`test-net-bytes-per-incoming-chunk-overhead` failing on Fedora 38

[targos]

$ out/Release/node --expose-gc /home/iojs/build/workspace/
node-test-commit-linux/test/pummel/test-net-bytes-per-incoming-chunk-overhead.js
node:assert:399
    throw err;
    ^

AssertionError [ERR_ASSERTION]: measured 4552.081408 bytes per chunk
    at process.<anonymous> (/home/iojs/build/workspace/node-test-commit-linux/test/pummel/test-net-bytes-per-incoming-chunk-overhead.js:49:3)
    at process.emit (node:events:523:35) {
  generatedMessage: false,
  code: 'ERR_ASSERTION',
  actual: false,
  expected: true,
  operator: '=='
}

Node.js v21.0.0-pre

$ uname -a
Linux test-digitalocean-fedora38-x64-1 6.3.8-200.fc38.x86_64 #1 SMP PREEMPT_DYNAMIC Thu Jun 15 02:15:40 UTC 2023 x86_64 GNU/Linux

$ cat /etc/os-release
NAME="Fedora Linux"
VERSION="38 (Cloud Edition)"
ID=fedora
VERSION_ID=38
VERSION_CODENAME=""
PLATFORM_ID="platform:f38"
PRETTY_NAME="Fedora Linux 38 (Cloud Edition)"
ANSI_COLOR="0;38;2;60;110;180"
LOGO=fedora-logo-icon
CPE_NAME="cpe:/o:fedoraproject:fedora:38"
HOME_URL="https://fedoraproject.org/"
DOCUMENTATION_URL="https://docs.fedoraproject.org/en-US/fedora/f38/system-administrators-guide/"
SUPPORT_URL="https://ask.fedoraproject.org/"
BUG_REPORT_URL="https://bugzilla.redhat.com/"
REDHAT_BUGZILLA_PRODUCT="Fedora"
REDHAT_BUGZILLA_PRODUCT_VERSION=38
REDHAT_SUPPORT_PRODUCT="Fedora"
REDHAT_SUPPORT_PRODUCT_VERSION=38
SUPPORT_END=2024-05-14
VARIANT="Cloud Edition"
VARIANT_ID=cloud

$ gcc --version
gcc (GCC) 13.1.1 20230614 (Red Hat 13.1.1-4)

Refs: nodejs/build#3350 (comment)

@nodejs/net

[ShogunPanda]

@targos Do you mind checking this once #48464 lands?

[targos]

@ShogunPanda I've applied the fix on top of main on the Fedora host and it still fails with the same error.

[ShogunPanda]

I see. If you disable network family autoselection does it work?

[targos]

It still doesn't work with --no-network-family-autoselection

[ShogunPanda]

Ok, I see. At least I know it's unrelated to my changes.
Seems like the memory overhead has changed. Do we have a lead on why?

[lpinca]

This also failed on macOS 11. See https://ci.nodejs.org/job/node-test-pull-request/52510/.

not ok 3596 pummel/test-net-bytes-per-incoming-chunk-overhead
  ---
  duration_ms: 73725.34500
  severity: crashed
  exitcode: -6
  stack: |-
    Assertion failed: (timeout != -1), function uv__io_poll, file kqueue.c, line 290.

[targos]

Not the same error, though.

[lpinca]

cc: @nodejs/libuv

[bnoordhuis]

Not a libuv issue. The test makes a shaky assumption about RSS:

node/test/pummel/test-net-bytes-per-incoming-chunk-overhead.js

Lines 45 to 49 in 951da52

	global.gc();
	const bytesPerChunk =
	(process.memoryUsage.rss() - baseRSS) / receivedChunks.length;
	// We should always have less than one page (usually ~ 4 kB) per chunk.
	assert(bytesPerChunk < 650, `measured ${bytesPerChunk} bytes per chunk`);

[lpinca]

@bnoordhuis I pinged libuv maintainers for this failed assertion. I think it is not related to the test but I guess it shouldn't happen.

Assertion failed: (timeout != -1), function uv__io_poll, file kqueue.c, line 290

[bnoordhuis]

Ah, right. That should have been fixed by #48078 but I guess not? cc @trevnorris

[trevnorris]

Want to confirm, does this still happen with the recent update to libuv v1.46.0? And does it happen consistently?