Node 14, 16 (Ubuntu) crash with --experimental-vm-modules if require and import are used in the same test file

[connorjclark]

Version

14.17.1, 16.4.2

Platform

Ubuntu 20.04.2 LTS

Subsystem

vm

What steps will reproduce the bug?

Full details: GoogleChrome/lighthouse#12702 (comment)

• Use both require and import in a jest test file
• Have a good number (~10-20+) of other test files (make things busy)
• Run tests with node --experimental-vm-modules ./node_modules/jest/bin/jest.js

How often does it reproduce? Is there a required condition?

~100% given enough tests are run / there is enough load on the machine

What is the expected behavior?

No crash.

What do you see instead?

Crash.

[
   "/home/runner/work/lighthouse/lighthouse/node_modules/segfault-handler/build/Release/segfault-handler.node(+0x3785) [0x7fbe3cdf2785]",
   "/lib/x86_64-linux-gnu/libpthread.so.0(+0x153c0) [0x7fbe410063c0]",
   "/opt/hostedtoolcache/node/14.17.1/x64/bin/node() [0x964366]",
   "/opt/hostedtoolcache/node/14.17.1/x64/bin/node(_ZN2v88internal7Isolate38RunHostImportModuleDynamicallyCallbackENS0_6HandleINS0_6ScriptEEENS2_INS0_6ObjectEEE+0x78) [0xcfdb48]",
   "/opt/hostedtoolcache/node/14.17.1/x64/bin/node(_ZN2v88internal25Runtime_DynamicImportCallEiPmPNS0_7IsolateE+0xaf) [0x1089b0f]",
   "/opt/hostedtoolcache/node/14.17.1/x64/bin/node() [0x1427254]"
]

example failure: https://github.com/GoogleChrome/lighthouse/runs/3025648367?check_suite_focus=true

Additional information

The crash only seems to happen when enough other tests are running. For the same set of tests, I noticed a big difference in crash rate at different times of the day. Given I'm running this on remote GitHub Action machines, and libpthread is the source of the crash, I assume the crash is caused by errant multithreading and is dependent on machine load.

I was never able to repro this locally on my Mac, Windows, or Debian machine. I do not have access to Ubuntu (besides via GHA).

[connorjclark]

This seems to be the same issue as #35889 (same segfault stacktrace). That issue has a better repro and more details, so this issue can probably be closed.