Repository navigation
Recreate update is not scaling down the statefulset to zero #39710
Description
Before reporting an issue
- I have read and understood the above terms for submitting issues, and I understand that my issue may be closed without action if I do not follow them.
Area
operator
Describe the bug
The recreate update logic is not scaling down the stateful set to zero, and it is doing a rolling update which is unsafe to do.
Version
26.2.4
Regression
- The issue is a regression
Expected behavior
Using version 26.2.2, I got the correct behavior with the following events when changing the image in the Keycloak CR
4m43s Normal SuccessfulDelete statefulset/keycloak delete Pod keycloak-3 in StatefulSet keycloak successful
4m43s Normal SuccessfulDelete statefulset/keycloak delete Pod keycloak-2 in StatefulSet keycloak successful
4m38s Normal SuccessfulDelete statefulset/keycloak delete Pod keycloak-1 in StatefulSet keycloak successful
4m36s Normal SuccessfulDelete statefulset/keycloak delete Pod keycloak-0 in StatefulSet keycloak successful
4m35s Normal SuccessfulCreate statefulset/keycloak create Pod keycloak-0 in StatefulSet keycloak successful
4m19s Normal SuccessfulCreate statefulset/keycloak create Pod keycloak-1 in StatefulSet keycloak successful
4m3s Normal SuccessfulCreate statefulset/keycloak create Pod keycloak-2 in StatefulSet keycloak successful
3m48s Normal SuccessfulCreate statefulset/keycloak create Pod keycloak-3 in StatefulSet keycloak successful
Actual behavior
With version 26.2.4, I got the following events when changing the image in the Keycloak CR:
3m12s Normal SuccessfulDelete statefulset/keycloak delete Pod keycloak-3 in StatefulSet keycloak successful
3m10s Normal SuccessfulCreate statefulset/keycloak create Pod keycloak-3 in StatefulSet keycloak successful
2m55s Normal SuccessfulDelete statefulset/keycloak delete Pod keycloak-2 in StatefulSet keycloak successful
2m55s Normal SuccessfulCreate statefulset/keycloak create Pod keycloak-2 in StatefulSet keycloak successful
2m35s Normal SuccessfulDelete statefulset/keycloak delete Pod keycloak-1 in StatefulSet keycloak successful
2m34s Normal SuccessfulCreate statefulset/keycloak create Pod keycloak-1 in StatefulSet keycloak successful
2m19s Normal SuccessfulDelete statefulset/keycloak delete Pod keycloak-0 in StatefulSet keycloak successful
2m18s Normal SuccessfulCreate statefulset/keycloak create Pod keycloak-0 in StatefulSet keycloak successful
As observed, each pod is restarted individually.
How to Reproduce?
- Deploy a simple Keycloak CR. I'm using the following
apiVersion: k8s.keycloak.org/v2alpha1
kind: Keycloak
metadata:
labels:
app: keycloak
name: keycloak
spec:
hostname:
hostname: "<redacted>"
resources:
requests:
cpu: "1"
memory: "1000M"
limits:
cpu: "1"
memory: "1000M"
db:
vendor: postgres
url: jdbc:postgresql://postgres:5432/keycloak
poolMinSize: 8
poolInitialSize: 8
poolMaxSize: 30
usernameSecret:
name: postgres-secret
key: username
passwordSecret:
name: postgres-secret
key: password
ingress:
enabled: true
image: image-registry.openshift-image-registry.svc:5000/training-base/keycloak-base:26.2
startOptimized: true
http:
tlsSecret: keycloak-tls-secret
instances: 4
unsupported:
podTemplate:
spec:
containers:
- env:
# We want to have an externally provided username and password, therefore, we override those two environment variables
- name: KC_BOOTSTRAP_ADMIN_USERNAME
valueFrom:
secretKeyRef:
name: keycloak-preconfigured-admin
key: username
optional: false
- name: KC_BOOTSTRAP_ADMIN_PASSWORD
valueFrom:
secretKeyRef:
name: keycloak-preconfigured-admin
key: password
optional: false
- On the OpenShift web console, change the image.
Anything else?
No response
Metadata
Metadata
Assignees
Labels
Type
Fields
No fields configured for issues without a type.
Activity
pruivo commentedon May 14, 2025
@shawkins Do you think it related to changes here #38844 by changing the replicas to zero without using the builder? It doesn't seem like it, but I'll dig into it more tomorrow.
fix: keeping the image consistent for the recreate strategy
shawkins commentedon May 14, 2025
This was due to the change to no longer base the desired state off of the actual, so the old image was no longer preserved. That was explicitly done in the old logic, so it needs added back.
#39714
pruivo commentedon May 14, 2025
Thanks for the quick turnaround.
fix: keeping the image consistent for the recreate strategy
fix: keeping the image consistent for the recreate strategy
fix: keeping the image consistent for the recreate strategy