[v20.x] parallel/test-buffer-tostring-range fails on SmartOS

[marco-ippolito]

Test

parallel/test-buffer-tostring-range

Platform

SmartOS

Console output

03:13:42 not ok 177 parallel/test-buffer-tostring-range
03:13:42   ---
03:13:43   duration_ms: 5550.52900
03:13:43   severity: fail
03:13:43   exitcode: 1
03:13:43   stack: |-
03:13:43     node:internal/buffer:961
03:13:43         super(bufferOrLength, byteOffset, length);
03:13:43         ^
03:13:43     
03:13:43     RangeError: Array buffer allocation failed
03:13:43         at new ArrayBuffer (<anonymous>)
03:13:43         at new Uint8Array (<anonymous>)
03:13:43         at new FastBuffer (node:internal/buffer:961:5)
03:13:43         at Function.alloc (node:buffer:395:10)
03:13:43         at Object.<anonymous> (/home/iojs/build/workspace/node-test-commit-smartos/nodes/smartos23-x64/test/parallel/test-buffer-tostring-range.js:107:28)
03:13:43         at Module._compile (node:internal/modules/cjs/loader:1468:14)
03:13:43         at Module._extensions..js (node:internal/modules/cjs/loader:1547:10)
03:13:43         at Module.load (node:internal/modules/cjs/loader:1287:32)
03:13:43         at Module._load (node:internal/modules/cjs/loader:1103:12)
03:13:43         at Function.executeUserEntryPoint [as runMain] (node:internal/modules/run_main:180:12)
03:13:43     
03:13:43     Node.js v20.18.3
03:13:43   ...

Build links

• https://ci.nodejs.org/job/node-test-commit-smartos/58746/console

Additional information

It's blocking v20 release #56699

@nodejs/platform-smartos

[joyeecheung]

This is v20.x only, it seems https://github.com/nodejs/reliability/blob/main/reports/2025-01-23.md (both PRs referenced are targetting v20.x)

[anonrig]

Cc @nodejs/platform-smartos

[mcollina]

@jclulow can you take a look?

[jclulow]

This is one of the ones I'm trying to reproduce, but I thought we already added it to the skip list to get past this? I'm sure I saw a change that did that go past

[marco-ippolito]

This is one of the ones I'm trying to reproduce, but I thought we already added it to the skip list to get past this? I'm sure I saw a change that did that go past

Yes we skipped the failing test from v20.x

[anonrig]

This is one of the ones I'm trying to reproduce, but I thought we already added it to the skip list to get past this? I'm sure I saw a change that did that go past

Please correct me if I'm wrong, but skipping a flaky test is not the ideal scenario here. Ideally smartOS node.js team should fix the flakiness and open a PR to node.js to make the tests more stable.

[jclulow]

This is one of the ones I'm trying to reproduce, but I thought we already added it to the skip list to get past this? I'm sure I saw a change that did that go past

Please correct me if I'm wrong, but skipping a flaky test is not the ideal scenario here. Ideally smartOS node.js team should fix the flakiness and open a PR to node.js to make the tests more stable.

I've been reproducing it, actually! It's actually not flaky, but also probably not a bug in Node per se. I believe the issue is in V8. I'm trying to isolate it, and then I'll need to go through the dubious CLA-signing hoops before I'm allowed to contribute a fix there.

I haven't finished writing up the details, but the short story is that V8 appears to be choosing addresses for mapped memory that collide with the sbrk heap occasionally. It happens maybe once in 50 or 100 runs on my test machine.