By default, when a client node is overloaded (typically measured as the sum of CPU load), the HaCluster Agent notifies the load balancer that the node should be temporarily decommissioned, and new connections should be routed to less loaded nodes.
While this technique allows for effective load distribution, it only applies to new connections.
Indeed, existing sessions remain bound to their original node.This feature is available on Artica v4.50 Service Pack 5 or Artica v4.50 Service Pack 4 Hotfix 20250628-23
This constraint is particularly impactful when using source IP stickiness with SSL-based websites, where the original routing decision will not change as long as the session remains active.
To address this scenario, a "forced failover" mechanism is automatically triggered when the node’s load becomes so high that even existing sessions are severely degraded — to the point that internet access may become impossible.
This failover is initiated by the client node itself, meaning it still retains enough resources to notify the load balancer of its degraded state.
For this reason, it’s important not to set the emergency threshold too high:
If, for example, the limit is set to 40, the node might be so overloaded that it becomes unable to communicate with the load balancer at all.
By default, this emergency switch is set at a load level of 15.
When the node’s system load exceeds 15, it informs the load balancer to enter emergency mode.
This mode decommissions the affected node and terminates all active connections.
Locally, the HaCluster client retains the state of this emergency procedure.
If the system load drops back below the defined threshold, the node informs the load balancer that it is ready to be reintegrated and accept new connections again.
Max Load ( Emergency )
fieldTo simulate the process, there is no need to overload the machine or set an artificially low threshold.
When the maximum load is reached, the HaCluster Agent creates the file:
/etc/artica-postfix/hacluster-client.emergency
0
, and informs the load balancer of the node's degraded state.0
indicates to Artica: "I am in a degraded state, but I haven't yet notified the load balancer."0
, Artica will notify the load balancer to perform a forced failover.1
.1
, Artica enters monitoring mode, checking system load every minute.1
.To simulate the HaCluster Client task:
echo 0 >/etc/artica-postfix/hacluster-client.emergency
artica-phpfpm-service -rest-api /hacluster/client/emergency
In production / Maintenance
mode inside the farm table status