Nodes fail to start up after complete failure - possible bug.

[Andrew Mcleod]

Hi,

I have discovered that if all node-groups break, I am unable to restart the cluster. For example, if all nodes in the cluster failed.

To replicate this problem, I can do a kill <pid> on the ndbd process on each node, then attempt to start the ndbd processes on each node.

The result is a completely broken cluster, the nodes fail to rejoin with a sequence of Phase 2: (system restart) Phase 3: (system restart) Node disconnected.

Now, I understand that the whole idea of a cluster is that there will never be a total failure, but what if a common peice of network hardware were to fail, or if a total power failure were to occur and backup power supplies didn't kick in straight away initiating a hard reboot?

I am a little concerned with trusting my configuraiton and finding that I am entirely unable to restart the cluster if such an event were to occur, which could certainly happen with a basic two node cluster.

Any thoughts or comments?

Thanks