Various flaky tests when running on test-osuosl-aix72-ppc64_be-3

There are a few tests which are showing high levels of inconsistency when run on the test-osuosl-aix72-ppc64_be-3 machine. While there are occasional (<1%) failures showing on some of the tests elsewhere they are showing enough problems on this machine that it should be investigated as it is causing problems showing up on the node reliability reports (and also from some of my own job runs!)

The ones I've noticed - which I'm currently running additional stress tests on are as follows, although this may not be an exhaustive list:

• parallel/test-repl-preview (Low failure rate - ~10% maybe)
• node-api/test_threadsafe_function (~90% failure rate on osuosl_be-3)
• async-hooks/test-emit-after-on-destroyed (~9% failure rate on osuosl_be-3)

1000 iterations of these on OSUOSL machines be-3 and be-4 are occurring in https://ci.nodejs.org/job/node-stress-single-test/638/console. (Note: I have modified that job to allow me to select the OSUOSL AIX machines specifically instead of just by the generic aix72-ppc64 label to allow explicit comparisons. I've also noted that the async-hooks test did fail once in a 100-iteration run on osuosl-be-4 (AssertionError [ERR_ASSERTION]: Expected values to be strictly equal: / 'SIGILL' !== null) and with a similar frequency on that machine on a 1000-iteration run.

Noting that the intermittent stringbytes-external-max was marked flaky last month and it potentially also primarily a problem on this machine (to be confirmed)

I had also raised https://github.com/nodejs/node/issues/61439 on the threadsafe failure but I will close that as this seems to be primarily an issue with the machine setup/configuration as opposed to a general problem with that test.

Commencez par le job 638 de 1000 itérations et comparez test-osuosl-aix72-ppc64_be-3 avec be-4. Reproduisez parallel/test-repl-preview, node-api/test_threadsafe_function et async-hooks/test-emit-after-on-destroyed, puis examinez les échecs signalés et les différences entre les machines. Le travail est terminé lorsqu’il est établi si la configuration du nœud AIX est à l’origine de l’instabilité et que le correctif qui en résulte est documenté ou pris en charge.