Serverless

Background

Databases are critical to modern IT systems. Creating a database requires careful resource configuration — CPU, memory, storage, and connections — to handle both peak and off-peak hours. Fixed provisioning wastes resources during low demand and risks insufficient capacity during spikes. Serverless databases solve this by automatically scaling resources based on current workload, eliminating complex capacity planning and O&M overhead.

The following figure compares resource usage between common clusters and Serverless clusters under fluctuating workloads.

Key differences under fluctuating workloads:

• Common clusters: Resources are wasted during off-peak periods and insufficient during peak periods, which impacts business continuity.

• Serverless clusters:

  • Adjust specifications based on demand, reducing resource waste and improving utilization.

  • Scale cluster resources quickly during peak hours to ensure business continuity and system stability.

  • Replace fixed-resource billing with pay-as-you-go, dynamically matching resources to workloads for significant cost savings.

  • Provide elastic scaling optimized for high-throughput writes and high concurrency, suitable for large data volumes and fluctuating access patterns.

  • Eliminate manual configuration adjustments, improving O&M efficiency and reducing labor costs.

Overview

The Serverless feature provides real-time elasticity for CPU, memory, storage, and network resources with vertical resource isolation for network resources, namespaces, and storage space. On-demand billing for compute and storage lets you independently adjust capacity to match business changes, optimizing costs and efficiency.

Types

Serverless feature for a cluster with defined specifications	Serverless cluster
Database proxy The database proxy has two parts: defined specifications (default for fixed-spec clusters) and Serverless. The Serverless part scales elastically based on workload. By default, scaling occurs in increments of 0.5 PCU. The scaling increment is dynamically adjusted based on current PCU usage. A higher PCU usage results in a larger scaling increment. Compute nodes The primary node (RW node) and read-only nodes (RO nodes) include fixed-specification resources and Serverless resources. Fixed-specification resources do not scale; Serverless resources scale elastically based on workload. When the primary node or read-only nodes scale out or in, the number of PCUs for the node increases or decreases accordingly. By default, scaling occurs in increments of 0.5 PCU. The scaling increment is dynamically adjusted based on current PCU usage. A higher PCU usage results in a larger scaling increment. You can set the elastic scaling range for a single node in PCUs. The system monitors the PCU of a compute node every second. Storage space The storage of the cluster with defined specifications is used. . Note After you enable the Serverless feature for a cluster with defined specifications, the maximum number of connections and the maximum IOPS for the cluster are proportional to the value of the Serverless Maximum Resources for Single Node parameter.	Database Proxy The database proxy is a Serverless service. Its resources are independent of compute nodes, and elastic scaling is automatic. By default, scaling occurs in increments of 0.5 PCU. The scaling increment is dynamically adjusted based on current PolarDB Capacity Unit (PCU) usage. A higher PCU usage results in a larger scaling increment. Compute nodes Primary nodes (RW nodes) and read-only nodes (RO nodes) are all Serverless, scaling elastically based on the workload and using single-zone shared storage. When the primary node or read-only nodes scale out or in, the number of PCUs for the node increases or decreases accordingly. By default, scaling occurs in increments of 0.5 PCU. The scaling increment is dynamically adjusted based on current PCU usage. A higher PCU usage results in a larger scaling increment. You can set the elastic scaling range for a single node in PCUs. The system monitors the PCU of a compute node every second. Storage space Storage is pay-as-you-go. You do not need to select capacity at purchase. Storage scales out automatically as data grows, and you pay only for actual usage. View Database Storage Usage on the Basic Information page of your cluster. . Note A serverless cluster supports a maximum of 100,000 connections and a maximum IOPS of 84,000.

Benefits

Serverless dynamically scales cluster resources in seconds based on workload. Core benefits:

• High availability

  Multi-node architecture ensures high availability and stability of Serverless clusters.

• High elasticity

  • Wide scaling range: Supports automatic vertical and horizontal scaling.

  • Scaling within seconds: Detects workload spikes in 5 seconds and completes a scale-out in 1 second. When workload decreases, resources are released in tiers.

• Strong data consistency

  Supports to ensure data written to the cluster is immediately readable on read-only nodes, with performance nearly identical to weak consistency.

  Note

  Global consistency is disabled by default. You can enable it for cluster endpoints. .

• Cost-effectiveness

  Serverless clusters are billed in PCUs on a pay-as-you-go basis. This can reduce your costs by up to 80%.

• Fully managed

  Alibaba Cloud handles all O&M work — version upgrades, system deployments, scaling, and alert processing — without affecting your services. This delivers a fully managed experience that lets you focus on your business.

Use cases

Supported versions

Limitations

Serverless clusters do not support custom cluster endpoints, manually adding nodes, or manual upgrades and downgrades.

Billing

• Serverless clusters

  Fees include compute node fees, storage fees, backup storage fees (charged only for usage exceeding the free quota), and SQL Explorer fees (optional). .

• Serverless-enabled clusters with defined specifications

  Fees include charges for the cluster with defined specifications and for the serverless feature. For cluster billing, see . For Serverless feature billing, see .

Video introduction