test: log before and after RSS in memory leak test

[bnoordhuis]

Refs: #21076

To help investigate said bug.

CI: https://ci.nodejs.org/job/node-test-pull-request/15212/

[tniessen]

Suggesting to fast-track this.

[Trott]

+1 to fast-track

[apapirovski]

@Trott can I take this as you approving fast tracking?

[Trott]

Re-run Linux CI because debugging another issue left around some root-owned files causing a build to fail: https://ci.nodejs.org/job/node-test-commit-linux/19255/

[Trott]

@Trott can I take this as you approving fast tracking?

Yes.

[bnoordhuis]

https://ci.nodejs.org/job/node-stress-single-test/1893/ once it starts (predicted: 2 hours from now.)

[Trott]

Linux re-run is green.
Stress test is green.
Only thing remaining is to kick off AIX again: https://ci.nodejs.org/job/node-test-commit-aix/15458/

AIX might take a while to start because I'm currently using one (and possibly the only right now) AIX hosts for a stress test. But it will start!

[Trott]

(By the way, I think @tniessen already got results from this in a CI run off a branch. So the delay in landing this hasn't affected it's usefulness.)

[tniessen]

This is the output from https://ci.nodejs.org/job/node-test-commit-linux-containered/4862/nodes=ubuntu1604_sharedlibs_debug_x64/console, its like 5.097MB, so just above the limit. My suggestion in the referenced issue should solve the problem, I can open a PR tomorrow. (Even though I am not sure whether the problem reproduces with the recent v8 update.)

15:20:45 not ok 355 parallel/test-crypto-dh-leak
15:20:45 ---
15:20:45 duration_ms: 2.231
15:20:45 severity: fail
15:20:45 exitcode: 1
15:20:45 stack: |-
15:20:45 assert.js:80
15:20:45 throw new AssertionError(obj);
15:20:45 ^
15:20:45 15:20:45 AssertionError [ERR_ASSERTION]: before=48488448 after=53833728
15:20:45 at Object.<anonymous> (/home/iojs/build/workspace/node-test-commit-linux-containered/nodes/ubuntu1604_sharedlibs_debug_x64/test/parallel/test-crypto-dh-leak.js:26:8)
15:20:45 at Module._compile (internal/modules/cjs/loader.js:702:30)
15:20:45 at Object.Module._extensions..js (internal/modules/cjs/loader.js:713:10)
15:20:45 at Module.load (internal/modules/cjs/loader.js:612:32)
15:20:45 at tryModuleLoad (internal/modules/cjs/loader.js:551:12)
15:20:45 at Function.Module._load (internal/modules/cjs/loader.js:543:3)
15:20:45 at Function.Module.runMain (internal/modules/cjs/loader.js:744:10)
15:20:45 at startup (internal/bootstrap/node.js:261:19)
15:20:45 at bootstrapNodeJSCore (internal/bootstrap/node.js:595:3)
15:20:45 ...

[bnoordhuis]

I added another commit that doubles the slop limit, PTAL.

https://ci.nodejs.org/job/node-test-pull-request/15221/
https://ci.nodejs.org/job/node-stress-single-test/1896/ (AIX)
https://ci.nodejs.org/job/node-stress-single-test/1897/ (Linux debug, pending at the time of writing)

[tniessen]

If I remember correctly, you originally wrote the regression test, so I'll trust your judgement on this one, but in my patch I also increased the iteration count: assuming there is a memory leak, it might go undetected in release builds if the amount of memory leaked per iteration is sufficiently small (which might not be the case, I didn't go into depth in the original issue). The patch which I showed to @Trott yesterday used 5e5 iterations and a limit of 8MB, which i assumed were reasonable choices.

[bnoordhuis]

Forgetting to free a DH key's memory leaks kilobytes at a time so 50k iterations should be plenty to flush that out.

I picked 50k because otherwise it gets really slow with FIPS builds, because you can't use keys < 1024 bits.

[tniessen]

Okay, thanks for the explanation, I wasn't sure which part was being leaked. In that case this sounds reasonable.

[Trott]

Landed in 7b6c428...997e97d