feat: Support new 2026 retries#3379
Conversation
|
Detected 1 possible performance regressions:
|
|
|
||
| # Minimum `aws-sdk-core` version for new gem builds | ||
| MINIMUM_CORE_VERSION = "3.247.0" | ||
| MINIMUM_CORE_VERSION = "3.248.0" |
There was a problem hiding this comment.
Note to self to check this before merging.
jterapin
left a comment
jterapin
left a comment
There was a problem hiding this comment.
Nice - you should check with everyone else when these changes should be going out. Also is there a blog post or documentation on how new standard retries work?
| if RetryErrors.new_retries? | ||
| backoff_scalar = if error_inspector.throttling_error? | ||
| 1 | ||
| elsif ['DynamoDB', 'DynamoDB Streams'].include?(service_id) |
There was a problem hiding this comment.
We can make a quick helper check since we do this multiple times - like dynamodb_service?(cfg)
There was a problem hiding this comment.
I can define a constant which contains the list of DynamoDB related services, I think the check is simple enough that we don't need any method overhead.
| # - Remove the old retries branch in #exponential_backoff | ||
| # - Remove LEGACY_RETRY_COST and TIMEOUT_RETRY_COST from RetryQuota | ||
| # @api private | ||
| def self.new_retries? |
There was a problem hiding this comment.
We have a quite a bit class methods here - we might as well do class << self to group there.
|
|
||
| * Feature - Add YJIT & ZJIT tracking to user agent. | ||
| * Issue - Fix error messaging in SSO OIDC. | ||
| * Feature - Add new retry behavior behind `AWS_NEW_RETRIES_2026` environment variable. When enabled, defaults to `standard` retry mode, service-specific tuning for DynamoDB, long-polling backoff support, and `x-amz-retry-after` header support. |
There was a problem hiding this comment.
This changelog entry is a dense read with internal details so maybe:
* Feature - Add `AWS_NEW_RETRIES_2026` environment variable to opt-in to updated
`standard` retry mode with reduced backoff intervals.
We may want to consider having changelog entries in DyanmoDB/DynamoDBStreams - customers who uses these service gems may not necessarily look at the core gem updates. I think we do generate a default "take a look at core for generated changes" but we could have more meaningful change entry for these services
|
|
||
| # Hard-coded combination of services and operations as having the | ||
| # longPoll trait. To be removed when trait is enabled. | ||
| LONG_POLLING_OPERATIONS = { |
There was a problem hiding this comment.
I'm not comfortable with service-specific logic living in core. I recognize this pattern exists in V3, but we should be more intentional about this in V4.
2 participants
This PR adds support for new 2026 retries behavior, including changing the default retry mode, updating retry quotas, and introducing new behavior for long-polling operations and retry backoff headers.
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.