Sizing question in AWS
noginaze is planning to host a Discourse web_only instance on Kubernetes with an externalized Postgres and Redis, and is seeking feedback on instance sizing for worker nodes, Postgres, and Redis to support around 2000 users.
supermathie advises to start with a size and adjust as needed, recommending 3-10 app nodes to handle normal traffic. They also mention that using the web_only sample config would be a supported configuration.
Jagster questions the need for such scaling ability with only 2000 users. noginaze responds that they have a team of Kubernetes enthusiasts and are taking the opportunity to use their cluster. They are considering a simplified deployment process with two machines and dedicated DB and Redis machines.
pfaffman notes that using Kubernetes will require either using the unsupported Bitnami image or figuring out how to use ./launcher to build and deploy the image.
Hello,
I would like to know if there are any feedback in terms of instance sizing. I would like to host discourse web_only on kubernetes, with an externalized postgres and redis hosted on rds and elasticache, and something like 2000 users. (not sure about post amount, we’re migrating from another system to which i don’t have access)
Does anyone have some feedbacks on the worker nodes sizing, postgres instance type & redis instance type ?
Thx,
so many piece for one puzzle. it just opens you up for sooo many things to go wrong.
My question is very naïve as I understand only half of the words you’ve used, but why not rely on the official way to install Discourse?
This strongly depends on your traffic load, but since these are all fairly easy to change, I’d recommend you pick a size and see how well it performs. Aiming for between 3-10 app nodes to handle normal traffic is probably a good plan.
As long as the deployed images are built with our launcher script, this would be a supported configuration. This would correspond to our web_only sample config which they’ve already said they’re using.
We wouldn’t be able to help with the Kubernetes side of things though.
My question is even naivier than @Canapin’s but are you expecting really big growing? Because with 2000 users you don’t need such scaling ability. Or are you planning ths just because you can (nothing wrong in that too)?
To answer to the various questions, i’d say that, first we have k8s big fans in the team, and this is a opportunity - as we can afford it its cool - to do stuff on our lovely cluster.
Indeed so many pieces to the puzzles, we’re also considering only bringing up like two machines with the web only template and keeping DB and redis on dedicated machines, would, really ease the deployment process.
And yes, we’ll adjust the worker nodes if we see them loaded, but we were a bit wondering on where to start, 2cpus 4gos looked a good start to me, but with feedback i would be more confortable 
You’ll need to either use the unsupported bitnami image or figure out how to use ./laucher to build your image and then push it somewhere that you can get k8s to crank it up. It’s not immediately apparent just how to do that.