Advanced Search with Enterprise Server

MongoDB enables AI-driven applications with enterprise-level search capabilities on self-managed infrastructures.

ユースケース: Artificial Intelligence, Intelligent Search

業種: 小売

製品: MongoDB Search with Enterprise Server, MongoDB Vector Search with Enterprise Server, MongoDB Enterprise Server

ソリューション概要

Even as cloud adoption accelerates, many customers opt for self-managed infrastructure and on-prem deployments, particularly in highly regulated industries. For instance, according to this survey, 93% of government bodies use on-prem deployments, followed by 91% of organizations in the IT industry and 90% in the finance industry. Moreover, many SMEs, such as boutique retailers or regional grocery chains, use on-prem deployments as the high cost to adopt cloud architecture is a financial barrier.

While self-managed infrastructure provides flexibility, compliance control, and data sovereignty, it introduces challenges for organizations that are looking to implement AI-powered solutions. Some of these challenges include:

Architectural complexity: Companies often combine separate search engines and vector databases to achieve advanced search capabilities like vector search.
Operational overwhelm: Teams can struggle to synchronize separate technologies, manage varied resources, and maintain consistent workflows across multiple systems.

To address these hurdles, MongoDB introduced the Search and Vector Search with MongoDB Enterprise Server. These search capabilities enable you to:

Build next generation AI apps on any infrastructure: Reduce architectural complexity, which makes it easier to develop AI-powered solutions and eliminates the need for multiple single-purpose technologies.
Enjoy a consistent experience everywhere: Use the same MQL API regardless of where the application is connected. This reduces friction and provides flexibility when moving between production and development environments.

This solution presents a demo which uses MongoDB Search and Vector Search to help users discover products in a grocery store using either full-text search or semantic search powered by vector embeddings. It includes pre-requisites, architecture requirements, and step-by-step instructions to run the demo in an environment with MongoDB Enterprise Server and search nodes.

Figure 1. The demo's landing page

リファレンスアーキテクチャ

To use advanced search capabilities with Enterprise Server, you must deploy MongoDB Search and Vector Search nodes in a Kubernetes cluster. This enables full-text and semantic search powered by vector embeddings, allowing users to find items based on descriptions or semantic similarity.

You don’t need to migrate your existing database. Use MongoDB Enterprise Edition running MongoDB 8.0.10 version or later, either inside or outside a Kubernetes cluster. The Kubernetes operator manages these search nodes and connects them securely to your database cluster, regardless of its deployment location.

In summary, you can use the following approaches to integrate MongoDB Search and Vector Search with Enterprise Server:

MongoDB Enterprise Edition deployed inside Kubernetes
MongoDB Enterprise Edition deployed outside Kubernetes

Both deployment architectures require MongoDB Search and Vector Search nodes deployed within Kubernetes. The following sections describe each approach.

Install Search With MongoDB Enterprise Edition Inside Kubernetes

This deployment option sets up MongoDB Search and Vector Search nodes directly within the Kubernetes cluster alongside MongoDB Enterprise Server instances, as used in the demo. This deployment ensures integrated operations and simplifies management by maintaining all components in the same environment. The image below illustrates this deployment.

Figure 2. MongoDB Enterprise Server inside of Kubernetes

In this solution, the Kubernetes cluster hosts Enterprise Server, the search nodes, and the inventory application. This architecture simplifies management and ensures efficient data exchange between components.

For step-by-step instructions on how to set up search nodes with an internal Enterprise Server configuration, refer to Install and Use Search With MongoDB Enterprise Edition.

Install Search With MongoDB Enterprise Edition Outside Kubernetes

This deployment option configures MongoDB Search and Vector Search nodes inside Kubernetes, which connect to an Enterprise Server replica set deployed outside the Kubernetes cluster. You can use this approach if you have infrastructure requirements or operational preferences outside Kubernetes. The image below illustrates this deployment.

Figure 3. MongoDB Enterprise Server outside of Kubernetes

This configuration allows you to integrate advanced search capabilities without changing your existing Enterprise Server infrastructure.

For step-by-step instructions on how to install MongoDB Search and Vector Search nodes with an external Enterprise Server configuration, refer to Install and Use MongoDB Search and Vector Search With External MongoDB Enterprise Edition.

ソリューションのビルド

Follow the steps below to set up and run the solution.

Deploy MongoDB Search and Vector Search nodes

Follow the steps in the Install and Use Search With MongoDB Enterprise Edition tutorial to deploy MongoDB Search and Vector Search nodes.

Clone the repository

For the necessary data dump and application configuration, download the demo repository:

git clone https://github.com/mongodb-industry-solutions/retail-unified-commerce

Go to the dump file directory:

cd ./retail-unified-commerce/docs/setup

Import the demo data

Import the sample grocery store inventory data into your MongoDB deployment.

Run the following command to retrieve the connection string. Replace <SECRET_NAME> with the appropriate Kubernetes secret name generated during deployment:

kubectl get secret -n ${MDB_NS} <SECRET_NAME> -o yaml

This command prints the secret metadata and its encoded contents. Copy the connectionStringStandard value and execute the following command to get the connection string:

echo "<Base64EncodedConnectionString>" | base64 --decode

Run the following command to restore the database from the collections dump folder:

mongorestore --uri "<CONNECTION_STRING>" ./collections

MongoDB Search およびベクトル検索インデックスの作成

Open mongosh with your connection string to connect to your cluster:

mongosh "<CONNECTION_STRING>"
use retail-unified-commerce

Run the following command to create the MongoDB Search index:

db.products.createSearchIndex("product_atlas_search", {
   "mappings": {
      "dynamic": false,
      "fields": {
         "brand": {
            "type": "string"
         },
         "category": {
            "type": "string"
         },
         "inventorySummary": {
            "fields": {
               "storeObjectId": {
                  "type": "objectId"
               }
            },
            "type": "document"
         },
         "productName": {
            "type": "string"
         },
         "subCategory": {
            "type": "string"
         }
      }
   }
})

Run the following command to create the MongoDB Vector Search index:

db.products.createSearchIndex({
   name: "product_text_vector_index",
   type: "vectorSearch",
   definition: {
      fields: [
         {
            type: "vector",
            path: "textEmbeddingVector",
            numDimensions: 1024,
            similarity: "cosine"
         },
         {
            type: "filter",
            path: "inventorySummary.storeObjectId"
         },
         {
            type: "filter",
            path: "inventorySummary.inStock"
         }
      ]
   }
})

Run the following command to confirm that both indexes were created correctly. If successful, the command returns metadata for product_atlas_search and product_text_vector_index.

JSON.stringify(db.runCommand({listSearchIndexes: "products"}), null, 2)

Test your indexes using the following query:

db.products.aggregate([
   {
      "$search": {
         "index": "product_atlas_search",
         "compound": {
            "should": [
               {
                  "text": {
                     "query": "Chicken Masala",
                     "path": ["productName", "brand", "category", "subCategory"],
                     "fuzzy": { "maxEdits": 2 },
                  }
               }
            ]
         },
      }
   },
   {
      $project: {
         productName: 1,
         score: { $meta: "searchScore" }
      }
   },
   {
      $limit: 2
   }
])

The query returns a list of product names and their search score:

[
   {
      _id: ObjectId("685bfe2d3d832cf7e1614edc"),
      productName: "Chicken Masala",
      score: 8.192384719848633,
   },
   {
      _id: ObjectId("685bfe2e3d832cf7e16159fe"),
      productName: "Shahi Chicken Masala",
      score: 7.7956156730651855,
   },
];

Configure and run the demo

Create a .env file in the backend/advanced-search-ms directory and another in the frontend directory. You can find a .env.template file in these directories. Update the placeholders with the connection string that you retrieved earlier.

Refer to the GitHub repository's README for additional context and optional setup configurations.

You can deploy the demo locally or on Kubernetes.

Run the following code:

make build

Ensure your frontend and backend services are accessible on localhost using your preferred method. Then, open your browser in http://localhost:3000/product-inventory.

Run the backend and frontend inside your Kubernetes cluster using the Docker images built for the demo. Follow these steps:

1. Build and push Docker images

Build and push the backend and frontend images to your container registry by executing the following commands:

#Backend
docker build --platform linux/amd64 -t retail-backend:latest ./backend/advanced-search-ms
docker tag retail-backend:latest <YOUR_REGISTRY>/retail-backend:latest
docker push <YOUR_REGISTRY>/retail-backend:latest
#Frontend
docker build --platform linux/amd64 -t retail-frontend:latest ./frontend
docker tag retail-frontend:latest <YOUR_REGISTRY>/retail-frontend:latest
docker push <YOUR_REGISTRY>/retail-frontend:latest

The --platform linux/amd64 flag may be required depending on your machine’s architecture.

2. Create the Kubernetes secrets

Store sensitive values in a Kubernetes secret file named retail-secrets.yaml within the retail-unified-commerce/k8s folder. Do not commit real credentials to source control. Below is an example template:

# retail-unified-commerce/k8s/retail-secrets.yaml
apiVersion: v1
kind: Secret
metadata:
  name: retail-secrets
  namespace: default
type: Opaque
data:
  NEXT_PUBLIC_GOOGLE_MAPS_API_KEY:
  VOYAGE_API_KEY:
stringData:
  MONGODB_URI: <YOUR_CONNECTION_STRING>
  DB_NAME: "retail-unified-commerce"
  NEXT_PUBLIC_COLLECTION_PRODUCTS: "products"
  NEXT_PUBLIC_COLLECTION_INVENTORY: "inventory"
  NEXT_PUBLIC_COLLECTION_STORES: "stores"
  SEARCH_INDEX: "product_atlas_search"
  NEXT_PUBLIC_BACKEND_ENDPOINT: "<YOUR_BACKEND_ENDPOINT>"
  NEXT_PUBLIC_ENABLE_ATLAS_SEARCH: "true"
  NEXT_PUBLIC_ENABLE_VECTOR_SEARCH: "true"
  NEXT_PUBLIC_ENABLE_HYBRID_SEARCH: "true"
  NEXT_PUBLIC_ENABLE_FULLTEXT_SEARCH: "true"
  MONGODB_DATABASE: "retail-unified-commerce"
  PRODUCTS_COLLECTION: "products"
  SEARCH_TEXT_INDEX: "product_atlas_search"
  SEARCH_VECTOR_INDEX: "product_text_vector_index"
  NEXT_PUBLIC_SEARCH_META_INDEX: "product_atlas_search_meta"
  EMBEDDING_FIELD_NAME: "textEmbeddingVector"
  VOYAGE_API_URL: "https://api.voyageai.com/v1"
  VOYAGE_MODEL: "voyage-3-large"

Apply the secrets to your cluster:

kubectl apply -f k8s/retail-secrets.yaml

3. Deploy the backend and the frontend

Place the backend-deployment.yaml template in your /k8s directory:

# retail-unified-commerce/k8s/backend-deployment.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: retail-backend
spec:
  replicas: 2
  selector:
    matchLabels:
      app: retail-backend
  template:
    metadata:
      labels:
        app: retail-backend
    spec:
      containers:
      - name: backend
        image: <YOUR_REGISTRY>/retail-backend:latest
        ports:
        - containerPort: 8000
        envFrom:
        - secretRef:
          name: retail-secrets
---
apiVersion: v1
kind: Service
metadata:
  name: retail-backend-service
spec:
  type: LoadBalancer
  selector:
    app: retail-backend
  ports:
  - port: 80
    targetPort: 8000

Place the frontend-deployment.yaml template in your /k8s directory:

# retail-unified-commerce/k8s/frontend-deployment.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: retail-frontend
spec:
  replicas: 2
  selector:
    matchLabels:
      app: retail-frontend
  template:
    metadata:
      labels:
        app: retail-frontend
    spec:
      containers:
      - name: frontend
        image: <YOUR_REGISTRY>/retail-frontend:latest
        ports:
        - containerPort: 3000
        envFrom:
        - secretRef:
          name: retail-secrets
---
apiVersion: v1
kind: Service
metadata:
  name: retail-frontend-service
spec:
  type: LoadBalancer
  selector:
    app: retail-frontend
  ports:
  - port: 80
    targetPort: 3000

Apply your Kubernetes deployments:

kubectl apply -f k8s/backend-deployment.yaml
kubectl apply -f k8s/frontend-deployment.yaml

Verify your deployments:

kubectl get pods -o wide
kubectl get svc -o wide

Wait until the frontend and backend services have an EXTERNAL-IP assigned.

4. Access the demo

Once the external IP appears for the frontend service, open it by using the following link: http://<EXTERNAL-IP>/product-inventory.

Explore the demo’s search capabilities

Select the MongoDB Search option and enter queries that don't exactly match product names. For example, use a term like green tea face mask or a typo like cukie. These queries demonstrate the system’s ability to return relevant results when the input is imprecise.

Figure 4. Search result for green tea face mask

Switch to the MongoDB Vector Search option to experience semantic understanding. Use descriptive or intent-based queries such as a gift for a teen girl or high-protein spicy snack. The system shows results aligned with the meaning of your query, not just the words used.

Figure 5. Vector search result for gift for a teen girl

In both results, you can view the search score displayed on the top-left corner of each product card. This score indicates the relevance of the result to your query.

キーポイント

On-premises adoption holds strategic importance: On-premises deployments remain critical for industries and organizations that prioritize data sovereignty and control. This public preview enables organizations to adopt AI in their on-premises environments, ensuring they stay competitive and meet compliance requirements.
Simplified tech stack accelerates AI adoption: MongoDB Search and Vector Search with Enterprise Server enable powerful AI applications by consolidating diverse technologies, reducing architectural complexity, and allowing smart search queries without migrating operational data.
Advanced search capabilities enhance user experience: Full-text search and vector search refine how users explore and access information, driving more intuitive experiences across different industries.

作成者

Angie Guemes Estrada, MongoDB
Rodrigo Leal, MongoDB
Prashant Juttukonda, MongoDB

詳細

To deploy MongoDB Search and Vector Search in your Kubernetes cluster, visit Deploy MongoDB Search and Vector Search.
To learn about search and vector search capabilities for use with MongoDB Community Edition and MongoDB Enterprise Server, read the Supercharge Self-Managed Apps With Search and Vector Search Capabilities.
MongoDB が小売のイノベーションを実現する方法については、 MongoDB for Retailをご覧ください。

戻る

統合トランザクション

製品説明の自動化