Description of problem: Master is failing to set up sdn due to service IP conflicting with the docker0 network. The customer has the following set: osm_cluster_network_cidr=10.0.0.0/14 openshift_portal_net=172.16.0.0/13 With a host IP of 10.230.60.38/22 which should not conflict but docker0 is getting configured with 172.17.0.1/16 which is conflicting. The thing is the atomic-openshift-node, if it was not failing, would run a script that removes this interface. Version-Release number of selected component (if applicable): 3.4 How reproducible: Not Steps to Reproduce: I'll attach the customer's inventory file in private [1]. Actual results: install failed. Apr 20 13:09:36 XXX atomic-openshift-node[4944]: F0420 13:09:36.345196 4944 start_node.go:131] master has not created a default cluster network, network plugin "redhat/openshift-ovs-multitenant" can not start Expected results: Successfully installed. Additional info: Ryan suggested a workaround, but it failed on the customer side. I'll attach the log in private [2].
(In reply to Takayoshi Tanaka from comment #0) > The thing is the > atomic-openshift-node, if it was not failing, would run a script that > removes this interface. Not true. As of 3.4, we do not remove docker0, and it does need to have its own reserved IP range, because build containers are created on the docker bridge, not the SDN.
So, yes, OpenShift will check at startup that the cluster and service networks don't overlap the networks on any local interfaces, and fail to start if there's a conflict. So, probably the installer should do that check itself too before installing.
This can be worked around at install time by setting openshift_docker_options='--bip=10.10.0.0/16' or whatever the desired settings are. Taking this as an RFE to add an install time check for conflicts.