Memory store limits in multi-threaded setups

[Buzut]

Describe the change

Even in stateful environments (bare-metal, VPS, VM), most production setups are multi-threaded through PM2 or node cluster NITRO_PRESET=node_cluster.

Usually, memory is bound to a process. So I infer that memory based stores aren't shared between processes.

It has important implications:

• bigger chance of cache misses
• unreliable rate-limiting…

Behaviour is totally fine, but it might be good to mention it on the specific drivers:

• memory
• LRU

Now if it's shared across processes (impressive!), it deserves to be mentioned as well as they suddenly become viable production stores for more use-cases.

URLs

No response

Additional information

• Would you be willing to help?

[FabienEssid]

I am exactly in this case: I have an app based on Nuxt v3.15 with cached routes. I used pm2 on my production app and the NITRO_PRESET=node_cluster.
Now, I noticed that the cache is not shared between the processes and I don't find anything about it in the docs 😄

Thank you for raising this issue, I can now update my cache management using an other driver than the Memory one