child_process: fix deoptimizing use of arguments #11535

vsemozhetbyt · 2017-02-24T08:59:02Z

Checklist

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

Affected core subsystem(s)

child_process, benchmarks

This is the last unaddressed case from mentioned in the #10323. PR fixes three functions.

1. Function normalizeExecArgs() is called by exports.exec() and exports.execSync() with a variable number of parameters.

To check deoptimizations, run this code with --trace_opt --trace_deopt flags:

const execSync = require('child_process').execSync;
for (let i = 0; i < 1e3; i++) execSync('echo');

You will see messages about normalizeExecArgs() deoptimization.

None of the current benchmarks checks this case: the only benchmark calling exports.exec() does it once and always with 2 parameters.

2. Function exports.execFile() is called by users with a variable number of parameters. See also #10323 (comment)

To check deoptimizations, run this code with --trace_opt --trace_deopt flags:

const execFile = require('child_process').execFile;
for (let i = 0; i < 1e3; i++) execFile('echo').kill();

You will see messages about exports.execFile() deoptimization.

None of the current benchmarks checks exports.execFile().

3. Function normalizeSpawnArguments() is called by exports.spawn(), exports.spawnSync(), and exports.execFileSync() with a variable number of parameters.

To check deoptimizations, run this code with --trace_opt --trace_deopt flags:

const spawnSync = require('child_process').spawnSync;
for (let i = 0; i < 1e3; i++) spawnSync('echo');

You will see messages about normalizeSpawnArguments() deoptimization.

None of the current benchmarks checks exports.spawnSync() or exports.execFileSync(). Two of them call exports.spawn() once with 3 params (not appropriate). One of them calls exports.spawn() 1000 times with 2 params, but it does not cover all sub-cases.

4. While fixing these cases, I've found an absolutely new deopt case in the exports.execFile() function introduced by v8 5.6. See the fourth commit and explanation in the upstream issue.

5. As changes affect almost all main child_process methods (except .fork()), I've added a benchmark calling these 6 methods with 1, 2, 3, and 4 params (when it is appropriate). Benchmarking is a bit murky in this case because of many external sync/async processes involved, but it seems no significant degradation is caused, moreover some tiny improvement is evident.

                                                                                  improvement confidence      p.value
 child_process\\child-process-params.js params=1 methodName="exec" n=1000              0.59 %            6.187539e-02
 child_process\\child-process-params.js params=1 methodName="execFile" n=1000          0.62 %          * 2.916666e-02
 child_process\\child-process-params.js params=1 methodName="execFileSync" n=1000      0.32 %            5.213650e-01
 child_process\\child-process-params.js params=1 methodName="execSync" n=1000          0.26 %          * 4.716863e-02
 child_process\\child-process-params.js params=1 methodName="spawn" n=1000             0.69 %        *** 2.204024e-04
 child_process\\child-process-params.js params=1 methodName="spawnSync" n=1000         0.19 %            7.165778e-01
 child_process\\child-process-params.js params=2 methodName="exec" n=1000              0.72 %          * 4.544538e-02
 child_process\\child-process-params.js params=2 methodName="execFile" n=1000          0.68 %          * 2.451089e-02
 child_process\\child-process-params.js params=2 methodName="execFileSync" n=1000     -0.30 %            5.470930e-01
 child_process\\child-process-params.js params=2 methodName="execSync" n=1000          0.30 %            1.907476e-01
 child_process\\child-process-params.js params=2 methodName="spawn" n=1000             0.40 %         ** 6.548929e-03
 child_process\\child-process-params.js params=2 methodName="spawnSync" n=1000         0.51 %            2.825198e-01
 child_process\\child-process-params.js params=3 methodName="exec" n=1000              0.37 %            2.182035e-01
 child_process\\child-process-params.js params=3 methodName="execFile" n=1000          0.86 %        *** 1.298733e-06
 child_process\\child-process-params.js params=3 methodName="execFileSync" n=1000      0.22 %            7.333930e-01
 child_process\\child-process-params.js params=3 methodName="spawn" n=1000             0.95 %         ** 7.015692e-03
 child_process\\child-process-params.js params=3 methodName="spawnSync" n=1000        -0.43 %            4.574352e-01
 child_process\\child-process-params.js params=4 methodName="execFile" n=1000          1.15 %        *** 6.094645e-07

ChALkeR · 2017-02-24T12:48:20Z

lib/child_process.js

@@ -195,6 +195,8 @@ exports.execFile = function(file /*, args, options, callback*/) {

  var ex = null;

+  var cmd = file;
+


Why is this change needed?

It prevents named parameter from leaking in a lower scope causing deopt. See https://bugs.chromium.org/p/v8/issues/detail?id=6010

I've just upped this line from the inner exithandler() function. It seems this does not cause any scope conflict.

Eh… Understood.
Perhaps this deserves some kind of comment there, /cc @mscdex.

Also /cc @nodejs/v8 and @targos.

I don't think that this is the only place that is affected by our update to V8 5.6, I expect more deopts like this.

This better should be fixed on the V8 side instead of rewriting all the places in Node.js where a named function parameter is used in a child scope.

It happens only if arguments[i] is read, so it could not be a huge number of cases.

I believe the whole arguments[..] and named parameter mixed usage deopt has been around for awhile, long before V8 5.6?

@mscdex These two factors do not cause deopt in:

v8 4.5.103.45 (Node.js 4.8.0 x64)
v8 5.1.281.93 (Node.js 6.10.0 x64)
v8 5.5.372.40 (Node.js 7.6.0 x64)

And I have not seen this case in any articles on v8 optimization.

@mscdex

I believe the whole arguments[..] and named parameter mixed usage deopt has been around for awhile, long before V8 5.6?

Only outside of the strict mode if I remember things correctly, but I could be mistaken.

The linked bug is about the code in the strict mode.

@ChALkeR There was a slightly different case in the sloppy mode: reassigning a defined parameter + mentioning arguments.

cjihrig · 2017-02-24T15:17:20Z

This seems like the wrong approach to me. It introduces more uses of arguments instead of fewer. Could the changes to normalizeExecArgs() and normalizeSpawnArgs() be made by adding named function parameters instead (I think you just made similar changes in dgram)?

vsemozhetbyt · 2017-02-24T15:27:55Z

@cjihrig I was not sure I should refactor it, I've just tried not to be very intrusive. But yes, I can try to add named parameters.

vsemozhetbyt · 2017-02-24T18:11:15Z

@cjihrig Fixed for normalizeExecArgs() and normalizeSpawnArgs(). I've tried to copy the approach in the dgram.

vsemozhetbyt · 2017-02-24T18:13:15Z

Build and tests are OK. I will launch a new benchmark run and will post the results a bit later.

cjihrig

This looks better to me. One last comment though. Is there a performance hit for reassigning named parameters. It looks like we might be able to get away with not having separate variables from the parameters. I know we do that in other places in core (and it looks cleaner), but again, I'm not sure about any performance hit.

cjihrig · 2017-02-24T18:14:13Z

lib/child_process.js

  let options;
  let callback;

-  if (typeof arguments[1] === 'function') {
+  if (options_ && typeof options_ === 'function') {


You shouldn't need the leading options_ now, right?

I've though it would eliminate unneeded type check / function call (isArray() below) on undefined and would be a bit faster. Should I delete them?

It's hard to know what V8 will do from version to version, but won't the leading options_ require a conversion to a boolean and comparison against true?

It seems so) So should I delete them?

Seems fine to me.

cjihrig · 2017-02-24T18:15:07Z

lib/child_process.js

-    options = arguments[2];
-  } else if (arguments[1] !== undefined &&
-             (arguments[1] === null || typeof arguments[1] !== 'object')) {
+  if (args_ && Array.isArray(args_)) {


Same comment here about the leading args_

vsemozhetbyt · 2017-02-24T18:33:02Z

@cjihrig If I remember right, in the ES6 named parameters and arguments are disconnected, so the redefining of named parameters has less performance impact. However, there is still a popular style rule for this case.

Maybe this simplification could be done as a separate PR by a more experienced contributor? It seems there are many other similar cases in the libs.

cjihrig · 2017-02-24T18:37:59Z

However, there is still a popular style rule for this case.

I don't think that matters. That's more of a style preference. Since we already do it in core, I think it only matters if it hurts performance.

Maybe this simplification could be done as a separate PR by a more experienced contributor?

I'd rather get this figured out in one PR than having a second PR to change the exact same thing (churn, backporting overhead, etc.).

vsemozhetbyt · 2017-02-24T18:40:35Z

So should I wait till some more opinions or should I remove the mediatorial params now?

cjihrig · 2017-02-24T18:44:36Z

@mscdex probably knows the answer :-)

jasnell · 2017-02-24T18:56:00Z

benchmark/child_process/child-process-params.js

+
+  const method = cp[methodName];
+
+  if (methodName === 'exec') {


This would be more manageable as a switch (methodName) {} structure... with each case handled in a separate function.

I will try.

jasnell · 2017-02-24T18:56:44Z

benchmark/child_process/child-process-params.js

+  const method = cp[methodName];
+
+  if (methodName === 'exec') {
+    if (params === 1) {


This would also be better as a switch (params) {} block

Hopefully fixed.

vsemozhetbyt · 2017-02-24T22:57:45Z

New benchmark with amended commits:

                                                                                  improvement confidence      p.value
 child_process\\child-process-params.js params=1 methodName="exec" n=1000              0.81 %        *** 0.0003641932
 child_process\\child-process-params.js params=1 methodName="execFile" n=1000          0.69 %          * 0.0463235196
 child_process\\child-process-params.js params=1 methodName="execFileSync" n=1000     -0.15 %            0.8497728229
 child_process\\child-process-params.js params=1 methodName="execSync" n=1000          0.02 %            0.8812585510
 child_process\\child-process-params.js params=1 methodName="spawn" n=1000             0.49 %            0.0689109073
 child_process\\child-process-params.js params=1 methodName="spawnSync" n=1000        -0.75 %            0.3482902808
 child_process\\child-process-params.js params=2 methodName="exec" n=1000              0.26 %            0.2333704471
 child_process\\child-process-params.js params=2 methodName="execFile" n=1000          0.84 %          * 0.0352815086
 child_process\\child-process-params.js params=2 methodName="execFileSync" n=1000     -0.51 %            0.4581887918
 child_process\\child-process-params.js params=2 methodName="execSync" n=1000          0.10 %            0.3326917806
 child_process\\child-process-params.js params=2 methodName="spawn" n=1000             0.47 %            0.1045985310
 child_process\\child-process-params.js params=2 methodName="spawnSync" n=1000         0.25 %            0.7458324210
 child_process\\child-process-params.js params=3 methodName="exec" n=1000              0.41 %            0.0730991610
 child_process\\child-process-params.js params=3 methodName="execFile" n=1000          0.70 %            0.1057560550
 child_process\\child-process-params.js params=3 methodName="execFileSync" n=1000      0.92 %            0.1793174691
 child_process\\child-process-params.js params=3 methodName="spawn" n=1000             0.79 %         ** 0.0079657734
 child_process\\child-process-params.js params=3 methodName="spawnSync" n=1000         0.27 %            0.6358194310
 child_process\\child-process-params.js params=4 methodName="execFile" n=1000          0.78 %            0.1491293323

vsemozhetbyt · 2017-02-28T12:14:32Z

@cjihrig, @jasnell How should I proceed?

cjihrig · 2017-02-28T13:22:35Z

@vsemozhetbyt would you be OK with using your benchmark to investigate the idea from #11535 (review). Then, we can get this through the CI and landed.

vsemozhetbyt · 2017-03-01T00:42:25Z

@cjihrig It seems OK:

                                                                                  improvement confidence      p.value
 child_process\\child-process-params.js params=1 methodName="exec" n=1000              0.58 %            1.169842e-01
 child_process\\child-process-params.js params=1 methodName="execFile" n=1000          0.74 %          * 1.589342e-02
 child_process\\child-process-params.js params=1 methodName="execFileSync" n=1000      1.04 %            1.439200e-01
 child_process\\child-process-params.js params=1 methodName="execSync" n=1000          0.30 %          * 3.973428e-02
 child_process\\child-process-params.js params=1 methodName="spawn" n=1000             0.23 %            2.825498e-01
 child_process\\child-process-params.js params=1 methodName="spawnSync" n=1000         0.86 %            3.056963e-01
 child_process\\child-process-params.js params=2 methodName="exec" n=1000              0.79 %         ** 3.069827e-03
 child_process\\child-process-params.js params=2 methodName="execFile" n=1000          1.25 %        *** 4.789426e-06
 child_process\\child-process-params.js params=2 methodName="execFileSync" n=1000      0.04 %            9.563199e-01
 child_process\\child-process-params.js params=2 methodName="execSync" n=1000          0.14 %            1.048517e-01
 child_process\\child-process-params.js params=2 methodName="spawn" n=1000             0.13 %            7.351656e-01
 child_process\\child-process-params.js params=2 methodName="spawnSync" n=1000         0.18 %            8.267698e-01
 child_process\\child-process-params.js params=3 methodName="exec" n=1000              0.55 %          * 4.067630e-02
 child_process\\child-process-params.js params=3 methodName="execFile" n=1000          1.27 %        *** 1.786565e-06
 child_process\\child-process-params.js params=3 methodName="execFileSync" n=1000      0.24 %            6.705638e-01
 child_process\\child-process-params.js params=3 methodName="spawn" n=1000             0.07 %            7.901901e-01
 child_process\\child-process-params.js params=3 methodName="spawnSync" n=1000        -0.61 %            3.780685e-01
 child_process\\child-process-params.js params=4 methodName="execFile" n=1000          1.30 %        *** 3.215549e-07

cjihrig · 2017-03-01T00:47:25Z

Awesome. The code looks cleaner, and has less use of arguments. LGTM. Can you squash it please?

Fixed functions: normalizeExecArgs(), exports.execFile(), normalizeSpawnArguments() Refs: #10323 Refs: https://bugs.chromium.org/p/v8/issues/detail?id=6010

vsemozhetbyt · 2017-03-01T00:57:09Z

Squashed.

cjihrig · 2017-03-01T01:00:43Z

CI: https://ci.nodejs.org/job/node-test-pull-request/6634/

cjihrig · 2017-03-01T15:30:28Z

Landed in dd81d53. Thanks for working through this.

Removed or fixed use of arguments in execFile(), normalizeExecArgs(), and normalizeSpawnArguments(). Refs: #10323 Refs: https://bugs.chromium.org/p/v8/issues/detail?id=6010 PR-URL: #11535 Reviewed-By: Colin Ihrig <cjihrig@gmail.com> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Michaël Zasso <targos@protonmail.com>

evanlucas · 2017-03-07T11:28:35Z

This is not landing cleanly on v7.x-staging. Mind backporting?

vsemozhetbyt · 2017-03-07T20:54:15Z

@evanlucas I never backported a PR. I shall try to learn how to do this and then shall try to apply if nobody does it sooner.

vsemozhetbyt · 2017-03-08T13:28:55Z

@evanlucas PTAL: #11748

Removed or fixed use of arguments in execFile(), normalizeExecArgs(), and normalizeSpawnArguments(). Refs: #10323 Refs: https://bugs.chromium.org/p/v8/issues/detail?id=6010 Backport-Of: #11535 PR-URL: #11748 Reviewed-By: James M Snell <jasnell@gmail.com>

gibfahn · 2017-06-17T20:24:15Z

See #11748 (comment)

Should this be backported to v6.x-staging? If yes please follow the guide and raise a backport PR, if no let me know or add the dont-land-on label.

vsemozhetbyt · 2017-06-17T22:58:47Z

@gibfahn Hopefully done: #13752

Remove use of arguments in normalizeExecArgs() and normalizeSpawnArguments(). Refs: #10323 PR-URL: #11535 Backport-PR-URL: #13752 Reviewed-By: Colin Ihrig <cjihrig@gmail.com> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Michaël Zasso <targos@protonmail.com>

nodejs-github-bot added the child_process Issues and PRs related to the child_process subsystem. label Feb 24, 2017

mscdex added the performance Issues and PRs related to the performance of Node.js. label Feb 24, 2017

ChALkeR reviewed Feb 24, 2017

View reviewed changes

cjihrig approved these changes Feb 24, 2017

View reviewed changes

jasnell requested changes Feb 24, 2017

View reviewed changes

jasnell approved these changes Feb 27, 2017

View reviewed changes

child_process: fix deoptimizing use of arguments

20480ff

Fixed functions: normalizeExecArgs(), exports.execFile(), normalizeSpawnArguments() Refs: #10323 Refs: https://bugs.chromium.org/p/v8/issues/detail?id=6010

targos approved these changes Mar 1, 2017

View reviewed changes

cjihrig closed this Mar 1, 2017

vsemozhetbyt deleted the fix-child-process-arguments-1 branch March 1, 2017 15:38

vsemozhetbyt mentioned this pull request Mar 2, 2017

Possibly deoptimizing use of arguments in libs #10323

Closed

evanlucas added the dont-land-on-v7.x label Mar 7, 2017

vsemozhetbyt mentioned this pull request Mar 8, 2017

(v7.x backport) child_process: fix deoptimizing use of arguments #11748

Closed

3 tasks

jasnell mentioned this pull request Apr 4, 2017

8.0.0 Release Proposal #12220

Closed

MylesBorins added backported-to-v7.x and removed dont-land-on-v7.x labels Apr 17, 2017

gibfahn mentioned this pull request Jun 15, 2017

Auditing for 6.11.1 nodejs/Release#230

Closed

3 tasks

gibfahn added the backport-requested-v6.x label Jun 17, 2017

vsemozhetbyt mentioned this pull request Jun 17, 2017

(v6.x backport) child_process: fix deoptimizing use of arguments #13752

Closed

3 tasks

gibfahn added backported-to-v6.x and removed backport-requested-v6.x labels Jun 20, 2017

MylesBorins mentioned this pull request Jul 18, 2017

v6.11.2 proposal #14356

Merged

		@@ -195,6 +195,8 @@ exports.execFile = function(file /, args, options, callback/) {

		var ex = null;

		var cmd = file;

child_process: fix deoptimizing use of arguments #11535

child_process: fix deoptimizing use of arguments #11535

Conversation

vsemozhetbyt commented Feb 24, 2017

Checklist

Affected core subsystem(s)

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

vsemozhetbyt Feb 24, 2017 • edited Loading

Choose a reason for hiding this comment

mscdex Feb 24, 2017 • edited Loading

Choose a reason for hiding this comment

vsemozhetbyt Feb 24, 2017 • edited Loading

Choose a reason for hiding this comment

ChALkeR Feb 24, 2017 • edited Loading

Choose a reason for hiding this comment

vsemozhetbyt Feb 24, 2017 • edited Loading

Choose a reason for hiding this comment

cjihrig commented Feb 24, 2017

vsemozhetbyt commented Feb 24, 2017 • edited Loading

vsemozhetbyt commented Feb 24, 2017

vsemozhetbyt commented Feb 24, 2017

cjihrig left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

vsemozhetbyt commented Feb 24, 2017

cjihrig commented Feb 24, 2017

vsemozhetbyt commented Feb 24, 2017

cjihrig commented Feb 24, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

vsemozhetbyt commented Feb 24, 2017

vsemozhetbyt commented Feb 28, 2017 • edited Loading

cjihrig commented Feb 28, 2017

vsemozhetbyt commented Mar 1, 2017

cjihrig commented Mar 1, 2017

vsemozhetbyt commented Mar 1, 2017

cjihrig commented Mar 1, 2017

cjihrig commented Mar 1, 2017

evanlucas commented Mar 7, 2017

vsemozhetbyt commented Mar 7, 2017

vsemozhetbyt commented Mar 8, 2017

gibfahn commented Jun 17, 2017

vsemozhetbyt commented Jun 17, 2017

vsemozhetbyt Feb 24, 2017 •

edited

Loading

mscdex Feb 24, 2017 •

edited

Loading

vsemozhetbyt Feb 24, 2017 •

edited

Loading

ChALkeR Feb 24, 2017 •

edited

Loading

vsemozhetbyt Feb 24, 2017 •

edited

Loading

vsemozhetbyt commented Feb 24, 2017 •

edited

Loading

vsemozhetbyt commented Feb 28, 2017 •

edited

Loading