[KUDU-2494] Support OR predicate pushdown

• Type: Sub-task
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 1.7.1
• Component/s: api
• Labels: roadmap-candidate
• Assignee: ZhangYao
• Reporter: Ricardo Gaspar
• Created: 10/Jul/18 12:27
• Updated: 18/Feb/21 17:47

Description

KuduScannerBuild API only supports the build of a scan using predicates in a conjuctive operation.
Meaning that when using more than one predicate (p1,p2), the scan will be done using a statement similar to:

WHERE p1 AND p2;

 

In Java code:
 

kc = new KuduClient.KuduClientBuilder(kuduMasters).build();
session = kc.newSession();
KuduScanner ks = kc.newScannerBuilder(table).addPredicate(p1).addPredicate(p2).build();

  

There is no possibility to specify a OR operation that applies an disjunction between all/some predicates. E.g.

WHERE p1 OR p2;

 

 

The only way to do this using the current API (1.7.1) is to use two (or more) scanners and iterate them separately:

 

KuduScanner scanner1 = kc.newScannerBuilder(table).addPredicate(p1).build();
KuduScanner scanner2 = kc.newScannerBuilder(table).addPredicate(p2).build();

while (scanner1.hasMoreRows()) {
    RowResultIterator results = scanner1.nextRows();
    while (results.hasNext()) {
        RowResult rowData = results.next();
        
        //Do stuff with rowData

    }
}

while (scanner2.hasMoreRows()) {
    RowResultIterator results = scanner2.nextRows();
    while (results.hasNext()) {
        RowResult rowData = results.next();
        //Do stuff with rowData

    }
}

 

Kudu API should provide "almost" the same query processing power as IMPALA or an SQL engine.
Programmers should rely on Kudu API to operations like these and others like: sorting, table joins, etc.

Issue Links

• is duplicated by: KUDU-2863 Support OR predicates
• links to: WIP Design Doc