I'm planning to self-host an OpenZiti network on a Linux VPS hosted in Canada, so a small team can reach a few internal apps without opening inbound ports. Some of the data behind those apps is covered by PIPEDA, one client is a BC public body, and another has Quebec Law 25 obligations, so I want the controller and edge routers to stay on Canadian servers.
Here's the plan so far. I'd appreciate feedback:
Controller: one controller on a 2 vCPU / 4 GB VPS, installed from the official Linux packages.
Edge router: a public edge router on a second VPS in another Canadian region for redundancy.
DNS and TLS: stable hostnames for both, with the PKI kept off the routers.
Firewall: allow only the controller and router ports, with SSH limited to a ZAC-managed admin identity.
Identities: enroll each laptop with the desktop tunneler and use attribute-based policies.
Private services: run a tunneler next to each internal app so nothing listens publicly.
Backups: back up the controller database nightly to separate Canadian storage.
Is one controller enough for about 30 identities, or should I set up an HA cluster from the start? And how do you usually back up and restore the controller?
Hi @4GoodHosting, welcome to the community and to OpenZiti!
I mean, I can't say 'for sure' that it's enough but I would expect one controller to be perfectly fine for hundreds to thousands of users unless you dialing 100's to 1000's and sessions for each user? (which is highly unlikely)
I use a free tier Oracle VPS for my controller and rotuer for my whole family and those things are one vcpu and 1gb ram. (I mean it's free, you can't beat it)
The rest of your plan seems good to me. You can also consider keeping ssh offline entirely and using OpenZiti itself as an ssh gateway. same for ZAC. That might be the only switch I might offer. "Split the API" (client/mgmt) and keep ZAC behind ziti itself once you get it cookign... (agents can do all this surprisingly well without a UI fwiw). there are numerous Discourse posts on that topic let me know if you can't find one and I'll search it up for you...
I would start with one, HA-enabled controller and deploy a second when one proves to be a 'problem'. Loads of people only deploy one controller and never notice a problem...
For only 30, I'd definitely stick with a single controller. The only reason to run more than one is to get HA on your controller layer, and for that you need at least 3 to form a quorum. You can run more than 3 controllers, but it has to be an odd number: 1, 3, 5, and so on.