#!/usr/bin/env bash
# 07-rag-setup.sh — Create Vertex AI RAG Engine corpus in Serverless mode
# Serverless mode does NOT require allowlist (unlike Spanner mode)
# Idempotent — safe to run multiple times
# Source .env before running: source .env

set -euo pipefail

: "${PROJECT_ID:?Set PROJECT_ID}"
: "${REGION:?Set REGION}"
: "${RAG_CORPUS_DISPLAY_NAME:?Set RAG_CORPUS_DISPLAY_NAME}"

RAG_REGION="${RAG_REGION:-${REGION}}"

echo "=== 07: Setting up Vertex AI RAG Engine (Serverless mode) ==="
echo "    Project  : ${PROJECT_ID}"
echo "    Region   : ${RAG_REGION}"
echo "    Corpus   : ${RAG_CORPUS_DISPLAY_NAME}"
echo ""

bash "$(dirname "$0")/00-authcheck.sh"

gcloud services enable aiplatform.googleapis.com \
  storage.googleapis.com \
  --project="${PROJECT_ID}" --quiet
echo "✓ APIs enabled"

# ── GCS bucket for corpus source documents (idempotent) ──────────────────
CORPUS_BUCKET="${PROJECT_ID}-agent-corpus"
if ! gsutil ls -b "gs://${CORPUS_BUCKET}" &>/dev/null; then
  gsutil mb -l "${REGION}" -b on "gs://${CORPUS_BUCKET}"
  echo "✓ GCS corpus bucket created: gs://${CORPUS_BUCKET}"
else
  echo "✓ GCS corpus bucket exists: gs://${CORPUS_BUCKET}"
fi

# ── Upload seed documents if present ─────────────────────────────────────
SEED_DIR="$(dirname "$0")/../docs/corpus-seed"
if [[ -d "${SEED_DIR}" ]] && ls "${SEED_DIR}"/*.md &>/dev/null; then
  gsutil -m cp "${SEED_DIR}"/*.md "gs://${CORPUS_BUCKET}/seed/" 2>/dev/null || true
  echo "✓ Seed documents uploaded to gs://${CORPUS_BUCKET}/seed/"
else
  echo "  No seed documents found in docs/corpus-seed/ — skipping upload"
fi

# ── Create corpus in Serverless mode ─────────────────────────────────────
python3 - << PYEOF
import os, sys

try:
    import vertexai
    from vertexai.preview import rag
except ImportError:
    print("ERROR: google-cloud-aiplatform not installed.")
    print("Run: pip install google-cloud-aiplatform>=1.87.0")
    sys.exit(1)

PROJECT_ID   = os.environ["PROJECT_ID"]
RAG_REGION   = os.environ.get("RAG_REGION", os.environ["REGION"])
DISPLAY_NAME = os.environ["RAG_CORPUS_DISPLAY_NAME"]
CORPUS_BUCKET = f"{PROJECT_ID}-agent-corpus"

vertexai.init(project=PROJECT_ID, location=RAG_REGION)

# Check for existing corpus
corpus = None
try:
    for c in rag.list_corpora():
        if c.display_name == DISPLAY_NAME:
            corpus = c
            print(f"✓ RAG corpus already exists: {c.name}")
            break
except Exception as e:
    print(f"WARNING: Could not list corpora: {e}")

if corpus is None:
    print(f"  Creating corpus in Serverless mode in {RAG_REGION}...")
    embedding_config = rag.EmbeddingModelConfig(
        publisher_model="publishers/google/models/text-embedding-005"
    )
    # Serverless mode: RagVectorDbConfig with rag_managed_db (no Spanner)
    vector_db = rag.RagVectorDbConfig(
        rag_managed_db=rag.RagManagedDb()
    )
    corpus = rag.create_corpus(
        display_name=DISPLAY_NAME,
        embedding_model_config=embedding_config,
        vector_db=vector_db,
    )
    print(f"✓ RAG corpus created (Serverless): {corpus.name}")

# Import seed documents (non-fatal)
gcs_uri = f"gs://{CORPUS_BUCKET}/seed/"
try:
    rag.import_files(
        corpus_name=corpus.name,
        paths=[gcs_uri],
        chunk_size=512,
        chunk_overlap=50,
        max_embedding_requests_per_min=900,
    )
    print(f"✓ Documents imported from {gcs_uri}")
except Exception as e:
    print(f"  WARNING: Document import skipped: {e}")

with open("/tmp/rag_corpus_name.txt", "w") as f:
    f.write(corpus.name)

print(f"")
print(f"  Corpus resource name : {corpus.name}")
print(f"  Region               : {RAG_REGION}")
print(f"")
print(f"  ACTION REQUIRED — add to .env:")
print(f"    export RAG_CORPUS_NAME=\"{corpus.name}\"")
PYEOF

echo ""
echo "=== 07: RAG Engine setup COMPLETE ==="
echo "    View: https://console.cloud.google.com/vertex-ai/rag?project=${PROJECT_ID}"
echo ""