Like IMPALA-10871, this patch adds MetastoreShim to support Apache
Hive 2.x. At the build time, based on the environment variable
IMPALA_HIVE_DIST_TYPE one of the three shims is added to as source
using the fe/pom.xml build plugin. And select the dependencies
related to Hive in the fe/pom.xml based on the environment variable
IMPALA_HIVE_MAJOR_VERSION. There are some duplicate classes under
compat-apache-hive2 directory, e.g. fe/src/compat-apache-hive-2/java/
org/apache/impala/catalog/events/MetastoreEvents.java duplicates
fe/src/main/java/org/apache/impala/catalog/events/MetastoreEvents.java
The class in compat-apache-hive2 is a simplified version that works
with Apache Hive 2.x. So we don't need to extract lots of Hive-dependent
codes in MetastoreEvents.java into the metastore shim. Due to this,
the build process simply remove the original source code when building
on Apache Hive 2. Additionally, it should be noted that all
the code in the fe/src/compat-apache-hive-2/java/org/apache/hadoop/hive
directory comes from Apache Hive 3.x, original
source: https://github.com/apache/hive/blob/branch-3.1
In order to reduce the unnecessary intrusion into the code, skip all
tests when building with Apache Hive 2.x. If wanting to build Impala
adapted to Apache Hive 2.x, please set the following environment variables
before `source bin/impala-config.sh`:
export USE_APACHE_COMPONENTS=true
export USE_APACHE_HIVE_2=true
TODO:
- IMPALA-14581: Support testing related to Apache Hive 2 in the minicluster.
Testing:
- Compile using the -package option to obtain the package. After
deployment, perform all types of query tests, including SELECT, INSERT,
CREATE TABLE, ALTER TABLE, COMPUTE STATUS, etc. In addition, comprehensive
testing has been conducted on the metadata auto-synchronization functionality.
The tests confirm that all event types are supported except for AlterDatabaseEvent,
AllocWriteIdEvent, AbortTxnEvent, PseudoAbortTxnEvent, and CommitCompactionEvent.
It is worth noting that these unsupported events are not generated in Apache Hive 2,
so their lack of processing support does not impact the functionality.
Change-Id: Ib5f104dc8d131835b8118b9d54077471db65681c
Reviewed-on: http://gerrit.cloudera.org:8080/21760
Reviewed-by: Impala Public Jenkins <impala-public-jenkins@cloudera.com>
Tested-by: Impala Public Jenkins <impala-public-jenkins@cloudera.com>
This adds the java/impala-package Maven project to make it easier
to ship / test the Calcite planner. impala-package has a dependency
on impala-frontend and calcite-planner, so its classpath requires
no extra work when constructing the classpath.
An additional cleanup is that this no longer puts the
impala-frontend-*-tests.jar on the classpath by default. This requires
updating the query event hooks test, as it relies on that jar being
present.
This does not change the default value for the use_calcite_planner
query option, so there is no change in behavior.
Testing:
- Ran a core job
- Built docker images and OS packages locally
Change-Id: I81dec2a5b59e279229a735c8bb1a23c77111a793
Reviewed-on: http://gerrit.cloudera.org:8080/23497
Reviewed-by: Impala Public Jenkins <impala-public-jenkins@cloudera.com>
Tested-by: Impala Public Jenkins <impala-public-jenkins@cloudera.com>
