Hello,
Background
I started learning UE5 a few weeks ago and I’m creating a game that has a large world but is made up of lots of small PCG levels linked together by entrances/exits (think like a chess board). I structured it like this because I’d like the game to be playable on mobile too!
Each level is just the same empty map populated by the various PCG graphs.
My initial version runs on Kubernetes on a single server and launches a Pod for each level that has a player in it, there is a database of all the areas that have been visited along with their PCG seeds (level is saved after first load) AND the K8s DNS name of the pod(if any) running the level. If a level doesn’t have a pod(nobody is already playing it) when a player tries to visit it, the backend API’s spin up a new instance via the K8’s api, load the level and update the database with the server address.
I’ve got a strong background in cloud/backend service development etc so although this seems quite big for a new person, actually, the server orchestration stuff is the easiest bit for me.
Question
It has occurred to me that by using Kubernetes I not only have its memory usage but also the overhead of a new Unreal instance for every level.
It then got me thinking, what about rather than using unique pods/UE instances for levels, I just spawn in the tile on a different Z access(with a suitable net cull distance) in a single server instance and move the players to the right one, when the level is empty just unload the tile!
Obviously, there’d come a point when I’d reach the maximum number of tiles UE5 can handle, but then I just spin up another “multi-level” Kubernetes Pod on another server.
Before I head off down that rabbit hole Is anyone else using the Kubernetes for orchestration? Are there any pitfalls to doing it this way? Any advice would be greatly appreciated - I had a browse through the suggested similar questions and couldn’t find any that quite covered this.