[PHOENIX-5307] Fix HashJoinMoreIT.testBug2961 failing after PHOENIX-5262

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Resolved
• Affects Version/s: 4.15.0
• Fix Version/s: 4.15.0, 5.1.0
• Assignee: chenglei
• Reporter: chenglei
• Created: 30/May/19 03:13
• Updated: 21/Dec/19 00:56
• Resolved: 03/Jun/19 17:45

Description

I noticed HashJoinMoreIT.testBug2961 always failed after PHOENIX-5262, and the error stack is different from PHOENIX-5290:

    java.lang.AssertionError
	at org.apache.phoenix.end2end.join.HashJoinMoreIT.testBug2961(HashJoinMoreIT.java:908)

I think this problem is caused by following line 453 modified in PHOENIX-5262:

445                if ( !isFixedWidth && ( sepByte == QueryConstants.DESC_SEPARATOR_BYTE 
446                                    || ( !exclusiveUpper 
447                                         && (fieldIndex < schema.getMaxFields() || inclusiveUpper || exclusiveLower) ) ) ) {
448                key[offset++] = sepByte;
449                // Set lastInclusiveUpperSingleKey back to false if this is the last pk column
450                // as we don't want to increment the null byte in this case
451                // To test if this is the last pk column we need to consider the span of this slot
452                // and the field index to see if this slot considers the last column
453                lastInclusiveUpperSingleKey &= (fieldIndex + slotSpan[i]) < schema.getMaxFields()-1;
454           }

It did not consider the case that the last field is variable length and also DESC, when the last field is variable length and also DESC, the trailiing 0xFF is not removed when stored in HBASE, so for such case, we should not set lastInclusiveUpperSingleKey back to false.

Issue Links

• is caused by: PHOENIX-5262 Wrong Result on Salted table with some Variable Length PKs