ReleaseBytes
aws AWS Developer Tools Blog ·

AWS SDKs Update Retry Behavior

bloginfraawsengineer
announcement

AWS is updating the default retry behavior in its SDKs and tools to improve error handling and latency. The changes standardize retry logic across SDKs, addressing issues where some retried too long or failed to fail fast during outages. This update affects applications using standard or adaptive retry modes and provides an opt-in mechanism before the changes become default in November 2026.

  • →Updated Retry Defaults for AWS SDKs and Tools
  • →Standard Retry Mode Enhancements
  • →Faster Transient Error Retries
  • →DynamoDB-Specific Tuning and Long-Polling Operations
  • →Impact and Opt-in/Revert Mechanisms
Enhancements (3) ›
  • Standard Retry Mode Enhancements

    The standard retry mode now features an updated retry quota, with higher costs for transient errors (14 tokens) and throttling errors (5 tokens), causing the quota to engage sooner during outages. Base backoff delays are also differentiated by error type: 50ms for transient errors and 1000ms for throttling errors, a change from previous SDKs' variable and undifferentiated delays.

  • Faster Transient Error Retries

    The first retry after a transient error is now significantly faster, averaging around 25ms, compared to hundreds of milliseconds previously. This aims to improve responsiveness for short-lived transient issues.

  • DynamoDB-Specific Tuning and Long-Polling Operations

    Amazon DynamoDB and DynamoDB Streams now use a shorter base backoff delay (25ms) and typically 4 max attempts to align with their low-latency profile. Long-polling operations, like SQS ReceiveMessage, now incorporate a backoff delay when the retry quota is depleted to prevent tight polling loops and excessive client-side load.

Notes (3) ›
  • Updated Retry Defaults for AWS SDKs and Tools

    AWS is standardizing retry behavior across its SDKs and tools to ensure consistent handling of transient errors and outages. The updated defaults aim to reduce latency by failing faster during sustained failures and prevent prolonged retries that can exacerbate service issues.

  • Impact and Opt-in/Revert Mechanisms

    Applications without explicit retry configurations will see the new behavior upon opting in or after the default rollout in November 2026. Those with explicit configurations will see updated defaults for settings not overridden. Users can opt in by setting the AWS_NEW_RETRIES_2026=true environment variable and revert by unsetting it or overriding specific configurations. Full default rollout is scheduled for November 2026.

  • SDK Availability for Opt-in

    Eight AWS SDKs (Java 2.x, Python, .NET, PowerShell, JavaScript 3.x, PHP, Kotlin, Rust) currently support the opt-in flag, with six more expected soon, including Go, C++, Ruby, and the AWS CLI. Each SDK's GitHub tracking issue provides version minimums, default comparisons, and examples.

Read the original announcement →

https://aws.amazon.com/blogs/developer/announcing-updated-retry-behavior-for-aws-sdks-and-tools/

Related releases