I’m going to install an application on a Data Center, my customer is providing me 2 physical servers for redundancy where implement our application.
All the infrastructure is redundant (2 switches, 2 firewalls, 2 power lines, 2 uplinks, 2 access to NAS).
My intention was to use Docker Swarm and run application in a docker for HA, but having only 2 phys servers I read that if one server fails we loose Manager quorum and application could not be moved to the healthy node even if it’s available.
Is it a way to overcome that with 2 physical nodes ?
Usually having 2 independent HW is enough for having a basic HA, wondering that’s not enough for Docker Swarm.
The typical HA approach uses an active/passive approach, where the active instance does the work, and falls back to the passive instance in case the active instance in unavailable.
Though, Docker Swarm (and Kubernetes as well) use the raft consensus algorithm, which requires (n/2)+1 active nodes for quorum (=vote for cluster changes on the state). It is designed for scale of up to 7 manager nodes and hundreds or thousands of worker nodes.
If you want HA with Docker Swarm, there is no way around a third manager node.