]> review.fuel-infra Code Review - openstack-build/cinder-build.git/commitdiff
Adding general notes about rolling upgrades
authorMichał Dulko <michal.dulko@intel.com>
Fri, 26 Feb 2016 09:27:07 +0000 (10:27 +0100)
committerMichał Dulko <michal.dulko@intel.com>
Fri, 26 Feb 2016 09:45:59 +0000 (10:45 +0100)
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

releasenotes/notes/general-upgrades-notes-120f022aa5bfa1ea.yaml [new file with mode: 0644]

diff --git a/releasenotes/notes/general-upgrades-notes-120f022aa5bfa1ea.yaml b/releasenotes/notes/general-upgrades-notes-120f022aa5bfa1ea.yaml
new file mode 100644 (file)
index 0000000..1daf74a
--- /dev/null
@@ -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.