On studying this quite a bit more ... I was confused about what caused SCSI host 23 to appear (syslog) at 10:55:26. It wasn't anything to do with the volume extend test - that was using host 22, which was created by the login at 10:55:19. It wasn't anything that c-vol was doing at the time ... so what was it? Then I realised that c-bak could have been doing something aswell. Sure enough:
So I tried adding a target manually, then I discovered it on the initiator, then did "iscsiadm -m node -l", whilst continually doing 'iscsiadm -m session'. It reproduced this:
iscsiadm: could not read session targetname: 5
iscsiadm: could not find session info for session519
for a few seconds, then went back to normal.
So, it appears that logging into a new target causes a temporary failure to list sessions. Seems like this may be a bug in the iscsi initiator implementation.
On studying this quite a bit more ... I was confused about what caused SCSI host 23 to appear (syslog) at 10:55:26. It wasn't anything to do with the volume extend test - that was using host 22, which was created by the login at 10:55:19. It wasn't anything that c-vol was doing at the time ... so what was it? Then I realised that c-bak could have been doing something aswell. Sure enough:
Sep 24 10:55:26.003013 ubuntu- xenial- rax-ord- 0002235107 cinder- backup[ 16673]: DEBUG oslo.privsep.daemon [-] privsep: request[ 140306203894800 ]: (3, 'os_brick. privileged. rootwrap. execute_ root', ('iscsiadm', '-m', 'node', '-T', u'iqn.2010- 10.org. openstack: volume- b8059959- d35e-4567- 86cd-a2094e8a5f ff', '-p', u'10.210. 68.187: 3260', '--login'), {'attempts': 1, 'check_exit_code': (0, 15, 255), 'delay_on_retry': True}) {{(pid=11439) loop /usr/local/ lib/python2. 7/dist- packages/ oslo_privsep/ daemon. py:443} }
So I tried adding a target manually, then I discovered it on the initiator, then did "iscsiadm -m node -l", whilst continually doing 'iscsiadm -m session'. It reproduced this:
iscsiadm: could not read session targetname: 5
iscsiadm: could not find session info for session519
for a few seconds, then went back to normal.
So, it appears that logging into a new target causes a temporary failure to list sessions. Seems like this may be a bug in the iscsi initiator implementation.