Management Server Upgrade don't work correctly (no replies)

I am trying to upgrade a MySQL Cluster from version 7.5.7 to 7.6.8. According to the Rolling Update instructions, the management node must be updated first.

I stop the management node, uninstall the old RPM, install the new RPM and restart it.

So far so good - unfortunately both the data nodes and the MySQL API nodes no longer connect to the management server:

ndb_mgm> show
Cluster Configuration
---------------------
[ndbd(NDB)] 2 node(s)
id=11 (not connected, accepting connect from datanode1d)
id=12 (not connected, accepting connect from datanode2d)

[ndb_mgmd(MGM)] 1 node(s)
id=1 @10.1.1.45 (mysql-5.7.24 ndb-7.6.8)

mysqld(API)] 2 node(s)
id=21 (not connected, accepting connect from apinode1d)
id=22 (not connected, accepting connect from apinode2d)

These messages are repeated endlessly in the log file of the management node:
2018-11-28 10:22:08 [MgmtSrvr] WARNING -- Failed to convert connection from '10.1.1.72:36378' to transporter: line: 457 : Incorrect reply from client: >11 1
<, node: 11
2018-11-28 10:22:08 [MgmtSrvr] WARNING -- Failed to convert connection from '10.1.1.73:45086' to transporter: line: 457 : Incorrect reply from client: >12 1
<, node: 12
2018-11-28 10:22:08 [MgmtSrvr] WARNING -- Failed to convert connection from '10.1.1.72:36380' to transporter: line: 457 : Incorrect reply from client: >11 1
<, node: 11
2018-11-28 10:22:08 [MgmtSrvr] WARNING -- Failed to convert connection from '10.1.1.73:45088' to transporter: line: 457 : Incorrect reply from client: >12 1
<, node: 12
2018-11-28 10:22:08 [MgmtSrvr] WARNING -- Failed to convert connection from '10.1.1.72:36382' to transporter: line: 457 : Incorrect reply from client: >11 1
<, node: 11
2018-11-28 10:22:08 [MgmtSrvr] WARNING -- Failed to allocate nodeid for API at 10.1.1.75. Returned error: 'Id 22 already allocated by another node.'
2018-11-28 10:22:08 [MgmtSrvr] WARNING -- Failed to convert connection from '10.1.1.73:45090' to transporter: line: 457 : Incorrect reply from client: >12 1

After a downgrade back to 7.5.7, all nodes connect again...