Heap reported by clinic is very different from that reported by node

[llimllib]

Expected Behavior

• When I run a test and jest reports that it's using 308 mb of memory, I expect clinic to report that I was using 308 mb of memory

Current Behavior

• Clinic is off by an order of magnitude

Steps to Reproduce (for bugs)

1. Run jest under clinic with --logHeapUsage: clinic doctor -- node node_modules/.bin/jest --logHeapUsage packages/models/__tests__/project
2. Note that jest reports 469 mb max heap usage

• PASS models packages/models/__tests__/project/model.test.ts (10.117 s, 469 MB heap size)

3. Note that clinic reports we used ~40:

What could be causing such a large difference? Jest uses process.memoryUsage().heapUsed to measure the heap size.

Sample upload

I don't feel comfortable uploading my work app's trace, unfortunately

Environment

• Clinic.js version: v13.0.0 Doctor v11.0.0

• Node.js version: v16.20.0 npm 8.19.4

• Browser name and version: N/A

• Operating system and version: macOS Ventura 13.4

[RafaelGSS]

I wasn't around when clinic heapprofiler was created so I don't know exactly if that's the expected behaviour. I'm guessing heapprofiler does count the amount of memory allocation during the profile while the process.memoryUsage().heapUsed gets the whole process memory (heap) used -- it's different because the process might hold 10mb but just allocate 1mb in objects.

@mcollina @jasnell would you be able to confirm that?

[llimllib]

I should have mentioned when I wrote this that --runInBand doesn't change the situation; I suspect it's something to do with the fact that node kicks off a child process. Here's the process tree, as reported by htop during a test run:

You can see there that the main jest process only has 86mb, while its child test-runner process is chomping up an order of magnitude more memory.

Is there any way to get at the child process?

[RafaelGSS]

Not really. Heapprofiler measures only the main process, in your case the main jest process.