Bug: Deployment time replicas are set wrong in case of fullnameOverride

[prakarsh-dt]

📜 Description

Devtron automatically detects current desiredReplicas from the HPA and sets the replicas according to it while sending a new deployment. This is done so that any application that has higher number of replicas brought up from an HPA isn't reset to the initial values specified in autoscaling configs/replicas.
If an HPA is not detected according to the HPA name template, it uses replica field from deployment-template ( Logic at https://github.com/devtron-labs/devtron/blob/5283e1c712c4a555c993a64247ecc058530ca1bf/pkg/app/AppService.go#LL2264C1-L2265C1 )

This logic needs to update to consider the HPA name template in case of fullnameOverride is used.

👟 Reproduction steps

1. Create an app with autoscaling enabled and autoscaling.MinReplicas=2 and replicas=1
2. Deploy this app and wait for it to get healthy.
3. Ensure that Rolling strategy is configured with default setting (ie MaxUnavailable=1, maxSurge=25% )
4. Redeploy this app with a different image and closely watch the app details while refreshing it on 10s intervals.
5. The deployment will not respect rolling strategy and both the old pods will terminate creating a disruption.

👍 Expected behavior

The deployment should respect deployment strategy rolling and should terminate the pods one by one instead. The logic should auto-detect any HPA objects created either from rawYaml or from KedaAutoscaling and use relevant desired replicas from these objects instead of using replica from deployment-template.

👎 Actual Behavior

The deployment will not respect rolling strategy and both the old pods will terminate creating a disruption. This happens because the new deployment happens with replica=1 which is passed on from the deployment-configuration if an hpa (with the specified format ie appName-envName is not detected).

☸ Kubernetes version

1.26

Cloud provider

AWS

🌍 Browser

Chrome

🧱 Your Environment

No response

✅ Proposed Solution

The hpa detection logic should auto-detect any HPA objects created either from rawYaml or from KedaAutoscaling and use relevant desired replicas from these objects instead of using replica from deployment-template.

👀 Have you spent some time to check if this issue has been raised before?

• I checked and didn't find any similar issue

🏢 Have you read the Code of Conduct?

• I have read the Code of Conduct

AB#3369