This ticket has evolved from general issues with master to the identified problem of the number of messages stored in etcd.

This ticket is complete when we have a way to control the number of kubernetes status messages stored for a stack service.

—
Historic information

Starting on 10/3 and about once per week, all kubernetes services stop on master1 on the beta cluster. This prevents users from starting/stopping services, although existing services run without problem. Rebooting the node resolves the problem.

This best-effort task is to investigate the root cause, which is expected to be out of memory error for etcd, and identify and implement a solution, if possible or to create additional tickets as needed.