App scaling

[guestisp]

Hi all
I thing that app scaling is something missing in coolify (or i didn't find anything similiar).

Would be great having the ability to scale an app over multiple hosts by setting the required number of instances, via web configuration or CLI.

In example, if you have a multi-node setup, setting 3 to an ipothetical app_count paramenter, would spin up 3 instances over 3 different hosts (if hosts are less than the required number of instances, 2 or more instances should be placed over the same host, like 2 in host 1, and 1 in host2)

[srinisubramanian]

This is a very important feature to get Coolify to handle a lot more use cases. Similar to docker compose scale, the setting should control the instances deployed and to servers. Ideally this should be a min - max setting so that when the load is negligible the app or docker instances can be scaled down also.

[AlejandroAkbal]

Yeah, one of my free services is very popular, and it's getting hammered, being slow for everyone, cache is implemented, but its not enough

Scaling is necessary in my case, even in the same VPS, just adding more service containers

[Wulfheart]

Is it this one? :D

[mo9a7i]

+1 for this feature.

[Andreas-Sujono]

need this feature +1

[maietta]

Scaling an app isn't always as straight forward as it might appear. Would this only apply to Docker Swarm configurations? How would the load balancing work in a scenario where apps could be scaled across hosts?

There are a lot of questions that need to be first asked before even considering this.

Like how would querying for server IP's work with the API if this was enabled? Would apps be limited to Docker Swarm? Would the schedular still work -- it would have to pick 1 of the instances to interact with, not all of them, etc.

[DavesBorges]

Would this only apply to Docker Swarm configurations? How would the load balancing work in a scenario where apps could be scaled across hosts?

Docker swarm already handles load balancing across hosts unless you are refering to some kind of coolify specific load balancing.

Would apps be limited to Docker Swarm?

Docker compose can accomplish has kind of similar feature for a single host environment, but at that point it's better to use docker swarm for single host environments anyway.

Like how would querying for server IP's work with the API if this was enabled?

I'm assuming that docker swarm is used in this case, the server ip would be the master node IP. I believe coolify uses volume so it will need to set a constraint in the coolify swarm service to only run on the master node.

One thing that will probably need to be workout is the health check with traffeik, because we might have 3 instances in which 2 are healthy and one is unhealthy.

I'm probably missing some other edge case

[ImBIOS]

My current plan, for my projects which need scaling. I'm using cloud engineering in cloud provider.
But I need this +1 for my medium traffic projects.

[Jagatpratap]

I also want this feature +1

[snoozedDev]

+1 for multiple container instances. Would also help with smooth container updates.
I miss that from my brief experience with railway.

[AymenElmehdhaoui]

+1

[robertherber]

This is the reason we're staying with Caprover for now. This together with having to manually configure traefik for SSL - but that's a one-time thing so we can live with that.

[Boscop]

@robertherber I'm using CapRover but I'd prefer a PaaS that lets me deploy docker-compose.yml files, but still allows scaling like CapRover. Do you know which PaaS does this? :)

[robertherber]

I'm afraid not! That sounds like a great feature though, for self-hosting things that require more advanced infra setups.

Although I don't think Caprover supports Docker Compose specifically some of the one-click apps do spin up related service dependencies (databases etc) automatically.

[Boscop]

Although I don't think Caprover supports Docker Compose specifically some of the one-click apps do spin up related service dependencies (databases etc) automatically.

Yeah, but it doesn't directly support docker-compose.yml files (it has its own files which follow a similar format), so it's not a drop-in solution..
E.g. I've been trying to figure out how to deploy EdgeDB on CapRover (which is not in the set of one-click apps) and it would be easy with docker compose:
https://docs.edgedb.com/guides/deployment/docker#docker-compose

[jtwebman]

My guess is you could still do this in a very manual way. Just setup the app multiple times and then setup an Nginx the load balances over those IP addresses and ports. Doesn't auto scale or anything like that but would let you have two hosts for the same app.

[jtwebman]

Also looks like you can use Hetzner load balancer and firewall as well if you wanted. https://coolify.io/docs/knowledge-base/how-to/hetzner-loadbalancing

[lindesvard]

This is an important feature for me as well, since I want to manage how many bullmq workers I want to spin up (on a single server). A feature CapRover does pretty well!

[nerdlibfront]

I also need that for the projects that need redundancy.
It is possible if you don't need zero downtime. One can use the multi server feature and map ports to host ports. Then e.g. the hetzner loadbalancer can be used to split the traffic to the instances.
This approach has caveats though:

1. Zero Downtime deployment is not possible (as container ports are mapped to host ports)
2. multiple containers per server are not possible (which is needed if you have e.g. a node application and want to utilize more cpus) as you only have one port which receives traffic (at least thats the case with hetzner loadbalancer).

Maybe i missed something, but for now i can't see how i could achieve zero downtime with the coolify multi server functionality.

[jtwebman]

You can do both.

1. The zero downtime just comes down to you would have to manually spin up another third one with the new changes then add it to the mix once it is healthy, set the old proxy to drain, and then reload nginx (nginx -s reload). Then you can wait about 30 seconds for the connections to drain or write code the gratefully drains requests process shutdown when getting the kill command. Then kill that old version docker container. Then remove the old proxy from the nginx config and reload. This is what K8s is doing for you but it is a bunch of work to do manually. You could do a blue green style deployment and that way you keep switch back on the ports and could write a bash script to do it for you. Google around I am sure there are examples. We did it all the time back in 2010s. Also here is how you can do it with hetzner load balancer if you are using them over your own nginx: https://community.hetzner.com/tutorials/load-balancer-blue-green-deployments
2. Easy you just have to have different internal ports or use a cluster tool like PM2. PM2 can also do zero downtime deployments if it is just the one server.

Though I do agree this tool having a basic load balancer and doing this stuff for you would be nice but it will be a bunch of work for them to do this and still be flexible for all the different possible stuff, like supporting web sockets, or just multiple ports. CapRover is also a good solution to not have to do this manually.

[nerdlibfront]

I know that CapRover does it, but i don't want to switch to another solution without giving coolify a chance here. And after all this is the discussion thread for the app scaling feature which isn't there yet as it seems :)

Thanks for your input on how to do it manually or with additional setup of other technologies, i will definitely have a look.

[IGIAG]

+1