clear-result-cache runs into out-of-memory when including big baseline

[j-koehler]

Bug report

Running PHPStan 1.4.2 (also occured on 1.3.3, but upgraded to check if it maybe is already fixed):

Running PHPStan through vendor/bin/phpstan clear-result-cache -c phpstan.neon runs into memory problems, when phpstan.neon includes a big baseline.

Code snippet that reproduces the problem

$ vendor/bin/phpstan clear-result-cache -c phpstan-debug.neon
PHP Fatal error:  Allowed memory size of 268435456 bytes exhausted (tried to allocate 20480 bytes) in phar:///export/vendor-bin/phpstan/vendor/phpstan/phpstan/phpstan.phar/vendor/nette/neon/src/Neon/Lexer.php on line 43
Fatal error: Allowed memory size of 268435456 bytes exhausted (tried to allocate 20480 bytes) in phar:///export/vendor-bin/phpstan/vendor/phpstan/phpstan/phpstan.phar/vendor/nette/neon/src/Neon/Lexer.php on line 43

Expected output

I run analyse with --memory-limit so it's not an issue there, but clear-result-cache does not support this parameter.

I don't understand why it's relevant to load the baseline, when I just want to clear the result cache? No matter how big the base line is - result cache should be cleared. Only relevant setting processed should be the tmpDir configuration.

A workaround could be supporting --memory-limit for clear-result-cache as well?

[ondrejmirtes]

Hi, how big is your baseline file? I've never seen Neon to consume so much memory!

I've added the option: phpstan/phpstan-src@156700b

[j-koehler]

how big is your baseline file?

9.5mb 🙈

I've added the option: phpstan/phpstan-src@156700b

Thank you very much! 🙏

[staabm]

@mw-jko could you create a blackfire or similar profile so we can see where all the memory is consumed with such a big baseline?

Maybe there is some low hanging fruit to optimize for memory in this case...

[j-koehler]

@staabm (I made the "business" github account my personal one, thus the handle change)

I will try, have no experience with blackfire (or services like this) so far.

But what I wonder is, why the clear-result-cache is even loading the baseline. I mean, in the end (ResultCacheClearer) it's just an unlink (as expected).

After getting some insight into the source code while fiddling around with the other issue this weekend my guess is, it's because of the centralized CommandHelper. Maybe this is already the room for improvement? Allowing to set the memory limit did help but the real issue is a different one.

Just not sure if it is important enough to change or how easy to change this part is.

[ondrejmirtes]

I don’t think it’s worth changing it. You’ll load the baseline often anyway, so it doesn’t matter you need to wait when clearing the result cache. Which shouldn’t be normally needed anyway - if you find the need to clear it often, that’s likely a bug to be opened about that.

[github-actions]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.