Open Stack

Hi guys,

I want to continue this discussion to get some agreement on it.  There're
three kind of HA solutions mentioned in this thread.

1.  Hypervisor level HA.  Based on the previous discussions,  it seems that
pacemaker is the best candidate. But it still lacks of success story and
document. I am wondering its response time on failure and if it could cause
mistaken fencing.  I will investigate it and give you an update if I could
make any progress.

2. Instance OS level HA.  libvirt watchdog can be used to detect instance
OS failure (kernel panic or hang). As Daniel said, currently nova doesn't
send out any notification when the watchdog is triggered. If it does, then
ceilometer can raise an alarm when get this kind of message. Probably the
instance could be recovered by pre-defined reboot action. I don't know if
this proposal is welcomed by ceilometer guys.

According to
http://www.vmware.com/files/pdf/VMware-High-Availability-DS-EN.pdf,
the above two cases are what VMware HA provides.

3. Instance App level HA.  I am not talking about any special application
HA solution, but a generic one, like keepalived.  It can provide more fine
grained check/recovery than instance OS level.  The following article
introduce how to do it:
http://blog.aaronorosen.com/implementing-high-availability-instances-with-neutron-using-vrrp/
  We could use heat to orchestrate the resources as the article describes.
But it can't work with neutron l2 population because neutron is not aware
of the vip switching inside instance, and therefore the pre populated arp
table got stale after switching. Maybe sending an gratuitous arp packet in
keepalive script could workaround it.  An alternative is letting keepalive
switch instance's floating ip address instead of virtual ip address.  The
other potential issue in this solution is that openstack still lacks of
multiple attached volume support. Like switching floating ip address,  the
data volume can be detached and attached in the keepavlie script.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/enterprise-wg/attachments/20141217/a4350f8c/attachment.html>