/**
* Check whether a join is valid. Invalid joins are CROSS JOIN, FULL OUTER JOIN, any join without
* criteria, any join with no equality criteria, and any outer join that has the outer side not
* the same as the dependent.
*
* @param joinNode The join node to check
* @param sourceNode The access node being considered
* @param analysisRecord TODO
* @return True if valid for making dependent
* @throws TeiidComponentException
* @throws QueryMetadataException
*/
boolean isValidJoin(PlanNode joinNode, PlanNode sourceNode, AnalysisRecord analysisRecord)
throws QueryMetadataException, TeiidComponentException {
JoinType jtype = (JoinType) joinNode.getProperty(NodeConstants.Info.JOIN_TYPE);
// Check that join is not a CROSS join or FULL OUTER join
if (jtype.equals(JoinType.JOIN_CROSS) || jtype.equals(JoinType.JOIN_FULL_OUTER)) {
sourceNode.recordDebugAnnotation(
"parent join is CROSS or FULL OUTER",
null,
"Rejecting dependent join",
analysisRecord,
null); //$NON-NLS-1$ //$NON-NLS-2$
return false;
}
if (!joinNode.getExportedCorrelatedReferences().isEmpty()) {
sourceNode.recordDebugAnnotation(
"parent join has a correlated nested table",
null,
"Rejecting dependent join",
analysisRecord,
null); //$NON-NLS-1$ //$NON-NLS-2$
return false;
}
// Check that join criteria exist
List jcrit = (List) joinNode.getProperty(NodeConstants.Info.JOIN_CRITERIA);
if (jcrit == null || jcrit.size() == 0) {
sourceNode.recordDebugAnnotation(
"parent join has has no join criteria",
null,
"Rejecting dependent join",
analysisRecord,
null); //$NON-NLS-1$ //$NON-NLS-2$
return false;
}
if (joinNode.getProperty(NodeConstants.Info.LEFT_EXPRESSIONS) == null) {
sourceNode.recordDebugAnnotation(
"parent join has no equa-join predicates",
null,
"Rejecting dependent join",
analysisRecord,
null); //$NON-NLS-1$ //$NON-NLS-2$
return false;
}
// Check that for a left or right outer join the dependent side must be the inner
if (jtype.isOuter() && JoinUtil.getInnerSideJoinNodes(joinNode)[0] != sourceNode) {
sourceNode.recordDebugAnnotation(
"node is on outer side of the join",
null,
"Rejecting dependent join",
analysisRecord,
null); //$NON-NLS-1$ //$NON-NLS-2$
return false;
}
return true;
}
@Override
public void processOp(Object row, int tag) throws HiveException {
try {
if (firstRow) {
// generate the map metadata
generateMapMetaData();
firstRow = false;
}
// get alias
alias = order[tag];
// alias = (byte)tag;
if ((lastAlias == null) || (!lastAlias.equals(alias))) {
nextSz = joinEmitInterval;
}
// compute keys and values as StandardObjects
AbstractMapJoinKey key =
JoinUtil.computeMapJoinKeys(
row, joinKeys.get(alias), joinKeysObjectInspectors.get(alias));
ArrayList<Object> value =
JoinUtil.computeValues(
row,
joinValues.get(alias),
joinValuesObjectInspectors.get(alias),
joinFilters.get(alias),
joinFilterObjectInspectors.get(alias),
noOuterJoin);
// Add the value to the ArrayList
storage.get((byte) tag).add(value);
for (Byte pos : order) {
if (pos.intValue() != tag) {
MapJoinObjectValue o = mapJoinTables.get(pos).get(key);
MapJoinRowContainer<ArrayList<Object>> rowContainer = rowContainerMap.get(pos);
// there is no join-value or join-key has all null elements
if (o == null || key.hasAnyNulls()) {
if (noOuterJoin) {
storage.put(pos, emptyList);
} else {
storage.put(pos, dummyObjVectors[pos.intValue()]);
}
} else {
rowContainer.reset(o.getObj());
storage.put(pos, rowContainer);
}
}
}
// generate the output records
checkAndGenObject();
// done with the row
storage.get((byte) tag).clear();
for (Byte pos : order) {
if (pos.intValue() != tag) {
storage.put(pos, null);
}
}
} catch (SerDeException e) {
e.printStackTrace();
throw new HiveException(e);
}
}
@Override
public void processOp(Object row, int tag) throws HiveException {
try {
reportProgress();
// get alias
alias = (byte) tag;
if ((lastAlias == null) || (!lastAlias.equals(alias))) {
nextSz = joinEmitInterval;
}
ArrayList<Object> nr =
JoinUtil.computeValues(
row,
joinValues.get(alias),
joinValuesObjectInspectors.get(alias),
joinFilters.get(alias),
joinFilterObjectInspectors.get(alias),
noOuterJoin);
if (handleSkewJoin) {
skewJoinKeyContext.handleSkew(tag);
}
// number of rows for the key in the given table
int sz = storage.get(alias).size();
StructObjectInspector soi = (StructObjectInspector) inputObjInspectors[tag];
StructField sf = soi.getStructFieldRef(Utilities.ReduceField.KEY.toString());
Object keyObject = soi.getStructFieldData(row, sf);
// Are we consuming too much memory
if (alias == numAliases - 1 && !(handleSkewJoin && skewJoinKeyContext.currBigKeyTag >= 0)) {
if (sz == joinEmitInterval) {
// The input is sorted by alias, so if we are already in the last join
// operand,
// we can emit some results now.
// Note this has to be done before adding the current row to the
// storage,
// to preserve the correctness for outer joins.
checkAndGenObject();
storage.get(alias).clear();
}
} else {
if (sz == nextSz) {
// Output a warning if we reached at least 1000 rows for a join
// operand
// We won't output a warning for the last join operand since the size
// will never goes to joinEmitInterval.
LOG.warn("table " + alias + " has " + sz + " rows for join key " + keyObject);
nextSz = getNextSize(nextSz);
}
}
// Add the value to the vector
storage.get(alias).add(nr);
// if join-key is null, process each row in different group.
if (SerDeUtils.hasAnyNullObject(keyObject, sf.getFieldObjectInspector())) {
endGroup();
startGroup();
}
} catch (Exception e) {
e.printStackTrace();
throw new HiveException(e);
}
}
public void testSimple() throws Exception {
final String idField = "id";
final String toField = "productId";
Directory dir = newDirectory();
RandomIndexWriter w =
new RandomIndexWriter(
random(),
dir,
newIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer(random()))
.setMergePolicy(newLogMergePolicy()));
// 0
Document doc = new Document();
doc.add(new TextField("description", "random text", Field.Store.NO));
doc.add(new TextField("name", "name1", Field.Store.NO));
doc.add(new TextField(idField, "1", Field.Store.NO));
w.addDocument(doc);
// 1
doc = new Document();
doc.add(new TextField("price", "10.0", Field.Store.NO));
doc.add(new TextField(idField, "2", Field.Store.NO));
doc.add(new TextField(toField, "1", Field.Store.NO));
w.addDocument(doc);
// 2
doc = new Document();
doc.add(new TextField("price", "20.0", Field.Store.NO));
doc.add(new TextField(idField, "3", Field.Store.NO));
doc.add(new TextField(toField, "1", Field.Store.NO));
w.addDocument(doc);
// 3
doc = new Document();
doc.add(new TextField("description", "more random text", Field.Store.NO));
doc.add(new TextField("name", "name2", Field.Store.NO));
doc.add(new TextField(idField, "4", Field.Store.NO));
w.addDocument(doc);
w.commit();
// 4
doc = new Document();
doc.add(new TextField("price", "10.0", Field.Store.NO));
doc.add(new TextField(idField, "5", Field.Store.NO));
doc.add(new TextField(toField, "4", Field.Store.NO));
w.addDocument(doc);
// 5
doc = new Document();
doc.add(new TextField("price", "20.0", Field.Store.NO));
doc.add(new TextField(idField, "6", Field.Store.NO));
doc.add(new TextField(toField, "4", Field.Store.NO));
w.addDocument(doc);
IndexSearcher indexSearcher = new IndexSearcher(w.getReader());
w.close();
// Search for product
Query joinQuery =
JoinUtil.createJoinQuery(
idField,
false,
toField,
new TermQuery(new Term("name", "name2")),
indexSearcher,
ScoreMode.None);
TopDocs result = indexSearcher.search(joinQuery, 10);
assertEquals(2, result.totalHits);
assertEquals(4, result.scoreDocs[0].doc);
assertEquals(5, result.scoreDocs[1].doc);
joinQuery =
JoinUtil.createJoinQuery(
idField,
false,
toField,
new TermQuery(new Term("name", "name1")),
indexSearcher,
ScoreMode.None);
result = indexSearcher.search(joinQuery, 10);
assertEquals(2, result.totalHits);
assertEquals(1, result.scoreDocs[0].doc);
assertEquals(2, result.scoreDocs[1].doc);
// Search for offer
joinQuery =
JoinUtil.createJoinQuery(
toField,
false,
idField,
new TermQuery(new Term("id", "5")),
indexSearcher,
ScoreMode.None);
result = indexSearcher.search(joinQuery, 10);
assertEquals(1, result.totalHits);
assertEquals(3, result.scoreDocs[0].doc);
indexSearcher.getIndexReader().close();
dir.close();
}
private void executeRandomJoin(
boolean multipleValuesPerDocument,
int maxIndexIter,
int maxSearchIter,
int numberOfDocumentsToIndex)
throws Exception {
for (int indexIter = 1; indexIter <= maxIndexIter; indexIter++) {
if (VERBOSE) {
System.out.println("indexIter=" + indexIter);
}
Directory dir = newDirectory();
RandomIndexWriter w =
new RandomIndexWriter(
random(),
dir,
newIndexWriterConfig(
TEST_VERSION_CURRENT,
new MockAnalyzer(random(), MockTokenizer.KEYWORD, false))
.setMergePolicy(newLogMergePolicy()));
final boolean scoreDocsInOrder = TestJoinUtil.random().nextBoolean();
IndexIterationContext context =
createContext(numberOfDocumentsToIndex, w, multipleValuesPerDocument, scoreDocsInOrder);
IndexReader topLevelReader = w.getReader();
w.close();
for (int searchIter = 1; searchIter <= maxSearchIter; searchIter++) {
if (VERBOSE) {
System.out.println("searchIter=" + searchIter);
}
IndexSearcher indexSearcher = newSearcher(topLevelReader);
int r = random().nextInt(context.randomUniqueValues.length);
boolean from = context.randomFrom[r];
String randomValue = context.randomUniqueValues[r];
FixedBitSet expectedResult =
createExpectedResult(randomValue, from, indexSearcher.getIndexReader(), context);
final Query actualQuery = new TermQuery(new Term("value", randomValue));
if (VERBOSE) {
System.out.println("actualQuery=" + actualQuery);
}
final ScoreMode scoreMode = ScoreMode.values()[random().nextInt(ScoreMode.values().length)];
if (VERBOSE) {
System.out.println("scoreMode=" + scoreMode);
}
final Query joinQuery;
if (from) {
joinQuery =
JoinUtil.createJoinQuery(
"from", multipleValuesPerDocument, "to", actualQuery, indexSearcher, scoreMode);
} else {
joinQuery =
JoinUtil.createJoinQuery(
"to", multipleValuesPerDocument, "from", actualQuery, indexSearcher, scoreMode);
}
if (VERBOSE) {
System.out.println("joinQuery=" + joinQuery);
}
// Need to know all documents that have matches. TopDocs doesn't give me that and then I'd
// be also testing TopDocsCollector...
final FixedBitSet actualResult = new FixedBitSet(indexSearcher.getIndexReader().maxDoc());
final TopScoreDocCollector topScoreDocCollector = TopScoreDocCollector.create(10, false);
indexSearcher.search(
joinQuery,
new Collector() {
int docBase;
@Override
public void collect(int doc) throws IOException {
actualResult.set(doc + docBase);
topScoreDocCollector.collect(doc);
}
@Override
public void setNextReader(AtomicReaderContext context) {
docBase = context.docBase;
topScoreDocCollector.setNextReader(context);
}
@Override
public void setScorer(Scorer scorer) throws IOException {
topScoreDocCollector.setScorer(scorer);
}
@Override
public boolean acceptsDocsOutOfOrder() {
return scoreDocsInOrder;
}
});
// Asserting bit set...
if (VERBOSE) {
System.out.println("expected cardinality:" + expectedResult.cardinality());
DocIdSetIterator iterator = expectedResult.iterator();
for (int doc = iterator.nextDoc();
doc != DocIdSetIterator.NO_MORE_DOCS;
doc = iterator.nextDoc()) {
System.out.println(
String.format(
Locale.ROOT,
"Expected doc[%d] with id value %s",
doc,
indexSearcher.doc(doc).get("id")));
}
System.out.println("actual cardinality:" + actualResult.cardinality());
iterator = actualResult.iterator();
for (int doc = iterator.nextDoc();
doc != DocIdSetIterator.NO_MORE_DOCS;
doc = iterator.nextDoc()) {
System.out.println(
String.format(
Locale.ROOT,
"Actual doc[%d] with id value %s",
doc,
indexSearcher.doc(doc).get("id")));
}
}
assertEquals(expectedResult, actualResult);
// Asserting TopDocs...
TopDocs expectedTopDocs = createExpectedTopDocs(randomValue, from, scoreMode, context);
TopDocs actualTopDocs = topScoreDocCollector.topDocs();
assertEquals(expectedTopDocs.totalHits, actualTopDocs.totalHits);
assertEquals(expectedTopDocs.scoreDocs.length, actualTopDocs.scoreDocs.length);
if (scoreMode == ScoreMode.None) {
continue;
}
assertEquals(expectedTopDocs.getMaxScore(), actualTopDocs.getMaxScore(), 0.0f);
for (int i = 0; i < expectedTopDocs.scoreDocs.length; i++) {
if (VERBOSE) {
System.out.printf(
Locale.ENGLISH,
"Expected doc: %d | Actual doc: %d\n",
expectedTopDocs.scoreDocs[i].doc,
actualTopDocs.scoreDocs[i].doc);
System.out.printf(
Locale.ENGLISH,
"Expected score: %f | Actual score: %f\n",
expectedTopDocs.scoreDocs[i].score,
actualTopDocs.scoreDocs[i].score);
}
assertEquals(expectedTopDocs.scoreDocs[i].doc, actualTopDocs.scoreDocs[i].doc);
assertEquals(expectedTopDocs.scoreDocs[i].score, actualTopDocs.scoreDocs[i].score, 0.0f);
Explanation explanation =
indexSearcher.explain(joinQuery, expectedTopDocs.scoreDocs[i].doc);
assertEquals(expectedTopDocs.scoreDocs[i].score, explanation.getValue(), 0.0f);
}
}
topLevelReader.close();
dir.close();
}
}
public void testSimpleWithScoring() throws Exception {
final String idField = "id";
final String toField = "movieId";
Directory dir = newDirectory();
RandomIndexWriter w =
new RandomIndexWriter(
random(),
dir,
newIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer(random()))
.setMergePolicy(newLogMergePolicy()));
// 0
Document doc = new Document();
doc.add(new TextField("description", "A random movie", Field.Store.NO));
doc.add(new TextField("name", "Movie 1", Field.Store.NO));
doc.add(new TextField(idField, "1", Field.Store.NO));
w.addDocument(doc);
// 1
doc = new Document();
doc.add(new TextField("subtitle", "The first subtitle of this movie", Field.Store.NO));
doc.add(new TextField(idField, "2", Field.Store.NO));
doc.add(new TextField(toField, "1", Field.Store.NO));
w.addDocument(doc);
// 2
doc = new Document();
doc.add(new TextField("subtitle", "random subtitle; random event movie", Field.Store.NO));
doc.add(new TextField(idField, "3", Field.Store.NO));
doc.add(new TextField(toField, "1", Field.Store.NO));
w.addDocument(doc);
// 3
doc = new Document();
doc.add(new TextField("description", "A second random movie", Field.Store.NO));
doc.add(new TextField("name", "Movie 2", Field.Store.NO));
doc.add(new TextField(idField, "4", Field.Store.NO));
w.addDocument(doc);
w.commit();
// 4
doc = new Document();
doc.add(
new TextField(
"subtitle", "a very random event happened during christmas night", Field.Store.NO));
doc.add(new TextField(idField, "5", Field.Store.NO));
doc.add(new TextField(toField, "4", Field.Store.NO));
w.addDocument(doc);
// 5
doc = new Document();
doc.add(new TextField("subtitle", "movie end movie test 123 test 123 random", Field.Store.NO));
doc.add(new TextField(idField, "6", Field.Store.NO));
doc.add(new TextField(toField, "4", Field.Store.NO));
w.addDocument(doc);
IndexSearcher indexSearcher = new IndexSearcher(w.getReader());
w.close();
// Search for movie via subtitle
Query joinQuery =
JoinUtil.createJoinQuery(
toField,
false,
idField,
new TermQuery(new Term("subtitle", "random")),
indexSearcher,
ScoreMode.Max);
TopDocs result = indexSearcher.search(joinQuery, 10);
assertEquals(2, result.totalHits);
assertEquals(0, result.scoreDocs[0].doc);
assertEquals(3, result.scoreDocs[1].doc);
// Score mode max.
joinQuery =
JoinUtil.createJoinQuery(
toField,
false,
idField,
new TermQuery(new Term("subtitle", "movie")),
indexSearcher,
ScoreMode.Max);
result = indexSearcher.search(joinQuery, 10);
assertEquals(2, result.totalHits);
assertEquals(3, result.scoreDocs[0].doc);
assertEquals(0, result.scoreDocs[1].doc);
// Score mode total
joinQuery =
JoinUtil.createJoinQuery(
toField,
false,
idField,
new TermQuery(new Term("subtitle", "movie")),
indexSearcher,
ScoreMode.Total);
result = indexSearcher.search(joinQuery, 10);
assertEquals(2, result.totalHits);
assertEquals(0, result.scoreDocs[0].doc);
assertEquals(3, result.scoreDocs[1].doc);
// Score mode avg
joinQuery =
JoinUtil.createJoinQuery(
toField,
false,
idField,
new TermQuery(new Term("subtitle", "movie")),
indexSearcher,
ScoreMode.Avg);
result = indexSearcher.search(joinQuery, 10);
assertEquals(2, result.totalHits);
assertEquals(3, result.scoreDocs[0].doc);
assertEquals(0, result.scoreDocs[1].doc);
indexSearcher.getIndexReader().close();
dir.close();
}
/**
* Mark the specified access node to be made dependent
*
* @param sourceNode Node to make dependent
* @param dca
* @throws TeiidComponentException
* @throws QueryMetadataException
*/
boolean markDependent(
PlanNode sourceNode,
PlanNode joinNode,
QueryMetadataInterface metadata,
DependentCostAnalysis dca,
Boolean bound)
throws QueryMetadataException, TeiidComponentException {
boolean isLeft = joinNode.getFirstChild() == sourceNode;
// Get new access join node properties based on join criteria
List independentExpressions =
(List)
(isLeft
? joinNode.getProperty(NodeConstants.Info.RIGHT_EXPRESSIONS)
: joinNode.getProperty(NodeConstants.Info.LEFT_EXPRESSIONS));
List dependentExpressions =
(List)
(isLeft
? joinNode.getProperty(NodeConstants.Info.LEFT_EXPRESSIONS)
: joinNode.getProperty(NodeConstants.Info.RIGHT_EXPRESSIONS));
if (independentExpressions == null || independentExpressions.isEmpty()) {
return false;
}
String id = "$dsc/id" + ID.getAndIncrement(); // $NON-NLS-1$
// Create DependentValueSource and set on the independent side as this will feed the values
joinNode.setProperty(NodeConstants.Info.DEPENDENT_VALUE_SOURCE, id);
PlanNode indNode = isLeft ? joinNode.getLastChild() : joinNode.getFirstChild();
if (bound == null) {
List<PlanNode> sources = NodeEditor.findAllNodes(indNode, NodeConstants.Types.SOURCE);
for (PlanNode planNode : sources) {
for (GroupSymbol gs : planNode.getGroups()) {
if (gs.isTempTable()
&& metadata.getCardinality(gs.getMetadataID())
== QueryMetadataInterface.UNKNOWN_CARDINALITY) {
bound = true;
break;
}
}
}
if (bound == null) {
bound = false;
}
}
PlanNode crit =
getDependentCriteriaNode(
id,
independentExpressions,
dependentExpressions,
indNode,
metadata,
dca,
bound,
(MakeDep) sourceNode.getProperty(Info.MAKE_DEP));
sourceNode.addAsParent(crit);
if (isLeft) {
JoinUtil.swapJoinChildren(joinNode);
}
return true;
}
