All Products
Search
Document Center

Realtime Compute for Apache Flink:Limitations

Last Updated:Mar 18, 2026

Realtime Compute for Apache Flink enforces limits on resources, workspace organization, and deployment configurations. Review these limits before designing your deployment.

Resource quotas

Compute units (CUs)

A single subscription workspace supports a maximum of 1,000 Compute Units (CUs) per request. To request more than 1,000 CUs, submit a ticket.

For information about what a CU represents in terms of computing resources, see Product billing.

Organizational limits

The following limits apply to workspace, project, and folder structure. To request an increase for any of these limits, submit a ticket.

ResourceLimit
Projects per workspace5,000
Folders per project100
Folder depth5 levels
Files per folder (including subfolders)10,000

Deployment constraints

Regions and zones

Realtime Compute for Apache Flink is available in multiple regions in and outside China. Each region has one or more zones. For a full list, see Regions and zones.

Cross-zone deployment

Cross-zone deployment is supported only for subscription workspaces that use the Intel x86 architecture in certain regions.

Browser

Access the Realtime Compute for Apache Flink console using a Chromium-based browser.

Network

Realtime Compute for Apache Flink does not support internet access. For workarounds, see How do I access the internet?

Performance reference

These figures come from stress testing and represent expected throughput under controlled conditions. Actual performance depends on your workload characteristics and runtime environment.

ScenarioThroughput per CU
Simple operations (single-stream filtering, string transformations)40,000–55,000 records/second
Complex operations (JOIN, GROUP BY, window functions)5,000–10,000 records/second

For detailed benchmark methodology, see Performance White Paper (Nexmark Performance Test).

Legal responsibility

Important

The compute instances that run your code are your assets. You are responsible for the legal liabilities arising from the code logic that runs on these instances.

What's next