On Fri, Apr 22, 2011 at 10:17 PM, Allen Wittenauer wrote:
Could someone actually take the branch and try to install from scratch? i.e., not copy a pre-existing config. I found a multitude of problems with 203 doing this, but haven't had a chance to file JIRAs for all of them. (Pay attention to the log messages in particular). I'm not going to be able to do this for various reasons.
Are the values in the default config broken and/or are some required
settings undocumented? The setup guide is almost surely out of sync.
Does the single-node setup still work? We should also test and
document an upgrade from an existing 0.20.2 cluster.
There are also a lot of places where hadoop-*-{core|hdfs|mapred} are referenced in the documentation, but since this release breaks with 0.20.{0-2}, all of those documentation references need to be fixed.
Are you talking about the jar naming since maven? I found a couple
places in the mapred-tutorial that use the old naming. Are there
others?
As it stands, I wouldn't want to use this release in production. In addition to the previously mentioned issues, the job size unpredictability/regression with capacity scheduler is particularly disturbing. The math we're seeing doesn't seem to match the documentation at all. (Are we missing a test here?)
The documentation has almost surely lagged behind the changes in the
CS, particularly w.r.t. limits and what a single user can claim of the
queue capacity. Can you file a JIRA w/ an example where the assigned
capacity doesn't match what was expected? That should definitely be
fixed. -C
