Tom Wieczorek b85ac7b548 Reconcile etcd endpoint's peer URL for single-member clusters ...

Previously, k0s only ever used the etcd peer address during local etcd
member initialization, either during etcd cluster bootstrap, or when
joining a new member into an existing cluster.

During k0s startup, allow for the peer address to be changed
automatically for single-node clusters. This addresses a design issue
in k0s restore where the etcd cluster is restored using either the
backed-up etcd peer address or the auto-detected node's peer address if
the backed-up node didn't specify one. Without this change, there was
no way of restoring a node while also changing the node IP. This is now
possible by simply modifying the local k0s config before starting the
restored node.

Note that this is _only_ done for single-member etcd clusters. For
multi-member etcd clusters, the join API performs a basic reachability
check before adding a new member, to prevent cluster outages caused by
bad configuration values. While this might be doable here as well, it
would require significantly more effort, like trying to communicate with
the join API, potentially adding a new endpoint. Postpone that decision
for now.

Signed-off-by: Tom Wieczorek <twieczorek@mirantis.com>

2026-08-27 16:14:36 +02:00

..

client.go

Reconcile etcd endpoint's peer URL for single-member clusters

2026-08-27 16:14:36 +02:00

health.go

Close etcd clients after use

2025-12-23 12:53:57 +01:00