Bug: Missing await in AdaptiveRetryStrategy causes unhandled promise rejection crash (@smithy/util-retry 4.3.6)

[Osama-Yusuf]

Describe the bug

AdaptiveRetryStrategy.refreshRetryTokenForRetry in @smithy/util-retry@4.3.6 has a missing await on the inner standardRetryStrategy.refreshRetryTokenForRetry() call, causing unhandled promise rejections that crash the Node.js process under sustained throttling.

// @smithy/util-retry 4.3.6, util-retry index.js:299-304
async refreshRetryTokenForRetry(tokenToRenew, errorInfo) {
    this.rateLimiter.updateClientSendingRate(errorInfo);
    const token = this.standardRetryStrategy.refreshRetryTokenForRetry(tokenToRenew, errorInfo); // <-- missing await
    await this.rateLimiter.getSendToken();
    return token;
}

When retry capacity is drained under throttling, the inner call returns a rejected promise with "No retry token available". Execution then suspends on getSendToken() (which sleeps via setTimeout under throttling). Node reaches the end of the event-loop tick with an unhandled rejected promise before the caller's try/catch can attach a handler — crashing the process.

Reproduction

Minimal repro using the exact installed package version — a loop calling refreshRetryTokenForRetry with every call wrapped in try/catch:

• AdaptiveRetryStrategy: 5th refresh crashes the process. The catch block never executes.
• StandardRetryStrategy: Same error is caught normally, process exits cleanly.

Stack trace (production)

Error: No retry token available
    at StandardRetryStrategy.refreshRetryTokenForRetry (node_modules/@smithy/util-retry/dist-cjs/index.js:265:13)
    at processTicksAndRejections (node:internal/process/task_queues)

This is an unhandledRejection — not an error propagating through any application call chain. The retry middleware's try/catch (middleware-retry index.js:313-322) never gets the chance to handle it because the promise rejects before return token attaches a handler.

Impact

• Production crash-loop across multiple environments (production, staging, UAT)
• Triggered by sustained ThrottlingException responses from @aws-sdk/client-translate
• Only workaround: revert retryMode from 'adaptive' to 'standard'

Versions

• @aws-sdk/client-translate: 3.1039.0
• @smithy/util-retry: 4.3.6
• Node.js: v20
• Runtime: NestJS application on EKS

Fix status

We understand the fix exists upstream in current @smithy/core (re-exported by @smithy/util-retry@4.4.x) where the await is added. Filing this because:

1. Version 4.3.6 is still installable via current @aws-sdk/* packages and causes silent production crashes
2. "No retry token available" is an internal SDK signal that should never surface as an unhandled rejection
3. No changelog entry or migration note warns about this behavior in adaptive retry mode

[Osama-Yusuf]

Closing — this was caused by the missing await in AdaptiveRetryStrategy.refreshRetryTokenForRetry in @smithy/util-retry@4.3.6, which is already fixed in 4.4.x. We've upgraded on our side. Apologies for the noise.

[github-actions]

This issue is now closed. Comments on closed issues are hard for our team to see.
If you need more assistance, please open a new issue that references this one.