Open Stack

Hi,

if an instance has attached EBS, you stop it and subsequently reboot
the whole compute node, then you won't be able to start the instance
again. iSCSI sessions are not re-established by nova-compute for
stopped instances (only for instances which should be running). If you
then invoke "euca-start-instance" it will produce a stracktrace about
a missing device.

This question is related to fix https://bugs.launchpad.net/nova/+bug/1036902
("Block storage connections are NOT restored on system reboot")

That fix will re-establish iSCSI sessions for instances that should be
in running state after system reboot and for instances that are to be
reboot-ed. So one workaround for this problem is to start instance by
"euca-reboot-instance".

The problem with automatically established sessions by "iscsi" service
is that it is not guaranteed that it will be able to relogin at boot
time. One example is when nova-compute and nova-volume are located on
the same box. iscsi will be starting much earlier than tgtd service to
which it is attempting to connect..

Would you consider the fact that iSCSI for stopped instances is not
re-established to be a feature or should I submit a bug to "extend"
the patch for #1036902 ?

Regards,

Brano Zarnovican

PS: This was tested on Folsom 2012.2.4