stasher maintains transactional integrity of the cluster only as long as the cluster remains in quorum (the majority of nodes in the cluster are up, connected and synchronized with each other). A special process is necessary to shut down the entire cluster in a manner that preserves the object repository's transactional integrity. Stopping each node in the cluster is not sufficient. Eventually, the cluster will no longer be in quorum, and some clients may not longer have a reliable indication whether their transactions were posted, or not.
Issuing the administrative HALT command
stop all the nodes in a cluster, in a manner that preserves
transactional integrity:
[root@octopus stasher]# stasher
Ready, EOF to exit.
> admin
Connected to objrepo.example.com, node octopus.objrepo.example.com.
Maximum 20 objects, 64 Mb aggregate object size, per transaction.
Maximum 10 concurrent subscriptions.
octopus> halt
Cannot halt the cluster, the cluster must be halted by monster.objrepo.example.com
octopus>
The HALT command must be issued to a node that's
the cluster's current master node. This error message reports that
another node in the cluster is the current master. The current
cluster master is also reported by the STATUS
command, too.
[root@monster ~]# stasher
Ready, EOF to exit.
> admin
Connected to objrepo.example.com, node monster.objrepo.example.com.
Maximum 20 objects, 64 Mb aggregate object size, per transaction.
Maximum 10 concurrent subscriptions.
monster> halt
Cluster is not in full quorum
monster>
The HALT command requires a full quorum (all nodes in the cluster are up, connected and synchronized with each other).
This error message indicates that not all nodes are up,
connected with each other, and fully synchronized. Use that
STATUS command to display the cluster's
current status, identify the unavailable nodes, and bring them
back up. In order to preserve transaction integrity, all nodes
in the cluster must be halted in full quorum, with each node's
copy of the object repository identical to every other node's.
monster> halt
Cluster halted
Disconnected from monster.objrepo.example.com
If all the requirements are met, the cluster gets halted. This stops all stasher node processes, on all the nodes.