events: improve events performance #50422

rluvaton · 2023-10-27T08:41:37Z

Benchmarks

Change `_events` to be a `Map`:

commit: 249c2b8

Notes:

I'm afraid this can break userland implementation of EventEmitter that are using node EventEmitter class functions (cc @ronag )
This also improves creation by a lot (and therefore improves streams creation)

15:31:01                                                                  confidence improvement accuracy (*)    (**)   (***)
15:31:01 events/ee-add-remove.js n=1000000 removeListener=0 newListener=0        ***     19.64 %       ±1.57%  ±2.09%  ±2.73%
15:31:01 events/ee-add-remove.js n=1000000 removeListener=0 newListener=1        ***     46.84 %       ±2.02%  ±2.70%  ±3.54%
15:31:01 events/ee-add-remove.js n=1000000 removeListener=1 newListener=0        ***     45.01 %       ±2.68%  ±3.60%  ±4.75%
15:31:01 events/ee-add-remove.js n=1000000 removeListener=1 newListener=1        ***     56.27 %       ±2.72%  ±3.64%  ±4.79%
15:31:01 events/ee-emit.js listeners=1 argc=0 n=2000000                          ***     32.42 %       ±9.82% ±13.07% ±17.02%
15:31:01 events/ee-emit.js listeners=1 argc=10 n=2000000                         ***     36.00 %       ±8.08% ±10.75% ±14.00%
15:31:01 events/ee-emit.js listeners=1 argc=2 n=2000000                          ***     23.57 %       ±9.57% ±12.75% ±16.62%
15:31:01 events/ee-emit.js listeners=1 argc=4 n=2000000                          ***     28.33 %       ±8.95% ±11.92% ±15.52%
15:31:01 events/ee-emit.js listeners=10 argc=0 n=2000000                                  4.48 %       ±7.03%  ±9.35% ±12.17%
15:31:01 events/ee-emit.js listeners=10 argc=10 n=2000000                                 0.02 %       ±4.28%  ±5.70%  ±7.42%
15:31:01 events/ee-emit.js listeners=10 argc=2 n=2000000                          **     -6.22 %       ±4.54%  ±6.06%  ±7.91%
15:31:01 events/ee-emit.js listeners=10 argc=4 n=2000000                          **      6.55 %       ±4.59%  ±6.13%  ±8.00%
15:31:01 events/ee-emit.js listeners=5 argc=0 n=2000000                                   1.39 %       ±9.30% ±12.37% ±16.11%
15:31:01 events/ee-emit.js listeners=5 argc=10 n=2000000                                 -0.34 %       ±7.02%  ±9.35% ±12.18%
15:31:01 events/ee-emit.js listeners=5 argc=2 n=2000000                                   5.45 %       ±8.85% ±11.78% ±15.33%
15:31:01 events/ee-emit.js listeners=5 argc=4 n=2000000                                   0.79 %       ±8.60% ±11.44% ±14.89%
15:31:01 events/ee-listen-unique.js n=1000000 events=1                           ***     30.56 %       ±2.05%  ±2.73%  ±3.56%
15:31:01 events/ee-listen-unique.js n=1000000 events=10                          ***     13.34 %       ±1.21%  ±1.62%  ±2.11%
15:31:01 events/ee-listen-unique.js n=1000000 events=2                           ***     18.32 %       ±1.99%  ±2.67%  ±3.51%
15:31:01 events/ee-listen-unique.js n=1000000 events=20                          ***     13.37 %       ±1.22%  ±1.62%  ±2.12%
15:31:01 events/ee-listen-unique.js n=1000000 events=3                           ***     19.46 %       ±1.15%  ±1.53%  ±2.00%
15:31:01 events/ee-listen-unique.js n=1000000 events=5                           ***     15.05 %       ±1.35%  ±1.80%  ±2.35%
15:31:01 events/ee-listener-count-on-prototype.js n=50000000                     ***     29.98 %       ±5.74%  ±7.70% ±10.14%
15:31:01 events/ee-listeners.js raw='false' listeners=5 n=5000000                        -2.69 %       ±7.61% ±10.13% ±13.18%
15:31:01 events/ee-listeners.js raw='false' listeners=50 n=5000000                        0.48 %       ±1.82%  ±2.42%  ±3.15%
15:31:01 events/ee-listeners.js raw='true' listeners=5 n=5000000                          6.34 %       ±8.35% ±11.12% ±14.49%
15:31:01 events/ee-listeners.js raw='true' listeners=50 n=5000000                        -1.34 %       ±4.45%  ±5.92%  ±7.71%
15:31:01 events/ee-once.js argc=0 n=20000000                                     ***     31.26 %       ±1.10%  ±1.47%  ±1.94%
15:31:01 events/ee-once.js argc=1 n=20000000                                     ***     26.62 %       ±0.60%  ±0.80%  ±1.04%
15:31:01 events/ee-once.js argc=4 n=20000000                                     ***     26.93 %       ±0.65%  ±0.86%  ±1.12%
15:31:01 events/ee-once.js argc=5 n=20000000                                     ***     27.19 %       ±0.76%  ±1.02%  ±1.32%
15:31:01 events/eventtarget-add-remove.js nListener=10 n=1000000                 ***      0.92 %       ±0.38%  ±0.51%  ±0.66%
15:31:01 events/eventtarget-add-remove.js nListener=5 n=1000000                    *      0.87 %       ±0.69%  ±0.91%  ±1.19%
15:31:01 events/eventtarget-creation.js n=1000000                                         0.60 %       ±5.04%  ±6.70%  ±8.73%
15:31:01 events/eventtarget.js listeners=1 n=1000000                                      0.90 %       ±6.28%  ±8.35% ±10.87%
15:31:01 events/eventtarget.js listeners=10 n=1000000                                     0.45 %       ±3.51%  ±4.68%  ±6.09%
15:31:01 events/eventtarget.js listeners=5 n=1000000                                      0.21 %       ±4.80%  ±6.39%  ±8.31%
15:31:01 
15:31:01 Be aware that when doing many comparisons the risk of a false-positive
15:31:01 result increases. In this case, there are 37 comparisons, you can thus
15:31:01 expect the following amount of false-positive results:
15:31:01   1.85 false positives, when considering a   5% risk acceptance (*, **, ***),
15:31:01   0.37 false positives, when considering a   1% risk acceptance (**, ***),
15:31:01   0.04 false positives, when considering a 0.1% risk acceptance (***)

I run it on streams as well to see the impact as this was my original intent: url

Change to `Map` todos:

move all our internal code away from _events and use [kEvents] instead
I saw undici used _events so need to migrate as well

~~change _events to be without __proto__: null~~

change `_events` to be without `proto: null`

Commit: eb29e97

I think the performance without proto is not fully representative as the event name can be anything so the dictionary mode is not used here

Benchmark URL

12:55:39 ++ Rscript benchmark/compare.R
12:55:40                                                                  confidence improvement accuracy (*)    (**)   (***)
12:55:40 events/ee-add-remove.js n=1000000 removeListener=0 newListener=0         **     -2.22 %       ±1.50%  ±2.00%  ±2.60%
12:55:40 events/ee-add-remove.js n=1000000 removeListener=0 newListener=1        ***     30.34 %       ±1.75%  ±2.33%  ±3.04%
12:55:40 events/ee-add-remove.js n=1000000 removeListener=1 newListener=0        ***     22.74 %       ±1.17%  ±1.56%  ±2.03%
12:55:40 events/ee-add-remove.js n=1000000 removeListener=1 newListener=1        ***     10.04 %       ±1.11%  ±1.47%  ±1.92%
12:55:40 events/ee-emit.js listeners=1 argc=0 n=2000000                          ***    159.54 %       ±8.72% ±11.64% ±15.20%
12:55:40 events/ee-emit.js listeners=1 argc=10 n=2000000                         ***    139.14 %      ±12.44% ±16.59% ±21.66%
12:55:40 events/ee-emit.js listeners=1 argc=2 n=2000000                          ***    159.56 %      ±13.83% ±18.46% ±24.15%
12:55:40 events/ee-emit.js listeners=1 argc=4 n=2000000                          ***    148.52 %      ±10.04% ±13.37% ±17.43%
12:55:40 events/ee-emit.js listeners=10 argc=0 n=2000000                         ***     13.34 %       ±7.48%  ±9.96% ±12.96%
12:55:40 events/ee-emit.js listeners=10 argc=10 n=2000000                        ***     11.22 %       ±5.66%  ±7.53%  ±9.81%
12:55:40 events/ee-emit.js listeners=10 argc=2 n=2000000                                  3.11 %       ±6.21%  ±8.27% ±10.77%
12:55:40 events/ee-emit.js listeners=10 argc=4 n=2000000                         ***     12.07 %       ±6.04%  ±8.04% ±10.47%
12:55:40 events/ee-emit.js listeners=5 argc=0 n=2000000                          ***     20.76 %       ±9.34% ±12.44% ±16.24%
12:55:40 events/ee-emit.js listeners=5 argc=10 n=2000000                         ***     17.29 %       ±8.71% ±11.59% ±15.10%
12:55:40 events/ee-emit.js listeners=5 argc=2 n=2000000                            *     13.10 %       ±9.93% ±13.21% ±17.20%
12:55:40 events/ee-emit.js listeners=5 argc=4 n=2000000                            *      9.82 %       ±9.51% ±12.66% ±16.48%
12:55:40 events/ee-listen-unique.js n=1000000 events=1                                    0.24 %       ±1.32%  ±1.76%  ±2.29%
12:55:40 events/ee-listen-unique.js n=1000000 events=10                          ***     -3.02 %       ±1.08%  ±1.44%  ±1.88%
12:55:40 events/ee-listen-unique.js n=1000000 events=2                           ***     -2.28 %       ±0.69%  ±0.92%  ±1.20%
12:55:40 events/ee-listen-unique.js n=1000000 events=20                           **     -2.16 %       ±1.56%  ±2.09%  ±2.73%
12:55:40 events/ee-listen-unique.js n=1000000 events=3                           ***     -2.03 %       ±0.98%  ±1.31%  ±1.71%
12:55:40 events/ee-listen-unique.js n=1000000 events=5                           ***     -4.36 %       ±0.89%  ±1.19%  ±1.55%
12:55:40 events/ee-listener-count-on-prototype.js n=50000000                     ***     -6.40 %       ±3.40%  ±4.53%  ±5.91%
12:55:40 events/ee-listeners.js raw='false' listeners=5 n=5000000                  *     -8.30 %       ±6.34%  ±8.45% ±11.02%
12:55:40 events/ee-listeners.js raw='false' listeners=50 n=5000000                       -1.08 %       ±1.71%  ±2.28%  ±2.97%
12:55:40 events/ee-listeners.js raw='true' listeners=5 n=5000000                         -3.00 %       ±5.69%  ±7.58%  ±9.89%
12:55:40 events/ee-listeners.js raw='true' listeners=50 n=5000000                         3.68 %       ±4.65%  ±6.18%  ±8.05%
12:55:40 events/ee-once.js argc=0 n=20000000                                             -0.21 %       ±0.64%  ±0.85%  ±1.10%
12:55:40 events/ee-once.js argc=1 n=20000000                                              0.06 %       ±0.44%  ±0.59%  ±0.76%
12:55:40 events/ee-once.js argc=4 n=20000000                                             -0.13 %       ±0.62%  ±0.83%  ±1.08%
12:55:40 events/ee-once.js argc=5 n=20000000                                              0.26 %       ±0.69%  ±0.91%  ±1.19%
12:55:40 events/eventtarget-add-remove.js nListener=10 n=1000000                  **      0.71 %       ±0.42%  ±0.56%  ±0.73%
12:55:40 events/eventtarget-add-remove.js nListener=5 n=1000000                           0.30 %       ±0.70%  ±0.94%  ±1.22%
12:55:40 events/eventtarget-creation.js n=1000000                                         1.95 %       ±3.80%  ±5.09%  ±6.72%
12:55:40 events/eventtarget.js listeners=1 n=1000000                                     -0.66 %       ±5.51%  ±7.34%  ±9.55%
12:55:40 events/eventtarget.js listeners=10 n=1000000                                    -0.32 %       ±3.15%  ±4.19%  ±5.46%
12:55:40 events/eventtarget.js listeners=5 n=1000000                                     -0.39 %       ±5.18%  ±6.90%  ±8.98%
12:55:40 
12:55:40 Be aware that when doing many comparisons the risk of a false-positive
12:55:40 result increases. In this case, there are 37 comparisons, you can thus
12:55:40 expect the following amount of false-positive results:
12:55:40   1.85 false positives, when considering a   5% risk acceptance (*, **, ***),
12:55:40   0.37 false positives, when considering a   1% risk acceptance (**, ***),
12:55:40   0.04 false positives, when considering a 0.1% risk acceptance (***)

lib/events.js

nodejs-github-bot · 2023-10-27T11:07:06Z

CI: https://ci.nodejs.org/job/node-test-pull-request/55268/

ronag · 2023-10-27T13:42:41Z

This is probably going to break someone somewhere. That being said, maybe there is a way to minimize it by returning a Proxy from the _events getter that more accurately imitates the original behavior.

Also, I'm surprised this is so much faster. Often the same events are used (e.g. for streams), so I would assume that V8 uses a shape rather than a map behind the scenes, which should be faster than a Map instance. @benjamingr wdyt?

ronag · 2023-10-27T13:44:50Z

I'm actually not that interested in ee-listen-unique and ee-once.

rluvaton · 2023-10-27T13:45:31Z

Also, I'm surprised this is so much faster. Often the same events are used (e.g. for streams), so I would assume that V8 uses a shape rather than a map behind the scenes, which should be faster than a Map instance.

{ __proto__: null } hint v8 to use the dictionary mode src/runtime/runtime-literals.cc in V8 (founded by @benjamingr)

rluvaton · 2023-10-27T13:47:51Z

This is probably going to break someone somewhere. That being said, maybe there is a way to minimize it by returning a Proxy from the _events getter that more accurately imitates the original behavior.

Often the same events are used (e.g. for streams)

you can see the results when removing __proto__ null at the bottom (collapsed)

rluvaton · 2023-10-27T13:58:32Z

This is concerning:

16:57:40                                                                                          confidence improvement accuracy (*)   (**)   (***)
16:57:40 streams/creation.js kind='duplex' n=50000000                                                     **      1.61 %       ±0.93% ±1.23%  ±1.61%
16:57:40 streams/creation.js kind='readable' n=50000000                                                  ***    -17.07 %       ±0.68% ±0.91%  ±1.19%
16:57:40 streams/creation.js kind='transform' n=50000000                                                 ***     24.20 %       ±1.46% ±1.95%  ±2.55%
16:57:40 streams/creation.js kind='writable' n=50000000                                                           0.54 %       ±0.92% ±1.23%  ±1.60%
16:57:40 streams/destroy.js kind='duplex' n=1000000                                                       **     -3.85 %       ±2.25% ±2.99%  ±3.90%
16:57:40 streams/destroy.js kind='readable' n=1000000                                                            -0.90 %       ±1.66% ±2.21%  ±2.88%
16:57:40 streams/destroy.js kind='transform' n=1000000                                                           -1.38 %       ±1.91% ±2.54%  ±3.31%
16:57:40 streams/destroy.js kind='writable' n=1000000                                                            -2.12 %       ±2.19% ±2.91%  ±3.80%
16:57:40 streams/pipe-object-mode.js n=5000000                                                             *      6.22 %       ±5.34% ±7.11%  ±9.26%
16:57:40 streams/pipe.js n=5000000                                                                       ***      7.73 %       ±3.87% ±5.15%  ±6.70%
16:57:40 streams/readable-async-iterator.js sync='no' n=100000                                                   -2.28 %       ±6.37% ±8.48% ±11.05%
16:57:40 streams/readable-async-iterator.js sync='yes' n=100000                                                  -0.80 %       ±5.49% ±7.30%  ±9.51%
16:57:40 streams/readable-bigread.js n=1000                                                                *      3.49 %       ±2.92% ±3.92%  ±5.17%
16:57:40 streams/readable-bigunevenread.js n=1000                                                                -0.62 %       ±3.22% ±4.33%  ±5.72%
16:57:40 streams/readable-boundaryread.js type='buffer' n=2000                                           ***     11.62 %       ±1.13% ±1.50%  ±1.95%
16:57:40 streams/readable-boundaryread.js type='string' n=2000                                                    2.67 %       ±3.46% ±4.61%  ±6.03%
16:57:40 streams/readable-from.js type='array' n=10000000                                                         4.68 %       ±5.29% ±7.05%  ±9.18%
16:57:40 streams/readable-from.js type='async-generator' n=10000000                                      ***      2.44 %       ±0.90% ±1.20%  ±1.56%
16:57:40 streams/readable-from.js type='sync-generator-with-async-values' n=10000000                     ***      4.27 %       ±0.99% ±1.32%  ±1.71%
16:57:40 streams/readable-from.js type='sync-generator-with-sync-values' n=10000000                               2.93 %       ±4.20% ±5.59%  ±7.28%
16:57:40 streams/readable-readall.js n=5000                                                                      -1.18 %       ±6.06% ±8.06% ±10.50%
16:57:40 streams/readable-unevenread.js n=1000                                                           ***      3.97 %       ±2.02% ±2.68%  ±3.49%
16:57:40 streams/writable-manywrites.js len=1024 callback='no' writev='no' sync='no' n=100000            ***     -3.08 %       ±1.04% ±1.38%  ±1.79%
16:57:40 streams/writable-manywrites.js len=1024 callback='no' writev='no' sync='yes' n=100000                    0.94 %       ±1.97% ±2.63%  ±3.43%
16:57:40 streams/writable-manywrites.js len=1024 callback='no' writev='yes' sync='no' n=100000           ***     -6.76 %       ±2.49% ±3.34%  ±4.39%
16:57:40 streams/writable-manywrites.js len=1024 callback='no' writev='yes' sync='yes' n=100000                  -3.66 %       ±4.38% ±5.84%  ±7.63%
16:57:40 streams/writable-manywrites.js len=1024 callback='yes' writev='no' sync='no' n=100000                   -2.13 %       ±3.01% ±4.01%  ±5.22%
16:57:40 streams/writable-manywrites.js len=1024 callback='yes' writev='no' sync='yes' n=100000                  -1.16 %       ±2.15% ±2.86%  ±3.72%
16:57:40 streams/writable-manywrites.js len=1024 callback='yes' writev='yes' sync='no' n=100000          ***     -4.85 %       ±2.67% ±3.56%  ±4.66%
16:57:40 streams/writable-manywrites.js len=1024 callback='yes' writev='yes' sync='yes' n=100000                 -1.77 %       ±3.70% ±4.95%  ±6.49%
16:57:40 streams/writable-manywrites.js len=32768 callback='no' writev='no' sync='no' n=100000                    0.89 %       ±3.28% ±4.39%  ±5.76%
16:57:40 streams/writable-manywrites.js len=32768 callback='no' writev='no' sync='yes' n=100000                  -1.94 %       ±3.64% ±4.85%  ±6.31%
16:57:40 streams/writable-manywrites.js len=32768 callback='no' writev='yes' sync='no' n=100000            *     -1.35 %       ±1.11% ±1.47%  ±1.92%
16:57:40 streams/writable-manywrites.js len=32768 callback='no' writev='yes' sync='yes' n=100000                 -2.63 %       ±3.85% ±5.12%  ±6.67%
16:57:40 streams/writable-manywrites.js len=32768 callback='yes' writev='no' sync='no' n=100000                   1.63 %       ±3.13% ±4.19%  ±5.53%
16:57:40 streams/writable-manywrites.js len=32768 callback='yes' writev='no' sync='yes' n=100000                 -1.37 %       ±2.82% ±3.75%  ±4.89%
16:57:40 streams/writable-manywrites.js len=32768 callback='yes' writev='yes' sync='no' n=100000          **     -1.50 %       ±1.09% ±1.44%  ±1.88%
16:57:40 streams/writable-manywrites.js len=32768 callback='yes' writev='yes' sync='yes' n=100000                -1.05 %       ±2.24% ±2.98%  ±3.88%
16:57:40 
16:57:40 Be aware that when doing many comparisons the risk of a false-positive
16:57:40 result increases. In this case, there are 38 comparisons, you can thus
16:57:40 expect the following amount of false-positive results:
16:57:40   1.90 false positives, when considering a   5% risk acceptance (*, **, ***),
16:57:40   0.38 false positives, when considering a   1% risk acceptance (**, ***),
16:57:40   0.04 false positives, when considering a 0.1% risk acceptance (***)

benjamingr

This is neat but I don't understand how it's faster since objects in dictionary mode and maps have basically the same performance.

Also the "polyfill" for ._eventsis incorrect since if someone sets a value on it it won't get reflected in the emitter.

ronag · 2023-10-27T14:06:55Z

{ proto: null } hint v8 to use the dictionary mod

What happens if you remove that hint? Sorry it's a bit unclear what your last benchmark results are from.

rluvaton · 2023-10-27T14:09:58Z

Also the "polyfill" for ._eventsis incorrect since if someone sets a value on it it won't get reflected in the emitter.

I know I still have a todo there to convert to Proxy so it support everything

This is neat but I don't understand how it's faster since objects in dictionary mode and maps have basically the same performance.

Maybe because it by default go to the dict mode?

{ proto: null } hint v8 to use the dictionary mod

What happens if you remove that hint? Sorry it's a bit unclear what your last benchmark results are from.

the benchmarks are from the linked commit

change `_events` to be without `proto: null`

Commit: eb29e97

I think the performance without proto is not fully representative as the event name can be anything so the dictionary mode is not used here

Benchmark URL

12:55:39 ++ Rscript benchmark/compare.R
12:55:40                                                                  confidence improvement accuracy (*)    (**)   (***)
12:55:40 events/ee-add-remove.js n=1000000 removeListener=0 newListener=0         **     -2.22 %       ±1.50%  ±2.00%  ±2.60%
12:55:40 events/ee-add-remove.js n=1000000 removeListener=0 newListener=1        ***     30.34 %       ±1.75%  ±2.33%  ±3.04%
12:55:40 events/ee-add-remove.js n=1000000 removeListener=1 newListener=0        ***     22.74 %       ±1.17%  ±1.56%  ±2.03%
12:55:40 events/ee-add-remove.js n=1000000 removeListener=1 newListener=1        ***     10.04 %       ±1.11%  ±1.47%  ±1.92%
12:55:40 events/ee-emit.js listeners=1 argc=0 n=2000000                          ***    159.54 %       ±8.72% ±11.64% ±15.20%
12:55:40 events/ee-emit.js listeners=1 argc=10 n=2000000                         ***    139.14 %      ±12.44% ±16.59% ±21.66%
12:55:40 events/ee-emit.js listeners=1 argc=2 n=2000000                          ***    159.56 %      ±13.83% ±18.46% ±24.15%
12:55:40 events/ee-emit.js listeners=1 argc=4 n=2000000                          ***    148.52 %      ±10.04% ±13.37% ±17.43%
12:55:40 events/ee-emit.js listeners=10 argc=0 n=2000000                         ***     13.34 %       ±7.48%  ±9.96% ±12.96%
12:55:40 events/ee-emit.js listeners=10 argc=10 n=2000000                        ***     11.22 %       ±5.66%  ±7.53%  ±9.81%
12:55:40 events/ee-emit.js listeners=10 argc=2 n=2000000                                  3.11 %       ±6.21%  ±8.27% ±10.77%
12:55:40 events/ee-emit.js listeners=10 argc=4 n=2000000                         ***     12.07 %       ±6.04%  ±8.04% ±10.47%
12:55:40 events/ee-emit.js listeners=5 argc=0 n=2000000                          ***     20.76 %       ±9.34% ±12.44% ±16.24%
12:55:40 events/ee-emit.js listeners=5 argc=10 n=2000000                         ***     17.29 %       ±8.71% ±11.59% ±15.10%
12:55:40 events/ee-emit.js listeners=5 argc=2 n=2000000                            *     13.10 %       ±9.93% ±13.21% ±17.20%
12:55:40 events/ee-emit.js listeners=5 argc=4 n=2000000                            *      9.82 %       ±9.51% ±12.66% ±16.48%
12:55:40 events/ee-listen-unique.js n=1000000 events=1                                    0.24 %       ±1.32%  ±1.76%  ±2.29%
12:55:40 events/ee-listen-unique.js n=1000000 events=10                          ***     -3.02 %       ±1.08%  ±1.44%  ±1.88%
12:55:40 events/ee-listen-unique.js n=1000000 events=2                           ***     -2.28 %       ±0.69%  ±0.92%  ±1.20%
12:55:40 events/ee-listen-unique.js n=1000000 events=20                           **     -2.16 %       ±1.56%  ±2.09%  ±2.73%
12:55:40 events/ee-listen-unique.js n=1000000 events=3                           ***     -2.03 %       ±0.98%  ±1.31%  ±1.71%
12:55:40 events/ee-listen-unique.js n=1000000 events=5                           ***     -4.36 %       ±0.89%  ±1.19%  ±1.55%
12:55:40 events/ee-listener-count-on-prototype.js n=50000000                     ***     -6.40 %       ±3.40%  ±4.53%  ±5.91%
12:55:40 events/ee-listeners.js raw='false' listeners=5 n=5000000                  *     -8.30 %       ±6.34%  ±8.45% ±11.02%
12:55:40 events/ee-listeners.js raw='false' listeners=50 n=5000000                       -1.08 %       ±1.71%  ±2.28%  ±2.97%
12:55:40 events/ee-listeners.js raw='true' listeners=5 n=5000000                         -3.00 %       ±5.69%  ±7.58%  ±9.89%
12:55:40 events/ee-listeners.js raw='true' listeners=50 n=5000000                         3.68 %       ±4.65%  ±6.18%  ±8.05%
12:55:40 events/ee-once.js argc=0 n=20000000                                             -0.21 %       ±0.64%  ±0.85%  ±1.10%
12:55:40 events/ee-once.js argc=1 n=20000000                                              0.06 %       ±0.44%  ±0.59%  ±0.76%
12:55:40 events/ee-once.js argc=4 n=20000000                                             -0.13 %       ±0.62%  ±0.83%  ±1.08%
12:55:40 events/ee-once.js argc=5 n=20000000                                              0.26 %       ±0.69%  ±0.91%  ±1.19%
12:55:40 events/eventtarget-add-remove.js nListener=10 n=1000000                  **      0.71 %       ±0.42%  ±0.56%  ±0.73%
12:55:40 events/eventtarget-add-remove.js nListener=5 n=1000000                           0.30 %       ±0.70%  ±0.94%  ±1.22%
12:55:40 events/eventtarget-creation.js n=1000000                                         1.95 %       ±3.80%  ±5.09%  ±6.72%
12:55:40 events/eventtarget.js listeners=1 n=1000000                                     -0.66 %       ±5.51%  ±7.34%  ±9.55%
12:55:40 events/eventtarget.js listeners=10 n=1000000                                    -0.32 %       ±3.15%  ±4.19%  ±5.46%
12:55:40 events/eventtarget.js listeners=5 n=1000000                                     -0.39 %       ±5.18%  ±6.90%  ±8.98%
12:55:40 
12:55:40 Be aware that when doing many comparisons the risk of a false-positive
12:55:40 result increases. In this case, there are 37 comparisons, you can thus
12:55:40 expect the following amount of false-positive results:
12:55:40   1.85 false positives, when considering a   5% risk acceptance (*, **, ***),
12:55:40   0.37 false positives, when considering a   1% risk acceptance (**, ***),
12:55:40   0.04 false positives, when considering a 0.1% risk acceptance (***)

ronag · 2023-10-27T14:19:11Z

change _events to be without proto: null

Looks great to me!

ronag · 2023-10-27T14:20:24Z

Also what about something like the following for streams?

diff --git a/lib/internal/streams/duplex.js b/lib/internal/streams/duplex.js
index 834d875be6..2b3fe64df9 100644
--- a/lib/internal/streams/duplex.js
+++ b/lib/internal/streams/duplex.js
@@ -63,6 +63,21 @@ function Duplex(options) {
   if (!(this instanceof Duplex))
     return new Duplex(options);
 
+  this._events = {
+    close: undefined,
+    error: undefined,
+    prefinish: undefined,
+    finish: undefined,
+    drain: undefined,
+    data: undefined,
+    end: undefined,
+    pause: undefined,
+    resume: undefined,
+    readable: undefined,
+    pipe: undefined,
+    unpipe: undefined,
+  };
+
   this._readableState = new Readable.ReadableState(options, this, true);
   this._writableState = new Writable.WritableState(options, this, true);
 
diff --git a/lib/internal/streams/readable.js b/lib/internal/streams/readable.js
index 80798a35dc..83f09194a6 100644
--- a/lib/internal/streams/readable.js
+++ b/lib/internal/streams/readable.js
@@ -316,6 +316,18 @@ function Readable(options) {
   if (!(this instanceof Readable))
     return new Readable(options);
 
+  this._events = {
+    close: undefined,
+    error: undefined,
+    data: undefined,
+    end: undefined,
+    pause: undefined,
+    resume: undefined,
+    readable: undefined,
+    pipe: undefined,
+    unpipe: undefined,
+  };
+
   this._readableState = new ReadableState(options, this, false);
 
   if (options) {
diff --git a/lib/internal/streams/writable.js b/lib/internal/streams/writable.js
index 74573033ea..69fb2be7f0 100644
--- a/lib/internal/streams/writable.js
+++ b/lib/internal/streams/writable.js
@@ -382,6 +382,14 @@ function Writable(options) {
   if (!(this instanceof Writable))
     return new Writable(options);
 
+    this._events = {
+      close: undefined,
+      error: undefined,
+      prefinish: undefined,
+      finish: undefined,
+      drain: undefined,
+    };
+
   this._writableState = new WritableState(options, this, false);
 
   if (options) {

rluvaton · 2023-10-27T14:20:33Z

Do you think it's representative, I think the reason someone put __proto__ null was to be in dict mode

rluvaton · 2023-10-27T14:21:26Z

Also what about something like the following for streams?

diff --git a/lib/internal/streams/duplex.js b/lib/internal/streams/duplex.js
index 834d875be6..2b3fe64df9 100644
--- a/lib/internal/streams/duplex.js
+++ b/lib/internal/streams/duplex.js
@@ -63,6 +63,21 @@ function Duplex(options) {
   if (!(this instanceof Duplex))
     return new Duplex(options);
 
+  this._events = {
+    close: undefined,
+    error: undefined,
+    prefinish: undefined,
+    finish: undefined,
+    drain: undefined,
+    data: undefined,
+    end: undefined,
+    pause: undefined,
+    resume: undefined,
+    readable: undefined,
+    pipe: undefined,
+    unpipe: undefined,
+  };
+
   this._readableState = new Readable.ReadableState(options, this, true);
   this._writableState = new Writable.WritableState(options, this, true);
 
diff --git a/lib/internal/streams/readable.js b/lib/internal/streams/readable.js
index 80798a35dc..83f09194a6 100644
--- a/lib/internal/streams/readable.js
+++ b/lib/internal/streams/readable.js
@@ -316,6 +316,18 @@ function Readable(options) {
   if (!(this instanceof Readable))
     return new Readable(options);
 
+  this._events = {
+    close: undefined,
+    error: undefined,
+    data: undefined,
+    end: undefined,
+    pause: undefined,
+    resume: undefined,
+    readable: undefined,
+    pipe: undefined,
+    unpipe: undefined,
+  };
+
   this._readableState = new ReadableState(options, this, false);
 
   if (options) {
diff --git a/lib/internal/streams/writable.js b/lib/internal/streams/writable.js
index 74573033ea..69fb2be7f0 100644
--- a/lib/internal/streams/writable.js
+++ b/lib/internal/streams/writable.js
@@ -382,6 +382,14 @@ function Writable(options) {
   if (!(this instanceof Writable))
     return new Writable(options);
 
+    this._events = {
+      close: undefined,
+      error: undefined,
+      prefinish: undefined,
+      finish: undefined,
+      drain: undefined,
+    };
+
   this._writableState = new WritableState(options, this, false);
 
   if (options) {

I thought about that, but since I changed to Map I did not do it

benjamingr · 2023-10-27T14:51:30Z

Eh, this makes one use case slower and another one faster (changing to no proto: null), making "EventEmitter events shape specific to common use cases for streams" could be worth it, with benchmarks

lpinca · 2023-10-27T14:54:23Z

Possible duplicate of #44627 and #39939. See also #11930.

rluvaton · 2023-10-29T10:21:47Z

Supersede by #50428

events: improve creation performance

eb29e97

nodejs-github-bot added events Issues and PRs related to the events subsystem / EventEmitter. needs-ci PRs that need a full CI run. labels Oct 27, 2023

rluvaton changed the title ~~events: improve creation performance~~ events: improve events performance Oct 27, 2023

rluvaton force-pushed the improve-ee-creation branch from 4c02bdb to bb44f6f Compare October 27, 2023 10:11

rluvaton added the performance Issues and PRs related to the performance of Node.js. label Oct 27, 2023

aduh95 reviewed Oct 27, 2023

View reviewed changes

lib/events.js Outdated Show resolved Hide resolved

aduh95 reviewed Oct 27, 2023

View reviewed changes

lib/events.js Outdated Show resolved Hide resolved

aduh95 reviewed Oct 27, 2023

View reviewed changes

lib/events.js Outdated Show resolved Hide resolved

events: replace _events with Map

249c2b8

rluvaton force-pushed the improve-ee-creation branch from bb44f6f to 249c2b8 Compare October 27, 2023 10:51

rluvaton added the request-ci Add this label to start a Jenkins CI on a PR. label Oct 27, 2023

rluvaton requested a review from ronag October 27, 2023 11:06

github-actions bot removed the request-ci Add this label to start a Jenkins CI on a PR. label Oct 27, 2023

rluvaton marked this pull request as draft October 27, 2023 11:22

rluvaton removed the request for review from ronag October 27, 2023 11:22

anonrig added needs-benchmark-ci PR that need a benchmark CI run. needs-citgm PRs that need a CITGM CI run. labels Oct 27, 2023

benjamingr reviewed Oct 27, 2023

View reviewed changes

events: remove __proto__ null from _events

9d626b9

rluvaton closed this Oct 29, 2023

rluvaton deleted the improve-ee-creation branch October 29, 2023 10:21

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

events: improve events performance #50422

events: improve events performance #50422

rluvaton commented Oct 27, 2023 •

edited

Loading

nodejs-github-bot commented Oct 27, 2023

ronag commented Oct 27, 2023 •

edited

Loading

ronag commented Oct 27, 2023

rluvaton commented Oct 27, 2023 •

edited

Loading

rluvaton commented Oct 27, 2023

rluvaton commented Oct 27, 2023

benjamingr left a comment

ronag commented Oct 27, 2023

rluvaton commented Oct 27, 2023 •

edited

Loading

ronag commented Oct 27, 2023

ronag commented Oct 27, 2023

rluvaton commented Oct 27, 2023

rluvaton commented Oct 27, 2023

benjamingr commented Oct 27, 2023

lpinca commented Oct 27, 2023

rluvaton commented Oct 29, 2023

events: improve events performance #50422

events: improve events performance #50422

Conversation

rluvaton commented Oct 27, 2023 • edited Loading

Benchmarks

Change _events to be a Map:

Change to Map todos:

change _events to be without __proto__: null

nodejs-github-bot commented Oct 27, 2023

ronag commented Oct 27, 2023 • edited Loading

ronag commented Oct 27, 2023

rluvaton commented Oct 27, 2023 • edited Loading

rluvaton commented Oct 27, 2023

rluvaton commented Oct 27, 2023

benjamingr left a comment

Choose a reason for hiding this comment

ronag commented Oct 27, 2023

rluvaton commented Oct 27, 2023 • edited Loading

change _events to be without __proto__: null

ronag commented Oct 27, 2023

ronag commented Oct 27, 2023

rluvaton commented Oct 27, 2023

rluvaton commented Oct 27, 2023

benjamingr commented Oct 27, 2023

lpinca commented Oct 27, 2023

rluvaton commented Oct 29, 2023

rluvaton commented Oct 27, 2023 •

edited

Loading

Change `_events` to be a `Map`:

Change to `Map` todos:

change `_events` to be without `proto: null`

ronag commented Oct 27, 2023 •

edited

Loading

rluvaton commented Oct 27, 2023 •

edited

Loading

rluvaton commented Oct 27, 2023 •

edited

Loading

change `_events` to be without `proto: null`