-
Notifications
You must be signed in to change notification settings - Fork 29.8k
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
timers: improve setTimeout/Interval performance #8661
timers: improve setTimeout/Interval performance #8661
Conversation
@@ -94,8 +94,8 @@ const TIMEOUT_MAX = 2147483647; // 2^31-1 | |||
// | |||
// - key = time in milliseconds | |||
// - value = linked list | |||
const refedLists = {}; | |||
const unrefedLists = {}; | |||
const refedLists = Object.create(null); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I remember we had a long discussion about this, I think for http. Have we ever come to a conclusion whether this is a good idea?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The objects here are not exposed, so it's not an issue in this case.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
OK. And it does improve performance? Because - at least for me - it decreases readability.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I didn't measure it specifically, but it does remove unnecessary hash table entries. It may also help if V8 sees only numeric keys being set on the object?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@fhinkel Just curious. Do you mean using a Map instead of Object.create? #7581 (comment)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
For me it increases readability, but please check performance - I recall there being a regression when using .create(null)
at one point.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@Florian-R Thanks, that's the discussion I was thinking of. But in this case I meant {} vs .create(null).
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@benjamingr I'm not sure what regression you're referring to. These objects are created at startup, not during runtime, so the time to create them doesn't matter as much.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I understand this is not about performance, but just out of curiosity, I measured the performance difference.
console.time("{}"); for (var i = 100000; i--;) { var a = {}; } console.timeEnd("{}"); var create = Object.create; console.time("create"); for (var i = 100000; i--;) { var b = create(null); } console.timeEnd("create");
{}: 5.792ms
create: 15.221ms
Probably, calling a function is heavier than {}
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Literals are always faster than instantiation and builder function calls, in JS and most languages.
@@ -317,51 +322,76 @@ exports.enroll = function(item, msecs) { | |||
*/ | |||
|
|||
|
|||
exports.setTimeout = function(callback, after) { | |||
exports.setTimeout = function(callback, after, arg1, arg2, arg3) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks. I like this much better than using arguments.
|
||
if (!(after >= 1 && after <= TIMEOUT_MAX)) { | ||
after = 1; // schedule on next tick, follows browser behaviour | ||
var len = arguments.length; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Can we check for typeof args1 == 'undefined' rather than using length? Or is undefined a valid value for args1?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I would say it's a valid value.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the speedup. LGTM.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Changes look good to me. Left one comment but it should not block merging.
@@ -94,8 +94,8 @@ const TIMEOUT_MAX = 2147483647; // 2^31-1 | |||
// | |||
// - key = time in milliseconds | |||
// - value = linked list | |||
const refedLists = {}; | |||
const unrefedLists = {}; | |||
const refedLists = Object.create(null); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
For me it increases readability, but please check performance - I recall there being a regression when using .create(null)
at one point.
} | ||
|
||
L.append(list, item); | ||
assert(!L.isEmpty(list)); // list is not empty | ||
} | ||
|
||
function createTimersList(msecs, unrefed) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Unsure why we'd want this in an extra function
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think extracting it into a function increases readability.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The main reason is to keep the insert()
size down, but it does also help a bit with readability I suppose.
@@ -317,51 +322,76 @@ exports.enroll = function(item, msecs) { | |||
*/ | |||
|
|||
|
|||
exports.setTimeout = function(callback, after) { | |||
exports.setTimeout = function(callback, after, arg1, arg2, arg3) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
...args
?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Rest args are still slow.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Great move.
|
||
if (!(repeat >= 1 && repeat <= TIMEOUT_MAX)) { | ||
function createRepeatTimeout(callback, repeat, args) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Why is this a new function?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
To keep function sizes down.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
IMHO, there's no repetead call, you add a short flow path (and creates a mini and fake cyclomatic complexity), function is in the same script and right under the function (so you just added brackets and more names to read). Function call is slower than direct structured code (and even minimal performance gains, in this call stack queen, is important).
I would not do this separation.
timer._onTimeout = wrapper; | ||
timer._repeat = ontimeout; | ||
var timer = new Timeout(repeat, callback, args); | ||
timer._repeat = repeat; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
unsure if changing this won't have side-effects somewhere
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I grepped lib/ and ._repeat
isn't used anywhere except timers.js.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm more concerned about user code, I do understand that it is unlikely but it has been the way it was an awfully long time in the wild.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Would you rather I leave the variable set to null
since it wouldn't be used anymore and add a new property that stores the interval value?
|
||
function rearm(timer) { | ||
// If timer is unref'd (or was - it's permanently removed from the list.) | ||
if (timer._handle && timer instanceof Timeout) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
should only be timer._handle
I think
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Unfortunately not, as _http_outgoing
and net
have a ._handle
and manually enroll()
pre-made objects. Node will crash since those ._handle
instances are not only not Timeout
instances, but they also do not have a .start()
method.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
And the reason this change is necessary now is that the rearming code no longer has access to the scope where the Timeout
instance was created.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
hmmmm ok
callback.apply(timer, args); | ||
} | ||
} | ||
if (timer._repeat != null) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
i would just check if (timer._repeat)
iirc that is what is done elsewhere.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I don't think it's safe really since ._repeat
is now being used to store the interval value, which would not work for the 0
interval.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@mscdex you can't have an interval of 0. It becomes 1.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Changed.
const refedLists = {}; | ||
const unrefedLists = {}; | ||
const refedLists = Object.create(null); | ||
const unrefedLists = Object.create(null); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I don't see any tangible benefit to doing this.
We only use Number value keys.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
As was previously mentioned, I didn't measure the change specifically, but it does remove unnecessary hash table entries. It may also help if V8 sees only numeric keys being set on the object? It has nothing to do with key name collisions in this case.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
@@ -229,7 +234,7 @@ function tryOnTimeout(timer, list) { | |||
timer._called = true; | |||
var threw = true; | |||
try { | |||
timer._onTimeout(); | |||
ontimeout(timer); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
if ontimeout()
throws will it deo-opt that function?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Which function? tryOnTimeout()
? ontimeout()
? The user's callback?
I don't know that throwing causes a deopt or not, but adding a try-*
block certainly does cause a deopt for the containing function.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
but adding a
try-*
block certainly does cause a deopt for the containing function.
Right, which is why it was constrained to within one separate function.
I'm curious if a throw fork a user callback de-opts ontimeout()
along the way though, since that could be somewhat unpleasant.
Maybe we shouldn't worry about it? Idk the sate of benchmarking this stuff is already so bad 😐
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I just wrote a little test and as far as I can tell a function (or any calling functions in the call stack) shouldn't get deopted if that function throws.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Try deopt any container function. Node.js should, IMHO, always test for nullity and use flags/optionals and never try-catch in the inner functions... This pattern is not really a requirement in functional programming and there's huge performance benefits on native calls (basically, in any platform too. try-catch system is expensive).
timer._onTimeout = wrapper; | ||
timer._repeat = ontimeout; | ||
var timer = new Timeout(repeat, callback, args); | ||
timer._repeat = repeat; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm more concerned about user code, I do understand that it is unlikely but it has been the way it was an awfully long time in the wild.
d7a7828
to
b8fcf3d
Compare
args = [arg1, arg2]; | ||
} else if (len > 4) { | ||
args = [arg1, arg2, arg3]; | ||
for (var i = 5; i < arguments.length; i++) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It probably doesn't matter much, but you could use len
here instead of arguments.length
.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Changed.
b8fcf3d
to
c6268a7
Compare
CI again after recent changes: https://ci.nodejs.org/job/node-test-pull-request/4188/ |
Still LGTM. |
@mscdex ... it would also be good to have a citgm run on this. |
citgm looks ok, osx and ppcle had a few |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
@Fishrock123 Thoughts? |
seems fine to me |
c6268a7
to
92c3623
Compare
One last CI before landing: https://ci.nodejs.org/job/node-test-pull-request/4347/ EDIT: CI is green |
This commit improves timers performance by making functions inlineable and avoiding the creation of extra closures/functions. This commit also makes setTimeout/Interval argument handling consistent with that of setImmediate. These changes give ~22% improvement in the existing 'breadth' timers benchmark. PR-URL: nodejs#8661 Reviewed-By: Franziska Hinkelmann <franziska.hinkelmann@gmail.com> Reviewed-By: Benjamin Gruenbaum <benjamingr@gmail.com> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Ilkka Myller <ilkka.myller@nodefield.com> Reviewed-By: Jeremiah Senkpiel <fishrock123@rocketmail.com>
92c3623
to
c8c2544
Compare
This commit improves timers performance by making functions inlineable and avoiding the creation of extra closures/functions. This commit also makes setTimeout/Interval argument handling consistent with that of setImmediate. These changes give ~22% improvement in the existing 'breadth' timers benchmark. PR-URL: #8661 Reviewed-By: Franziska Hinkelmann <franziska.hinkelmann@gmail.com> Reviewed-By: Benjamin Gruenbaum <benjamingr@gmail.com> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Ilkka Myller <ilkka.myller@nodefield.com> Reviewed-By: Jeremiah Senkpiel <fishrock123@rocketmail.com>
adding LTS watch for now, although it will need quite a bit of time to bake before landing. @mscdex do you see any reason we should not land it in the future? |
@thealphanerd No, it should be ok. |
This commit improves timers performance by making functions inlineable and avoiding the creation of extra closures/functions. This commit also makes setTimeout/Interval argument handling consistent with that of setImmediate. These changes give ~22% improvement in the existing 'breadth' timers benchmark. PR-URL: #8661 Reviewed-By: Franziska Hinkelmann <franziska.hinkelmann@gmail.com> Reviewed-By: Benjamin Gruenbaum <benjamingr@gmail.com> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Ilkka Myller <ilkka.myller@nodefield.com> Reviewed-By: Jeremiah Senkpiel <fishrock123@rocketmail.com>
* fs: - `SyncWriteStream` now inherits from `Stream.Writable`. (Anna Henningsen) #8830 - Practically, this means that when stdio is piped to a file, stdout and stderr will still be `Writable` streams. - `fs.existsSync()` has been undeprecated. `fs.exists()` remains deprecated. (Dan Fabulich) #8364 * http: `http.request()` now accepts a `timeout` option. (Rene Weber) #8101 * module: The module loader now maintains its own realpath cache. (Anna Henningsen) #8100 * npm: Upgraded to 3.10.8 (Kat Marchán) #8706 * stream: `Duplex` streams now show proper `instanceof Stream.Writable`. (Anna Henningsen) #8834 * timers: Improved `setTimeout`/`Interval` performance by up to 22%. (Brian White) #8661 PR-URL: #9034
* fs: - `SyncWriteStream` now inherits from `Stream.Writable`. (Anna Henningsen) #8830 - Practically, this means that when stdio is piped to a file, stdout and stderr will still be `Writable` streams. - `fs.existsSync()` has been undeprecated. `fs.exists()` remains deprecated. (Dan Fabulich) #8364 * http: `http.request()` now accepts a `timeout` option. (Rene Weber) #8101 * module: The module loader now maintains its own realpath cache. (Anna Henningsen) #8100 * npm: Upgraded to 3.10.8 (Kat Marchán) #8706 * stream: `Duplex` streams now show proper `instanceof Stream.Writable`. (Anna Henningsen) #8834 * timers: Improved `setTimeout`/`Interval` performance by up to 22%. (Brian White) #8661 PR-URL: #9034
* fs: - `SyncWriteStream` now inherits from `Stream.Writable`. (Anna Henningsen) nodejs/node#8830 - Practically, this means that when stdio is piped to a file, stdout and stderr will still be `Writable` streams. - `fs.existsSync()` has been undeprecated. `fs.exists()` remains deprecated. (Dan Fabulich) nodejs/node#8364 * http: `http.request()` now accepts a `timeout` option. (Rene Weber) nodejs/node#8101 * module: The module loader now maintains its own realpath cache. (Anna Henningsen) nodejs/node#8100 * npm: Upgraded to 3.10.8 (Kat Marchan) nodejs/node#8706 * stream: `Duplex` streams now show proper `instanceof Stream.Writable`. (Anna Henningsen) nodejs/node#8834 * timers: Improved `setTimeout`/`Interval` performance by up to 22%. (Brian White) nodejs/node#8661 Signed-off-by: Ilkka Myller <ilkka.myller@nodefield.com>
* fs: - `SyncWriteStream` now inherits from `Stream.Writable`. (Anna Henningsen) nodejs/node#8830 - Practically, this means that when stdio is piped to a file, stdout and stderr will still be `Writable` streams. - `fs.existsSync()` has been undeprecated. `fs.exists()` remains deprecated. (Dan Fabulich) nodejs/node#8364 * http: `http.request()` now accepts a `timeout` option. (Rene Weber) nodejs/node#8101 * module: The module loader now maintains its own realpath cache. (Anna Henningsen) nodejs/node#8100 * npm: Upgraded to 3.10.8 (Kat Marchan) nodejs/node#8706 * stream: `Duplex` streams now show proper `instanceof Stream.Writable`. (Anna Henningsen) nodejs/node#8834 * timers: Improved `setTimeout`/`Interval` performance by up to 22%. (Brian White) nodejs/node#8661 Signed-off-by: Ilkka Myller <ilkka.myller@nodefield.com>
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There's some point to enhance. Object.create is bad at that first lines (literal is faster than Object.create() and I do not see real memory benefits).
There's space for micro optimizations, but the overall improvements are great and makes the commit rich enough.
@@ -94,8 +94,8 @@ const TIMEOUT_MAX = 2147483647; // 2^31-1 | |||
// | |||
// - key = time in milliseconds | |||
// - value = linked list | |||
const refedLists = {}; | |||
const unrefedLists = {}; | |||
const refedLists = Object.create(null); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Literals are always faster than instantiation and builder function calls, in JS and most languages.
@@ -229,7 +234,7 @@ function tryOnTimeout(timer, list) { | |||
timer._called = true; | |||
var threw = true; | |||
try { | |||
timer._onTimeout(); | |||
ontimeout(timer); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Try deopt any container function. Node.js should, IMHO, always test for nullity and use flags/optionals and never try-catch in the inner functions... This pattern is not really a requirement in functional programming and there's huge performance benefits on native calls (basically, in any platform too. try-catch system is expensive).
@@ -317,51 +322,76 @@ exports.enroll = function(item, msecs) { | |||
*/ | |||
|
|||
|
|||
exports.setTimeout = function(callback, after) { | |||
exports.setTimeout = function(callback, after, arg1, arg2, arg3) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Great move.
|
||
if (!(repeat >= 1 && repeat <= TIMEOUT_MAX)) { | ||
function createRepeatTimeout(callback, repeat, args) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
IMHO, there's no repetead call, you add a short flow path (and creates a mini and fake cyclomatic complexity), function is in the same script and right under the function (so you just added brackets and more names to read). Function call is slower than direct structured code (and even minimal performance gains, in this call stack queen, is important).
I would not do this separation.
this._repeat = null; | ||
} | ||
|
||
|
||
function unrefdHandle() { | ||
this.owner._onTimeout(); | ||
ontimeout(this.owner); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This is always better for internal workings. ALWAYS, Node community.
Avoid this, globals, function calls, function separation in minimal pieces and inner functions... and we'll always have fine code with max JS performance.
@leodutra this was already merged. If you want, you could submit a new PR. |
I think performance-related improvements to |
Checklist
make -j4 test
(UNIX), orvcbuild test nosign
(Windows) passesAffected core subsystem(s)
Description of change
This commit improves timers performance by making functions inlineable and avoiding the creation of extra closures/functions.
This commit also makes setTimeout/Interval argument handling consistent with that of setImmediate.
These changes give ~22% improvement in the existing 'breadth' timers benchmark.