The ADK supports different types of session state persistence. We currently use the InMemory implementation.
With this, state is lost when agents are restarted. Also, we can not scale agents beyond one instance.
We need a kubernetes-native way to configure the kind of storage and wire it into the agent container.