Apache Tez and Hive LLAP

Cloudera Data Warehouse supports low-latency analytical processing (LLAP) of Hive queries.

Hive uses Apache Tez to execute queries internally. Apache Tez provides the following execution modes:
  • Container mode

    Every time you run a Hive query, Tez requests a container from YARN.

  • LLAP mode

    Every time you run a Hive query, Tez asks the LLAP daemon for a free thread, and starts running a fragment.

Apache Tez provides the framework to run a job that creates a graph with vertexes and tasks. SQL semantics for deciding the query physical plan, which identifies how to execute the query in a distributed fashion, is based on Apache Tez. The entire execution plan is created under this framework. SQL syntax in Hive is the same irrespective of execution engine (mode) used in Hive.

Apache Tez does not have to start from the ground up, requesting a container and waiting for a JVM, to run a fragment in LLAP mode. LLAP mode provides dedicated capacity. Caching is automated. The primary difference between LLAP mode and container mode, is that in LLAP mode the LLAP executors are used to run query fragments.

In Cloudera Data Warehouse (CDW), the Hive execution mode is LLAP. In Cloudera Data Hub on CDP Public Cloud and CDP Private Cloud Base, the Hive execution mode is container, and LLAP mode is not supported. When Apache Tez runs Hive in container mode, it has traditionally been called Hive on Tez.