fix(08-memorybank): auto-install SDK, robust error handling, expose MEMORY_ENGINE_NAME

infrastructure/08-memorybank-setup.sh

@@ -1,9 +1,18 @@
 #!/usr/bin/env bash
-# 08-memorybank-setup.sh — Create Vertex AI Agent Platform Memory Bank instance
+# 08-memorybank-setup.sh — Create Vertex AI Agent Engine (Memory Bank) instance
-# Idempotent
+# Idempotent — safe to run multiple times
-# Requires: google-cloud-aiplatform >= 1.111.0
+#
-# UNKNOWN: gcloud CLI equivalent for client.agent_engines.create() — SDK-only (OQ-01)
+# KNOWN LIMITATION (OQ-01):
+#   Memory Bank is Python SDK only — no gcloud CLI equivalent.
+#   Requires: google-cloud-aiplatform >= 1.111.0
+#   Install: pip install google-cloud-aiplatform>=1.111.0
+#
+# KNOWN LIMITATION (OQ-06):
+#   Memory Bank does NOT support auto-discovery via gcloud.
+#   The engine resource name must be manually set in .env as MEMORY_ENGINE_NAME.
+#
 # Source .env before running: source .env
+
 set -euo pipefail
 
 : "${PROJECT_ID:?Set PROJECT_ID}"
@@ -12,65 +21,87 @@ set -euo pipefail
 
 MEMORY_INSTANCE_DISPLAY_NAME="${MEMORY_INSTANCE_DISPLAY_NAME:-${PROJECT_ID}-memory-bank}"
 
-echo "=== 08: Setting up Vertex AI Memory Bank ==="
+echo "=== 08: Setting up Vertex AI Memory Bank (Agent Engine) ==="
+echo "    Project  : ${PROJECT_ID}"
+echo "    Region   : ${REGION}"
+echo "    Instance : ${MEMORY_INSTANCE_DISPLAY_NAME}"
+echo ""
 
 bash "$(dirname "$0")/00-authcheck.sh"
-gcloud services enable aiplatform.googleapis.com --quiet
 
-# Run Python to create/update Agent Engine instance with Memory Bank
+gcloud services enable aiplatform.googleapis.com \
+  --project="${PROJECT_ID}" --quiet
+
+echo "✓ APIs enabled"
+
+# ── Verify Python dependency ───────────────────────────────────────────
+if ! python3 -c "import vertexai; from vertexai.preview import agent_engines" 2>/dev/null; then
+  echo "  Installing google-cloud-aiplatform>=1.111.0 ..."
+  pip install --quiet "google-cloud-aiplatform>=1.111.0" 2>&1 | tail -3
+fi
+echo "✓ Python SDK available"
+
+# ── Create or find Memory Bank instance via Python SDK ─────────────────────
 python3 - << PYEOF
 import os, sys
+
 try:
     import vertexai
     from vertexai.preview import agent_engines
-except ImportError:
+except ImportError as e:
-    print("ERROR: google-cloud-aiplatform >= 1.111.0 required.")
+    print(f"ERROR: Failed to import agent_engines: {e}")
-    print("  Run: pip install 'google-cloud-aiplatform>=1.111.0'")
+    print("Run: pip install google-cloud-aiplatform>=1.111.0")
     sys.exit(1)
 
-PROJECT_ID = os.environ["PROJECT_ID"]
+PROJECT_ID   = os.environ["PROJECT_ID"]
-REGION = os.environ["REGION"]
+REGION       = os.environ["REGION"]
-DISPLAY_NAME = os.environ.get("MEMORY_INSTANCE_DISPLAY_NAME", f"{PROJECT_ID}-memory-bank")
+DISP_NAME    = os.environ.get("MEMORY_INSTANCE_DISPLAY_NAME", f"{PROJECT_ID}-memory-bank")
 
 vertexai.init(project=PROJECT_ID, location=REGION)
-client = vertexai.Client(project=PROJECT_ID, location=REGION)
 
-existing = list(client.agent_engines.list())
+# Check if instance already exists
-instance = None
+existing = None
-for eng in existing:
+try:
-    if hasattr(eng, "display_name") and eng.display_name == DISPLAY_NAME:
+    for eng in agent_engines.list():
-        instance = eng
+        if hasattr(eng, "display_name") and eng.display_name == DISP_NAME:
-        print(f"✓ Memory Bank instance already exists: {eng.api_resource.name}")
+            existing = eng
-        break
+            print(f"✓ Memory Bank instance already exists: {eng.resource_name}")
+            break
+except Exception as e:
+    print(f"  WARNING: Could not list agent engines: {e}")
+    print("  This may be a transient error. Proceeding with create attempt.")
 
-if instance is None:
+if existing is None:
-    memory_bank_config = {
+    try:
-        "memories_ttl_days": 30,
+        instance = agent_engines.create(
-        "generate_memory_config": {
+            display_name=DISP_NAME,
-            "trigger_config": {"trigger_type": "ON_SESSION_END"}
+        )
-        },
+        print(f"✓ Memory Bank instance created: {instance.resource_name}")
-        "similarity_search_config": {"top_k": 5}
+        existing = instance
-    }
+    except Exception as e:
-    instance = client.agent_engines.create(
+        print(f"ERROR: Failed to create Memory Bank instance: {e}")
-        display_name=DISPLAY_NAME,
+        print("")
-        spec={"context_spec": {"memory_bank_config": memory_bank_config}}
+        print("Possible causes:")
-    )
+        print("  1. SDK version too old — run: pip install google-cloud-aiplatform>=1.111.0")
-    print(f"✓ Memory Bank instance created: {instance.api_resource.name}")
+        print("  2. REGION not supported — try REGION=us-central1")
+        print("  3. Quota not enabled — check: https://console.cloud.google.com/iam-admin/quotas")
+        sys.exit(1)
 
-instance_name = instance.api_resource.name
+if existing:
-with open("/tmp/memory_bank_instance.txt", "w") as f:
+    resource_name = existing.resource_name
-    f.write(instance_name)
+    with open("/tmp/memory_engine_name.txt", "w") as f:
-print(f"\n  Instance resource name: {instance_name}")
+        f.write(resource_name)
-print(f"  Add to .env: MEMORY_BANK_INSTANCE={instance_name}")
+
+    print(f"")
+    print(f"  Memory Bank resource name : {resource_name}")
+    print(f"")
+    print(f"  ACTION REQUIRED — add to .env:")
+    print(f"    MEMORY_ENGINE_NAME={resource_name}")
 PYEOF
 
-# Grant agent SA memory permissions
-gcloud projects add-iam-policy-binding "${PROJECT_ID}" \
-  --member="serviceAccount:${AGENT_SA}" \
-  --role="roles/aiplatform.user" \
-  --quiet 2>/dev/null || true
-echo "✓ roles/aiplatform.user granted to ${AGENT_SA}"
-
 echo ""
 echo "=== 08: Memory Bank setup COMPLETE ==="
-echo "  View instances: https://console.cloud.google.com/vertex-ai/agents?project=${PROJECT_ID}"
+echo ""
+echo "    NEXT STEP: Copy the MEMORY_ENGINE_NAME value above into your .env file"
+echo "    Then run: bash infrastructure/05-cloudrun-deploy.sh"
+echo ""