Performance degradation in vitest when using --import dd-trace/register.js

[niklaas]

We're investigating a performance issue that we see if --import dd-trace/register.js is imported to trace a vitest test suite. The issue is noticeable both locally and in GitHub actions. Please note the Duration lines and the summary below:

GitHub actions ubuntu-latest-4core with --import dd-trace/register.js:

Test Files  373 passed (373)
    Tests  4411 passed | 1 skipped (4412)
  Start at  13:19:10
  Duration  1203.67s (transform 12.73s, setup 331.96s, collect 1857.33s, tests 293.27s, environment 274.86s, prepare 227.77s)

GitHub actions ubuntu-latest-4core without --import dd-trace/register.js:

Test Files  373 passed (373)
    Tests  4411 passed | 1 skipped (4412)
  Start at  12:10:05
  Duration  513.13s (transform 9.51s, setup 161.73s, collect 657.62s, tests 267.97s, environment 241.71s, prepare 38.61s)

Apple M3 Max with --import dd-trace/register.js:

 Test Files  373 passed (373)
      Tests  4411 passed | 1 skipped (4412)
   Start at  17:04:12
   Duration  131.36s (transform 11.57s, setup 144.47s, collect 1009.37s, tests 196.96s, environment 134.83s, prepare 241.18s)

Apple M3 Max without --import dd-trace/register.js:

 Test Files  373 passed (373)
      Tests  4411 passed | 1 skipped (4412)
   Start at  17:07:26
   Duration  62.56s (transform 10.46s, setup 75.92s, collect 453.93s, tests 162.53s, environment 122.32s, prepare 17.95s)

In general, we notice that the execution time can take more than twice the time with the import:

Env	with import	without import	ratio
GH ubuntu-lates-4core	~20 mins	~9 mins	2.35
Apple M3 Max	~2 mins	~1 min	2.01

Is this a known issue? Maybe even expected due to the way vitest works? Somebody else experiencing the same?

For what it's worth, there is another performance related discussion at vitest-dev/vitest#6223 about the big difference between local execution and execution with GitHub runners without mentioning dd-trace. However, since I can demonstrate a similar impact to the execution time no matter the environment, I think I have ruled out that it's an environment issue alone.

[BridgeAR]

Hey @niklaas thank you for reporting this. I believe this is not yet known, so if you could open an issue for it, that would be great!

[niklaas]

👍🏼 Thanks! I have contacted the DataDog Helpdesk because the issue blocks us. I'll feed back what I learn to the community as soon as I know more -- or will open a public issue if the helpdesk asks me to.

I wasn't sure whether it's best to submit the issue publicly or privately. In part because the "Create new issue" modal is a bit contradictory: