I am running a large setup with many VMs connecting as workers to a manager node. The docker manager is not able to properly schedule all services evenly throughout the VMs. I experience some nodes having 550 containers, while others only run 300 containers. A VM crashes when more than 450 containers are deployed, which is why I need to avoid such a scenario. I am using the default scheduling mechanism atm.
I need to manually overwrite the deployment choice that the docker manager makes. While there is a --placement-pref attribute, it only allows a spread throughout machines that carry the same label. I simply need the ability to define where a service is going to be deployed. It would be perfect to specify the node directly, which does not seem to be possible.
Does anyone have a bright idea for a workaround?