`R CMD check` consumes a large amount of memory

[cicdguy]

What happened?

Running R CMD check on this package consumes a large amount of memory, almost to a point where total amount of available memory gets saturated.

See memory consumption graph when R CMD check --ignore-vignettes teal.modules.hermes_0.1.6.9001.tar.gz is run on a R 4.3.1 + BioC 3.18 container:

sessionInfo()

I can supply the renv.lock for reference. Cannot post it here due to the character limit for issue bodies imposed by GitHub (maximum is 65536 characters).

Relevant log output

The actual check runs fine when run in isolation.

The real issue happens when there are other R CMD checks or other processes that run alongside this one. The resulting issue is a memory leak and the following error in the log:

  In callback(...) :
    Chromote has received a Inspector.targetCrashed event. This means that the ChromoteSession has probably crashed.

Code of Conduct

• [X] I agree to follow this project's Code of Conduct.

Contribution Guidelines

• [X] I agree to follow this project's Contribution Guidelines.

Security Policy

• [X] I agree to follow this project's Security Policy.

[danielinteractive]

Thanks @cicdguy , how about other teal.modules.* packages? Do we see a similarly increasing, almost linear over time consumption of memory?

For now let's try to put https://rstudio.github.io/shinytest2/reference/AppDriver.html#method-AppDriver-stop into each shiny test.

[cicdguy]

Other teal.modules.* do not have this issue, for some reason. Sure, we can give it a shot and we'll let you know how it goes.

[BFalquet]

The 2 first test files and the 11th (which could correspond to the plateau in mid-curve) are the only test file where we are not performing snapshot tests. Maybe there is a glitch there.

[danielinteractive]

Hm interesting ... let's see after we add the stop commands what happens

[danielinteractive]

If this does not help, we could try

• reproduce locally
• interrogate locally where the memory is consumed
• find out root cause like that

[cicdguy]

Agreed with the approach above. If possible we can try profiling the tests in isolation using a memory profiler (eg profmem) and see where the issue could be.

[danielinteractive]

@cicdguy is it better now?

[cicdguy]

@walkowif - do you mind testing it on the integration test suite?

[walkowif]

This is the current memory consumption of tests running only for teal.modules.hermes:

Here's the test log.

[cicdguy]

Nice! That looks much better. Looks like app$stop did the trick. Thanks @danielinteractive and @BFalquet!

[danielinteractive]

good to know and learn about it! thanks @cicdguy and @walkowif !