Perf: better to require autoloader inside or outside worker loop?

[summersab]

I'm wanting to work on adding support for RoadRunner to a major project (Nextcloud). This project has several thousand classes in its autoloader. This made me wonder - in terms of performance, is it better to have the autoloader called outside the worker loop or does it not really matter? It would be somewhat easier to call the autoloader inside the worker loop based on the current Nextcloud structure. However, if there is a performance benefit, that information would be helpful so I can justify proposing code changes.

Thanks!

[rauanmayemir]

Does it really matter if you use opcode cache?

[wolfy-j]

You can find some diagrams explaining the working model here - https://spiral.dev/docs/framework-lifecycle/3.6/en

[wolfy-j]

Indeed, it was a rhetorical question. 😄

I actually meant to say preload, not regular opcode cache. https://stitcher.io/blog/preloading-in-php-74 I believe it actually targets use cases like yours.

Preloading also have little effect, it will only speed up the bootstrapping phase which you do once. Nothing can really beat having code in RAM.

[wolfy-j]

However, you do not need to FORCE loading code in your worker bootload phase. You can keep reading classes you need as you go, it called worker "warming up" and it will take some time for worker to reach it's optimal memory state and performance, but it also much easier to implement since it requires no changes in autoloading.

[summersab]

This is all excellent information - thank you! I may revisit this and ask more questions down the road if I have more questions. Adding RR support to Nextcloud is a side project for me at the moment.

Is there documentation regarding best practices for RR, how to structure projects and code to improve performance, and what settings to tune and tweak? I've seen a few posts regarding the number of workers and RAM, but I haven't seen much regarding best practices from an actual coding perspective. Things like:

• What should/shouldn't be put in the worker loop
• How best to efficiently load the application into memory to improve worker performance
• The operations that should be avoided because they have a negative impact on performance
• Tradeoffs and pitfalls, etc.

A nice bulleted checklist on a wiki page would be a great starting point.

[rustatian]

We have this advice for a production use: https://roadrunner.dev/docs/app-server-production/2.x/en