Overview
As of version 26.2, administrators can trigger a coordinated restart of the entire cluster themselves. The restart is available via the System, DMZ, Loadbalancing page and via the REST API.
Click path
Navigate to Administration > Base Settings > System, DMZ, Loadbalancing.
How it works
When you trigger a restart, the following happens:
The system immediately stops accepting new jobs.
Running jobs get a configurable grace period. Default: 1 minute.
After the grace period expires, the restart is forced. Jobs that are still running are stopped.
The cluster components restart in a defined order.
Restart order
The order depends on your cluster topology. The system detects the configuration automatically.
Without load balancer, without DMZ
The Node Controller restarts directly after the grace period expires.
Without load balancer, one DMZ
Restart the primary DMZ and wait until it is available.
Restart the Node Controller.
Without load balancer, multiple DMZs
Restart all secondary DMZs in parallel.
Wait until at least one secondary DMZ is available.
Restart the primary DMZ.
Restart the Node Controller.
With load balancer and DMZs
DMZs: Restart the secondary DMZs. At least one must become available. Then restart the primary DMZ.
Worker Nodes: Split into two groups. Restart the first group. At least one node must become available. Then restart the second group.
Node Controller: Only restarts if more than half (
ceil(N/2)) of all Worker Nodes are available.
The active DMZ is always the last DMZ to restart.
With load balancer, without DMZs
Same as above, starting with the Worker Nodes.
Cancellation behavior
The restart is canceled if a component does not become available within the time window. Subsequent components are then not restarted.
System, DMZ, Loadbalancing: If the primary DMZ does not become available, a secondary DMZ takes over its role. The restart continues.