[Openstack-operators] [openstack-dev] [openstack-operators] [glance] Austin summit summary: Rolling upgrades
nik.komawar at gmail.com
Thu May 5 06:06:40 UTC 2016
Just wanted to send a brief summary of the discussions at the summit.
This list is not holistic however, it covers the relevant aspects that
various stakeholders need to be aware of.
* The intent is that we want operators to be able to upgrade from one
Glance release to the next with minimal (ideally, zero) downtime.
* Nova's been working on this, so there's a good example of how to
accomplish this. Also, the product working group has a cross-project
spec on this topic.
* Glance DB is the one component that would require some sophisticated
logic to avoid the downtime. Other services are being handled by
upgrade & swap mechanism.
* Some discussion was around relative comparison on what other
services are doing:In terms of performance there's simple schemain
Glance thoughcan have massive data.
The different approaches today include:
oslo versioned objects
neutron: expansion/contraction scheme; expand, lazy updates, force
contract after some time
ironic: not using versioned objects, pinning the version
cinder: split across multiple releases - add table, code can handle both
* Besides these there was some discussion around best practices for
upgrades: The preferred upgrade scheme is first DB, then registry
and then API.
* There were some research items: documenting the draining protocol
for Glance nodes handling uploads. Write up alternatives in the spec
based on findings what other projects are achieving; sign-ups
include mclaren for cinder, rosmaita for neutron, nikhil for nova.
More volunteers are welcome.
For more information please reach out to me on #openstack-glance, email,
reply here etc.
More information about the OpenStack-operators