feat(files): remove spawning a thread for each file handling, as it does more harm than actually blocking

[joelwurtz]

PR Type

Fix/Feat (not always sure with those kind of patch)

PR Checklist

• Tests for the changes have been added / updated.
• Documentation comments have been added / updated.
• A changelog entry has been made for the appropriate packages.
• Format code with the latest stable rustfmt.
• (Team) Label with affected crates and semver status.

Overview

I was looking at a recent comment that indicates than actix files was having bad perf, after some investigation it seems that using the sync api make it way better, than having to spawn a thread to make it async.

In my local environment i want from 11k req/s to 600k req/s (so roughly a x54 increase of performance)

It may be a problem if you only handle big files with poor fs i/io speed (so all actix web threads will be blocked).

Maybe we could still spawn this thread if the size of the file is superior to some limit ?

[robjtede]

Having some threshold on the spawn might be a good idea. I can see this current patch being quite risky for high concurrency situations.

[joelwurtz]

Maybe, we can provide a default strategy that tell that half of worker can do sync file serving (using a sémaphore) and allow people to change this strategy if they want.