stringbytes-external fails on AIX.

[mhdawson]

This test was recently updated as part of #3005

and it now fails on AIX:

not ok 697 test-stringbytes-external.js
#buffer.js:98
#    const ui8 = new Uint8Array(size);
#                ^
#
#RangeError: Invalid array buffer length
#    at new ArrayBuffer (native)
#    at new Uint8Array (native)
#    at allocate (buffer.js:98:17)
#    at new Buffer (buffer.js:49:12)
#    at /scratch/jenkins-local/workspace/convergance-test-aix/arch/ppc64/build/build/gcc48/gcc48/os/aix/node/test/parallel/test-stringbytes-external.js:125:16
#    at Object. (/scratch/jenkins-local/workspace/convergance-test-aix/arch/ppc64/build/build/gcc48/gcc48/os/aix/node/test/parallel/test-stringbytes-external.js:178:3)
#    at Module._compile (module.js:425:26)
#    at Object.Module._extensions..js (module.js:432:10)
#    at Module.load (module.js:356:32)
#    at Function.Module._load (module.js:311:12)
  ---

[mhdawson]

@evanlucas any thoughts

[mhdawson]

@jasnell FYI

[Fishrock123]

cc @trevnorris

[Trott]

I suspect I'll be fixing this as part of #3287 although I don't think that specific issue is fixed yet.

[evanlucas]

@Trott will that actually fix this though? This is failing on the initial allocation of buf0 (https://github.com/nodejs/node/blob/master/test/parallel/test-stringbytes-external.js#L125). Why would it throw there, when it passes the try/catch that allocates a larger buffer?

[Trott]

@evanlucas Yeah, I realized I was mistaken and edited my comment to backtrack a bit...maybe I need to backtrack more...

[trevnorris]

Problem probably is that v8 doesn't bother to GC after the first large allocation, then the subsequent malloc() fails since the needed memory hasn't been handed back.

@Trott I'd suggest two things. Once is that we force GC after the try/catch. Second is to possibly also wrap the second allocation in a try/catch.

@mhdawson How much memory do those machines have? We're allocating under 1GB for the test.

[Trott]

@mhdawson In #3287, this test file is split up into five different test files. Can you run that PR on AIX and confirm that all but one of the stringbytes test files still fail (which is what I suspect) or not? (Do we have AIX on ci.nodejs.org and I just need to know the special secret handshake to get the CI to run on AIX?)

[jasnell]

I don't want a failing test to hold up v4.2.0 so if we can't get a fix in for this by tomorrow we should likely just revert this change in v4.x

[Fishrock123]

I'm not sure we officially support AIX yet (no CI), so I move we don't hold up releases for any significant length of time on it until it is.

[jasnell]

Definitely don't want to hold it up, but if a fix is trivial, then there's no reason not to land it before we cut. Hopefully #3287 does the job and we can just pick it over.

[mhdawson]

After the split they all did fail except:

for ok 699 test-stringbytes-external.js

Turns out that it was the ulimit set for AIX. On linux by default the ulimit is high enough but the same is not true on AIX. Changing the ulimit seems resolves and all of the tests are passing.

Closing this issues as its related to machine config as opposed to the tests themselves.