child_process: swallow errors in internal communication

[apapirovski]

Much like with NODE_HANDLE_ACK, the internal protocol for communication about the sent socket should not expose its errors to the users when those async calls are not initiated by them. This also switches to use the internal _send api instead of the public one which performs a bunch of args validation which is not necessary in core code.

This will fix the currently reasonably frequent flakes on Windows in relation to child_process. It might also make it possible to remove some of our special case handling of Windows in our tests.

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• commit message follows commit guidelines

[Trott]

LGTM if CI is green. The relevant test or tests will need to be removed from the flaky lists in the relevant *.status files, although that could happen in a subsequent PR.

[apapirovski]

CI: https://ci.nodejs.org/job/node-test-pull-request/15239/
Stress test CI: https://ci.nodejs.org/job/node-stress-single-test/1904/

[jasnell]

I'd prefer this as a options object rather than a boolean argument but given that this is internal, I won't block on it.

[mscdex]

Since this is internal, can't we re-position this parameter so that callback is last for consistency?

[apapirovski]

CI: https://ci.nodejs.org/job/node-test-pull-request/15258/

[apapirovski]

@nodejs/child_process Please have a look, if possible. Would be nice to land soon to fix some of the unnecessary Windows failures.

[apapirovski]

CI: https://ci.nodejs.org/job/node-test-pull-request/15289/

[Trott]

Any remote chance this sole Linux failure in CI is potentially related?:

14:50:31 not ok 2178 sequential/test-inspector-port-zero-cluster
14:50:31   ---
14:50:31   duration_ms: 0.628
14:50:31   severity: fail
14:50:31   exitcode: 1
14:50:31   stack: |-
14:50:31     Debugger listening on ws://127.0.0.1:37475/fa5dcb1a-cc33-4e75-8731-79284e135a0f
14:50:31     For help, see: https://nodejs.org/en/docs/inspector
14:50:31     Debugger listening on ws://127.0.0.1:37476/8c20a652-a9c7-463c-938b-776a2fd29e33
14:50:31     For help, see: https://nodejs.org/en/docs/inspector
14:50:31     Debugger listening on ws://127.0.0.1:37477/51411750-a9a2-4451-8ecb-e849550121aa
14:50:31     For help, see: https://nodejs.org/en/docs/inspector
14:50:31     Starting inspector on 127.0.0.1:37478 failed: address already in use
14:50:31     node: ../deps/uv/src/unix/core.c:117: uv_close: Assertion `!(((handle)->flags & (UV_CLOSING | UV_CLOSED)) != 0)' failed.
14:50:31     assert.js:80
14:50:31       throw new AssertionError(obj);
14:50:31       ^
14:50:31     
14:50:31     AssertionError [ERR_ASSERTION]: code: null, signal: SIGABRT
14:50:31         at Worker.worker.on.common.mustCall (/home/iojs/build/workspace/node-test-commit-linux/nodes/centos7-64-gcc6/test/sequential/test-inspector-port-zero-cluster.js:20:16)
14:50:31         at Worker.<anonymous> (/home/iojs/build/workspace/node-test-commit-linux/nodes/centos7-64-gcc6/test/common/index.js:451:15)
14:50:31         at Worker.emit (events.js:182:13)
14:50:31         at ChildProcess.worker.process.once (internal/cluster/master.js:193:12)
14:50:31         at Object.onceWrapper (events.js:273:13)
14:50:31         at ChildProcess.emit (events.js:182:13)
14:50:31         at Process.ChildProcess._handle.onexit (internal/child_process.js:237:12)

[Trott]

Linux re-run: https://ci.nodejs.org/job/node-test-commit-linux/19369/

[apapirovski]

@Trott I doubt it. It failed elsewhere earlier and it's on our known flaky tests list. Also, it fails with: "14:50:31 Starting inspector on 127.0.0.1:37478 failed: address already in use" but no processes were alive at the start or the end of that run.

[apapirovski]

Landed in af2a104