We have released an interim release of MR3 1.3 (MR3 1.3-SNAPSHOT) before the final 1.3 release. The release includes only Hive 3 and does not include Hive 2 and Hive 4-SNAPSHOT.
For binary tar balls:
For docker files:
For the scripts for running Hive on MR3 on Kubernetes:
For the source repository of Tez-MR3:
For the source repository of Hive 3 on MR3:
Previously Hive 3 on MR3 was mostly identical to Hive 3.1.2, with only a few additional patches. We are in the middle of applying new patches, and this interim release has applied additional 223 patches. The final release of MR3 1.3 will apply additional 100+ patches to Hive 3 on MR3, making it as mature as HDP 3.1.5.
The source repository of Hive 3 on MR3 has also been reorganized. The commit history is a sequence of Hive patches, followed by the last single commit specific to MR3. Now, the user can quickly find out which patches have been backported to Hive 3 on MR3.
One more thing -- we have implemented Spark on MR3. It's implemented as an add-on, so no change to Spark is necessary. Spark on MR3 uses MR3 as the resource manager, and different Spark drivers can share Yarn containers or K8s pods. So, it's very useful for multiple Spark users sharing a cluster.
Ultimately we would like to run both Hive and Spark using a shared MR3 master, facilitating the exchange between Hive LLAP I/O and Spark RDD.