Comment 3 for bug 1549828

Thanks for the feedback!

console-log is a good hint on how to mitigate this issue, prevents the user from having to manually find the file and correct this. Nice.

Nevertheless live-migrations or starting an instance after a stopped instance migration can fail because of this.

I can reproduce this with a CentOS 7, RDO / Kilo release OpenStack installation running with Quobyte storage. Cinder is running with nas_secure_file_permissions and nas_secure_file_operations, libvirt is running qemu with dynamic_ownership = 0 (off) and nova:cinder ownership. If you want me to try any reconfigurations, etc. please feel free to contact me (kaisers@irc) or comment in here.

I can mitigate this by running e.g.: 'nova console-log <instanceid> && nova live-migration <instanceid> <targethost>' which works more often than not. But this is not reliable. I'd prefer to have the (live-)migration commands to work reliably on itself.

I'd think this issue should come up with GlusterFS setups running nas_secure_file_* = true, too. Has anybody experiences with that kind of setup?