BOOKKEEPER-172<https://issues.apache.org/jira/browse/BOOKKEEPER-172> Provide the implementation for upgrades between versions.
Here fileSystem layout will refer the bookie data layout format which is persisted on the directories or ZKs. The new version of software(BK server) should be able to load them to memory correctly. So, It is better to do the upgrades as suggested in doc. That will take care of upgrading your software and also can finalyze if every thing ok. If there are any issues, it will roll back as well. Let Ivan/Flavio conform about the layout changes between versions and if i am wrong BK case. Thanks & Regards, Uma ________________________________ From: John Nagro [[email protected]] Sent: Thursday, May 03, 2012 7:51 PM To: [email protected] Subject: Re: rolling restarts Thanks Uma - what I actually need to know is if its okay doing the restarts on a live cluster. I think my filesystem is okay at this point, but I want to upgrade the server software on a live, running cluster. Is it fine to ^C, upgrade, start again - as long as I only do it to 1 of the 5 nodes at a time? Thanks! -John On Thu, May 3, 2012 at 10:10 AM, Uma Maheswara Rao G <[email protected]<mailto:[email protected]>> wrote: Hi John, Seems Ivan already worked on upgradation framework. I am not sure about the filesystem layout changes with previous version as i was not aware of older versions code changes. Pls Refer * BOOKKEEPER-172<https://issues.apache.org/jira/browse/BOOKKEEPER-172> You can find some details here: http://zookeeper.apache.org/bookkeeper/docs/trunk/bookkeeperConfig.html Hope this will help you. Regards, Uma ________________________________ From: John Nagro [[email protected]<mailto:[email protected]>] Sent: Thursday, May 03, 2012 7:09 PM To: [email protected]<mailto:[email protected]> Subject: rolling restarts hello - I have a cluster with 5 bk nodes, my client settings are to use 3 at a time with a quorum of 2. I want to upgrade the software of a live cluster from an older 4.1-SNAPSHOT to a newer one. I am under the impression that I should not need to upgrade the filesystem since I've already done that when moving to 4.1. Am I correct in that I should be able to bring down the nodes one at a time, replace the jar, and bring them back up. As long as I do it one at a time the cluster should continue to operate properly? also - there doesn't appear to be any sort of graceful shutdown code, and the docs show ^C in a few places - so thats how i've been bringing nodes down. that is still the standard procedure? thanks! -John Nagro
