fix(rag): use ragManagedDbConfig.basic in corpus body; fix engine config PATCH with scaled→basic switch

The v1beta1 API uses ragManagedDbConfig (not vectorDbConfig) on the RagCorpus
resource itself. Omitting it causes the engine to default to Spanner. Also the
engine config PATCH must use {"ragManagedDbConfig":{"basic":{}}} with the
correct field name matching v1beta1 schema.

agents/rag/setup_corpus.py

@@ -10,9 +10,10 @@ REST API directly for corpus creation, then use the SDK for everything else.
 Key design decisions:
 - ragEngineConfig PATCH always targets us-central1 — it is a project-level
   control-plane endpoint that does NOT exist in other regions.
-- Corpus creation payload omits vectorDbConfig entirely. Sending
+- Corpus creation uses ragManagedDbConfig.basic in the RagCorpus body.
-  vectorDbConfig.ragManagedDb forces Spanner mode which is allowlist-restricted
+  The v1beta1 API uses ragManagedDbConfig (not vectorDbConfig) on the corpus
-  for new projects. Omitting it honours the project-level basic (serverless) config.
+  resource. Sending it explicitly with {"basic":{}} forces serverless mode
+  and overrides any project-level Spanner default.
 """
 
 import os
@@ -46,15 +47,18 @@ def get_token() -> str:
 def ensure_serverless_engine_config() -> None:
     """
     Set project-level RAG Engine Config to basic (serverless) tier.
-    Always targets us-central1 — the ragEngineConfig endpoint is global/control-plane
+    Always targets us-central1. The endpoint does not exist in other regions.
-    and does not exist in other regions (patching europe-west1 etc. returns 'Invalid
+
-    endpoint name').
+    The v1beta1 ragEngineConfig schema:
+      { "ragManagedDbConfig": { "basic": {} } }   <- serverless
+      { "ragManagedDbConfig": { "scaled": {} } }  <- Spanner / managed DB
     """
     print("Ensuring RAG Engine Config is set to serverless (basic) tier...")
     endpoint = (
         f"https://{_ENGINE_CONFIG_LOCATION}-aiplatform.googleapis.com/v1beta1"
         f"/projects/{PROJECT_ID}/locations/{_ENGINE_CONFIG_LOCATION}/ragEngineConfig"
     )
+    # Explicitly set basic and unset spanner by sending only basic.
     payload = json.dumps({"ragManagedDbConfig": {"basic": {}}})
     token = get_token()
 
@@ -69,7 +73,7 @@ def ensure_serverless_engine_config() -> None:
 
     if "error" in resp:
         err = resp["error"]
-        print(f"  ⚠️  Engine config warning ({err.get('code')}): {err.get('message')} — continuing.")
+        print(f"  \u26a0\ufe0f  Engine config warning ({err.get('code')}): {err.get('message')} \u2014 continuing.")
         return
 
     op_name = resp.get("name", "")
@@ -89,10 +93,10 @@ def ensure_serverless_engine_config() -> None:
             if op.get("done"):
                 break
         else:
-            print("  ⚠️  Operation timed out — continuing anyway.")
+            print("  \u26a0\ufe0f  Operation timed out \u2014 continuing anyway.")
             return
 
-    print("  ✓ RAG Engine Config set to basic tier.")
+    print("  \u2713 RAG Engine Config set to basic tier.")
 
 
 def create_corpus_rest() -> str:
@@ -100,10 +104,10 @@ def create_corpus_rest() -> str:
     Create corpus via REST API directly, bypassing SDK backend_config bug.
     Returns the corpus resource name.
 
-    IMPORTANT: vectorDbConfig is intentionally omitted from the payload.
+    Uses ragManagedDbConfig.basic explicitly in the corpus body.
-    Sending vectorDbConfig.ragManagedDb explicitly requests Spanner mode,
+    The v1beta1 RagCorpus resource accepts ragManagedDbConfig directly
-    which is allowlist-restricted for new projects. Omitting it causes the
+    (not vectorDbConfig). Sending {"basic":{}} forces serverless mode
-    API to honour the project-level ragEngineConfig (basic/serverless).
+    regardless of the project-level engine config default.
     """
     url = (
         f"https://{LOCATION}-aiplatform.googleapis.com/v1beta1"
@@ -115,8 +119,13 @@ def create_corpus_rest() -> str:
             "vertexPredictionEndpoint": {
                 "model": "publishers/google/models/text-embedding-004"
             }
+        },
+        # Explicitly request serverless (basic) mode at the corpus level.
+        # The v1beta1 API uses ragManagedDbConfig on the RagCorpus body,
+        # NOT vectorDbConfig. This overrides any Spanner project default.
+        "ragManagedDbConfig": {
+            "basic": {}
         }
-        # vectorDbConfig intentionally omitted — see docstring above.
     })
     token = get_token()
     result = subprocess.run(
@@ -163,7 +172,7 @@ def get_or_create_corpus() -> RagCorpus:
 
     print(f"Creating corpus '{CORPUS_DISPLAY_NAME}' in {LOCATION} via REST...")
     corpus_name = create_corpus_rest()
-    print(f"✓ Corpus created: {corpus_name}")
+    print(f"\u2713 Corpus created: {corpus_name}")
 
     # Re-fetch via SDK so we get a proper RagCorpus object
     for c in rag.list_corpora():
@@ -185,7 +194,7 @@ def import_documents(corpus: RagCorpus) -> None:
             )
         ),
     )
-    print("✓ Import complete.")
+    print("\u2713 Import complete.")
 
 
 def test_retrieval(corpus: RagCorpus) -> None:
@@ -195,7 +204,7 @@ def test_retrieval(corpus: RagCorpus) -> None:
         text="test query",
         rag_retrieval_config=rag.RagRetrievalConfig(top_k=3),
     )
-    print(f"✓ Test retrieval returned {len(response.contexts.contexts)} chunk(s).")
+    print(f"\u2713 Test retrieval returned {len(response.contexts.contexts)} chunk(s).")
 
 
 def main() -> None: