Result cache seems to kill Elasticsearch

[shadowhand]

Support question

This really makes no sense to me, but here it is...

I've been attempting to get phpstan upgraded to v0.12 for a couple of days. Everything worked fine locally but when running vendor/bin/phpstan analyse --memory-limit=512M on CircleCI, our elasticsearch container would be killed with the following error:

Exited with code 137
CircleCI received exit code 137

And then because elasticsearch is dead, our integration tests fail and the build fails.

Nothing else in our environment changed that should result in ES being terminated this way. The only change is upgrading phpstan and phpstan extensions. I attempted to upgrade ES to a newer version, hoping that it was some kind of version specific problem, but got the same result. While trying to figure out exactly which analysis was killing ES, I added --debug to the phpstan command and suddenly the builds started working. 😕

The only thing I can guess is happening here is that somehow phpstan's result cache is flooding the filesystem and/or memory in a way that causes elasticsearch to terminate. The whole situation is incredibly confusing to me.

[ondrejmirtes]

Hi, I don't think it can be because of result cache as that's just a file on disk. I'd suspect parallel running more. Can you tune some of the settings (https://phpstan.org/config-reference#parallel-processing) to see if it's really the cause?

[shadowhand]

@ondrejmirtes does --debug disable parallel processing?

Edit: I see that the documentation states that it does. I will try running with maximumNumberOfProcesses: 1 and see if that solves it.

[shadowhand]

@ondrejmirtes using a single process solved the issue. It also made phpstan about 4x slower, but such is life.

[ondrejmirtes]

You can try to use a smaller number of threads, or get a bigger server :)

[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.