04-10-2014 05:21 PM
I had Apache 2.3.0 installed and decided to try CDH 5. I uninstalled Apache and installed CDH, thinking that there would be no issues, but its telling me I need to upgrade. What is the compatibility policy for CDH. I seem to remember back in CDH3 that this switch was possible with no upgrade.
04-14-2014 04:39 PM
Sorry about the delay in response. While our CDH5.0.0's Hadoop is based on top of Apache Hadoop 2.3.0, we pulled in certain features from later Hadoop branches (e.g. ACL support from Apache Hadoop 2.4.0) in CDH 5.0.0 that require an fsimage upgrade when moving from Apache Hadoop 2.3.0 to CDH 5.0.0.
I consider this no different than having to upgrade fsimage when moving from Apache Hadoop 2.3.0 to Apache Hadoop 2.4.0.
In particular, what I can say is the fsImage for CDH5.0.0's Hadoop matches up with Apache Hadoop 2.4.0 release.
Here is the fsImage version for CDH5.0.0 Hadoop and here it is for Apache Hadoop 2.4.0 and as you can see, they are the same. However, here it is for Apache Hadoop 2.3.0, and it's different than the former two.
While we did backport this feature in CDH 5.0.0 hadoop, we still call our hadoop based on Apache Hadoop 2.3.0, because we didn't necessarily backport ever single feature/fix in Apache Hadoop 2.4.0.
Hope that helps.