|
- adding standard Doctor / Vitrage Fault Mgmt Architecture Diagram, and
- refering to Doctor SB API
-------------
Updating HA GUEST API OVERVIEW Proposal after review with OPNFV DOCTOR Team.
Minutes from meeting, that were addressed as part of this update are below.
- VM Heartbeating & Health Checking
* libvirt watchdog and its integration / use with OpenStack
> https://blueprints.launchpad.net/nova/+spec/libvirt-watchdog
> https://wiki.openstack.org/wiki/LibvirtWatchdog#Notifications
> Need to review behaviour of this capability as compared to the
proposed VM Heartbeating/Health-checking
> i believe there is a bit of overlap
but believe that VM Heartbeating / Health-checking provides a
more complete solution
> i'll update document with a NOTE on comparing the proposal with
the libvirt watchdog
* Update the architecture diagram to be consistent with the most current
DOCTOR Architecture Diagrams
> e.g. Vitrage & Congress are not necessrily deployed at same time
> use DOCTOR terminology where applicable, e.g. "inspector modules"
> OPNFV DOCTOR includes patches/components of AODH, NOVA, NEUTRON
... not just Vitrage and Congress, as shown in diagram
- although also had comment that should remove OPNFV DOCTOR outline
from diagram as OPNFV DOCTOR is a reuirement specification and
not an implementation
> indicate (possibly just in text below diagram) that the "Guest
Heartbeat / Health-check Server' on Controller Node is possibly not
required, as the Vitrage data source interface can be remotely
reached by "Guest Heartbeat / Health-check Compute" on the Compute
Node
* In text, provide a little more detail on content of actual messaging
e.g. PDUs and rough message content
* OVERALL
> believe there was a general agreement in the way that VM Heartbeating
& Health-checking was integrated / inter-worked with OPNFV DOCTOR's
Vitrage / Congress and overall OPNFV fault reporting architecture
> key feedback was to understand and highlight the additional value of
the VM Heartbeating & Health-checking functionality over the existing
libvirt watchdog integration into OpenStack.
- Server Group Messaging
* suggestion that Rabbit MQ pub/sub messaging could be an alternative for
routing of messages
> this is an implementation detail though
* general discussions on "HA use cases" for how this messaging could be
leveraged
> e.g. split-brain avoidance, faster peer VM state change notifications
* OVERALL
> agreement that the Server Group Messaging Architecture did NOT conflict
with Doctor Architecture
> need further review with OPNFV MANO Team as to how they would position
this functionality
e.g.
- position it as an alternative for various HA use cases ?
- versus
- mandating that this service group messaging be used for specific
HA user cases
Change-Id: Icd54bbf8889017cfe3f617656ddf483cbb171e63
Signed-off-by: gwaines <greg.waines@windriver.com>
|
gwaines
fuqiao