Rate and concurrency limits
Rate and concurrency limits
Anyscale enforces default limits on how frequently you can submit jobs and create clusters, and on how many clusters you can run at once. These limits protect platform stability when many workloads start in a short window. This page describes the default limits and how to work within them.
Admins configure resource quotas to cap CPU, GPU, and instance usage. Anyscale applies the limits on this page by default across your organization. See Resource quotas.
Default limits
Anyscale applies two kinds of limits: a rate on how often you create resources, and a cap on how many clusters run concurrently.
The following creation-rate limits apply by default:
| Limit | Scope | Default |
|---|---|---|
| Job submissions | Per user | 500 per 30 minutes |
| Job submissions (burst) | Per organization | 300 per 30 seconds |
| Cluster creations | Per user | 1000 per 30 minutes |
Cluster creations count workspaces, services, and jobs together, because each starts a cluster.
Anyscale also limits concurrent clusters. Each user can run up to 5000 active clusters at any point in time.
Work within the limits
When a workload submits jobs or creates clusters faster than these limits allow, some requests fail. To submit large batches of jobs reliably, add client-side retries so your code resubmits requests that don't succeed on the first attempt.
If your workloads require higher limits, contact Anyscale Support.
Related limits
Anyscale enforces a separate rate limit on log ingestion volume. For details, see Accessing logs.