Closed oliverbusch closed 1 week ago
Hi @oliverbusch, Kubernetes cluster created by kind should not be restarted. If you somehow restart the cluster, it may not work as expected, and you need to re-create the cluster. Please see below references: kubernetes-sigs/kind#148 https://discuss.kubernetes.io/t/restart-a-kind-cluster/5413/4
Hello @hemantxpatel, I use minikube for the kubernetes cluster. And from my point of view the error is not related to the restart. The error mesaage
Message : ERROR: column "expiration_timestamp" of relation "edc_edr_cache" already exists
The message indicates an error in the flyway setup.
Hi @oliverbusch , this issue is best suited for tractusx-edc. We don't configure anything related to flyway. DB Migration scripts are part of EDC internally and we don't have any control / configurations on it. I found a similar old issue related to EDR eclipse-tractusx/tractusx-edc#792
Ok, I will monitor https://github.com/eclipse-tractusx/tractusx-edc/issues/792.
Current Behavior
After restarting Kubernetes, the alice-tractusx-connector-controlplane cannot be started due to a flyway error.
Expected Behavior
After restarting Kubernetes, all containers should start without any problems.
Steps To Reproduce
Restart the kubernetes cluster Main Branch (pull date 05.06.24)