[Openstack-operators] automatically evacuate an instance when a compute node dies

Tzach Shefi tshefi at redhat.com
Wed Dec 10 15:02:43 UTC 2014


Hi, 

First of all for this to work (if possible..) guessing you need to configure live migration, else instance disk is saved on local hostA's disk /var/lib/nova/instances. 
As this hostA is down you can't reach instance's disk -> end of story :( 
On the other hand if you configure live migration with remote shared storage, instance's disk will be saved on a remote shared NFS (or other options maybe ceph)
And even if hostA is down the instance's disk would still be accessible for hostB. 

https://www.mirantis.com/blog/tutorial-openstack-live-migration-with-kvm-hypervisor-and-nfs-shared-storage/
http://blog.zhaw.ch/icclab/setting-up-live-migration-in-openstack-icehouse/
And many more just google "configure nova live migration"

Maybe you could setup pacemaker to check access to hostA if it can't be reached then issue #nova evacuate instanceID hostB then nove reboot --hard instanceID
http://serverfault.com/questions/413566/openstack-is-it-possible-to-migrate-an-instance-from-a-dead-compute-server-to-a
No idea how to go about doing this from pacemaker's side.
Interesting to check I'll try to play around with it. 


One question comes to mind thought we are talking about Openstack right :)
By design/definition basically if an instance dyes (or it's host) just boot up other instances on other hosts in it's place. 
Remember instances should be regarded as temporary dispensable instances. 

Recall Openstack can but isn't meant to replace virtualization solution like Ovirt/rhev/vmware..
These solutions have built in automatic virtual machine fail-over just what your looking for. 
You'll still need shared storage and proper license needed for such "enterprise options" 

Understanding when you need an "instance" vs. "virtual machine" is hard to grasp but fundamental to selecting the proper tool for the job at hand.
Not to mention may save you lots of hassle and frustration later on. 

Tshefi

----- Original Message -----
From: "Pedro Sousa" <pgsousa at gmail.com>
To: openstack-operators at lists.openstack.org
Sent: Tuesday, December 9, 2014 4:46:59 PM
Subject: [Openstack-operators] automatically evacuate an instance when a	compute node dies

Hi all, 

is there a working solution in nova to automatically restart an instance when a compute node dies in a healthy node? 

I've heard about pacemaker, any good howto to help with this? 

Regards 

_______________________________________________
OpenStack-operators mailing list
OpenStack-operators at lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators



More information about the OpenStack-operators mailing list