RDO tickets are now tracked in Jira https://issues.redhat.com/projects/RDO/issues/
Bug 1049877 - cinder: data is not saved on a volume created from an image after instance termination
Summary: cinder: data is not saved on a volume created from an image after instance te...
Keywords:
Status: CLOSED INSUFFICIENT_DATA
Alias: None
Product: RDO
Classification: Community
Component: openstack-cinder
Version: unspecified
Hardware: x86_64
OS: Linux
unspecified
urgent
Target Milestone: ---
: ---
Assignee: RHOS Maint
QA Contact: Dafna Ron
URL:
Whiteboard: storage
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2014-01-08 12:02 UTC by Dafna Ron
Modified: 2016-04-26 14:44 UTC (History)
5 users (show)

Fixed In Version:
Clone Of:
Environment:
Last Closed: 2015-03-20 15:58:49 UTC
Embargoed:


Attachments (Terms of Use)
logs (1.34 MB, application/x-gzip)
2014-01-08 12:02 UTC, Dafna Ron
no flags Details

Description Dafna Ron 2014-01-08 12:02:15 UTC
Created attachment 847097 [details]
logs

Description of problem:

I booted an instance from a volume created from an image with linux os. 
after creating a directory and a file under / I terminated the instance and booted a new instance from the same volume. 
The file and dir I created were gone. 

Version-Release number of selected component (if applicable):

openstack-cinder-2014.1-0.2.b1.el6.noarch
openstack-nova-compute-2014.1-0.5.b1.el6.noarch

How reproducible:

100%

setup: 

cinder stand alone with gluster backend
glance stand alone with gluster backend 
nova + all other components installed on stand alone
+ one other compute.

Steps to Reproduce:
1. create an image with linux os
2. create a volume from the image
3. boot an instance from the volume using horizon
4. create a file/dir under / and under /tmp
5. terminate the instance
6. boot a second instance from the same volume
7. once the instance starts ls -l to see if the file/dir exists

Actual results:

the dir and file do not exist

Expected results:

volume is not stateless and all information should not be deleted 

Additional info: logs 

I am not sure if this is nova or cinder related, but since the consistency of the data is a cinder issue I am opening for cinder

Comment 1 Giulio Fidente 2014-01-08 15:46:16 UTC
this is happening for me with regular empty volumes too; after formatting and populating them from a random instance, they appear to be unfortmatted on another

Comment 2 Eric Harney 2014-01-13 18:31:24 UTC
Note that terminating an instance as part of this test means there could be cache data in the guest that isn't written to the backing storage.  Would be much better to poweroff the guest from within.

Comment 3 Lars Kellogg-Stedman 2014-02-11 20:49:41 UTC
Dafna, can you perform the test that eharney suggested (powering down the instance before deleting it)?

Comment 4 Dafna Ron 2014-02-11 21:18:48 UTC
(In reply to Eric Harney from comment #2)
> Note that terminating an instance as part of this test means there could be
> cache data in the guest that isn't written to the backing storage.  Would be
> much better to poweroff the guest from within.


1. while i was doing this test I shut off the instance internally but that moves the instance to shut off state and does not destroy it.
2. if there is internal cache lost during a destroy of instance I think its a bug - do not destroy the instance until all data is copied (that is the whole point of cinder)
3. I don't have HW in the next few days to try it again. we can try to investigate together in a few days if you like.


Note You need to log in before you can comment on or make changes to this bug.