On-Demand DQL Operations

Public Preview

DQL operations in collections for on-demand computing, such as search, query, get, and hybrid search, require attaching compute resources from an on-demand cluster. Zilliz Cloud allows you to create a session to meet your on-demand compute needs.

This article assumes that you have created a collection in a database using the project endpoint. For details, refer to Create an External Collection.

Connect to a project endpoint

A project endpoint is designed to provide your access to on-demand compute resources. You can use it to manage on-demand clusters and databases, as well as manipulate data stored in collections.

The following code example assumes that you have an external collection named my_collection in the default database. And you should always use a valid API key with sufficient permissions to set up the connection.

client = MilvusClient(
    uri="https://{project-id}.{region}.vectordb.zillizcloud.com",
    token="YOUR_API_KEY"
)

client.has_collection(
    collection_name="my_collection"
)

const client = new MilvusClient({
    address: "https://{project-id}.{region}.vectordb.zillizcloud.com",
    token: "YOUR_API_KEY"
});

client.has_collection({
    collection_name: "my_collection"
});

export PROJECT_ENDPOINT='https://{project-id}.{region}.vectordb.zillizcloud.com'
export TOKEN="YOUR_API_KEY"

curl --request POST \
--url "${CLUSTER_ENDPOINT}/v2/vectordb/collections/has" \
--header "Authorization: Bearer ${TOKEN}" \
--header "Content-Type: application/json" \
-d '{
    "collectionName": "my_collection"
}'

Create a session

Once you set up a connection to the project endpoint, create a session to attach the compute resource from a specified on-demand cluster.

The following example assumes that you have already created an on-demand cluster, whose ID is inxx-xxxxxxxxxxxxxxxxx.

📘Notes

For RESTful requests, instead of creating a session, you should pass the cluster ID as a query parameter to DQL calls.

session = client.session(
    cluster_id="inxx-xxxxxxxxxxxxxxxxx"
)

const session = client.session("inxx-xxxxxxxxxxxxxxxxx");

export CLUSTER_ID="inxx-xxxxxxxxxxxxxxxxx"

Conduct DQL operations

Once the session is ready, you can conduct searches. The following example uses a basic vector search as an example. This also applies to query, get, and hybrid search.

query_vector = [0.3580376395471989, -0.6023495712049978, 0.18414012509913835, -0.26286205330961354, ..., 0.9029438446296592]
res = session.search(
    db_name="my_database",
    collection_name="my_collection",
    anns_field="vector",
    data=[query_vector],
    limit=3,
    output_fields=["product_id", "title", "main_category", "price", "average_rating", "rating_number"],
    search_params={"metric_type": "COSINE"}
)

const query_vector = [0.3580376395471989, -0.6023495712049978, 0.18414012509913835, -0.26286205330961354, ..., 0.9029438446296592];
const res = session.search({
    db_name: "my_database",
    collection_name: "my_collection",
    anns_field: "vector",
    data: [query_vector],
    limit: 3,
    output_fields: ["product_id", "title", "main_category", "price", "average_rating", "rating_number"],
});

curl --request POST \
--url "${PROJECT_ENDPOINT}/v2/vectordb/entities/search?cluster_id=inxx-xxxxxxxxxxxxxxxxx" \
--header "Authorization: Bearer ${TOKEN}" \
--header "Content-Type: application/json" \
-d '{
    "dbName": "my_database",
    "collectionName": "my_collection",
    "data": [
        [
            0.3580376395471989,
            -0.6023495712049978,
            0.18414012509913835,
            -0.26286205330961354,
            0.9029438446296592
        ]
    ],
    "annsField": "vector",
    "limit": 3,
    "outputFields": [
        "product_id",
        "title",
        "main_category",
        "price",
        "average_rating",
        "rating_number"
    ]
}'

Close a session

Once your on-demand computing tasks are complete, you can close the session. A closed session cannot be used for further DQL operations.

📘Notes

RESTful calls do not need this.

session.close()

session.close();