-
Notifications
You must be signed in to change notification settings - Fork 29.6k
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: cross JS/C++ border less frequently for setImmediate() #17064
Conversation
Hmmm, I'm pretty @mscdex has touched the immediates code more than I have. It lives in the same file and uses some the same general structure but I haven't really touched the details much at all. |
|
||
if (async_hook_fields[kDestroy] > 0) { | ||
emitDestroy(immediate[async_id_symbol]); | ||
} | ||
} | ||
|
||
immediate._onImmediate = null; | ||
|
||
immediateQueue.remove(immediate); |
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.
Does it make sense to move this into the if (or even flip the if to if (immediate._destroyed) return;
)
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.
You can try it but all of this code is so fragile I’d rather not touch more than necessary
uv_check_start(env->immediate_check_handle(), CheckImmediate); | ||
// Idle handle is needed only to stop the event loop from blocking in poll. | ||
uv_idle_start(env->immediate_idle_handle(), | ||
[](uv_idle_t*){ /* do nothing, just keep the loop running */ }); |
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 see that the comment came from the previous impl, but it seems redundant with the comment above the line...
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.
Maybe. One comment is describing the what, one comment is describing the why, so I’m okay with having two.
Edit: On second thought, yes, I guess this can be removed if you like
Local<Value> value, | ||
const PropertyCallbackInfo<void>& info) { | ||
Environment* env = Environment::GetCurrent(info); | ||
if (MaybeStopImmediate(env)) |
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.
What's the case here? A callback from uv after the Immediate was cleared on the JS side?
If so maybe add a 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.
Yes, that is one way to trigger this, but the check here isn’t semantically tied to that – the general idea is that if there’s nothing to run, don’t bother calling into JS, no matter why there is nothing to run.
(In particular, it’s not quite worked out yet how this is going to work with other native code also using this mechanism, as hinted at in the commit message, and it’s quite possible that that may not be the only way to trigger it.)
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 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.
Woo... Nice
@@ -512,6 +513,11 @@ inline void Environment::set_fs_stats_field_array(double* fields) { | |||
fs_stats_field_array_ = fields; | |||
} | |||
|
|||
inline AliasedBuffer<uint32_t, v8::Uint32Array>& | |||
Environment::scheduled_immediate_count() { |
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 not really a fan of returning a mutable reference. I realize other AliasedBuffer things do that too but IMO they shouldn't, it obscures whether call sites deal with a copy or a reference.
(I'll allow that it looks like a little nicer coupled with array syntax.)
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 know. :) I don’t like it when one can’t tell the difference either, but in these cases it should be pretty clear.
All the code being removed/moved in |
This removes the `process._needImmediateCallback` property and its semantics of having a 1/0 switch that tells C++ whether immediates are currently scheduled. Instead, a counter keeping track of all immediates is created, that can be increased on `setImmediate()` or decreased when an immediate is run or cleared. This is faster, because rather than reading/writing a C++ getter, this operation can be performed as a direct memory read/write via a typed array. The only C++ call that is left to make is activating the native handles upon creation of the first `Immediate` after the queue is empty. One other (good!) side-effect is that `immediate._destroyed` now reliably tells whether an `immediate` is still scheduled to run or not. Also, as a nice extra, this should make it easier to implement an internal variant of `setImmediate` for C++ that piggybacks off the same mechanism, which should be useful at least for async hooks and HTTP/2. Benchmark results: $ ./node benchmark/compare.js --new ./node --old ./node-master-1b093cb93df0 --runs 10 --filter immediate timers | Rscript benchmark/compare.R [00:08:53|% 100| 4/4 files | 20/20 runs | 1/1 configs]: Done improvement confidence p.value timers/immediate.js type="breadth" thousands=2000 25.61 % ** 1.432301e-03 timers/immediate.js type="breadth1" thousands=2000 7.66 % 1.320233e-01 timers/immediate.js type="breadth4" thousands=2000 4.61 % 5.669053e-01 timers/immediate.js type="clear" thousands=2000 311.40 % *** 3.896291e-07 timers/immediate.js type="depth" thousands=2000 17.54 % ** 9.755389e-03 timers/immediate.js type="depth1" thousands=2000 17.09 % *** 7.176229e-04 timers/set-immediate-breadth-args.js millions=5 10.63 % * 4.250034e-02 timers/set-immediate-breadth.js millions=10 20.62 % *** 9.150439e-07 timers/set-immediate-depth-args.js millions=10 17.97 % *** 6.819135e-10
6e26656
to
8310986
Compare
@Trott Thanks for the suggestion! The coverage for existing code remains the same with this patch, so that’s nice. That being said – 100 % coverage for timers would be really nice… |
Landed in ae558af |
This removes the `process._needImmediateCallback` property and its semantics of having a 1/0 switch that tells C++ whether immediates are currently scheduled. Instead, a counter keeping track of all immediates is created, that can be increased on `setImmediate()` or decreased when an immediate is run or cleared. This is faster, because rather than reading/writing a C++ getter, this operation can be performed as a direct memory read/write via a typed array. The only C++ call that is left to make is activating the native handles upon creation of the first `Immediate` after the queue is empty. One other (good!) side-effect is that `immediate._destroyed` now reliably tells whether an `immediate` is still scheduled to run or not. Also, as a nice extra, this should make it easier to implement an internal variant of `setImmediate` for C++ that piggybacks off the same mechanism, which should be useful at least for async hooks and HTTP/2. Benchmark results: $ ./node benchmark/compare.js --new ./node --old ./node-master-1b093cb93df0 --runs 10 --filter immediate timers | Rscript benchmark/compare.R [00:08:53|% 100| 4/4 files | 20/20 runs | 1/1 configs]: Done improvement confidence p.value timers/immediate.js type="breadth" thousands=2000 25.61 % ** 1.432301e-03 timers/immediate.js type="breadth1" thousands=2000 7.66 % 1.320233e-01 timers/immediate.js type="breadth4" thousands=2000 4.61 % 5.669053e-01 timers/immediate.js type="clear" thousands=2000 311.40 % *** 3.896291e-07 timers/immediate.js type="depth" thousands=2000 17.54 % ** 9.755389e-03 timers/immediate.js type="depth1" thousands=2000 17.09 % *** 7.176229e-04 timers/set-immediate-breadth-args.js millions=5 10.63 % * 4.250034e-02 timers/set-immediate-breadth.js millions=10 20.62 % *** 9.150439e-07 timers/set-immediate-depth-args.js millions=10 17.97 % *** 6.819135e-10 PR-URL: #17064 Reviewed-By: Refael Ackermann <refack@gmail.com> Reviewed-By: Minwoo Jung <minwoo@nodesource.com> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
|
This removes the `process._needImmediateCallback` property and its semantics of having a 1/0 switch that tells C++ whether immediates are currently scheduled. Instead, a counter keeping track of all immediates is created, that can be increased on `setImmediate()` or decreased when an immediate is run or cleared. This is faster, because rather than reading/writing a C++ getter, this operation can be performed as a direct memory read/write via a typed array. The only C++ call that is left to make is activating the native handles upon creation of the first `Immediate` after the queue is empty. One other (good!) side-effect is that `immediate._destroyed` now reliably tells whether an `immediate` is still scheduled to run or not. Also, as a nice extra, this should make it easier to implement an internal variant of `setImmediate` for C++ that piggybacks off the same mechanism, which should be useful at least for async hooks and HTTP/2. Benchmark results: $ ./node benchmark/compare.js --new ./node --old ./node-master-1b093cb93df0 --runs 10 --filter immediate timers | Rscript benchmark/compare.R [00:08:53|% 100| 4/4 files | 20/20 runs | 1/1 configs]: Done improvement confidence p.value timers/immediate.js type="breadth" thousands=2000 25.61 % ** 1.432301e-03 timers/immediate.js type="breadth1" thousands=2000 7.66 % 1.320233e-01 timers/immediate.js type="breadth4" thousands=2000 4.61 % 5.669053e-01 timers/immediate.js type="clear" thousands=2000 311.40 % *** 3.896291e-07 timers/immediate.js type="depth" thousands=2000 17.54 % ** 9.755389e-03 timers/immediate.js type="depth1" thousands=2000 17.09 % *** 7.176229e-04 timers/set-immediate-breadth-args.js millions=5 10.63 % * 4.250034e-02 timers/set-immediate-breadth.js millions=10 20.62 % *** 9.150439e-07 timers/set-immediate-depth-args.js millions=10 17.97 % *** 6.819135e-10 PR-URL: #17064 Reviewed-By: Refael Ackermann <refack@gmail.com> Reviewed-By: Minwoo Jung <minwoo@nodesource.com> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
This removes the `process._needImmediateCallback` property and its semantics of having a 1/0 switch that tells C++ whether immediates are currently scheduled. Instead, a counter keeping track of all immediates is created, that can be increased on `setImmediate()` or decreased when an immediate is run or cleared. This is faster, because rather than reading/writing a C++ getter, this operation can be performed as a direct memory read/write via a typed array. The only C++ call that is left to make is activating the native handles upon creation of the first `Immediate` after the queue is empty. One other (good!) side-effect is that `immediate._destroyed` now reliably tells whether an `immediate` is still scheduled to run or not. Also, as a nice extra, this should make it easier to implement an internal variant of `setImmediate` for C++ that piggybacks off the same mechanism, which should be useful at least for async hooks and HTTP/2. Benchmark results: $ ./node benchmark/compare.js --new ./node --old ./node-master-1b093cb93df0 --runs 10 --filter immediate timers | Rscript benchmark/compare.R [00:08:53|% 100| 4/4 files | 20/20 runs | 1/1 configs]: Done improvement confidence p.value timers/immediate.js type="breadth" thousands=2000 25.61 % ** 1.432301e-03 timers/immediate.js type="breadth1" thousands=2000 7.66 % 1.320233e-01 timers/immediate.js type="breadth4" thousands=2000 4.61 % 5.669053e-01 timers/immediate.js type="clear" thousands=2000 311.40 % *** 3.896291e-07 timers/immediate.js type="depth" thousands=2000 17.54 % ** 9.755389e-03 timers/immediate.js type="depth1" thousands=2000 17.09 % *** 7.176229e-04 timers/set-immediate-breadth-args.js millions=5 10.63 % * 4.250034e-02 timers/set-immediate-breadth.js millions=10 20.62 % *** 9.150439e-07 timers/set-immediate-depth-args.js millions=10 17.97 % *** 6.819135e-10 PR-URL: #17064 Reviewed-By: Refael Ackermann <refack@gmail.com> Reviewed-By: Minwoo Jung <minwoo@nodesource.com> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Should this land on 6.x and 8.x? I'm leaning towards leaving it to bake for a while first, thoughts? |
This removes the `process._needImmediateCallback` property and its semantics of having a 1/0 switch that tells C++ whether immediates are currently scheduled. Instead, a counter keeping track of all immediates is created, that can be increased on `setImmediate()` or decreased when an immediate is run or cleared. This is faster, because rather than reading/writing a C++ getter, this operation can be performed as a direct memory read/write via a typed array. The only C++ call that is left to make is activating the native handles upon creation of the first `Immediate` after the queue is empty. One other (good!) side-effect is that `immediate._destroyed` now reliably tells whether an `immediate` is still scheduled to run or not. Also, as a nice extra, this should make it easier to implement an internal variant of `setImmediate` for C++ that piggybacks off the same mechanism, which should be useful at least for async hooks and HTTP/2. Benchmark results: $ ./node benchmark/compare.js --new ./node --old ./node-master-1b093cb93df0 --runs 10 --filter immediate timers | Rscript benchmark/compare.R [00:08:53|% 100| 4/4 files | 20/20 runs | 1/1 configs]: Done improvement confidence p.value timers/immediate.js type="breadth" thousands=2000 25.61 % ** 1.432301e-03 timers/immediate.js type="breadth1" thousands=2000 7.66 % 1.320233e-01 timers/immediate.js type="breadth4" thousands=2000 4.61 % 5.669053e-01 timers/immediate.js type="clear" thousands=2000 311.40 % *** 3.896291e-07 timers/immediate.js type="depth" thousands=2000 17.54 % ** 9.755389e-03 timers/immediate.js type="depth1" thousands=2000 17.09 % *** 7.176229e-04 timers/set-immediate-breadth-args.js millions=5 10.63 % * 4.250034e-02 timers/set-immediate-breadth.js millions=10 20.62 % *** 9.150439e-07 timers/set-immediate-depth-args.js millions=10 17.97 % *** 6.819135e-10 PR-URL: nodejs#17064
This removes the `process._needImmediateCallback` property and its semantics of having a 1/0 switch that tells C++ whether immediates are currently scheduled. Instead, a counter keeping track of all immediates is created, that can be increased on `setImmediate()` or decreased when an immediate is run or cleared. This is faster, because rather than reading/writing a C++ getter, this operation can be performed as a direct memory read/write via a typed array. The only C++ call that is left to make is activating the native handles upon creation of the first `Immediate` after the queue is empty. One other (good!) side-effect is that `immediate._destroyed` now reliably tells whether an `immediate` is still scheduled to run or not. Also, as a nice extra, this should make it easier to implement an internal variant of `setImmediate` for C++ that piggybacks off the same mechanism, which should be useful at least for async hooks and HTTP/2. Benchmark results: $ ./node benchmark/compare.js --new ./node --old ./node-master-1b093cb93df0 --runs 10 --filter immediate timers | Rscript benchmark/compare.R [00:08:53|% 100| 4/4 files | 20/20 runs | 1/1 configs]: Done improvement confidence p.value timers/immediate.js type="breadth" thousands=2000 25.61 % ** 1.432301e-03 timers/immediate.js type="breadth1" thousands=2000 7.66 % 1.320233e-01 timers/immediate.js type="breadth4" thousands=2000 4.61 % 5.669053e-01 timers/immediate.js type="clear" thousands=2000 311.40 % *** 3.896291e-07 timers/immediate.js type="depth" thousands=2000 17.54 % ** 9.755389e-03 timers/immediate.js type="depth1" thousands=2000 17.09 % *** 7.176229e-04 timers/set-immediate-breadth-args.js millions=5 10.63 % * 4.250034e-02 timers/set-immediate-breadth.js millions=10 20.62 % *** 9.150439e-07 timers/set-immediate-depth-args.js millions=10 17.97 % *** 6.819135e-10 PR-URL: nodejs#17064 Backport-PR-URL: nodejs#18179 Reviewed-By: Refael Ackermann <refack@gmail.com> Reviewed-By: Minwoo Jung <minwoo@nodesource.com> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
This removes the `process._needImmediateCallback` property and its semantics of having a 1/0 switch that tells C++ whether immediates are currently scheduled. Instead, a counter keeping track of all immediates is created, that can be increased on `setImmediate()` or decreased when an immediate is run or cleared. This is faster, because rather than reading/writing a C++ getter, this operation can be performed as a direct memory read/write via a typed array. The only C++ call that is left to make is activating the native handles upon creation of the first `Immediate` after the queue is empty. One other (good!) side-effect is that `immediate._destroyed` now reliably tells whether an `immediate` is still scheduled to run or not. Also, as a nice extra, this should make it easier to implement an internal variant of `setImmediate` for C++ that piggybacks off the same mechanism, which should be useful at least for async hooks and HTTP/2. Benchmark results: $ ./node benchmark/compare.js --new ./node --old ./node-master-1b093cb93df0 --runs 10 --filter immediate timers | Rscript benchmark/compare.R [00:08:53|% 100| 4/4 files | 20/20 runs | 1/1 configs]: Done improvement confidence p.value timers/immediate.js type="breadth" thousands=2000 25.61 % ** 1.432301e-03 timers/immediate.js type="breadth1" thousands=2000 7.66 % 1.320233e-01 timers/immediate.js type="breadth4" thousands=2000 4.61 % 5.669053e-01 timers/immediate.js type="clear" thousands=2000 311.40 % *** 3.896291e-07 timers/immediate.js type="depth" thousands=2000 17.54 % ** 9.755389e-03 timers/immediate.js type="depth1" thousands=2000 17.09 % *** 7.176229e-04 timers/set-immediate-breadth-args.js millions=5 10.63 % * 4.250034e-02 timers/set-immediate-breadth.js millions=10 20.62 % *** 9.150439e-07 timers/set-immediate-depth-args.js millions=10 17.97 % *** 6.819135e-10 Backport-PR-URL: #18179 PR-URL: #17064 Reviewed-By: Refael Ackermann <refack@gmail.com> Reviewed-By: Minwoo Jung <minwoo@nodesource.com> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
This removes the
process._needImmediateCallback
propertyand its semantics of having a 1/0 switch that tells C++ whether
immediates are currently scheduled.
Instead, a counter keeping track of all immediates is created,
that can be increased on
setImmediate()
or decreased when animmediate is run or cleared.
This is faster, because rather than writing/reading a C++ getter,
this operation can be performed as a direct memory read/write via
a typed array. The only C++ call that is left to make is
activating the native handles upon creation of the first
Immediate
after the queue is empty.One other (good!) side-effect is that
immediate._destroyed
nowreliably tells whether an
immediate
is still scheduled to run or not.Also, as a nice extra, this should make it easier to implement
an internal variant of
setImmediate
for C++ that piggybacksoff the same mechanism, which should be useful at least for
async hooks and HTTP/2.
(/cc @jasnell so you’re aware of this last thing 🙂)
Benchmark results:
Checklist
make -j4 test
(UNIX), orvcbuild test
(Windows) passesAffected core subsystem(s)
timers