On Wed, Jul 27, 2016 at 1:00 PM, Brown, Scott M (IS) <[email protected]> wrote: > Yes I only sent the one set from 1.6.2 that doesn’t work.. I can paste the 2 > files from 1.6.1 that will resolve if you want them, let me know, glad to do > it. >
Yes, please include the configs you are using in both cases. > I'm downloading the yum install packages from Cloudera now, I'll install and > look at the conf files and see how general_classpath is set. The setting is > looking in the correct paths for commons-io, as both /usr/lib/Hadoop and > /usr/lib/Hadoop/lib are defined in there, so its like you referred to, the > setting of classpath has changed from 1.6.1 to 1.6.2+ > The configs you posted from the 1.6.2 home do not include the /usr/lib/hadoop and /usr/lib/hadoop/lib directories. Those paths are in a commented out example for HDP2.0. The only un-commented out hadoop set is the Hadoop 2 based on the packaging that Apache Hadoop itself does for their convenience binary tarballs. Those paths don't work on CDH. > Could you clarify this " but I wouldn't expect a deployment with these > configs to work."? If there is a bug in the way classpath works after 1.6.1, > would it not have a workaround or fix? > As mentioned above, the config that was posted on pastebin doesn't contain a path to the jars Accumulo needs in a CDH installation. So even if a particular command like "classpath" works, I wouldn't expect a cluster deployment based on it to be production worthy. > Typically this has always been minimal setup, and init works, but there seems > to be a difference in 1.6.2+ The example configs that ship with Accumulo are just examples. As mentioned in them, you have to customize them to your environment (such as the zookeeper quorum). The classpath config will vary greatly depending on what you're using to satisfy your runtime dependencies, since Hadoop locations on disk a different for different packaging options (and some folks would rather avoid getting them from Hadoop entirely). I believe in 1.6.1 the example configs contained both the hadoop 2 packaging from the Hadoop PMC and the HDP 2.0 layout. That's probably why it happened to work. -- busbey
