iSCSI gateways become unavailable when third unit of ceph-iscsi is deployed

Steps to reproduce:

1. Deploy two ceph-iscsi gateway units
2. Create a target with 'target-create' action
3. Discover and login to target on the initiator
4. Create a filesystem and mount the shared disk on the initiator machine
5. Deploy additional (third) unit of ceph-iscsi
6. Wait until new unit of ceph-iscsi settles
7. Watch what happens now:

A. All three units go to blocked state with message "3 is an invalid unit count".

B. The disk on the initiator becomes read-only:
root@node05:/mnt/volume# echo "test" > test.txt
-bash: test.txt: Read-only file system

C. Both iSCSI gateways become unavailable. 'gwcli' utility reports "2 gateways are inaccessible - updates will be disabled".

D. multipath -ll on the initiator reports faulty disks:
root@node05:/mnt/volume# multipath -ll
mpatha (36001405bb2233f3e2224572ac97fed60) dm-0 LIO-ORG,TCMU device
size=1.0G features='0' hwhandler='1 alua' wp=rw
`-+- policy='round-robin 0' prio=0 status=enabled
  |- 6:0:0:0 sdd 8:48 failed faulty running
  `- 5:0:0:0 sde 8:64 failed faulty running

Expected result:

The shared disk should still be available on the initiator. Existing gateways should still be available.