🤔 What is the purpose of prefork?

[peterbourgon]

Question description

What is the intent behind the prefork option?

[welcome]

Thanks for opening your first issue here! 🎉 Be sure to follow the issue template!

[Fenny]

Prefork enables use of the SO_REUSEPORT socket option, which is available in newer versions of many operating systems, including DragonFly BSD and Linux (kernel version 3.9 and later). This socket option allows multiple sockets to listen on the same IP address and port combination. The kernel then load balances incoming connections across the sockets.

SO_REUSEPORT scales perfectly when a lot of concurrent client connections (at least thousands) are established over the real network (preferrably 10Gbit with per-CPU hardware packet queues). When small number of concurrent client connections are established over localhost, then SO_REUSEPORT usually doesn't give any performance gain.

Benchmarks where preforking is enabled.
https://www.techempower.com/benchmarks/#section=test&runid=350f0783-cc9b-4259-9831-28987799782a&hw=ph&test=json&l=zijocf-1r

NGINX on socket sharding
https://www.nginx.com/blog/socket-sharding-nginx-release-1-9-1/

[peterbourgon]

Thanks for that explanation. Do you have evidence to suggest a single Go process cannot support "[thousands] of concurrent client connections" without SO_REUSEPORT?

[Fenny]

A single go process can easily support thousands of concurrent connections. Preforking makes use of single go processes but will load balance connections on OS level.

It's up to you if preforking has an advantage for your web app, we only provide the experimental option to enable it.

Feel free to re-open this issue if you have further questions!

[Sahil624]

So prefork runs multiple worker processes?
If each worker is a different process then memory will not be shared in each worker, If I'm not wrong?

[calbot]

Does using fiber behind a reverse proxy like nginx reduce the possible benefit by doing the same thing or is there still possibly a benefit?

I assume there are fewer tcp connections between the reverse proxy and fiber.

[ozkansen]

When fiber prefork is active, database automigrate works in every process, how can I make it work only once?

[ReneWerner87]

database automigrate works in every process

https://docs.gofiber.io/api/fiber#ischild

if fiber.isChild() == false {
// make it work only once
}

[ozkansen]

Thanks @ReneWerner87

[para-d]

I also have a question about sharing memory.

We have implemented a FastCache instance to use for our Fiber application. When application starts, the data is pulled from an external server with a HTTP get request and the cache is updated. So far so good as both child processes gets their own update.

We also have a feature to update the cache by pushing it through HTTP POST from the external server to make it not require a server restart on Fiber application's side. Will it update every process? I assume it will not. My questions are,

1. Is there a way to make it work?
2. If we run a separate fiber instance as a cache server on the same machine, outside of the pre-forked application, will there be a performance penalty?
3. Also if a separate cache server is an option, what protocol(s) would be the most efficient?

If we can share cache between forks within the application it should be a solution. But after the research I don't see it's possible.

Sharing memory is also beneficial as it reduces the memory requirements as we need only one instance for all the processes.

[ReneWerner87]

well thats the problem with inmemory caches when you want to use multiple threads

to 1. you have to inform all threads to do this, for this there are concepts like message queues or pub/sub mechanisms
in kubernetes there is often the same problem, you have a deployment with multiple pods and there you sometimes have to establish a communication if you want to update something over all pods

i personally use redis and the pub/sub concept for this purpose

to 2. and 3. i think no, but you loose the benefit of processing over several threads
instead of building your own solution, i.e. server endpoint with control and inmemory instance, i would recommend you to use a redis server
it's fast to install (at least with docker) and really fast + many features and the possibility of scaling through a master/slave cluster

[ReneWerner87]

https://redis.io/docs/manual/pubsub/
https://redis.com/redis-best-practices/communication-patterns/pub-sub/

https://hub.docker.com/_/redis/ -> alpine 28-32mb
or helm https://artifacthub.io/packages/helm/bitnami/redis

go client
https://redis.io/resources/clients/#go
-> https://github.com/go-redis/redis