Bad args to iscsiadm for iSCSI multipath target

Cinder 3rd party CI tests fail for multipath iSCSI targets due to an incorrect invocation of iscsiadm:

Called from connect_volume() in a multipath, multiattach configuration, ISCSIConnector._recover_node_startup_values() passes an iqn wrapped in a list to _iscsiadm_update() via the connection properties 'target_iqn' field, but _iscsiadm_update() expects a plain string which will be used as a command line argument to iscsiadm's -T option. This results in an unexpected iscsiadm failure which aborts the connection attempt.

The system is running Ubuntu 16.04 with open-iscsi 2.0.873+git0.3b4b4500-14ubuntu3.7.

Patching _recover_node_startup_values() to pass a scalar instead of a list appears to solve the problem:

diff --git a/os_brick/initiator/connectors/iscsi.py b/os_brick/initiator/connectors/iscsi.py
index a875b76..c428772 100644
--- a/os_brick/initiator/connectors/iscsi.py
+++ b/os_brick/initiator/connectors/iscsi.py
@@ -1178,7 +1178,7 @@ class ISCSIConnector(base.BaseLinuxConnector, base_iscsi.BaseISCSIConnector):
                 # as discovering target.
                 # So target_iqn is updated, and other values aren't updated.
                 recover_connection = copy.deepcopy(connection_properties)
- recover_connection['target_iqn'] = [iqn]
+ recover_connection['target_iqn'] = iqn
                 self._iscsiadm_update(recover_connection,
                                       "node.startup",
                                       old_node_startup)

Attached is a stack trace showing the symptom. In this scenario a volume from the target has been attached to one instance and the test case is attempting to attach the same volume to a second instance.