with an invalid value.

    • OpenShift may be trying to pull the image from
    • If using ImageStreams, leave image blank and just set your imagestream details in the triggers section of your DeploymentConfig object.

Q. I can’t log in as system:admin

  • Probably some mismatch in your local .kube configuration. Delete the .kube directory and start again.

Q. I do oc new-app from a template but nothing happens. oc status says “deployment #1 waiting on image or update”

  • The deployment is waiting for the image to exist first.
  • Check the image: attribute in the Deployment Config. Does it point to an Image Stream Tag which exists?
  • Is the image: pointing to the correct namespace? If the namespace is omitted, it will assume the default namespace (openshift). For example, the spec image: myimage:1.1 expects an Image Stream Tag called myimage:1.1 to exist in the openshift namespace.
  • If the app is an S2I/binary build, then you need to wait for the image to be built, before it can be deployed automatically.

Q. I get this error: “No API token found for service account “deployer”, retry after the token is automatically created and added to the service account”

  • Delete the service account and it should be recreated again, e.g.: oc delete sa deployer

Q. I can’t use --as=system:admin in commands: “Error from server (Forbidden): users “system:admin” is forbidden: User “developer” cannot impersonate users at the cluster scope: no RBAC policy matched”

  • oc create clusterrolebinding developer-as-admin --clusterrole=sudoer --user=developer

Q. oc cluster down leaves some directories mounted, which means that openshift.local.clusterup cannot be deleted.

  • You can use the mount command to see that some directories are still mounted: mount | grep openshift
  • Unmount using: for i in $(mount | grep openshift | awk '{ print $3}'); do sudo umount "$i"; done && sudo rm -rf ./openshift.local.clusterup

Q. Binary builds fail when they pull images from Docker Hub, due to reaching rate limits.

  • Set the build secret explicitly using oc set build-secret --pull bc/<build config name> <secret name>

oc new-app doesn’t deploy your app

error: unable to locate any images in image streams, image stream images with a ‘supports’ annotation, local docker images with name “jee”:

  • OpenShift is trying to find an image stream in the cluster which will let it build your app, which it’s detected as being a Java app (jee).
  • Look in the openshift project for any Image Stream Tags that have an annotation supports which contains the string jee.
  • TODO: Complete this bit.

General troubleshooting tips

If something’s not working, or not deploying:

oc status -v

If something’s still not working:

oc get events

Changing the log level of a build:

oc set env bc/my-build-name BUILD_LOGLEVEL=[1-5]

Problems pulling images? Check the integrated Docker registry logs:

oc logs docker-registry-n-{xxxxx} -n default | less

Get diagnostics information on your OpenShift cluster:

oc adm diagnostics

Administration and security

Grant the admin user permissions to administer the cluster (e.g. to create a PersistentVolume):

oc adm policy add-cluster-role-to-user cluster-admin admin

Grant edit permissions in the current namespace to the service account called jenkins:

oc policy add-role-to-user edit -z jenkins

Check which users can perform a certain action (useful e.g. when debugging why Jenkins can’t create slave pods):

oc policy who-can create pod

Bonus section: Docker!

Pull an image by its SHA256 digest; sometimes useful to inspect the same image that was used in a build. (In OpenShift, the SHA256 digest of the image used in a Build is given on the Build’s details page, under Builder Image)

docker pull

Delete all exited Docker containers:

docker rm $(docker ps -aq)

View the size of the Docker storage file (Docker for Mac):

du -h -d 1 ~/Library/Containers/com.docker.docker

View stats on all containers with their names:

docker stats $(docker ps | awk '{if(NR>1) print $NF}')