benchmark: support for multiple http benchmarkers #8140

bzoz · 2016-08-17T13:08:21Z

Checklist

make -j4 test (UNIX), or vcbuild test nosign (Windows) passes
documentation is changed or added
commit message follows commit guidelines

Affected core subsystem(s)

benchmark

Description of change

Continued from #7180

Add support for multiple HTTP benchmarkers. Adds autocannon as the secondary benchmarker.

This allows for all http benchmarks to be executed under Windows. All available tools (wrk and autocannon) will be used to run HTTP benchmarks. It will fail if none of the tools is installed.

cc @nodejs/benchmarking

This adds support for multiple HTTP benchmarkers. Adds autocannon as the secondary benchmarker.

jasnell · 2016-08-17T16:52:47Z

benchmark/common.js

+      return false;
+    else
+      return true;
+  };


Perhaps simplify this to just:

return !(result.error && result.error.code === 'ENOENT');

jasnell · 2016-08-17T16:57:35Z

Good stuff. Left a few comments. @mscdex @nodejs/benchmarking

AndreasMadsen · 2016-08-17T19:38:17Z

benchmark/README.md

+* When running the benchmakrs, set `NODE_HTTP_BENCHMARKER` environment variable
+to desired benchmarker.
+* To select the default benchmarker for a particular benchmark, specify it as
+`benchmarker` key (e.g. `benchmarker: 'wrk'`) in configuration passed to


I don't think this is a good idea. The options to createBenchmark should just be the benchmark parameters. If you want this feature, I think it would be much better to add an option object to bench.http.

var bench = common.createBenchmark(main, { num: [1, 4, 8, 16], size: [1, 64, 256], c: [100], benchmarker: ['wrk'] }); function main(conf) { bench.http({ url: '/', duration: 10, connections: conf.c, benchmarker: conf.benchmarker }, function () { ... }); }

bzoz · 2016-08-18T08:29:28Z

Updated, PTAL.

I've moved the benchmarkers code to http-benchmarkers.js.

bzoz · 2016-08-23T09:03:31Z

Updated the PR again, @AndreasMadsen PTAL

Benchmark options are now passed as object. For each benchmarker node will be restarted. By default only one benchmarker will be used.

AndreasMadsen · 2016-08-23T09:53:59Z

benchmark/common.js

-      process.exit(1);
+    self.report(result, elapsed);
+    if (cb) {
+      cb(0);


This is a little odd. I think you should make the .run signature

http_benchmarkers.run(options, function(error, results) { ... })

that way you can check the error and call the callback appropriately. Actually I don't care so much if the behaviour is the same. It is just that we should avoid calling process.exit() from more than one file, as that makes the program difficult to reason about. This way the process.exit() logic can be in common.js.

AndreasMadsen · 2016-08-23T10:10:48Z

@bzoz Great. I think this is much better.

AndreasMadsen · 2016-08-23T10:11:17Z

benchmark/http-benchmarkers.js

+                            new WrkBenchmarker() ];
+
+var default_http_benchmarker;
+var supported_http_benchmarkers = [];


Use const

bzoz · 2016-08-23T13:41:57Z

@AndreasMadsen updated with your suggestions

AndreasMadsen · 2016-08-23T13:57:20Z

benchmark/_http-benchmarkers.js

+  this.name = 'autocannon';
+}
+
+AutocannonBenchmarker.prototype.autocannon_exe = process.platform === 'win32'


I don't think strings should be put on the prototype. I would just evaluate it in the constructor.

bzoz · 2016-08-24T15:36:13Z

Updated, PTAL

As for ENV - the --set thing will not work. If the overriden key is not it configuration already, there will be TypeError in common.js here. Not all http benchmarks use benchmarking tool (e. g. bench-parser.js), so it won't work there.

AndreasMadsen · 2016-08-24T15:47:54Z

If the overriden key is not it configuration already, there will be TypeError in common.js here.

That we could fix, either such it defaults to a string or simply skip the property. I think the latter would be best.

bzoz · 2016-08-26T08:59:31Z

I think it would be better to just add this as string. Otherwise I think it would be confusing for users - it would seem that sometimes it does not work. Also, no feedback when one would misspell config option.

Anyhow - I'll change common.js so that it will assume string type and drop the ENV thing.

bzoz · 2016-08-26T11:09:48Z

Updated, PTAL

bzoz · 2016-08-26T11:17:55Z

@AndreasMadsen BTW, why the "dont-land-on-v*.x" labels?

AndreasMadsen · 2016-08-26T11:41:44Z

benchmark/_http-benchmarkers.js

+                       'instructions.'));
+    return;
+  }
+  var benchmarker = benchmarkers[options.benchmarker];


use const

AndreasMadsen · 2016-08-26T11:55:13Z

BTW, why the "dont-land-on-v*.x" labels?

Some major changes to the benchmark suite has been made, this PR depends on those changes and thus it can't land on v6 or earlier. See #7890

AndreasMadsen · 2016-08-26T13:00:31Z

I don't want to hold this any more and I don't have this much time to review code. So except for the minor const and let issues, it LGTM.

The current version is much better than original, however I still think there are some very implicit things that I don't like (but can tolerate). What is implicit:

adding benchmark as an config option when bench.http is executed.
- this is added such the used benchmarker is shown in the output.
--set benchmarker= also has an effect when there is no explicit benchmarker option.
- this is to support --set benchmarker when there is no explicit benchmarker option.
misspelled options are not shown in the output or options object just the extra_options object.
- I know I sugested they should just be ignored, I was wrong.

I think all this can be solved by just enforcing explicitly setting the benchmarker parameter in the createBenchmark options object.

var bench = common.createBenchmark(main, {
  num: [1, 4, 8, 16],
  size: [1, 64, 256],
  c: [100],
  benchmarker: bench.default_http_benchmarker
});

function main(conf) {
  bench.http({
    url: '/', 
    duration: 10,
    connections: conf.c,
    benchmarker: conf.benchmarker
  }, function () { ... });
}

This way the benchmarker is added to the output and no special logic is needed for --set benchmarker.

Yes, if --set benchmarker is used, benchmarker= will be also be added to the output of benchmarks that doesn't use it. I don't think this is a big issue. If it turns out to be an issue, we can filter the output to only show options passed explicitly to createBenchmark. But this is a tradeoff between readability and transparency.

bzoz · 2016-08-26T13:25:27Z

I would like to keep it the way it is now - without explicitly adding default_http_benchmarker and with proper output for other benchmarks. I've updated that var/let/const thingy.

Thanks for all the suggestions!

AndreasMadsen · 2016-08-26T13:39:28Z

/cc @jasnell @mscdex @mcollina, please review

AndreasMadsen · 2016-08-26T13:46:39Z

Let's also cc @nodejs/collaborators as this is a fairly big addition and someone might have stronger opinions.

mcollina · 2016-08-26T18:07:34Z

LGTM

bzoz · 2016-08-29T14:35:36Z

Any more opinions?

If not, I would like to land this tomorrow.

AndreasMadsen · 2016-08-29T15:57:10Z

Maybe you shouldn't put me under Reviewed-By since I'm not 100% onboard. I don't know what the policy is.

mcollina · 2016-08-29T16:26:40Z

@AndreasMadsen can you please recap why you are not onboard? Just to understand for someone jumping into this later on.

AndreasMadsen · 2016-08-29T16:37:38Z

@mcollina I've covered that in #8140 (comment). tl;dr: there are a few implicit things which I think are confusion/surprising, and they could become explicit with minimal effort.

mcollina · 2016-08-29T16:51:57Z

can we get you to to agree, or some other opinion?

AndreasMadsen · 2016-08-29T17:03:10Z

@mcollina I do these reviews in my spare time. I'm sure we could agree if given enough time, but my time is quite limited these days (exams, assignments, etc.).

edit: I hope you don't take it the wrong way. I can understand that someone suddenly deciding not to spend more spare time on someone else can be offensive. I had reasonable time two weeks ago, but very little these weeks. I'm sorry.

This adds support for multiple HTTP benchmarkers. Adds autocannon as the secondary benchmarker. PR-URL: #8140 Reviewed-By: Matteo Collina <matteo.collina@gmail.com>

bzoz · 2016-08-31T16:02:43Z

CI: https://ci.nodejs.org/job/node-test-pull-request/3903/ (failures unrealated)

Landed in b1bbc68

@AndreasMadsen I did as you asked. Anyhow, thanks for all your input! It was very helpful in improving the quality of this PR.

benchmark: support for multiple http benchmarkers

3d14782

This adds support for multiple HTTP benchmarkers. Adds autocannon as the secondary benchmarker.

bzoz added windows Issues and PRs related to the Windows platform. benchmark Issues and PRs related to the benchmark subsystem. labels Aug 17, 2016

nodejs-github-bot added the build Issues and PRs related to build files or the CI. label Aug 17, 2016

bzoz mentioned this pull request Aug 17, 2016

benchmark: use autocannon instead of wrk #7180

Closed

2 tasks

jasnell reviewed Aug 17, 2016
View reviewed changes

AndreasMadsen reviewed Aug 17, 2016
View reviewed changes

fixup: nits, benchmarkers in separate file

6113cec

AndreasMadsen mentioned this pull request Aug 18, 2016

benchmark: http/simple is unreliable #8139

Closed

Options as object, benchmarker as configuration

721a1c5

AndreasMadsen reviewed Aug 23, 2016
View reviewed changes

Rename file, style changes, wrk as default

52521b9

AndreasMadsen reviewed Aug 23, 2016
View reviewed changes

AndreasMadsen added the dont-land-on-v6.x label Aug 24, 2016

Move stderr.pipe, clarify documentation

687a64f

Drop ENV, support --set, remove names tables

3a8a9c8

fixup: documentation wording

ed3ad0f

AndreasMadsen reviewed Aug 26, 2016
View reviewed changes

benchmark/_http-benchmarkers.js

'instructions.'));

return;

}

var benchmarker = benchmarkers[options.benchmarker];

Copy link

Member

AndreasMadsen Aug 26, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

use const

Change var to let or const

7cd3daa

bzoz added a commit that referenced this pull request Aug 31, 2016

benchmark: support for multiple http benchmarkers

b1bbc68

This adds support for multiple HTTP benchmarkers. Adds autocannon as the secondary benchmarker. PR-URL: #8140 Reviewed-By: Matteo Collina <matteo.collina@gmail.com>

bzoz closed this Aug 31, 2016

gibfahn mentioned this pull request Jun 15, 2017

Auditing for 6.11.1 nodejs/Release#230

Closed

3 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

benchmark: support for multiple http benchmarkers #8140

benchmark: support for multiple http benchmarkers #8140

bzoz commented Aug 17, 2016

jasnell Aug 17, 2016

jasnell commented Aug 17, 2016

AndreasMadsen Aug 17, 2016 •

edited

Loading

bzoz commented Aug 18, 2016

bzoz commented Aug 23, 2016

AndreasMadsen Aug 23, 2016 •

edited

Loading

AndreasMadsen commented Aug 23, 2016

AndreasMadsen Aug 23, 2016

bzoz commented Aug 23, 2016

AndreasMadsen Aug 23, 2016 •

edited

Loading

bzoz commented Aug 24, 2016

AndreasMadsen commented Aug 24, 2016

bzoz commented Aug 26, 2016

bzoz commented Aug 26, 2016

bzoz commented Aug 26, 2016

AndreasMadsen Aug 26, 2016

AndreasMadsen commented Aug 26, 2016 •

edited

Loading

AndreasMadsen commented Aug 26, 2016 •

edited

Loading

bzoz commented Aug 26, 2016

AndreasMadsen commented Aug 26, 2016

AndreasMadsen commented Aug 26, 2016

mcollina commented Aug 26, 2016

bzoz commented Aug 29, 2016

AndreasMadsen commented Aug 29, 2016

mcollina commented Aug 29, 2016

AndreasMadsen commented Aug 29, 2016 •

edited

Loading

mcollina commented Aug 29, 2016

AndreasMadsen commented Aug 29, 2016 •

edited

Loading

bzoz commented Aug 31, 2016

benchmark: support for multiple http benchmarkers #8140

benchmark: support for multiple http benchmarkers #8140

Conversation

bzoz commented Aug 17, 2016

Checklist

Affected core subsystem(s)

Description of change

jasnell Aug 17, 2016

Choose a reason for hiding this comment

jasnell commented Aug 17, 2016

AndreasMadsen Aug 17, 2016 • edited Loading

Choose a reason for hiding this comment

bzoz commented Aug 18, 2016

bzoz commented Aug 23, 2016

AndreasMadsen Aug 23, 2016 • edited Loading

Choose a reason for hiding this comment

AndreasMadsen commented Aug 23, 2016

AndreasMadsen Aug 23, 2016

Choose a reason for hiding this comment

bzoz commented Aug 23, 2016

AndreasMadsen Aug 23, 2016 • edited Loading

Choose a reason for hiding this comment

bzoz commented Aug 24, 2016

AndreasMadsen commented Aug 24, 2016

bzoz commented Aug 26, 2016

bzoz commented Aug 26, 2016

bzoz commented Aug 26, 2016

AndreasMadsen Aug 26, 2016

Choose a reason for hiding this comment

AndreasMadsen commented Aug 26, 2016 • edited Loading

AndreasMadsen commented Aug 26, 2016 • edited Loading

bzoz commented Aug 26, 2016

AndreasMadsen commented Aug 26, 2016

AndreasMadsen commented Aug 26, 2016

mcollina commented Aug 26, 2016

bzoz commented Aug 29, 2016

AndreasMadsen commented Aug 29, 2016

mcollina commented Aug 29, 2016

AndreasMadsen commented Aug 29, 2016 • edited Loading

mcollina commented Aug 29, 2016

AndreasMadsen commented Aug 29, 2016 • edited Loading

bzoz commented Aug 31, 2016

AndreasMadsen Aug 17, 2016 •

edited

Loading

AndreasMadsen Aug 23, 2016 •

edited

Loading

AndreasMadsen Aug 23, 2016 •

edited

Loading

AndreasMadsen commented Aug 26, 2016 •

edited

Loading

AndreasMadsen commented Aug 26, 2016 •

edited

Loading

AndreasMadsen commented Aug 29, 2016 •

edited

Loading

AndreasMadsen commented Aug 29, 2016 •

edited

Loading