Re: General Confusion (cluster crashing/unstable)

[Harrison Fisk]

Hi,

Generally you only need one ndb_mgmd server, you should run that on a separate machine, such as the load balancer in this case. That way if you lose one system you won't lose a data node and the managment server, which might cause problems if arbitration was required. You can point all of the ndbd's and mysqld's at the single ndb_mgmd using the --connect-string option in a my.cnf or at startup.

You didn't mention anything about crashing though like your topic indicated. Are things shutting down/segfaulting/etc...? Have you checked in the logs for the ndbd nodes or the cluster log in general? If they are crashing, it should also generate a trace output which could be helpful.

With that said, I would definitely try upgrading to 4.1.8, as there have been many bugs found and fixed.