Looking at io.undertow.server.handlers.proxy.mod_cluster.Node#state this is declared as:

    private volatile int state = ERROR; // This gets cleared with the first status report

Which means the node is going to be in error until retried or receives first status update. Discovered while authoring a test [1] for ranked routing – ranked routing only takes candidates from available contexts and thus until first status, ranked routing is going to be broken. Workaround for the test is to wait until first status.

[1] org.jboss.as.test.clustering.cluster.affinity.RankedAffinityTestCase in WF