For the corruption test, try corrupting the records, not the sync marker. The features added to DataFileReader for corruption recovery were for the case when decoding a record fails (corrupted record), not for when a sync marker is corrupted. Perhaps we should add that too, but it does not surprise me that that case has a bug.
On 1/6/13 7:38 PM, "Russell Jurney" <russell.jur...@gmail.com> wrote: >We are trying to recover, report bad record, and move to the next record >of an Avro file in PIG-3015 and PIG-3059. It seems that sync blocks don't >exist between files, however. > >How should we recover from a bad record using Avro's DataFileReader? > >https://issues.apache.org/jira/browse/PIG-3015 >https://issues.apache.org/jira/browse/PIG-3059 > >Russell Jurney http://datasyndrome.com