I'm finding the preparation before upgrading quite difficult since I'm already new to hdp 2.3, I want to install hdp 2.4 since it has certain features I would like to use. So I was wondering would it be easier for me to uninstall and delete hdp 2.3 then install hdp 2.4 rather than upgrading?
Depends. How much stuff do you want to keep? Upgrading a hadoop cluster can be challenging but if you need to keep the data ...
I just upgraded to 2.3.4 from 2.3.2 and that was pretty straightforward with the new express upgrade but the jump to 2.4 might be bigger. Still I would say if you have data, hive tables etc. give update a shot. But for a clean system? Starting fresh might be easier.
This largely depend on what is the cluster purpose.
If you cluster is just a dev/test cluster and nobody care about the data/processes on the cluster. Start from scratch with new OS images maybe the easier.
If you do need to keep the data/process on your cluster, but fine with same downtime, then doing a express upgrade should be your best choice.
When talking about a production system where data lose and downtime are your concerns, rolling upgrade with minimum impact should be your choice.
@ A Ismail
It's depend on you if you need to install HDP 2.4 without upgrading then you can do it.
To uninstall the HDP follow the steps:-
How to uninstall hdp from scratch
Delete all database from mysql or whatever database you use
First unsitall hdp packages and zookeeper with
Yum remove command
Verify pakages are removed are not woth command:-
yum list installed |grep HDP
If not then remove all packages
With yum remove command.
rm –fr /usr/hdp on all nodes
Run following command :-
Now run the command to delete all the user which are created by hdp
python /usr/lib/python2.6/site-packages/ambari_agent/HostCleanup.py --silent --skip=users
Once you unistall the package now you can start installation of HDP 2.4.
The clean-up that Ashnee proposes may be the fastest way, but it may also not be a guaranteed success for installing from scratch the new version. The upgrade that Bemn and Qi proposed is also a viable option. If it is done right, it should not be that difficult to achieve. As everyone stated in this forum, it depends on your environment layout and objectives. If you are talking about a small cluster and you don't care about the data, then you may be better taking Ashnee's approach or even more, by wiping-out the OS and reinstalling the minimum version (you may even do an upgrade of the OS since you are at it, e.g. RHEL 7 seems to give HDP a boost when compared with RHEL 6) and go ahead with installation.