An architecture question: When Rancher’s support for Swarm Mode in Docker 1.12 lands, what is the feasibility of creating a Rancher HA cluster via Swarm Mode’s services construct? In other words, currently Rancher HA requires a lot of tooling that could be thrown out in favor of the native cluster management in Docker engine.
My goal is to leverage the best aspects of both technologies. Docker Swarm > 1.12 has some amazing features re: Raft consensus and cluster management, but Rancher still has some advantages in how it builds and manages service stacks, especially since DABs are still experimental.
So what do the developers think of creating a Rancher HA cluster as a Docker Swarm Mode service group and simplifying the sidecar services that Rancher HA currently must start and maintain on its own?
It’s actually the lack of sidekick/sidecar/pod support in Swarm that makes it hard to do such a thing. When you run Rancher in non-HA the Rancher Server container is running quite a few processes. For better availability in HA we separate those processes out into different containers. We can also get better throughput on certain tasks because of how the API requests are routed different in HA mode. In short, Rancher HA is a bit complicated and in 1.2 hopefully there will be some improvements there, but Swarm Mode is not rich enough to do what we need.
Thanks for the awesome response. If I could ask a follow-up, then, since you’re very familiar with the product - Say I have a group of 3 nodes (At the moment Docker engine 1.11) running as a Rancher Swarm production environment, is there any reason not to use those same nodes for Rancher HA? Or is there a compelling reason to run the HA nodes as a separate, out-of-band cluster independent of any managed environment nodes?
was hoping for the same.
been trying to deploy Rancher cluster as is on 1.12 and 1.12.1, failed entirely errors in HA servers trying to connect to each other… same even without HA, just adding agents, some fail when trying to form the dhcp and virtual network to the ‘master’ rancher server. that was a long and depressing experience…
errors in HA servers trying to connect to each other… same even without HA, just adding agents, some fail when trying to form the dhcp and virtual network to the ‘master’ rancher server. that was a long and depressing experience…