Hi Guys,

Just went through the upgrade.  It took a while, but mostly because the Hadoop 
(Storm, Hbase, Yarn) upgrades failed to upgrade properly via the automated 
Ambari upgrade from this link: 

http://docs.hortonworks.com/HDPDocuments/Ambari-2.4.0.1/bk_ambari-upgrade/content/ambari_upgrade_guide.html

The upgrade of Ambari itself wasn't necessary, but upgrading Hadoop took some 
fumbling around.  Once I got Hadoop updated, upgrade of Metron to 0.3.0 was 
fairly straightforward.

-  I switched git branches from 0.2.1 to 0.3.0 and did a new maven build
-  Stop all sensors via Monit
-  Then copied the new jars from my local target directories of all the maven 
modules up to the server that had the old jars staged
- I backed up all my old jars off my deployment box and renamed the new jars 
(with 0.3.0 in their name) back to 0.2.1.  Make sure to grab the uber jars.  
- I then checked my configs and they survived the upgrade.  I did the dump from 
zookeeper to make sure
- I then killed all my topologies.  After you go through the upgrade for some 
reason it re-launches your old storm topologies, even though prior to the 
upgrade it asks you to kill them.  I am not sure why it does that.  But 
obviously  0.2.1 topologies were built against storm 0.10 so they fail after 
storm is upgraded to 1.0
- I then used the old 0.2.1 launching scripts to launch all my topologies 
(newer jars renamed back to 0.2.1).  Everything came up and worked fine
- ES service survived the upgrade and did not need to be rebooted.  Kibana came 
up fine.  ES indexes were there
- Old data in HDFS that I had from the running 0.2.1 system was there after the 
0.3.0 upgrade

Has anyone else attempted the upgrade?  Any problems?

------------------- 
Thank you,

James Sirota
PPMC- Apache Metron (Incubating)
jsirota AT apache DOT org

Reply via email to