Placement determines where replicas may run. An ordinary application can use a scheduling label and a per-server replica cap. Stateful applications instead select a fixed server.
Configure
- Ask an Admin to add the intended server label.
- Select that label in application settings.
- Set a per-server replica cap if needed.
- Check reservations and desired replicas.
- Save and inspect the actual task distribution.
The label is a membership condition, not a complex selector language. A machine must also be ready, schedulable, and able to satisfy resource requirements.
For example, three replicas with a one-per-server cap need at least three eligible servers. Adding CPU and memory reservations can further reduce the usable set.
Change placement carefully
Changing labels or constraints can replace or relocate tasks. A start-first update may require spare capacity beyond the steady-state replica count.
Placement does not copy local files or volumes. Giving another server the same label or volume name does not make data available there. Use stateful deployment for local-data workloads.
Diagnose pending tasks
Check the selected label on the actual servers, their availability and runtime state, remaining capacity, reservations, and the per-server cap. Inspect task errors before repeatedly restarting.
After a server is removed and rejoined, its ID can change. Recheck fixed-server selections and labels before expecting previous placement to resume. See server management.