Spark SQL supports most commonly used features of HiveQL. However, different HiveQL statements are executed in different manners:
-
1. DDL statements (e.g.
CREATE TABLE
,DROP TABLE
, etc.) and commands (e.g.SET <key> = <value>
,ADD FILE
,ADD JAR
, etc.)2. In most cases, Spark SQL simply delegates these statements to Hive, as they don’t need to issue any distributed jobs and don’t rely on the computation engine (Spark, MR, or Tez).
-
SELECT
queries,CREATE TABLE ... AS SELECT ...
statements and insertionsThese statements are executed using Spark as the execution engine.
The Hive classes packaged in the assembly jar are used to provide entry points to Hive features, for example:
- 1. HiveQL parser
- 2. Talking to Hive metastore to execute DDL statements
- 3. Accessing UDF/UDAF/UDTF
As for the differences between Hive on Spark and Spark SQL’s Hive support, please refer to this article by Reynold: https://databricks.com/blog/2014/07/01/shark-spark-sql-hive-on-spark-and-the-future-of-sql-on-spark.html