Add section: "Loco in production", and include tips about Rust default allocators (malloc) and memory after heavy loading about loco HOT 4 OPEN

The default allocator does not return unused memory to avoid the performance loss and memory fragmentation that additional system calls would cause. Reddit

Additionally, there is a PR in rust-analyzer repository suggesting a revisit to the use of jemalloc in release build. issue

Despite the advantages, introducing jemalloc into Loco might lead to observation issue resulting in misleading or out of sync memory usage reports in htop.

In my opinion, adding a bref guide in the documentation about switching to different allocator could benefit users seeking max performance. We should also include warning for the observation issue mentioned above.

from loco.

This is a great suggestion. I think it calls for some kind of "running in production" section in the docs 👍

from loco.

This is very interesting! thanks!
I'm wondering what is the reason for the difference? probably that the default allocator prefers keeping things in memory until memory has pressure and then it will release?

from loco.

Just to add to this, as mentioned above, Rust used to ship with the jemalloc allocator by default but it got removed (possibly due to binary size bloat and not supporting as many platforms).

The foundations crate from cloudflare, which has some tools for building networked services, uses jemalloc (https://docs.rs/foundations/latest/foundations/static.JEMALLOC_MEMORY_ALLOCATOR.html).

I point this out just to say that it's not uncommon for rust projects to use an allocator different from the system allocator, and for networked or long running services to use the jemalloc allocator more specifically (where it excels).

(No affiliation with Cloudflare, just an example I found)

from loco.