Comment 3 for bug 1632677

I also see this issue. I'm using an 'interesting' environment, but I believe the issue is still valid.

OS: CentOS
Kolla release: Fork of Mitaka circa April 2016, back ported to deploy Kilo OpenStack(!)
Ironic release: kilo
Ironic driver: pxe_ssh

I believe that the key piece of information here is the driver. The deployment method used by the pxe_* drivers involves the conductor connecting via iSCSI to a target on the node being deployed. AFAIK this doesn't require the iscsid container to be running on the conductor node as it is an initiator rather than a target.

To make this work, I did the following on the conductor nodes:

$ modprobe iscsi_tcp

I also made the following change to the Dockerfile:

$ git diff
diff --git a/docker/ironic/ironic-conductor/Dockerfile.j2 b/docker/ironic/ironic-conductor/Dockerfile.j2
index cc39356..4590cdc 100644
--- a/docker/ironic/ironic-conductor/Dockerfile.j2
+++ b/docker/ironic/ironic-conductor/Dockerfile.j2
@@ -8,6 +8,7 @@ RUN yum -y install \
         openstack-ironic-conductor \
         gdisk \
         ipmitool \
+ iscsi-initiator-utils \
         parted \
         psmisc \