Rolling Schema Upgrade does not work as described in the dokumentation

MySQL & MariaDB Percona XtraDB Cluster 8.x
1

The documentation states

RSU or Rolling Schema Upgrade
The node desynchronizes with the cluster and disables flow control during the execution of the DDL statement. The rest of the cluster is not affected. After the statement execution, the node applies delayed events and synchronizes with the cluster. Although the cluster is active, during the process some nodes have the newer schema and some nodes have the older schema. The RSU method is a manual operation. For this method, the gcache must be large enough to store the data for the duration of the DDL change.

https://docs.percona.com/percona-xtradb-cluster/8.0/online-schema-upgrade.html

But when I do this the “ALTER TABLE” will be replicated to all my nodes.
Is this only a viewing problem? Do I really need to do the “ALTER TABLE” on all cluster nodes?
SHOW FULL COLUMNS for the table seems to be correct on all nodes.

2

OK found one more thing about this.

It makes a difference if you do
set global wsrep_OSU_method=RSU; (alter table will be replicated)
or
set wsrep_OSU_method=RSU; (alter table is only on the executed node)

I think this is a Bug.

3

Hello @joerg.frede,
This is the correct behavior and correct steps to using RSU. You should set this mode at a session level on node1, and execute the ALTER. It will not replicate to other nodes. Then repeat this on node2, then finally on node3.

4

We have figured it out where the problem is.

set global wsrep_OSU_method=RSU;

Sets wsrep_OSU_method only for new sessions not for the current one so my “ALTER TABLE” is still running in TOI when I put it in the same session.

5

Right, you should not be using SET GLOBAL to change session-level parameters, as I described above. Glad it is working.