Different instance_name_template config in controller node and compute node will lead to VM not be deleted actually

Bug #1270726 reported by Feng Ju
12
This bug affects 2 people
Affects Status Importance Assigned to Milestone
OpenStack Compute (nova)
Opinion
Low
Unassigned

Bug Description

If we config instance_name_template differently in controller node and compute node will lead to VM not be deleted in the compute node, but the instance is deleted in the OpenStack.

The reason of the issue are:
When boot instance, instance's name generated on controller node using controller's instance_name_template config and is passed to nova-comute and used to boot a VM as VM's name.
But When delete instance, instance's name do not pass from controller node to compute node, the instance's name is generated on compute node using compute's instance_name_template config, so the nova-compute will not find the VM to delete when compute node try to delete the VM using the instance name genenrated on compute node.

The nova should use instance_name_template either on controller node or compute node, and should be consistent.

Matt Riedemann (mriedem)
Changed in nova:
status: New → Confirmed
jichenjc (jichenjc)
Changed in nova:
assignee: nobody → jichencom (jichenjc)
Joe Gordon (jogo)
Changed in nova:
importance: Undecided → Wishlist
Revision history for this message
jichenjc (jichenjc) wrote :

I analyzed this problem and looks to me compute/manager.py and virt/*.py highly depends on the instance['name']
which means it's not easy to transfer instance_name from api to compute node unless we not only transfer uuid of the instance but also transfer the instance name , it's complicated and not worthy

Also, if we think there are multiple nova-api service with different instance_name_template, it also can lead to problem and modification on compute layer won't help

Is there a way to do offline check to help admin on this? would think a healthy checker will be much helpful

Revision history for this message
jichenjc (jichenjc) wrote :

I posted a mail to ask opinion, seems no one pay attention to it
will wait for more time and see what I can do
any opinion or suggestion is appreciated

Revision history for this message
jichenjc (jichenjc) wrote :

no one replied my mail :)

looks to me there are lots of options need to be sync between api and compute
e.g reclaim_instance_interval, it's used in api and compute layer , we need to sync them too

so I don't think it's a defect, instead an enhancement on other module such as chef or puppet might be better
to mandatory sync those values between api and compute

Changed in nova:
assignee: jichencom (jichenjc) → nobody
Revision history for this message
Blair Bethwaite (blair-bethwaite) wrote :

Arguably a simple improvement on the current situation would be to pass the name from the API in all the relevant calls and simply have the local nova-compute ERROR when they don't match, so at least it gets flagged and identified somewhere.

Revision history for this message
Markus Zoeller (markus_z) (mzoeller) wrote :

Sounds like a real bug, therefore setting the importance to "low". But the report is pretty old, so I'm closing it. If you could reproduce this issue on a supported release [1], please reopen this bug report.

References:
[1] http://releases.openstack.org/

Changed in nova:
importance: Wishlist → Low
status: Confirmed → Opinion
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.