How to use image stream in deploy configuration for OpenShift - openshift

Question

I want my deploy configuration to use an image that was the output of a build configuration.
I am currently using something like this:
- apiVersion: v1
kind: DeploymentConfig
metadata:
annotations:
openshift.io/generated-by: OpenShiftNewApp
creationTimestamp: null
labels:
app: myapp
name: myapp
spec:
replicas: 1
selector:
app: myapp
deploymentconfig: myapp
strategy:
resources: {}
template:
metadata:
annotations:
openshift.io/container.myapp.image.entrypoint: '["python3"]'
openshift.io/generated-by: OpenShiftNewApp
creationTimestamp: null
labels:
app: myapp
deploymentconfig: myapp
spec:
containers:
- name: myapp
image: 123.123.123.123/myproject/myapp-staging:latest
resources: {}
command:
- scripts/start_server.sh
ports:
- containerPort: 8000
test: false
triggers: []
status: {}
I had to hard-code the integrated docker registry's IP address; otherwise Kubernetes/OpenShift is not able to find the image to pull down. I would like to not hard-code the integrated docker registry's IP address, and instead use something like this:
- apiVersion: v1
kind: DeploymentConfig
metadata:
annotations:
openshift.io/generated-by: OpenShiftNewApp
creationTimestamp: null
labels:
app: myapp
name: myapp
spec:
replicas: 1
selector:
app: myapp
deploymentconfig: myapp
strategy:
resources: {}
template:
metadata:
annotations:
openshift.io/container.myapp.image.entrypoint: '["python3"]'
openshift.io/generated-by: OpenShiftNewApp
creationTimestamp: null
labels:
app: myapp
deploymentconfig: myapp
spec:
containers:
- name: myapp
from:
kind: "ImageStreamTag"
name: "myapp-staging:latest"
resources: {}
command:
- scripts/start_server.sh
ports:
- containerPort: 8000
test: false
triggers: []
status: {}
But this causes Kubernetes/OpenShift to complain with:
The DeploymentConfig "myapp" is invalid.
spec.template.spec.containers[0].image: required value
How can I specify the output of a build configuration as the image to use in a deploy configuration?
Thank you for your time!
Also, oddly enough, if I link the deploy configuration to the build configuration with a trigger, Kubernetes/OpenShift knows to look in the integrated docker for the image:
- apiVersion: v1
kind: DeploymentConfig
metadata:
annotations:
openshift.io/generated-by: OpenShiftNewApp
creationTimestamp: null
labels:
app: myapp-staging
name: myapp-staging
spec:
replicas: 1
selector:
app: myapp-staging
deploymentconfig: myapp-staging
strategy:
resources: {}
template:
metadata:
annotations:
openshift.io/container.myapp.image.entrypoint: '["python3"]'
openshift.io/generated-by: OpenShiftNewApp
creationTimestamp: null
labels:
app: myapp-staging
deploymentconfig: myapp-staging
spec:
containers:
- name: myapp-staging
image: myapp-staging:latest
resources: {}
command:
- scripts/start_server.sh
ports:
- containerPort: 8000
test: false
triggers:
- type: "ImageChange"
imageChangeParams:
automatic: true
containerNames:
- myapp-staging
from:
kind: ImageStreamTag
name: myapp-staging:latest
status: {}
But I don't want the automated triggering...
Update 1 (11/21/2016):
Configuring the trigger but having the trigger disabled (hence manually triggering the deploy), still left the deployment unable to find the image:
$ oc describe pod myapp-1-oodr5
Name: myapp-1-oodr5
Namespace: myproject
Security Policy: restricted
Node: node.url/123.123.123.123
Start Time: Mon, 21 Nov 2016 09:20:26 -1000
Labels: app=myapp
deployment=myapp-1
deploymentconfig=myapp
Status: Pending
IP: 123.123.123.123
Controllers: ReplicationController/myapp-1
Containers:
myapp:
Container ID:
Image: myapp-staging:latest
Image ID:
Port: 8000/TCP
Command:
scripts/start_server.sh
State: Waiting
Reason: ImagePullBackOff
Ready: False
Restart Count: 0
Volume Mounts:
/var/run/secrets/kubernetes.io/serviceaccount from default-token-goe98 (ro)
Environment Variables:
ALLOWED_HOSTS: myapp-myproject.url
Conditions:
Type Status
Ready False
Volumes:
default-token-goe98:
Type: Secret (a volume populated by a Secret)
SecretName: default-token-goe98
QoS Tier: BestEffort
Events:
FirstSeen LastSeen Count From SubobjectPath Type Reason Message
--------- -------- ----- ---- ------------- -------- ------ -------
42s 42s 1 {scheduler } Scheduled Successfully assigned myapp-1-oodr5 to node.url
40s 40s 1 {kubelet node.url} implicitly required container POD Pulled Container image "openshift3/ose-pod:v3.1.1.7" already present on machine
40s 40s 1 {kubelet node.url} implicitly required container POD Created Created with docker id d3318e880e4a
40s 40s 1 {kubelet node.url} implicitly required container POD Started Started with docker id d3318e880e4a
40s 24s 2 {kubelet node.url} spec.containers{myapp} Pulling pulling image "myapp-staging:latest"
38s 23s 2 {kubelet node.url} spec.containers{myapp} Failed Failed to pull image "myapp-staging:latest": Error: image library/myapp-staging:latest not found
35s 15s 2 {kubelet node.url} spec.containers{myapp} Back-off Back-off pulling image "myapp-staging:latest"
Update 2 (08/23/2017):
In case, this helps others, here's a summary of the solution.
triggers:
- type: "ImageChange"
imageChangeParams:
automatic: true # this is required to link the build and deployment
containerNames:
- myapp-staging
from:
kind: ImageStreamTag
name: myapp-staging:latest
With the trigger and automatic set to true, the deployment should use the build's image in the internal registry.
The other comments relating to making the build not trigger a deploy relates to a separate requirement of wanting to manually deploy images from the internal registry. Here's more information about that portion:
The build needs to trigger the deployment at least once before automatic is set to false. So far a while, I was:
setting automatic to true
initiate a build and deploy
after deployment finishes, manually change automatic to false
manually, trigger a deployment later (though I did not verify if this deployed the older, out-of-date image or not)
I was initially trying to use this manual deployment as a way for a non-developer to go into the web console and make deployments. But this requirement has since been removed, so having build trigger deployments each time works just fine for us now. Builds can build at different branches and then tag the images differently. Deployments can then just use the appropriately tagged images.
Hope that helps!

Answer 1

Are you constructing the resource definitions by hand?
It would be easier to use oc new-build and then oc new-app if you really need to set this up as two steps for some reason. If you just want to setup the build and deployment in one go, just use oc new-app.
For example, to setup build and deployment in one go use:
oc new-app --name myapp <repository-url>
To do it in two steps use:
oc new-build --name myapp <repository-url>
oc new-app myapp
If you still rather use hand created resources, at least use the single step variant with the --dry-run -o yaml options to see what it would create for the image stream, plus build and deployment configuration. That way you can learn from it how to do it. The bit you currently have missing is an image stream.
BTW. It looks a bit suspicious that you have the entry point set to python3. That is highly unusual. What are you trying to do as right now it looks like you may be trying to do something in a way which may not work with how OpenShift works. OpenShift is mainly about long running processes and not for doing single docker run. You can do the latter, but not how you are currently doing it.