Several openstack services (nova-compute, glance-api, cinder-backup, etc.) create iSCSI connections to a storage backend provided by the cinder-volume service. These connections update information maintained in /var/lib/iscsi, and that data isn't propagated to the host when the openstack service is running in a container.
This can lead to problems during system shut down, when the host needs to log off all iSCSI connections (this is done by the host's iscsi-shutdown systemd service). However, connections created by containerized openstack services aren't "visible" to the host, and the connection's aren't terminated. This causes the shutdown sequence to hang.
The problem can be reproduced using the following steps:
1. Create a nova instance and attach it to a cinder volume via iSCSI
2. Stop the instance ("openstack server stop") but do not delete the instance
3. Reboot the Compute node, and observe it hang on shutdown
After step 1, you can see the nova_compute container's iSCSI node info:
[root@overcloud-compute-0 ~]# docker exec -ti nova_compute ls /var/lib/iscsi/nodes
iqn.2010-10.org.openstack:volume-440caa2b-1bfe-4a13-a0c4-0d86a096d524
[root@overcloud-compute-0 ~]#
But the node is not visible on the host:
[root@overcloud-compute-0 ~]# ls /var/lib/iscsi/nodes
[root@overcloud-compute-0 ~]#
Fix proposed to branch: master
Review: https:/