Object.setPrototypeOf performance regression after Node.js 12.18.0.

[schamberg97]

• Version: All versions past 12.18.0, starting with 12.18.1 (not sure about 13.x)
• Platform: At least macOS Big Sur (Darwin Nikolays-MacBook-Pro.local 20.3.0 Darwin Kernel Version 20.3.0: Thu Jan 21 00:07:06 PST 2021; root:xnu-7195.81.3~1/RELEASE_X86_64 x86_64), likely other platforms
• Subsystem: probably v8

What steps will reproduce the bug?

The issue is hard to reproduce with a code snippet without external modules (I've been unable too), but occurs 100% of the time, if Express.js is used. Here is the snippet:

'use strict'

var express = require('express')

var app = express()

app.set('etag', false)

app.get('/hi', function (req, res) {
  res.send('')
})

app.listen(3005)

Under 12.18.0, the wrk test wrk -t8 -c64 -d10s http://localhost:3005/hi/ shows around 28K req/s. However, under 12.18.1 and later it only shows 13K req/s.

Normally, I'd assume this is a problem with Express.js, however:

1. Using res.end('') instead of res.send('') doesn't eliminate the performance regression
2. The regression is fully eliminated, when using res.end('') IF Object.setPrototypeOf(obj, prototype) is eliminated from Express's init middleware.

I have always been able to replicate the issue with Express's req and res addition Objects, but I haven't been able to do so with my own self-made Objects. Perhaps an object needs to be large enough to trigger problems with Object.setPrototypeOf(obj, prototype). @uwinkelvos in related problem #34322 required

two objects that have at least 50 keys and the keys of the first one need to be non trivial.

How often does it reproduce? Is there a required condition?

What is the expected behavior?

No performance regression.

$ wrk -t8 -c64 -d10s http://localhost:3005/hi/

Running 10s test @ http://localhost:3005/hi/
  8 threads and 64 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     2.54ms  820.10us  25.94ms   95.71%
    Req/Sec     3.22k   434.07     3.89k    93.18%
  258324 requests in 10.10s, 39.91MB read
Requests/sec:  25568.63
Transfer/sec:      3.95MB

What do you see instead?

$ wrk -t8 -c64 -d10s http://localhost:3005/hi/

Running 10s test @ http://localhost:3005/hi/
  8 threads and 64 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     4.72ms    1.33ms  43.35ms   91.04%
    Req/Sec     1.71k   235.66     1.94k    91.34%
  137911 requests in 10.10s, 24.33MB read
Requests/sec:  13649.13
Transfer/sec:      2.41MB

Additional information

I haven't been able to dig into this issue deeper, but judging by Node.JS 12.18.1 Changelog, I strongly suspect [af95bd70bd] to be the culprit

[schamberg97]

Sorry for the edits, got confused by exact version numbers. My bad

[schamberg97]

Update: modifying v8 dep to revert changes introduced by [af95bd70bd] indeed solves this problem, should also solve #34322 but should reintroduce #32049. Until a better solution can be found, I am uncertain what course of action should be taken. Unfortunately, I cannot offer one, since I am not really familiar with v8 internals. Moreover, Object.setPrototypeOf(obj, prototype) usage is generally discouraged. However, it is still relied upon by a lot of users, including corporates

[schamberg97]

https://bugs.chromium.org/p/v8/issues/detail?id=11482

[targos]

Still open on the V8 side, but I'll close the issue here as not actionable.