[ORC-491] PPD: Column name lookups need to look a struct deeper for ACID

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.5.3
• Fix Version/s: 1.5.6, 1.6.0
• Assignee: Vineet Garg
• Reporter: Gopal Vijayaraghavan
• Created: 10/Apr/19 21:40
• Updated: 06/Apr/20 13:47
• Resolved: 12/Apr/19 23:56

Description

  public static int[] mapSargColumnsToOrcInternalColIdx(
                            List<PredicateLeaf> sargLeaves,
                            SchemaEvolution evolution) {
    int[] result = new int[sargLeaves.size()];
    Arrays.fill(result, -1);
    for(int i=0; i < result.length; ++i) {
      String colName = sargLeaves.get(i).getColumnName();
      result[i] = findColumns(evolution, colName);
    }
    return result;
  }

returns -1 for all data column names, because it was one level deeper.

sarg = [(IS_NULL date_time_date)]
colName = date_time_date
struct<operation:int,originalTransaction:bigint,bucket:int,rowId:bigint,currentTransaction:bigint,row:struct<date_time_date:date,term:string>>

Issue Links

• causes: HIVE-23143 Transactions: PPD in Delete deltas is broken
• depends upon: ORC-477 BloomFilter for ACID table does not get created
• links to: GitHub Pull Request #381
• links to: GitHub Pull Request #382
• links to: GitHub Pull Request #383