Having a CXF REST or WS application on Karaf (such as karaf-cxf-rest), clients accessing the service will receive transient connection resets and HTTP 404 errors when scaling up to more replicas.

The problem is that the newly created pods are put into service too soon before the actual application bundle is deployed. There needs to be a readiness probe defined to make sure only a properly deployed karaf is put into service.

While customers can potentially define their own application-specific probes, there should be a general readiness probe for Karaf that would mark the Karaf container "Ready" when all the bundles have finished deploying (see also the readinessProbe.sh in the EAP openshift image, which handles the same problem in EAP, or the A-MQ readiness probe which uses Jolokia in https://github.com/jboss-openshift/application-templates/blob/master/amq/amq6-basic.json )

This probe should be defined by default in Karaf archetypes and any karaf templates shipped with Fuse on OSE.