I have been reading through the docs, and playing around with Rancher through Vagrant. I really like what I see. I have also found confirmation here on the forums, that I should be able to use Rancher for my HA setup across multiple locations. However, I’m curious to know what would happen if I only deploy a single Rancher server (to avoid having to battle with the seemingly rather complicated process of setting up HA for the Rancher server), and then that server suddenly crashes. Will that effectively leave all my services unable to perform, or will they still run?
yes, the started services will just continue to run.
Rancher is “just” a scheduler and orchestrator like a mix of docker-compose and swarm kind of, with a sweet UI and catalog on top, it is not a “runtime engine” like docker engine … so Rancher schedules containers and deploys them across machines as desired and let them running by the docker engine that is on each hist., it tries to keep them up to the desired scale defined in the rancher-compose.yml. The containers run by docker engine itself that is on every host, not by rancher . If Rancher server is down, the containers keep running because the docker engines on the hosts are still running.
Thank you so much for clarifying that. Guess it’s time for me to get my lab setup operational, so I can start testing all the details.
DNS, metadata, the overlay network, etc run on each host so those continue to work if the server is gone. But those are all updated when states change by the server, so obviously that doesn’t happen if it’s not running. That means if a health check starts failing or a host disappears, metadata/DNS won’t be updated to remove it, and replacement containers won’t be started (until the server is back).
For many uses a single server with an external database (e.g. Amazon RDS) is a very reasonable compromise. Then you can easily startup a new server pointed to the same database if it dies, and you’re not worrying about care and feeding of a reliable database.
Vincent, thank you for further expanding on the impact of losing the Rancher server. I think the compromise you mention is quite a reasonable one, from my perspective, although I will have to worry about that reliable database anyways, but that’s af different story entirely.
My rancher just crash last night. Still don’t understanding what happens. Swarm and Kubernetes environments are lost.