An Azure service that provides a registry of Docker and Open Container Initiative images.
Hello Matthias,
Thank you for reaching out.
Based on the error message shared, the registry is returning HTTP 429 (TooManyRequests) responses because it has exceeded the request rate limit for a specific operation category:
Registry exceeded the registry rate limit of 2000 requests in a 60 second window.
Although your registry is using the Premium SKU, rate limits are still enforced for individual operation types. In this case, the request is related to the referrers API (/referrers/...), which is used for retrieving artifact metadata such as signatures, SBOMs, and related OCI artifacts. A high volume of these requests within a short period can trigger throttling.
A few common scenarios that can lead to this behavior include:
- High-frequency metadata/referrer queries from CI/CD pipelines or automation workflows.
- Multiple clients or identities accessing the same registry concurrently, causing the combined request rate to exceed the per-registry limit.
- Large bursts of requests that temporarily exhaust the available request capacity, resulting in 429 responses until the limit window resets.
As an immediate mitigation while we investigate further, we recommend the following best practices:
- Ensure that client applications honor the
Retry-Afterheader returned by Azure Container Registry and retry requests only after the specified interval. - Implement an exponential backoff strategy with jitter for all retry operations to prevent repeated request bursts during throttling events.
- Review workflows that perform frequent metadata, tag, manifest, or referrer queries, and reduce the request frequency where possible.
- Cache artifact metadata locally when appropriate to minimize repetitive calls to the registry.
- Evaluate deployment and automation processes for excessive parallelism, and consider reducing the number of concurrent requests to the registry.
- If the issue is associated with large-scale deployments or image distribution activities, consider staggering operations to distribute requests more evenly over time.