instance cleanup needs to be more defensive
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
nova-lxd |
Confirmed
|
High
|
Unassigned |
Bug Description
If an instance moves to ERROR or STOPPED state in the underlying LXD hypervisor, the driver appears to sometimes have trouble cleaning it up, resulting in instance leakage - the instance is gone in nova, but is clearly still present in lxc list:
+------
| NAME | STATE | IPV4 | IPV6 | TYPE | SNAPSHOTS |
+------
| instance-
+------
| instance-
+------
Profiles also get left behind when this occurs (as the can't be deleted because the container is present, and LXD blocks due to the dependency).
This is pretty easy to reproduce on a standard tempest run with devstack+nova-lxd
Changed in nova-lxd: | |
assignee: | nobody → Alex Kavanagh (ajkavanagh) |
Changed in nova-lxd: | |
importance: | Undecided → Critical |
importance: | Critical → High |
Changed in nova-lxd: | |
assignee: | Alex Kavanagh (ajkavanagh) → nobody |
This bug is also now blocking the nova-lxd gate