[nova] Instance is not shutoff after nova-compute start for a long time
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Mirantis OpenStack |
Invalid
|
Medium
|
Sergey Arkhipov |
Bug Description
Instance is not shutoff after nova-compute start for a long time.
Steps to reproduce:
* Choose instance and detect compute which manages it
* SSH to compute and stop nova-compute with 'service nova-compute stop'
* Wait for 2-3 minutes
* Proceed to Horizon and click 'Shutoff instance'
* Check that no error is shown, instance has not changed its state
* Wait for 1 minute, proceed to the nova host again (where 'nova-compute' is not running)
* Start nova-compute with 'service nova-compute start'
* Wait for 5 minutes, check that instance has not changed its state and still accessible using webconsole at least.
* Try to shutoff it again, with working nova-compute
Expected result:
One of 2 results were expected:
* Warning or error message that it is not possible complete operation right now
* Instance is shut after nova-compute is up and running again
Real result:
* No error in Horizon
* Instance is still running and accessible
I've checked with instance '39d5f023-
VERSION:
feature_groups:
- mirantis
production: "docker"
release: "8.0"
openstack_
api: "1.0"
build_number: "199"
build_id: "199"
fuel-nailgun_sha: "a65871a955dd08
python-
fuel-agent_sha: "88c993eab990dd
fuel-
astute_sha: "c8400f51b0b922
fuel-library_sha: "33c0fa3aada734
fuel-ostf_sha: "11afd5743a12b1
fuel-
fuelmenu_sha: "d12061b1aee82f
shotgun_sha: "c377d163519f6d
network-
fuel-upgrade_sha: "1e894e26d4e142
fuelmain_sha: "22fe551f5525d1
Changed in mos: | |
milestone: | none → 8.0 |
assignee: | nobody → MOS Nova (mos-nova) |
importance: | Undecided → Medium |
status: | New → Confirmed |
I checked this both on env #307 and in the code:
The shutoff action is sent as an RPC CAST to nova-compute. Even if a particular nova-compute service is down at the moment of time, the message will be stored in the message queue until nova-compute is up again, and then be consumed properly, shutting the instance down.
The status of instance is updated properly in the DB (i.e. nova list will show it as SHUTOFF) and there is no qemu process running on the compute node.