test-http-max-sockets fails in node-test-binary-windows-js-suites

[tniessen]

Test

test-http-max-sockets

Platform

Windows

Console output

Socket: 1/10 queued: 0
Socket: 2/10 queued: 0
Socket: 3/10 queued: 0
Socket: 4/10 queued: 0
Socket: 5/10 queued: 0
Socket: 6/10 queued: 0
Socket: 7/10 queued: 0
Socket: 8/10 queued: 0
Socket: 9/10 queued: 0
Socket: 10/10 queued: 0
Socket: 10/10 queued: 1
Socket: 10/10 queued: 2
Socket: 10/10 queued: 3
Socket: 10/10 queued: 4
Socket: 10/10 queued: 5
Socket: 10/10 queued: 6
Socket: 10/10 queued: 7
Socket: 10/10 queued: 8
Socket: 10/10 queued: 9
Socket: 10/10 queued: 10

Build links

• https://ci.nodejs.org/job/node-test-binary-windows-js-suites/19661/
• https://ci.nodejs.org/job/node-test-binary-windows-js-suites/19664/
• https://ci.nodejs.org/job/node-test-binary-windows-js-suites/19672/
• ...

(Many recent Windows builds.)

Additional information

Build trend:

Refs: #47091 (comment)

cc @nodejs/http @nodejs/platform-windows

[mcollina]

Ouch! Should we skip on that platform?

cc @ShogunPanda

[tniessen]

I don't know why it only started failing (frequently) recently. I don't see any related changes in the commit history.

[mcollina]

Where do we source our Windows machines? I've seen a significant degradation in performance for GitHub Actions on Windows (which uses Azure), leading to many of my project to start failing on Windows.

My 2 cents is something has changed on the host and the test is now flaky enough to be a problem (while before it failed spuriously)

[tniessen]

cc @nodejs/build

[tniessen]

Marking it: #47134

[targos]

Azure and Rackspace (~50%/50%)

[targos]

The three failures linked by @tniessen happened on Rackspace machines

[tniessen]

@targos Almost all recent builds have failed. Are all of these on rackspace?

[targos]

I don't know how to check without opening each run one by one.

[tniessen]

I don't expect you to do that, of course 😄 And I assume that we couldn't do anything about it even if it was rackspace machines, so let's go ahead with marking it as flaky (or skip if necessary) for now.

[tniessen]

According to nodejs/reliability#527 and nodejs/reliability#526, the test did indeed only fail on Rackspace machines. However, given the build trend (see #47116 (comment)), that would mean that Rackspace picked up almost every build.