Apologies off the line for this one.  This is an annoying one.

I'm trying to hack tez into Cloudera's fairly gimped version of hive.
Upgrading Hive to 1.2 would
fix the problem (I know, I'm running it off to the side), but that
would mean metadata changes that would make reverting quite difficult.

So here's some basic information:
1-  Compiling MR down to tez works (ie,
mapreduce.framework.name=yarn-tez).  This works for all regular MR as
well as using MR in hive, but compiled to tez.
2-  Running tez on hive 1.2.1 works against the same cluster (separate
metastore/hive/etc)
3-  Tez itself seems fine.

I'm convinced this is a hive issue, but I'm sending it here because
you folks might have a good idea on what the issue is.  It appears
that the tez package from hdfs is not being localized when children
are spun up.  The AM does work.

Configured with minimal + local hadoop jars does NOT fix this.  I've
also had to add the following to tez-site.xml to get compilation to
tez to work (there's some stubs that cloudera adds to hive that I
needed to populate):

<property>
  <name>tez.cluster.additional.classpath.prefix</name>
  
<value>/u/cloudera/parcels/CDH-5.4.1-1.cdh5.4.1.p0.6/jars/hadoop-mapreduce-client-common-2.6.0-cdh5.4.1.jar</value>
</property>

So... dag compilation to tez works.  The session loads.  The tasks
themselves fail because:
Error: Could not find or load main class org.apache.tez.runtime.task.TezChild

Yet... this works for every other execution of tez.  Is there
something I could look into here?  I could in theory populate all
nodes with the tez libraries, but I feel like that would just lead me
down a bad path.  Suggestions?

Thanks,
-Aaron

Reply via email to