[HAWQ-1228] Use profile based on file format in HCatalog integration(HiveRC, HiveText profiles)

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.1.0.0-incubating
• Component/s: PXF
• Assignee: Oleksandr Diachenko
• Reporter: Oleksandr Diachenko
• Created: 20/Dec/16 22:23
• Updated: 18/Feb/17 05:26
• Resolved: 31/Jan/17 20:21

Description

To leverage changes introduced in HAWQ-1177, expand optimization for other Hive profiles. Additional information needs to be included in user metadata(e.g. DELIMITER etc).

Changes needed:

• Enhance the Metadata API, to add new attributes: outputFormats, outputParameters;
• Hive, HiveORC profiles should support just GPDBWritable format;
• HIveText, HiveRC profiles should support both TEXT and GPDBWritable formats;
• Unify HiveUserData data structures to be same among all Hive- profiles;
• Bridge should read fragments using optimal profile read from fragment information;
• Optimal profile should be determined based on file's input format(org.apache.hadoop.hive.ql.io.orc.OrcInputFormat - HiveORC, org.apache.hadoop.hive.ql.io.RCFileInputFormat - HiveRC, org.apache.hadoop.mapred.TextInputFormat - HiveText);
• Default profile is Hive;
• If Hive table has org.apache.hadoop.mapred.TextInputFormat but also has some comlex types - Hive profile should be used(limitation should be addressed in HAWQ-1265);
• If table is homogeneous(all input file have the same output format) Bridge uses the same format which table has. Otherwise, if table is heterogeneous, GPDBWritable should be used;
• Add new property outputFormat to pxf-profiles-default.xml, which means default output format of profile.

Issue Links

• links to: GitHub Pull Request #1076