Using VertexAiSessionService with Cloud Run (or GKE) deployed agent

[jnuvola]

The ADK Cloud Run deployment docs lists --agent_engine_id as an option for adk deploy cloud_run, implying I can deploy an agent to Cloud Run while using Vertex AI Agent Engine Sessions as a managed session store.

However, obtaining an agent_engine_id currently requires creating an "empty" Agent Engine instance (via agent_engines.create()) or deploying an agent to Agent Engine runtime, which is not explained on Cloud Run page and feels confusing and unintuitive.

Request: Please document a preferred, production-grade workflow for Cloud Run (or any other runtime) + VertexAiSessionService setup.

[aleks-andr]

I believe this same issue also applies to using VertexAiMemoryBankService. That also requires an existing Agent Engine ID. For sessions there's technically still the option of using a SQL database instead of Vertex, but for Agent Memory the only way to get persistence is through Vertex AI 😬