As part of the effort to make the PXF build and rpm OSS friendly which was initiated as part of https://issues.apache.org/jira/browse/HAWQ-1297 and documented here https://cwiki.apache.org/confluence/display/HAWQ/PXF+Build+and+Install, we wanted to make the usage of PXF rpms clear.
Below is the summary of the expected behavior as an outcome of the minor change in https://github.com/apache/incubator-hawq/pull/1208 1. The default outcome of make rpm will be to generate apache hadoop compliant PXF rpms. 2. If a user needs to generate PXF rpms that need to be compliant to a particular distribution say hdp, they will simply need to set HD=hdp prior to running make rpm For both scenarios the component level PXF rpms (pxf-service, pxf-hdfs, pxf-hive, etc), will have a dependancy on the generic hadoop/hive rpms (hadoop-hdfs, hadoop-common, hive,etc) with a minimum specified version as shown in https://github.com/apache/incubator-hawq/blob/master/pxf/gradle.properties. If the user uses a distribution which has different rpm names, they will need to make sure to have virtual rpms that refer to the custom hadoop rpm names. 3. Alternatively if you are not a believer of rpms, feel free to simply build and configure PXF using make install as shown in the PXF Build Install document. -- shivram mani
