[CRUNCH-627] Shard API doesn't work well with parquet target

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 0.13.0
• Fix Version/s: 0.13.0
• Component/s: MapReduce Patterns
• Labels: patch
• Environment: Linux X86
• Flags: Important
• Assignee: Unassigned
• Reporter: Tony Wu
• Created: 18/Nov/16 02:05
• Updated: 30/Nov/16 07:42

Description

PCollection<User> outTable = oldTable.union(newTable);
Shard.shard(outTable,10).write(new AvroParquetFileTarget(tempOut+path), Target.WriteMode.OVERWRITE);

However, I have another job which would read the output of above target output and use a field as the key , the job output looks like below
3.0.3.1.2.CH24_RELEASE 2
3.0.3.1.2.CH24_RELEASEE 1
3.0.3.1.2.CH24_RELEASEEA 1
3.0.3.1.2.CH24_RELEASEEAS 1
3.0.3.1.2.CH24_RELEASEEASE 29
3.0.3.1.2.CH24_RELEASEEASES 160
3.0.3.1.2.CH24_RELEASEEASESE 85
3.0.3.1.2.CH24_RELEASEEASESEE 14
3.0.3.1.2.CH24_RELEASEEASESEEE 4
3.0.3.1.2.CH24_RELEASEEASESEEES 1
there is extra suffix added to the key of the PTable, all of them
should be RELEASE but not the RELEASEEASE bra bra

If I remove the Shard, and keeps all the same, the output looks like normal
3.0.0.1.2.CH.1.4_RELEASE 1
3.0.1.1.2.CH22_RELEASE 1622
3.0.1.1.2.CH23_RELEASE 10607
3.0.14.1.2.CH.1.3_RELEASE 18080
3.0.19.1.2.TC21_RELEASE 5
3.0.2.1.2.CH11_RELEASE 3
3.0.2.1.2.TC21_RELEASE 4
3.0.20.1.2.TC21_RELEASE 247
3.0.20.7.2.SX.1.2A_RELEASE 2
3.0.20.8.2.SX.1.3A_RELEASE 1