How to do manual rolling upgrade?

[#1]

I use Cloudera Manager Express.

I am trying to upgrade from 5.3.8 to 5.11.1

When I use CME I get an error:

"File system image contains an old layout version -59. An upgrade to version -60 is required."

As CME does not support rolling restart I will need to do this manually. Can anyone point me to documentation on how to do a rolling restart from the command line because CME does not support it.

Thanks

Paul

[#2]

I can not do this from the command line because the packages aren't installed with yum.

eg; sudo service hadoop-hdfs-namenode upgrade 

Does not work

I can not do this with Cloudera Manager Express because rolling restart is not supported.

So, catch-22. Do I give up on Cloudera Manager and go back to a manual configuration?  

Surely this is possible??

[#3]

I don't know of a doc or if it is not possible. You should be able to do it all using the CM api. You just need to figure out the order in which to do it. Then you could build a script that does it all with CM in place. This is a lot of work and this is why rolling upgrades is an Enterprise feature.

What I would do is get the quickstart VM or install a single node version and use the trial license. Then you can conduct rolling restart, get the ordering, commands from the logs, etc. But then you need to build and test your version.

Good luck.

[#4 · Accepted solution]

Thanks very much for replying. I was hoping these forums might be a bit busier but I guess there aren't so many people using community edition these days.

Anyway, I did get this working in the end, but I was very disappointed in the lack of clear documentation. From what I remember this is what I did:

**BACKUP NAMENODE BEFORE TOUCHING ANYTHING**

1) Shutdown the cluster

2) Upgraded to 5.11 using Parcels in CM. Did *NOT* restart cluster

3) Started ZooKeepers only

4) Went to HDFS and choose "Upgrade HDFS Metadata". The new binaries must be in place, and the cluster must be down

5) Then try to finalize the upgrade. I am not sure this worked as expected. I think there was an issue with one BP

6) re-started cluster

7) When to NameNode page and selected "Finalize Metadata Upgrade"

After a long time it all sorted itself out and is up and running on 5.11

What a painful process though.

[#5]

Thanks for sharing, it's helpful.