investigate flaky parallel/test-crypto-dh-leak

[Trott]

• Version: v11.0.0-pre (master)
• Platform: ubuntu1604_sharedlibs_debug_x64
• Subsystem: test crypto

https://ci.nodejs.org/job/node-test-commit-linux-containered/4846/nodes=ubuntu1604_sharedlibs_debug_x64/console

03:13:14 not ok 414 parallel/test-crypto-dh-leak
03:13:14 ---
03:13:14 duration_ms: 2.717
03:13:14 severity: fail
03:13:14 exitcode: 1
03:13:14 stack: |-
03:13:14 assert.js:270
03:13:14 throw err;
03:13:14 ^
03:13:14 
03:13:14 AssertionError [ERR_ASSERTION]: The expression evaluated to a falsy value:
03:13:14 
03:13:14 assert(after - before < 5 << 20)
03:13:14 
03:13:14 at Object.<anonymous> (/home/iojs/build/workspace/node-test-commit-linux-containered/nodes/ubuntu1604_sharedlibs_debug_x64/test/parallel/test-crypto-dh-leak.js:26:1)
03:13:14 at Module._compile (internal/modules/cjs/loader.js:702:30)
03:13:14 at Object.Module._extensions..js (internal/modules/cjs/loader.js:713:10)
03:13:14 at Module.load (internal/modules/cjs/loader.js:612:32)
03:13:14 at tryModuleLoad (internal/modules/cjs/loader.js:551:12)
03:13:14 at Function.Module._load (internal/modules/cjs/loader.js:543:3)
03:13:14 at Function.Module.runMain (internal/modules/cjs/loader.js:744:10)
03:13:14 at startup (internal/bootstrap/node.js:261:19)
03:13:14 at bootstrapNodeJSCore (internal/bootstrap/node.js:595:3)
03:13:14 ...

[Trott]

By the way, that was the fourth time in a row that happened on test-digitalocean-ubuntu1604_sharedlibs_container-x64-7. Other three are at:

https://ci.nodejs.org/job/node-test-commit-linux-containered/nodes=ubuntu1604_sharedlibs_debug_x64/4846/console
https://ci.nodejs.org/job/node-test-commit-linux-containered/nodes=ubuntu1604_sharedlibs_debug_x64/4845/console
https://ci.nodejs.org/job/node-test-commit-linux-containered/nodes=ubuntu1604_sharedlibs_debug_x64/4841/console

[Trott]

@nodejs/crypto @nodejs/build @nodejs/testing

[Trott]

This comment above the assertion that's failing suggests that this may be inherently unreliable?

// RSS should stay the same, ceteris paribus, but allow for
// some slop because V8 mallocs memory during execution.

[ryzokuken]

@Trott if definitely doesn't fail on MacOS. I'll try to look into it.

[ryzokuken]

// RSS should stay the same, ceteris paribus, but allow for
// some slop because V8 mallocs memory during execution.

Shouldn't this comment be improved? Of all things: why is there latin in there?

[bnoordhuis]

Because being a programmer doesn't mean you can't be well-versed in the liberal arts too. Ceteris paribus means "all other things being equal" but with fewer words.

Is the ./configure --debug buildbot the only one where it fails? If yes, I would guess that V8's debug bookkeeping slanting the numbers.

[tniessen]

We might be able to circumvent that by massively increasing the number of iterations and slightly increasing the amount of permitted RSS difference.

[Trott]

Other options:

We could check process.config.target_defaults.default_configuration and skip the test if it's not 'Release'. h/t @danbev

Or @tniessen is suggesting we could remove the test entirely as it was for a very specific regression.

Thoughts? @bnoordhuis

[tniessen]

On my maschine, the debug build allocates 4.8MB of the 5MB limit, but that value does not increase when I increase the number of iterations, so #21076 (comment) could work.