Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

child_process: improve ipc performance #13459

Merged
merged 2 commits into from
Jun 9, 2017

Conversation

mscdex
Copy link
Contributor

@mscdex mscdex commented Jun 5, 2017

These commits reduce/improve nextTick() usage in the child_process IPC implementation.

I should point out that I wasn't sure if we should ever need nextTick() when receiving/emitting messages, so for now I just made sure that both all internal messages and only the first non-internal message in a group are emitted within the same tick.

There is a substantial performance improvement when removing nextTick() altogether for emitted messages. My thought was that these messages should always happen on at least the next tick anyway, so perhaps it would be safe to remove it completely? All tests pass with nextTick() completely removed from handleMessage().

Results with the current changes with the included benchmark:

                                                                           improvement confidence      p.value
 cluster/echo.js n=100000 sendsPerBroadcast=1 payload="object" workers=1      15.62 %        *** 1.440552e-08
 cluster/echo.js n=100000 sendsPerBroadcast=1 payload="string" workers=1      18.59 %        *** 3.478095e-10
 cluster/echo.js n=100000 sendsPerBroadcast=10 payload="object" workers=1     25.71 %        *** 6.436977e-26
 cluster/echo.js n=100000 sendsPerBroadcast=10 payload="string" workers=1     33.88 %        *** 1.812799e-22

CI: https://ci.nodejs.org/job/node-test-pull-request/8473/

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

@mscdex mscdex added child_process Issues and PRs related to the child_process subsystem. performance Issues and PRs related to the performance of Node.js. labels Jun 5, 2017
@nodejs-github-bot nodejs-github-bot added the child_process Issues and PRs related to the child_process subsystem. label Jun 5, 2017
return (message !== null &&
typeof message === 'object' &&
typeof message.cmd === 'string' &&
message.cmd.length > INTERNAL_PREFIX.length &&
Copy link
Contributor

Choose a reason for hiding this comment

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

Shave off a few more millis with const INTERNALPREFIX_LENGTH = INTERNAL_PREFIX.length

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Again, I know it can be simplified and probably optimized, but I opted to leave it as-is for now.

function isInternal(message) {
return (message !== null &&
typeof message === 'object' &&
typeof message.cmd === 'string' &&
Copy link
Contributor

Choose a reason for hiding this comment

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

[question] since internal messages are rare, will adding a Object.hasOwnProperty(message, 'cmd') && fail faster than typeof message.cmd === 'string'?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

I know this could be simplified but I opted to leave it as-is for now.

@@ -456,16 +456,22 @@ function setupChannel(target, channel) {
}
chunks[0] = jsonBuffer + chunks[0];

var nextTick = false;
for (var i = 0; i < numCompleteChunks; i++) {
var message = JSON.parse(chunks[i]);
Copy link
Contributor

Choose a reason for hiding this comment

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

Why not check for isInternal on the raw chunk?

Copy link
Contributor Author

@mscdex mscdex Jun 5, 2017

Choose a reason for hiding this comment

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

It has to be parsed first. chunks here is just an array of (complete) JSON strings.

Copy link
Contributor

Choose a reason for hiding this comment

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

I meant implement a string based check. Can't we assume "innerMessages" will have a {cmd: NODE_ prefix? or at least include a , cmd: NODE_ substring? Might also be able to distinguish between three cases: user/inner/inner_with_handle, and turn the following compound if into a switch

Copy link
Contributor Author

Choose a reason for hiding this comment

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

I'd rather not make assumptions about JSON.stringify() output.

target.emit(eventName, message, handle);
});
function isInternal(message) {
return (message !== null &&
Copy link
Contributor

Choose a reason for hiding this comment

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

even if we don't do a string based check, returning 0,1,2 (for user / inner / NODE_HANDLE) will enable using a switch in L466

Copy link
Contributor Author

@mscdex mscdex Jun 5, 2017

Choose a reason for hiding this comment

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

It's unlikely it'll matter much, I'll let someone else work on it if they want. My main focus on this PR is nextTick() usage.

@refack
Copy link
Contributor

refack commented Jun 5, 2017

@mscdex Thanks for the answers 🎩

@mscdex
Copy link
Contributor Author

mscdex commented Jun 8, 2017

Results with V8 5.9 now in master:

                                                                           improvement confidence      p.value
 cluster/echo.js n=100000 sendsPerBroadcast=1 payload="object" workers=1      11.22 %        *** 4.196927e-10
 cluster/echo.js n=100000 sendsPerBroadcast=1 payload="string" workers=1      14.63 %        *** 4.344752e-10
 cluster/echo.js n=100000 sendsPerBroadcast=10 payload="object" workers=1     17.93 %        *** 1.171912e-15
 cluster/echo.js n=100000 sendsPerBroadcast=10 payload="string" workers=1     26.25 %        *** 6.978421e-19

Copy link
Member

@bnoordhuis bnoordhuis left a comment

Choose a reason for hiding this comment

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

Nice optimization, LGTM.

return;
}
for (id in cluster.workers) {
const worker = cluster.workers[id];
Copy link
Member

Choose a reason for hiding this comment

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

Nothing wrong with this but you could ES6-ify it with for (const worker of Object.values(cluster.workers)). Whatever your preference.

I suppose it could even be a little faster because Object.values() doesn't walk the prototype chain but that's probably only a marginal benefit (and I didn't test so my intuition might very well be wrong.)

Copy link
Contributor Author

@mscdex mscdex Jun 8, 2017

Choose a reason for hiding this comment

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

FWIW I literally copied it from the cluster docs. I know we were trying to be more on the conservative side when it comes to introducing ES* features to benchmarks because of possible incompatibilities with older branches. For example, Object.values() does not exist in node v4.x or v6.x (although it's apparently behind a flag in recent v6.x releases).

PR-URL: nodejs#13459
Reviewed-By: Refael Ackermann <refack@gmail.com>
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
PR-URL: nodejs#13459
Reviewed-By: Refael Ackermann <refack@gmail.com>
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
@mscdex mscdex merged commit 8208fda into nodejs:master Jun 9, 2017
@mscdex mscdex deleted the child_process-ipc-perf branch June 9, 2017 05:41
addaleax pushed a commit that referenced this pull request Jun 10, 2017
PR-URL: #13459
Reviewed-By: Refael Ackermann <refack@gmail.com>
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
addaleax pushed a commit that referenced this pull request Jun 10, 2017
PR-URL: #13459
Reviewed-By: Refael Ackermann <refack@gmail.com>
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
@addaleax addaleax mentioned this pull request Jun 10, 2017
@gibfahn gibfahn mentioned this pull request Jun 15, 2017
3 tasks
cjihrig added a commit to cjihrig/node that referenced this pull request Jun 24, 2017
This commit fixes a regression related to IPC 'message'
events. When messages are not emitted in the next tick,
a 'message' handler that throws can break the IPC read
loop.

Refs: nodejs#6909
Refs: nodejs#13459
Refs: nodejs#13648
PR-URL: nodejs#13856
Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
addaleax pushed a commit that referenced this pull request Jun 29, 2017
This commit fixes a regression related to IPC 'message'
events. When messages are not emitted in the next tick,
a 'message' handler that throws can break the IPC read
loop.

Refs: #6909
Refs: #13459
Refs: #13648
PR-URL: #13856
Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
addaleax pushed a commit that referenced this pull request Jul 11, 2017
This commit fixes a regression related to IPC 'message'
events. When messages are not emitted in the next tick,
a 'message' handler that throws can break the IPC read
loop.

Refs: #6909
Refs: #13459
Refs: #13648
PR-URL: #13856
Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
@MylesBorins MylesBorins added baking-for-lts PRs that need to wait before landing in a LTS release. lts-watch-v6.x labels Jul 17, 2017
@MylesBorins
Copy link
Contributor

should this land in LTS? If so it will need to bake a bit longer.

Please change labels as appropriate

addaleax pushed a commit that referenced this pull request Jul 18, 2017
This commit fixes a regression related to IPC 'message'
events. When messages are not emitted in the next tick,
a 'message' handler that throws can break the IPC read
loop.

Refs: #6909
Refs: #13459
Refs: #13648
PR-URL: #13856
Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
@MylesBorins MylesBorins removed the baking-for-lts PRs that need to wait before landing in a LTS release. label Aug 17, 2018
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
child_process Issues and PRs related to the child_process subsystem. performance Issues and PRs related to the performance of Node.js.
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants