[CASSANDRA-9381] AbstractColumnFamilyInputFormat is incorrectly interpreting token values when building ColumnFamilySplit

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 2.2.0 beta 1
• Severity: Normal
• Assignee: Philip Thompson
• Reporter: Mike Adamson

Authors

Philip Thompson

Reviewers

Mike Adamson

• Created: 14/May/15 12:40
• Updated: 16/Apr/19 09:31
• Resolved: 15/May/15 11:27

Description

The SplitCallable.call method in AbstractColumnFamilyInputFormat has the following code for building ColumnFamilySplits:

            for (TokenRange subSplit : subSplits.keySet())
            {
                List<TokenRange> ranges = subSplit.unwrap();
                for (TokenRange subrange : ranges)
                {
                    ColumnFamilySplit split =
                            new ColumnFamilySplit(
                                    subrange.getStart().toString().substring(2),
                                    subrange.getEnd().toString().substring(2),
                                    subSplits.get(subSplit),
                                    endpoints);

                    logger.debug("adding {}", split);
                    splits.add(split);
                }
            }

The subrange.getStart().toString().substring(2) works for the OrderPreservingPartitioner because the toString for the OPPToken in the java driver returns a hex value preceded by 0x. It doesn't work for the Murmur3Partitioner because the toString for the M3PToken returns Long.toString.

As a result the tokens lose their first 2 digits.