[NUTCH-814] SegmentMerger bug

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.1
• Fix Version/s: 1.1
• Assignee: Andrzej Bialecki
• Reporter: Dennis Kubes
• Created: 27/Apr/10 10:45
• Updated: 27/Apr/10 15:24
• Resolved: 27/Apr/10 15:24

Description

Dennis reported:

In the SegmentMerger.java file about line 150 we have this:

final SequenceFile.Reader reader =
new SequenceFile.Reader(FileSystem.get(job), fSplit.getPath(),
job);

Then about line 166 in the record reader we have this:

boolean res = reader.next(key, w);

If I am reading that right, that would mean that the map tap would loop
over all records for a given file and not just a given split.

Right, this should instead use SequenceFileRecordReader that already has the logic to handle splits. Patch coming shortly - thanks for spotting this! This could be the reason for "out of disk space" errors that many users reported.