Further investigate flaky sequential/test-performance

[gireeshpunathil]

• Version: master
• Platform: arm
• Subsystem: test

#23291 reported a flake in sequential/test-performance that was subsequently identified as an improvement to the test and was fixed in #25695 . However, the fact that the test started failing all of sudden and that too very frequently warrants an investigation on the root cause of it - what caused to the change in timing w.r.t bootup. commit d3806f9 is identified as the change agent, and the change envisaged in that commit was a positive change. Effort is required to analyse that commit and generate inference about its effects.

[gireeshpunathil]

./node -e "console.log(require('perf_hooks').performance.nodeTiming.bootstrapComplete)"

without d3806f9
51.95341110229492
with d3806f9
38.98078536987305

so basically this commit improved the bootup time, not decreased!

but then looking at what happened in the test, I am not fully comprehending it, but able to see at least one inconsistency:

node/test/sequential/test-performance.js

Line 15 in 5cb1964

assert(Math.abs(performance.timeOrigin - Date.now()) < 15000);

where the result is always going to be negative, so the assertion always passes.

So in short, the said commit's purpose was to improve the bootup time and it is confirmed to do so. No more action needed on this; so closing.

[ofrobots]

FYI, I am continuing to see a flake in this test: https://ci.nodejs.org/job/node-test-binary-arm/5816/RUN_SUBSET=3,label=pi1-docker/testReport/junit/(root)/test/sequential_test_performance/

IIRC, the test asserts that Node.js can complete bootstrap in 1110ms. I don't know if this is a reasonable assertion to have in the CI. It will always depend on machine capability and load.

[Trott]

FYI, I am continuing to see a flake in this test: https://ci.nodejs.org/job/node-test-binary-arm/5816/RUN_SUBSET=3,label=pi1-docker/testReport/junit/(root)/test/sequential_test_performance/

That CI run was triggered by https://ci.nodejs.org/job/node-test-pull-request/20435/ which is itself a Resume Build of https://ci.nodejs.org/job/node-test-pull-request/20310/ which was started on January 24 before the fix landed on January 25. So that's a failure from before the thing was fixed. Use 'Rebuild' instead of 'Resume Build' or start a whole new CI to get the fix into your CI run.