Adding general notes about rolling upgrades

There are some known problems that may happen during rolling (live)
upgrade. Even though feature is considered a tech-preview there's no
harm in communicating them in the release notes.

This commits adds also a note on the fact that now Cinder services
detect lowest RPC version and that it is needed to keep the `services`
table up to date.

Change-Id: Ic4c9188fb0bbd375d809c7dedd9e07df049ae648
Related-Blueprint: rpc-object-compatibility
This commit is contained in:
Michał Dulko 2016-02-26 10:27:07 +01:00
parent 68e7ae469d
commit 2b4b4883a3

View File

@ -0,0 +1,26 @@
---
issues:
- Cinder services are now automatically downgrading RPC
messages to be understood by the oldest version of a
service among all the deployment. Disabled and dead
services are also taken into account. It is important to
keep service list up to date, without old, unused
records. This can be done using ``cinder-manage service
remove`` command. Once situation is cleaned up services
should be either restarted or ``SIGHUP`` signal should
be issued to their processes to force them to reload
version pins. Please note that cinder-api does not
support ``SIGHUP`` signal.
upgrade:
- If during a *live* upgrade from Liberty a backup service
will be killed while processing a restore request it may
happen that such backup status won't be automatically
cleaned up on the service restart. Such orphaned backups
need to be cleaned up manually.
- When performing a *live* upgrade from Liberty it may
happen that retype calls will reserve additional quota.
As by default quota reservations are invalidated after
24 hours (config option ``reservation_expire=86400``),
we recommend either decreasing that time or watching for
unused quota reservations manually during the upgrade
process.