Database Explained

A database is a logical container for collections within a project.

Zilliz Cloud supports two types of databases, depending on how they are hosted and accessed.

Database in serving cluster

A cluster database is created in a specific serving cluster. When a serving cluster is created, a default cluster database is automatically created with it. You can create additional cluster databases in the same serving cluster as needed.

A cluster database has full access to all operations — DDL, DML (insert, upsert, delete), and DQL (search, query) — through the serving cluster endpoint.

The lifecycle of a cluster database is tied to its serving cluster:

• If the serving cluster is suspended, all cluster databases and collections in it become unavailable until the cluster is resumed.

• If the serving cluster is dropped, all cluster databases and collections in it are deleted as well.

Cluster databases are suited for production workloads that require always-on, low-latency access to data.

The following diagram shows how projects, serving clusters, databases, and collections are organized.

Project                                                                                                                                                                                                   
   └── Serving Cluster                       
        ├── Database (default)                                                                                                                                                                   
        │    ├── Collection_01
        │    └── Collection_02                                                                                                                                                                              
        │                                                       
        └── Database
             ├── Collection_03                                                                                                                                                                              
             └── Collection_04

Database in on-demand compute

In addition to the cluster database, there is another type of project-level database that is not tied to any cluster. It is managed by the platform and does not require you to provision or maintain a cluster for it. You specify on-demand compute to perform query search on data in this type of database.

This type of databases support the following operations:

Operations	Supported
Create/drop database	Yes
Create/drop collection	Yes
Load/release collection	No need
Search, query	Yes
Import	Yes (Import is only supported for managed collections in on-demand compute databases. For details, see External Collection Limits.)
Insert, upsert, delete	No

This type of database is suited for large-scale datasets with infrequent queries.

Project
 ├── Serving Cluster 
 │    └── Database (default)
 │         ├── Collection_01 
 │         └── Collection_02                                                                                                                                                            
 │                                 
 └── Databases in on-demand compute
      ├── External_Collection_01     
      └── External_Collection_02

Comparison

The following table compares the 2 types of databases.

	Database in Serving Cluster	Database in On-Demand Compute
Best for	Production workloads that require always-on, low-latency access to data.	Large-scale datasets with bursty searches and queries.
Hosted on	User-created serving cluster	Platform-managed
Compute resource	Served by the hosting serving cluster	Served by a specified on-demand cluster
Insert/upsert/delete	Yes	No
Import/Truncate	Yes	Yes
Search and query	Yes	Yes
Lifecycle	Tied to serving cluster	Independent of any cluster

📘Note

Use different connection endpoints for the two types of databases. For details, see Connection Endpoints.

Next steps

• Create an External Collection

• Create a Collection