[
https://issues.redhat.com/browse/ISPN-12266?page=com.atlassian.jira.plugi...
]
Dan Berindei commented on ISPN-12266:
-------------------------------------
Caches with local storage could also be made available before all the members of the
stable topology join, with the same rules as for keeping a cache AVAILABLE (need a
majority of the nodes + at least one owner of each segment).
We could also keep the shutdown command as-is and add a new command to force the cluster
to become available if no data would be lost, and an optional {{--force}} argument
skipping the data loss checks.
shutdown cluster should have an option to not block restart if there
is no or a shared persistence
--------------------------------------------------------------------------------------------------
Key: ISPN-12266
URL:
https://issues.redhat.com/browse/ISPN-12266
Project: Infinispan
Issue Type: Feature Request
Reporter: Wolf-Dieter Fink
Priority: Minor
Labels: graceful, graceful-start-stop
The current implementation of a full cluster restart will save the CH and state for all
caches.
If caches does not have a persistence or the persistence is a shared one this might not
be necesarry as the cache can start correctly with one node (empty or consistent
persistence).
In that case it should be allowed to start the cache and make it available.
There are two approaches
- use extra parameter for shutdown to allow it
This will leave the default behavior the same and the admin need to decide
- change the current behavior and add a parameter to flag a FULL cluster restart
The benefit of the full cluster restart is that the caches will be available for clients
only if the cluster is completely restarted, so no further state-transfer is needed for
optimal performance
--
This message was sent by Atlassian Jira
(v7.13.8#713008)