updatescattercastendpoints updates the endpoints for currently running nodes.

In order to update scattercast endpoints, the SLEE must be stopped clusterwide. If this is successful, it triggers an immediate cluster restart to reload scattercast state.

Update commands make a best-efforts attempt to validate that the updated value will be usable after cluster reboot. This is done by attempting to bind the new address.

Updates cannot be applied to non-running cluster nodes. To update a node that is out of the cluster, simply delete the node and add with the new address.

Command

updatescattercastendpoints <node,ip-address[,port]>*
  Description
    Update scattercast endpoints for existing cluster members. If port is omitted,
    one will be assigned automatically. WARNING: This command will cause a cluster
    restart.

Examples

Update with the whole cluster in a stopped state:

[Rhino@localhost (#0)] updatescattercastendpoints 101,192.168.0.127,18000
Update executed successfully, cluster shutting down now.

Update while the SLEE is in a running state:

[Rhino@localhost (#3)] updatescattercastendpoints 101,192.168.0.127,12000
Failed to update scattercast endpoints due to: Cannot update scattercast endpoints while SLEE is running.

Update a non-running node:

[Rhino@localhost (#5)] updatescattercastendpoints 102,192.168.0.127,12000
Failed to update scattercast endpoints due to: 102 is not currently alive. Updates can only be done against currently live nodes

Errors

If the update command fails part way through execution, it is likely that the update will not have been applied. Under some rare circumstances, such as multiple disk errors or filesystem-access problems, the update will have only been applied to some nodes. Check the current configuration by running the getscattercastendpoints command to verify that the on-disk config is either coherent with the in-memory config (the command rolled back cleanly) or consistent across all nodes (the command failed after writing all changes). Identify and fix the fault that caused the command to fail, then reboot the cluster. If the update failed before writing the new configuration, rerun the update after fixing the fault that caused the initial attempt to fail.

Alarms

If the updatescattercastendpoints command is unable to reboot the cluster automatically, for example due to a timeout writing state to the persistent database, it raises a CRITICAL alarm of type rhino.scattercast.update-reboot-required and a message:

Scattercast endpoints have been updated. A cluster reboot is required
to apply the update as soon as possible otherwise a later partial reboot
e.g. due to network segmentation could result in a split-brain cluster.
Previous page Next page
Rhino Version 3.2