@Override
public Result getNextTuple() throws ExecException {
res = super.getNextTuple();
if (writer == null) { // In the case of combiner
return res;
}
try {
switch (res.returnStatus) {
case POStatus.STATUS_OK:
if (illustrator == null) {
Tuple result = (Tuple) res.result;
Byte index = (Byte) result.get(0);
PigNullableWritable key = HDataType.getWritableComparableTypes(result.get(1), keyType);
NullableTuple val = new NullableTuple((Tuple) result.get(2));
// Both the key and the value need the index. The key needs it so
// that it can be sorted on the index in addition to the key
// value. The value needs it so that POPackage can properly
// assign the tuple to its slot in the projection.
key.setIndex(index);
val.setIndex(index);
if (isSkewedJoin) {
// Wrap into a NullablePartitionWritable to match the key
// of the right table from POPartitionRearrangeTez for the skewed join
NullablePartitionWritable wrappedKey = new NullablePartitionWritable(key);
wrappedKey.setPartition(-1);
key = wrappedKey;
}
writer.write(key, val);
} else {
illustratorMarkup(res.result, res.result, 0);
}
res = RESULT_EMPTY;
break;
case POStatus.STATUS_EOP:
case POStatus.STATUS_ERR:
case POStatus.STATUS_NULL:
default:
break;
}
} catch (IOException ioe) {
int errCode = 2135;
String msg = "Received error from POLocalRearrage function." + ioe.getMessage();
throw new ExecException(msg, errCode, ioe);
}
return inp;
}
@Override
public void collect(Context oc, Tuple tuple) throws InterruptedException, IOException {
Byte index = (Byte) tuple.get(0);
PigNullableWritable key = HDataType.getWritableComparableTypes(tuple.get(1), keyType);
NullableTuple val = new NullableTuple((Tuple) tuple.get(2));
// Both the key and the value need the index. The key needs it so
// that it can be sorted on the index in addition to the key
// value. The value needs it so that POPackage can properly
// assign the tuple to its slot in the projection.
key.setIndex(index);
val.setIndex(index);
oc.write(key, val);
}
@Override
public void collect(Context oc, Tuple tuple) throws InterruptedException, IOException {
Byte tupleKeyIdx = 2;
Byte tupleValIdx = 3;
Byte index = (Byte) tuple.get(0);
Integer partitionIndex = -1;
// for partitioning table, the partition index isn't present
if (tuple.size() == 3) {
// super.collect(oc, tuple);
// return;
tupleKeyIdx--;
tupleValIdx--;
} else {
partitionIndex = (Integer) tuple.get(1);
}
PigNullableWritable key =
HDataType.getWritableComparableTypes(tuple.get(tupleKeyIdx), keyType);
NullablePartitionWritable wrappedKey = new NullablePartitionWritable(key);
NullableTuple val = new NullableTuple((Tuple) tuple.get(tupleValIdx));
// Both the key and the value need the index. The key needs it so
// that it can be sorted on the index in addition to the key
// value. The value needs it so that POPackage can properly
// assign the tuple to its slot in the projection.
wrappedKey.setIndex(index);
// set the partition
wrappedKey.setPartition(partitionIndex);
val.setIndex(index);
oc.write(wrappedKey, val);
}
private void runTest(Object key, boolean inner[], byte keyType)
throws ExecException, IOException {
Random r = new Random();
DataBag db1 = GenRandomData.genRandSmallTupDataBag(r, 10, 100);
DataBag db2 = GenRandomData.genRandSmallTupDataBag(r, 10, 100);
List<NullableTuple> db = new ArrayList<NullableTuple>(200);
Iterator<Tuple> db1Iter = db1.iterator();
if (!inner[0]) {
while (db1Iter.hasNext()) {
NullableTuple it = new NullableTuple(db1Iter.next());
it.setIndex((byte) 0);
db.add(it);
}
}
Iterator<Tuple> db2Iter = db2.iterator();
while (db2Iter.hasNext()) {
NullableTuple it = new NullableTuple(db2Iter.next());
it.setIndex((byte) 1);
db.add(it);
}
// ITIterator iti = new TestPackage.ITIterator(db.iterator());
POPackage pop = new POPackage(new OperatorKey("", r.nextLong()));
pop.setNumInps(2);
pop.getPkgr().setInner(inner);
PigNullableWritable k = HDataType.getWritableComparableTypes(key, keyType);
pop.attachInput(k, db.iterator());
if (keyType != DataType.BAG) {
// test serialization
NullablePartitionWritable wr;
if (keyType == DataType.TUPLE) {
BinSedesTuple tup =
(BinSedesTuple) binfactory.newTupleNoCopy(((Tuple) k.getValueAsPigType()).getAll());
wr = new NullablePartitionWritable(new NullableTuple(tup));
} else {
wr = new NullablePartitionWritable(k);
}
ByteArrayOutputStream baos = new ByteArrayOutputStream();
DataOutputStream out = new DataOutputStream(baos);
wr.write(out);
byte[] arr = baos.toByteArray();
ByteArrayInputStream bais = new ByteArrayInputStream(arr);
DataInputStream in = new DataInputStream(bais);
NullablePartitionWritable re = new NullablePartitionWritable();
re.readFields(in);
assertEquals(re, wr);
}
// we are not doing any optimization to remove
// parts of the "value" which are present in the "key" in this
// unit test - so set up the "keyInfo" accordingly in
// the POPackage
Map<Integer, Pair<Boolean, Map<Integer, Integer>>> keyInfo =
new HashMap<Integer, Pair<Boolean, Map<Integer, Integer>>>();
Pair<Boolean, Map<Integer, Integer>> p =
new Pair<Boolean, Map<Integer, Integer>>(false, new HashMap<Integer, Integer>());
keyInfo.put(0, p);
keyInfo.put(1, p);
pop.getPkgr().setKeyInfo(keyInfo);
Tuple t = null;
Result res = null;
res = pop.getNextTuple();
if (res.returnStatus == POStatus.STATUS_NULL && inner[0]) return;
assertEquals(POStatus.STATUS_OK, res.returnStatus);
t = (Tuple) res.result;
Object outKey = t.get(0);
DataBag outDb1 = (DataBag) t.get(1);
DataBag outDb2 = (DataBag) t.get(2);
assertEquals(key, outKey);
assertTrue(TestHelper.compareBags(db1, outDb1));
assertTrue(TestHelper.compareBags(db2, outDb2));
}
