Getting systematic Timeout Error on PUT requests after bumping @smithy/node-http-handler from v4.3.0 to v4.4.2

[MathieuGuillet]

Checkboxes for prior research

• I've gone through Developer Guide and API reference
• I've checked AWS Forums and StackOverflow.
• I've searched for previous similar issues and didn't find any solution.

Describe the bug

All the file uploads to S3 are failing after the dependency @smithy/node-http-handler moved from v4.3.0 to v4.4.2 (renovate bump).
Without changing anything than just reverting to v4.3.0 fixes the issue, so no doubt about the source of these timeout errors.
We couldn't understand what could be the issue though (looking at the changelogs between the v4.3.0 and the v4.4.2).
Testing with the version v4.4.3 didn't fix the issue.

Our use case is simple : just uploading a file to S3 using the S3Client. The file is not big (a text file with a maximum of 10k rows).
The read accesses to S3 are working well though.

Regression Issue

• Select this option if this issue appears to be a regression.

SDK version number

@smithy/node-http-handler@4.4.2, @smithy/node-http-handler@4.4.3

Which JavaScript Runtime is this issue in?

Node.js

Details of the browser/Node.js/ReactNative version

Node v22.14.0

Reproduction Steps

Here is how we create our S3Clients using node-http-handler:

import { S3Client } from '@aws-sdk/client-s3'
import { NodeHttpHandler } from '@smithy/node-http-handler'
import { HttpsProxyAgent } from 'https-proxy-agent'
import http from 'node:http'
import https from 'node:https'

type CreateS3ClientParams = {
  region: string
  stsRegion?: string
  endpoint?: string
}

export function createS3Client({ region, stsRegion, endpoint }: CreateS3ClientParams): S3Client {
  const httpProxy = process.env.HTTP_PROXY ?? process.env.http_proxy
  const httpsProxy = process.env.HTTPS_PROXY ?? process.env.https_proxy

  const config: Record<string, unknown> = {
    region,
    requestHandler: new NodeHttpHandler({
      httpAgent: httpProxy ? new HttpsProxyAgent(httpProxy) : new http.Agent({ keepAlive: true }),
      httpsAgent: httpsProxy ? new HttpsProxyAgent(httpsProxy) : new https.Agent({ keepAlive: true }),
    }),
    followRegionRedirects: true,
  }

  if (stsRegion && stsRegion !== region) {
    config.clientConfig = { region: stsRegion }
  }

  if (endpoint) {
    config.endpoint = endpoint
  }

  return new S3Client(config)
}

How we upload files using this S3Client:

new Upload({
    partSize: 1024 * 1024 * 5,
    client: s3Client,
    params: {
      Bucket: bucketName,
      Body: body,
      Key: fileName,
      ContentType: contentType,
    },
  }).done()

Here is our dependency list at the moment:

"dependencies": {
    "@aws-sdk/client-s3": "^3.901.0",
    "@aws-sdk/credential-providers": "^3.901.0",
    "@aws-sdk/signature-v4": "^3.374.0",
    "@aws-sdk/util-format-url": "^3.901.0",
    "@smithy/fetch-http-handler": "^5.3.1",
    "@smithy/hash-node": "^4.2.0",
    "@smithy/node-http-handler": "^4.3.0",
    "@smithy/protocol-http": "^5.3.3",
    "@smithy/types": "^4.6.0",
    "https-proxy-agent": "^7.0.6"
  }

Observed Behavior

Here is the error we get systematically when doing these Uploads with the node-http-handler v4.4.2 or v4.4.3 :

Error type : TimeoutError

AggregateError [ETIMEDOUT]: 
    at internalConnectMultiple (node:net:1139:18)
    at afterConnectMultiple (node:net:1712:7)
    at TCPConnectWrap.callbackTrampoline (node:internal/async_hooks:130:17)

Expected Behavior

No timeout on uploads, as with the version 4.3.0.

Possible Solution

No response

Additional Information/Context

For now, we just froze the version to v4.3.0 to fix the issue.

[kuhe]

Using the latest version of @aws-sdk/client-s3 initialize with expectContinueHeader: false in the client constructor.

This is a workaround. I'll comment with more information later because my phone battery is running out.

[kuhe]

I've opened a fix for this issue in smithy-lang/smithy-typescript#1757 and will notify here when published.

---

The cause of the error is a problematic interaction between S3 and Node.js http(s) modules where a delay in writing the request body Buffer can cause a request to fail when making a high volume of requests using a single Agent instance. The fix was to initialize a separate instance when making a PUT request with the expect: 100-continue header, which is why setting

new S3({
  expectContinueHeader: false
});

is a workaround. However, given this regression for http Proxy agents, it would be better to ignore the expect: 100-continue header altogether when using a custom Agent. The side effect is that the request body will be sent without waiting for server confirmation, but this side effect is better than the request failing, and requiring every proxy user to use the expectContinueHeader configuration.

[kuhe]

The fix for proxy agents is published to https://www.npmjs.com/package/@smithy/node-http-handler/v/4.4.4.

[MathieuGuillet]

Thank you very much for your reactivity. I'll test with this version tomorrow and let you know if it's fixed ! 👍