parallel.test-worker-message-port-transfer-terminate is flaky on smartOS

[anonrig]

Test

parallel.test-worker-message-port-transfer-terminate

Platform

SmartOS

Console output

---
duration_ms: 362.04
exitcode: -6
severity: crashed
stack: |-
  terminate called after throwing an instance of 'std::bad_alloc'
    what():  std::bad_alloc
...

Build links

• https://ci.nodejs.org/job/node-test-commit-smartos/58872/nodes=smartos23-x64/testReport/junit/(root)/parallel/test_worker_message_port_transfer_terminate/

Additional information

cc @nodejs/platform-smartos

[mcollina]

This has been found flaky a few times before: https://github.com/search?q=repo%3Anodejs%2Freliability+test-worker-message-port-transfer-terminate&type=issues.

I have a rough issue this would have some timing issues that are unrelated to SmartOS

[jclulow]

My assumption, without having dug in too far just yet, is that std::bad_alloc is presumably just a failure to allocate memory. As part of nodejs/build#4011 we're looking at increasing the memory available to the CI workers, and perhaps most critically here we're going to substantially increase the available swap.

Unlike Linux systems, illumos systems do not overcommit memory; if we don't have a place to put the memory you allocate with malloc(3C), whether in physical memory or in a swap file, we fail the allocation. This is to avoid the need for dubious mechanisms like an OOM killer, etc.

My hope is that once we sort out nodejs/build#4011 that issues around memory allocation like this one will just go away!