config: add master_node_uuids to EdgeNodeCluster

[naiming-zededa]

  In a 3-master k3s cluster, replacing a master node leaves a stale
  Node object that blocks the new master from joining — k3s rejects
  a new server while an old member is still registered.

  Add master_node_uuids (field 13) to EdgeNodeCluster so the controller
  can ship the authoritative list of current master UUIDs to surviving
  devices. EVE uses this list to delete stale control-plane Node
  objects, which causes k3s to automatically drop the corresponding
  member and allows the replacement master to join cleanly.

[zedi-pramodh]

I am trying to understand, so in a replace node operation, eve will get three uuid of the two old nodes + one new node ?
Then the eve code will delete the node that is missing from the list ? Which eve node will handle that in a cluster ? Will there be a race ? Does this also mean our replace node never worked earlier ?

[naiming-zededa]

I am trying to understand, so in a replace node operation, eve will get three uuid of the two old nodes + one new node ? Then the eve code will delete the node that is missing from the list ? Which eve node will handle that in a cluster ? Will there be a race ? Does this also mean our replace node never worked earlier ?

right, two old nodes uuid, plus one new node uuid.
it is the elected lease leader, who is responsible for deleting the stale node
our replace node works (without counting for corner cases), when the deleted node is alive during the operation

[eriknordmark]

LGTM.
Is this tied to using etcd as the database, or should the comment be more general?

[naiming-zededa]

LGTM. Is this tied to using etcd as the database, or should the comment be more general?

@eriknordmark removed wording on etcd in the commit message.