All Products
Search
Document Center

Hologres:Select an instance type for Hologres

Last Updated:Feb 28, 2026

Hologres offers five instance types, each designed for different data volume, concurrency, and resource isolation requirements. This topic describes each instance type and the scenarios it supports.

Which instance type to use

Use the following guidance to determine which instance type fits your requirements. Virtual warehouse instances cover the broadest range of production scenarios and are the default recommendation.

Virtual warehouse instance (recommended)

A virtual warehouse instance provides dedicated computing and storage resources. A single instance supports a read/write splitting architecture through virtual warehouses, without requiring separate instances.

Choose this when:

  • Data volumes are in the terabyte (TB) or petabyte (PB) range and you run frequent complex queries.

  • You run high-concurrency online services with millisecond-level responses (over 1,000 QPS).

  • You need payload isolation between workloads, such as separating Online Analytical Processing (OLAP) queries from real-time writes, or isolating resources across lines-of-business.

  • Traffic fluctuates significantly with periodic peaks that require rapid scaling.

For more information, see Virtual warehouses.

Serverless instance

A serverless instance uses shared computing resources with dedicated storage resources for high-performance, large-scale, and low-frequency data manipulation and queries. You do not need to reserve computing capacity in advance.

Choose this when:

  • You are running development or staging environments.

  • You have low-frequency queries that still require high performance when they run.

  • Business payloads fluctuate significantly, peaks are hard to predict, and you cannot reserve dedicated computing resources in advance.

For more information, see Serverless instances.

General-purpose instance

A general-purpose instance provides dedicated computing and storage resources for cost control and lightweight analytics. General-purpose instances can be upgraded to virtual warehouse instances as requirements grow.

Choose this when:

  • Data volume is less than 1 TB.

  • You run lightweight analytics such as low-frequency ad hoc queries or small- to medium-scale business intelligence (BI) report generation.

  • You have limited budgets and need a cost-effective solution, or you are in a trial phase for non-core services.

  • Concurrency requirements are low, such as internal data service platforms or scheduled task processing.

Instance type comparison

The following table compares the active instance types across five capability dimensions.

Feature

Virtual warehouse instance

Serverless instance

General-purpose instance

Seamless scaling

Supported

N/A

Not supported

Time-based elasticity

Supported

N/A

Not supported

Payload isolation

Supported

Supported

Not supported

Automatic throttling

Supported

Supported

Not supported

Serverless computing

Supported

Supported

Supported

Legacy instance types (to be discontinued)

The following instance types are scheduled for discontinuation. If you currently use one of these types, migrate to a virtual warehouse instance.

Read-only replica instance

A read-only replica instance provides dedicated computing and storage resources and operates by attaching to a general-purpose instance to create a multi-instance read/write splitting architecture. The general-purpose primary instance handles data writes and data manipulation, while the read-only replica instance handles data analytics.

Note

Read-only replica instances are to be discontinued. Use virtual warehouse instances instead, which support read/write splitting within a single instance.

Legacy comparison attributes:

Feature

Primary/replica instance

Seamless scaling

Not supported

Time-based elasticity

Not supported

Payload isolation

Supported

Automatic throttling

Not supported

Serverless computing

General-purpose instance: Supported / Read-only replica instance: Not supported

Shared cluster (Lakehouse Acceleration Edition)

A shared cluster uses shared computing resources and does not provide storage resources. No need to read from or write to Hologres internal tables.

Previous use cases:

  • Accelerating queries on data stored in OSS or MaxCompute through Hologres.

  • Running large burst queries that exceed the resources of reserved dedicated instances.

  • Connecting BI tools such as Tableau, Quick BI, and FineReport to accelerate queries on data lakehouse data.

Note

Shared clusters (Lakehouse Acceleration Edition) are to be discontinued. Use virtual warehouse instances instead.

For more information, see Shared cluster (Lakehouse Acceleration Edition).