[ClusterLabs] Antw: [EXT] What/how to clean up when bootstrapping new cluster (or: I have a phantom node)

[Andreas Hasenack]

Hi,

On Wed, May 25, 2022 at 5:43 AM Ulrich Windl
<Ulrich.Windl at rz.uni-regensburg.de> wrote:

> > But now I have this phantom "node1" in the cluster, and the cluster
> > thinks it has two nodes:
>
> I guess when starting pacemaker it creates a CIB (containing the nodes).

Correct, the cib files where the ones still referencing the old node
name. " crm_node --remove node1" did the cleanup nicely.

> Changing corosync.conf does not affect the CIB, so you'll have to clean it up
> (as Ken noted).
> The point is: Why did you start with some fake node?

It's a packaging choice, you get a working cluster right out of the
box to play with. All crm and pcs commands work, etc. It's also very
useful for simple automated smoke tests (keeping in mind it's a
single-node cluster). There are pros and cons, but I think it's a
valid choice.

But then comes the time when you want to make it a bit more real, and
add more nodes, give them meaningful names, etc. How to go from this
single-node-cluster to something bigger? Definitely just changing a
config file and restarting isn't enough, although I learned through
experimentation that depending on what you change, you might get away
with it.

A related question would be "how to rename a node in an existing
cluster", or "what are the key identifiers of a node": nodeid, name,
ringN_address, all of them, a combination?

Automated deployments also have to deal with this: they install the
packages, which gives that default configuration, but then have to
configure it to be part of a real cluster. The configuration made by
the package for the default install has to be undone, instead of
"adjusted".

I looked at what `pcs cluster destroy` does, and it's quite the
cleanup, I guess those steps are the most correct answer if you want
to start over. Really cleanup, instead of trying to massage the
existing single-node-cluster into something else.