Server's host not changed but actually on dest node after live-migrating
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack Compute (nova) |
Expired
|
Undecided
|
Unassigned |
Bug Description
Description
===========
The insances has been migrating for more than two hours. Then it got error 'Unauthorized'.
The host of output of CLI `openstack server show` was stall the old one.
But the instances had already been running on dest node.
Steps to reproduce
==================
1. Create one instance with large mem
2. Run some application which cosume mem, like `memtester`
3. Execute live-migrate
Expected result
===============
Rollback instance to old one, or update instance's host to dest node
Actual result
=============
Instance on dest node but the host is src node in DB
Environment
===========
$ git log -1
commit ee6af34437069a2
Author: Luigi Toscano <email address hidden>
Date: Wed Dec 18 00:28:15 2019 +0100
Zuul v3: use devstack-
... and replace its legacy ancestor.
Change-Id: Ifd4387a02b3103
(cherry picked from commit e7e39b8c2e20f5d
(cherry picked from commit e82e1704caa1c2b
(cherry picked from commit 99aa8ebc12949f9
Logs & Configs
==============
2020-04-02 21:08:32,890.890 6358 INFO nova.virt.
254a85a4130a268
es processed=
2020-04-02 23:08:05,165.165 6358 INFO nova.virt.
2020-04-02 23:08:05,166.166 6358 INFO nova.compute.
2020-04-02 23:08:05,535.535 6358 WARNING nova.virt.
2020-04-02 23:08:05,537.537 6358 ERROR nova.compute.
Looks like something bad happened to you when finishing the live migration :
The exception you see comes from https:/ /github. com/openstack/ nova/blob/ 07939afe3e60ef6 061aa89ef408dc7 2785f3d8dd/ nova/virt/ libvirt/ driver. py#L9225- L9228 which means something got bad during the live migration process.
Based on the log above, we can see it was happening in https:/ /github. com/openstack/ nova/blob/ 07939afe3e60ef6 061aa89ef408dc7 2785f3d8dd/ nova/virt/ libvirt/ driver. py#L9130- L9135 which is at the very end of the migration itself.
As you can also see, we call post_live_ migration( ) which runs https:/ /github. com/openstack/ nova/blob/ master/ nova/compute/ manager. py#L8384 and later fails when calling either Neutron or Cinder.
Could you please check the neutron-api and cinder-api logs to see a potential HTTP401 failure at the same time, and then you'll see that probably there is some misconfiguration.