Investigate flaky parallel/test-crypto-keygen

[Trott]

https://ci.nodejs.org/job/node-test-binary-arm/3782/RUN_SUBSET=5,label=pi1-docker/console

20:47:00 not ok 58 parallel/test-crypto-keygen
20:47:00   ---
20:47:00   duration_ms: 360.113
20:47:00   severity: fail
20:47:00   exitcode: -15
20:47:00   stack: |-
20:47:00     timeout
20:47:00   ...

[Trott]

If we continue to see this: The duration this test takes to run on Pi1 devices suggests that maybe it should be broken up into smaller tests if possible? Or skipped on devices that don't have a lot of memory and/or CPU?

https://ci.nodejs.org/job/node-test-binary-arm/3787/RUN_SUBSET=1,label=pi1-docker/console

00:56:36 ok 58 parallel/test-crypto-keygen
00:56:36   ---
00:56:36   duration_ms: 86.2
00:56:36   ...

[Trott]

Previous: #22978

[sam-github]

I don't think its memory intensive, but keygen is cpu intensive. crypto in general can be.

Breaking it up might be the right thing to do, if we can't annotate the test to allow more time.

[bnoordhuis]

That test searches for suitable large primes, which is CPU-intensive and non-determinate. e4e5b13 fixed the same issue for the Diffie-Hellman tests (where it also manifested on the Pi1s) by narrowing the search space from 1024 to 256 bits.

[tniessen]

cc @nodejs/crypto (mostly to test something on GitHub, sorry for the notification)

[sam-github]

I started a branch to reduce key strengths in test locally, but didn't have enough time and laptop power to build.

[Trott]

https://ci.nodejs.org/job/node-test-binary-arm/3870/RUN_SUBSET=1,label=pi1-docker/console

00:25:11 not ok 58 parallel/test-crypto-keygen
00:25:11   ---
00:25:11   duration_ms: 360.106
00:25:11   severity: fail
00:25:11   exitcode: -15
00:25:11   stack: |-
00:25:11     timeout
00:25:11   ...