One significant problem is that if you hit the scenario in example 6 (management node is up, both data nodes are down) which can happen if you reboot the switch, MySQL Cluster by default doesn’t attempt to automatically restart the cluster when the switch returns.

Is there a way to configure MySQL Cluster so that when the data nodes can see each other and the management node too (e.g. when the network connection is restored to all 3), then it automatically restarts the cluster? You’d hope it would be something perhaps the management node would do if the data nodes can’t manage this.

My quick and dirty workaround is to check the output of “ndb_mgm -e show” every 30 seconds, parse it and decide whether the cluster needs restarting on one or both of the data nodes. Surely there’s a better solution than this (I know, redundant switches are one possible fix, but I’d still like a software solution too)?

I tried building a Cluster with your config.ini and the same command to launch the ndbds and Nodegroups 0 & 1 were created automatically.

Just to check, are you starting up the Cluster from scratch or are you trying to add a new node group to an existing Cluster?

One other thing you can try is that you can explicitly specify NG by adding “NodeGroup=1″ to the relevant ndbd sections of config.ini (if you’ve run the Cluster previously then you’ll need to start the ndbds with the –initial option which will also erase any data you’ve added to the database).

Andrew.

/usr/sbin/ndbd -c 10.1.101.23:1186 –nodaemon –ndb-nodeid=2

and config.ini is;

[ndbd default]
noofreplicas=2
datadir=/var/lib/mysql/data/mysqlcluster/

[ndb_mgmd default]
datadir=/var/lib/mysql/data/mysqlcluster/

[ndbd]
hostname=10.1.101.23
id=2

[ndbd]
hostname=10.1.101.31
id=3

[ndbd]
hostname=10.1.101.23
id=22

[ndbd]
hostname=10.1.101.31
id=33

[ndb_mgmd]
id = 1
hostname=10.1.101.23

[mysqld]
id=4
hostname=10.1.101.23

[mysqld]
id=5
hostname=10.1.101.31

I’d have expected Node Group 1 to be formed automatically – could you please post your config.ini file and the command lines you used to launch your ndbd processes?

ndb_mgm> show
Cluster Configuration
———————
[ndbd(NDB)] 4 node(s)
id=2 @10.1.101.23 (mysql-5.1.34 ndb-7.0.6, Nodegroup: 0)
id=3 @10.1.101.31 (mysql-5.1.34 ndb-7.0.6, Nodegroup: 0, Master)
id=22 @10.1.101.23 (mysql-5.1.34 ndb-7.0.6, no nodegroup)
id=33 @10.1.101.31 (mysql-5.1.34 ndb-7.0.6, no nodegroup)

[ndb_mgmd(MGM)] 1 node(s)
id=1 @10.1.101.23 (mysql-5.1.34 ndb-7.0.6)

[mysqld(API)] 2 node(s)
id=4 @10.1.101.23 (mysql-5.1.34 ndb-7.0.6)
id=5 @10.1.101.31 (mysql-5.1.34 ndb-7.0.6)

I then execute;

create nodegroup 22,33

and it creates the nodegroup 1 just fine.

Is this correct ? I had thought I could create the 2 nodegroups just from the static configuration without having to issue the create nodegroup command?

You can take a look at http://www.clusterdb.com/mysql-cluster/deploying-mysql-cluster-over-multiple-hosts/ for an example of how to configure a node group. In that example, the node group is defined implicitly…

By setting noofreplicas=2 in config.ini I specify that there should be two data nodes in each node group.

I then define [ndbd] sections in config.ini, for each one specifying which host it should run on. The bit that’s implicit is that Cluster looks at the order of those [ndbd] sections to decide which data nodes to place in which node groups – as there were only 2 specified, they’re both placed into Node Group 0. If there were 4 [ndbd] sections then the first 2 would form NG0 and the second 2 NG1. If there were 6 [ndbd] sections then the fith and sixth sections would form NG2….

If you want to play around with the make up of nodegroups (or when using on-line add-node) there are CREATE NODEGROUP and DROP NODEGROUP commands within the ndb_mgm tool.

Can you point me at a link that describes how to configure nodegroups ? For example, how exactly do you set up the configuration shown in Fig. 3 ?

In example 7, should’n it read

“The surviving data node from NG2 forms a community with the one from NG1″ ?