Question about main.cf modification

[Assaf_Leibovitc]

Hi,

How VCS handles a case where:

1. one out of 2 nodes cluster is stopped 

2. main.cf changed (removal of cluster parameter)

3. node starts.

will there be a collision between the 2 nodes? something else?

Thanks

[E_Dipko]

Since the node will be joining an active cluster (even though there is only one remaining node), it will receive the "in-memory" configuration from the running node over the cluster interconnect.

Ernie

[Hywel_Mallett]

What Ernie says! Because the node will be joining an already-running cluster, the main.cf file won't be read. If you are using 5.0, I'm not sure if it'll be overwritten by the new auto-save-type option.

If you wanted to force the edited main.cf to be read, then you would need to stop the cluster (hastop -all -force), then hastart on the edited node first, then start the other nodes. You could also start the other nodes in the stale state (hastart -stale (IIRC)), then startup order wouldn't matter.

All of this is from memory, so may not be right ;)

[Gene_Henriksen]

Good answer Ernie, you learned well.

VCS will NEVER read from local main.cf (local build) if another node is running in the cluster. To get the latest copy, it will request a copy from a running node (this is called a remote build). If somehow two nodes realize they have different versions, one will kill HAD and restart to get the same copy. If the heartbeats die and both nodes are running independently and you plug them back in, one node will kill HAD and rejoin the cluster. The node that continues is always the lowest numbered node in the cluster (see the llthost file).

In 5.0 their is no hastart -stale.