use of org.apache.hadoop.hive.serde2.ByteStream.Output in project hive by apache.
the class MapJoinKey method serializeVector.
/**
* Serializes row to output for vectorized path.
* @param byteStream Output to reuse. Can be null, in that case a new one would be created.
*/
public static Output serializeVector(Output byteStream, VectorHashKeyWrapper kw, VectorExpressionWriter[] keyOutputWriters, VectorHashKeyWrapperBatch keyWrapperBatch, boolean[] nulls, boolean[] sortableSortOrders, byte[] nullMarkers, byte[] notNullMarkers) throws HiveException, SerDeException {
Object[] fieldData = new Object[keyOutputWriters.length];
List<ObjectInspector> fieldOis = new ArrayList<ObjectInspector>();
for (int i = 0; i < keyOutputWriters.length; ++i) {
VectorExpressionWriter writer = keyOutputWriters[i];
fieldOis.add(writer.getObjectInspector());
// This is rather convoluted... to simplify for perf, we could call getRawKeyValue
// instead of writable, and serialize based on Java type as opposed to OI.
fieldData[i] = keyWrapperBatch.getWritableKeyValue(kw, i, writer);
if (nulls != null) {
nulls[i] = (fieldData[i] == null);
}
}
return serializeRow(byteStream, fieldData, fieldOis, sortableSortOrders, nullMarkers, notNullMarkers);
}
use of org.apache.hadoop.hive.serde2.ByteStream.Output in project hive by apache.
the class TestVectorMapJoinFastRowHashMap method addAndVerifyRows.
private void addAndVerifyRows(VectorRandomRowSource valueSource, Object[][] rows, VectorMapJoinFastHashTable map, HashTableKeyType hashTableKeyType, VerifyFastRowHashMap verifyTable, String[] keyTypeNames, boolean doClipping, boolean useExactBytes) throws HiveException, IOException, SerDeException {
final int keyCount = keyTypeNames.length;
PrimitiveTypeInfo[] keyPrimitiveTypeInfos = new PrimitiveTypeInfo[keyCount];
PrimitiveCategory[] keyPrimitiveCategories = new PrimitiveCategory[keyCount];
ArrayList<ObjectInspector> keyPrimitiveObjectInspectorList = new ArrayList<ObjectInspector>(keyCount);
for (int i = 0; i < keyCount; i++) {
PrimitiveTypeInfo primitiveTypeInfo = (PrimitiveTypeInfo) TypeInfoUtils.getTypeInfoFromTypeString(keyTypeNames[i]);
keyPrimitiveTypeInfos[i] = primitiveTypeInfo;
PrimitiveCategory primitiveCategory = primitiveTypeInfo.getPrimitiveCategory();
keyPrimitiveCategories[i] = primitiveCategory;
keyPrimitiveObjectInspectorList.add(PrimitiveObjectInspectorFactory.getPrimitiveWritableObjectInspector(primitiveTypeInfo));
}
boolean[] keyColumnSortOrderIsDesc = new boolean[keyCount];
Arrays.fill(keyColumnSortOrderIsDesc, false);
byte[] keyColumnNullMarker = new byte[keyCount];
Arrays.fill(keyColumnNullMarker, BinarySortableSerDe.ZERO);
byte[] keyColumnNotNullMarker = new byte[keyCount];
Arrays.fill(keyColumnNotNullMarker, BinarySortableSerDe.ONE);
BinarySortableSerializeWrite keySerializeWrite = new BinarySortableSerializeWrite(keyColumnSortOrderIsDesc, keyColumnNullMarker, keyColumnNotNullMarker);
TypeInfo[] valueTypeInfos = valueSource.typeInfos();
final int columnCount = valueTypeInfos.length;
SerializeWrite valueSerializeWrite = new LazyBinarySerializeWrite(columnCount);
final int count = rows.length;
for (int i = 0; i < count; i++) {
Object[] valueRow = rows[i];
Output valueOutput = new Output();
((LazyBinarySerializeWrite) valueSerializeWrite).set(valueOutput);
for (int index = 0; index < columnCount; index++) {
VerifyFastRow.serializeWrite(valueSerializeWrite, valueTypeInfos[index], valueRow[index]);
}
byte[] value = Arrays.copyOf(valueOutput.getData(), valueOutput.getLength());
// Add a new key or add a value to an existing key?
byte[] key;
if (random.nextBoolean() || verifyTable.getCount() == 0) {
Object[] keyRow = VectorRandomRowSource.randomWritablePrimitiveRow(keyCount, random, keyPrimitiveTypeInfos);
Output keyOutput = new Output();
keySerializeWrite.set(keyOutput);
for (int index = 0; index < keyCount; index++) {
VerifyFastRow.serializeWrite(keySerializeWrite, keyPrimitiveTypeInfos[index], keyRow[index]);
}
key = Arrays.copyOf(keyOutput.getData(), keyOutput.getLength());
verifyTable.add(key, keyRow, value, valueRow);
} else {
key = verifyTable.addRandomExisting(value, valueRow, random);
}
// Serialize keyRow into key bytes.
BytesWritable keyWritable = new BytesWritable(key);
BytesWritable valueWritable = new BytesWritable(value);
map.putRow(keyWritable, valueWritable);
// verifyTable.verify(map);
}
verifyTable.verify(map, hashTableKeyType, valueTypeInfos, doClipping, useExactBytes, random);
}
use of org.apache.hadoop.hive.serde2.ByteStream.Output in project SQLWindowing by hbutani.
the class MRUtils method initialize.
/**
* Construct the data structures containing ExprNodeDesc for partition
* columns and order columns. Use the input definition to construct the list
* of output columns for the ReduceSinkOperator
*
* @throws WindowingException
*/
public void initialize() throws WindowingException {
TableFuncDef tabDef = RuntimeUtils.getFirstTableFunction(qdef);
hiveTableDef = tabDef.getHiveTableDef();
InputInfo inputInfo;
ArrayList<ColumnDef> partColList = tabDef.getWindow().getPartDef().getColumns();
TableFunctionEvaluator tEval = tabDef.getFunction();
/*
* If the query has a map phase, the inputInfo is retrieved from the map
* output info of the table function definition. This is constructed
* using the map output oi of the table function definition. If the
* query does not have a map phase, the inputInfo is retrieved from the
* QueryInputDef (either HiveTableDef or HiveQueryDef) of the query.
*/
if (tEval.isTransformsRawInput()) {
inputInfo = qdef.getTranslationInfo().getMapInputInfo(tabDef);
} else {
inputInfo = qdef.getTranslationInfo().getInputInfo(hiveTableDef);
}
for (ColumnDef colDef : partColList) {
partCols.add(colDef.getExprNode());
}
ArrayList<OrderColumnDef> orderColList = tabDef.getWindow().getOrderDef().getColumns();
for (OrderColumnDef colDef : orderColList) {
Order order = colDef.getOrder();
if (order.name().equals("ASC")) {
orderString.append('+');
} else {
orderString.append('-');
}
orderCols.add(colDef.getExprNode());
outputColumnNames.add(colDef.getAlias());
}
RowResolver rr = inputInfo.getRowResolver();
ArrayList<ColumnInfo> colInfoList = rr.getColumnInfos();
for (ColumnInfo colInfo : colInfoList) {
String internalName = colInfo.getInternalName();
TypeInfo type = colInfo.getType();
valueCols.add(TranslateUtils.getExprDesc(internalName, type));
outputColumnNames.add(internalName);
}
}
use of org.apache.hadoop.hive.serde2.ByteStream.Output in project hive by apache.
the class DDLSemanticAnalyzer method getFullPartitionSpecs.
/**
* Get the partition specs from the tree. This stores the full specification
* with the comparator operator into the output list.
*
* @param ast Tree to extract partitions from.
* @param tab Table.
* @return Map of partitions by prefix length. Most of the time prefix length will
* be the same for all partition specs, so we can just OR the expressions.
*/
private Map<Integer, List<ExprNodeGenericFuncDesc>> getFullPartitionSpecs(CommonTree ast, Table tab, boolean canGroupExprs) throws SemanticException {
String defaultPartitionName = HiveConf.getVar(conf, HiveConf.ConfVars.DEFAULTPARTITIONNAME);
Map<String, String> colTypes = new HashMap<String, String>();
for (FieldSchema fs : tab.getPartitionKeys()) {
colTypes.put(fs.getName().toLowerCase(), fs.getType());
}
Map<Integer, List<ExprNodeGenericFuncDesc>> result = new HashMap<Integer, List<ExprNodeGenericFuncDesc>>();
for (int childIndex = 0; childIndex < ast.getChildCount(); childIndex++) {
Tree partSpecTree = ast.getChild(childIndex);
if (partSpecTree.getType() != HiveParser.TOK_PARTSPEC) {
continue;
}
ExprNodeGenericFuncDesc expr = null;
HashSet<String> names = new HashSet<String>(partSpecTree.getChildCount());
for (int i = 0; i < partSpecTree.getChildCount(); ++i) {
CommonTree partSpecSingleKey = (CommonTree) partSpecTree.getChild(i);
assert (partSpecSingleKey.getType() == HiveParser.TOK_PARTVAL);
String key = stripIdentifierQuotes(partSpecSingleKey.getChild(0).getText()).toLowerCase();
String operator = partSpecSingleKey.getChild(1).getText();
ASTNode partValNode = (ASTNode) partSpecSingleKey.getChild(2);
TypeCheckCtx typeCheckCtx = new TypeCheckCtx(null);
ExprNodeConstantDesc valExpr = (ExprNodeConstantDesc) TypeCheckProcFactory.genExprNode(partValNode, typeCheckCtx).get(partValNode);
Object val = valExpr.getValue();
boolean isDefaultPartitionName = val.equals(defaultPartitionName);
String type = colTypes.get(key);
PrimitiveTypeInfo pti = TypeInfoFactory.getPrimitiveTypeInfo(type);
if (type == null) {
throw new SemanticException("Column " + key + " not found");
}
// Create the corresponding hive expression to filter on partition columns.
if (!isDefaultPartitionName) {
if (!valExpr.getTypeString().equals(type)) {
Converter converter = ObjectInspectorConverters.getConverter(TypeInfoUtils.getStandardJavaObjectInspectorFromTypeInfo(valExpr.getTypeInfo()), TypeInfoUtils.getStandardJavaObjectInspectorFromTypeInfo(pti));
val = converter.convert(valExpr.getValue());
}
}
ExprNodeColumnDesc column = new ExprNodeColumnDesc(pti, key, null, true);
ExprNodeGenericFuncDesc op;
if (!isDefaultPartitionName) {
op = makeBinaryPredicate(operator, column, new ExprNodeConstantDesc(pti, val));
} else {
GenericUDF originalOp = FunctionRegistry.getFunctionInfo(operator).getGenericUDF();
String fnName;
if (FunctionRegistry.isEq(originalOp)) {
fnName = "isnull";
} else if (FunctionRegistry.isNeq(originalOp)) {
fnName = "isnotnull";
} else {
throw new SemanticException("Cannot use " + operator + " in a default partition spec; only '=' and '!=' are allowed.");
}
op = makeUnaryPredicate(fnName, column);
}
// If it's multi-expr filter (e.g. a='5', b='2012-01-02'), AND with previous exprs.
expr = (expr == null) ? op : makeBinaryPredicate("and", expr, op);
names.add(key);
}
if (expr == null) {
continue;
}
// We got the expr for one full partition spec. Determine the prefix length.
int prefixLength = calculatePartPrefix(tab, names);
List<ExprNodeGenericFuncDesc> orExpr = result.get(prefixLength);
// If we don't, create a new separate filter. In most cases there will only be one.
if (orExpr == null) {
result.put(prefixLength, Lists.newArrayList(expr));
} else if (canGroupExprs) {
orExpr.set(0, makeBinaryPredicate("or", expr, orExpr.get(0)));
} else {
orExpr.add(expr);
}
}
return result;
}
use of org.apache.hadoop.hive.serde2.ByteStream.Output in project hive by apache.
the class TaskCompiler method compile.
@SuppressWarnings({ "nls", "unchecked" })
public void compile(final ParseContext pCtx, final List<Task<? extends Serializable>> rootTasks, final HashSet<ReadEntity> inputs, final HashSet<WriteEntity> outputs) throws SemanticException {
Context ctx = pCtx.getContext();
GlobalLimitCtx globalLimitCtx = pCtx.getGlobalLimitCtx();
List<Task<MoveWork>> mvTask = new ArrayList<>();
List<LoadTableDesc> loadTableWork = pCtx.getLoadTableWork();
List<LoadFileDesc> loadFileWork = pCtx.getLoadFileWork();
boolean isCStats = pCtx.getQueryProperties().isAnalyzeRewrite();
int outerQueryLimit = pCtx.getQueryProperties().getOuterQueryLimit();
if (pCtx.getFetchTask() != null) {
if (pCtx.getFetchTask().getTblDesc() == null) {
return;
}
pCtx.getFetchTask().getWork().setHiveServerQuery(SessionState.get().isHiveServerQuery());
TableDesc resultTab = pCtx.getFetchTask().getTblDesc();
// then either the ThriftFormatter or the DefaultFetchFormatter should be used.
if (!resultTab.getSerdeClassName().equalsIgnoreCase(ThriftJDBCBinarySerDe.class.getName())) {
if (SessionState.get().isHiveServerQuery()) {
conf.set(SerDeUtils.LIST_SINK_OUTPUT_FORMATTER, ThriftFormatter.class.getName());
} else {
String formatterName = conf.get(SerDeUtils.LIST_SINK_OUTPUT_FORMATTER);
if (formatterName == null || formatterName.isEmpty()) {
conf.set(SerDeUtils.LIST_SINK_OUTPUT_FORMATTER, DefaultFetchFormatter.class.getName());
}
}
}
return;
}
optimizeOperatorPlan(pCtx, inputs, outputs);
/*
* In case of a select, use a fetch task instead of a move task.
* If the select is from analyze table column rewrite, don't create a fetch task. Instead create
* a column stats task later.
*/
if (pCtx.getQueryProperties().isQuery() && !isCStats) {
if ((!loadTableWork.isEmpty()) || (loadFileWork.size() != 1)) {
throw new SemanticException(ErrorMsg.INVALID_LOAD_TABLE_FILE_WORK.getMsg());
}
LoadFileDesc loadFileDesc = loadFileWork.get(0);
String cols = loadFileDesc.getColumns();
String colTypes = loadFileDesc.getColumnTypes();
String resFileFormat;
TableDesc resultTab = pCtx.getFetchTableDesc();
if (resultTab == null) {
resFileFormat = HiveConf.getVar(conf, HiveConf.ConfVars.HIVEQUERYRESULTFILEFORMAT);
if (SessionState.get().getIsUsingThriftJDBCBinarySerDe() && (resFileFormat.equalsIgnoreCase("SequenceFile"))) {
resultTab = PlanUtils.getDefaultQueryOutputTableDesc(cols, colTypes, resFileFormat, ThriftJDBCBinarySerDe.class);
// Set the fetch formatter to be a no-op for the ListSinkOperator, since we'll
// read formatted thrift objects from the output SequenceFile written by Tasks.
conf.set(SerDeUtils.LIST_SINK_OUTPUT_FORMATTER, NoOpFetchFormatter.class.getName());
} else {
resultTab = PlanUtils.getDefaultQueryOutputTableDesc(cols, colTypes, resFileFormat, LazySimpleSerDe.class);
}
} else {
if (resultTab.getProperties().getProperty(serdeConstants.SERIALIZATION_LIB).equalsIgnoreCase(ThriftJDBCBinarySerDe.class.getName())) {
// Set the fetch formatter to be a no-op for the ListSinkOperator, since we'll
// read formatted thrift objects from the output SequenceFile written by Tasks.
conf.set(SerDeUtils.LIST_SINK_OUTPUT_FORMATTER, NoOpFetchFormatter.class.getName());
}
}
FetchWork fetch = new FetchWork(loadFileDesc.getSourcePath(), resultTab, outerQueryLimit);
boolean isHiveServerQuery = SessionState.get().isHiveServerQuery();
fetch.setHiveServerQuery(isHiveServerQuery);
fetch.setSource(pCtx.getFetchSource());
fetch.setSink(pCtx.getFetchSink());
if (isHiveServerQuery && null != resultTab && resultTab.getSerdeClassName().equalsIgnoreCase(ThriftJDBCBinarySerDe.class.getName()) && HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVE_SERVER2_THRIFT_RESULTSET_SERIALIZE_IN_TASKS)) {
fetch.setIsUsingThriftJDBCBinarySerDe(true);
} else {
fetch.setIsUsingThriftJDBCBinarySerDe(false);
}
pCtx.setFetchTask((FetchTask) TaskFactory.get(fetch));
// For the FetchTask, the limit optimization requires we fetch all the rows
// in memory and count how many rows we get. It's not practical if the
// limit factor is too big
int fetchLimit = HiveConf.getIntVar(conf, HiveConf.ConfVars.HIVELIMITOPTMAXFETCH);
if (globalLimitCtx.isEnable() && globalLimitCtx.getGlobalLimit() > fetchLimit) {
LOG.info("For FetchTask, LIMIT " + globalLimitCtx.getGlobalLimit() + " > " + fetchLimit + ". Doesn't qualify limit optimization.");
globalLimitCtx.disableOpt();
}
if (outerQueryLimit == 0) {
// Believe it or not, some tools do generate queries with limit 0 and than expect
// query to run quickly. Lets meet their requirement.
LOG.info("Limit 0. No query execution needed.");
return;
}
} else if (!isCStats) {
for (LoadTableDesc ltd : loadTableWork) {
Task<MoveWork> tsk = TaskFactory.get(new MoveWork(null, null, ltd, null, false));
mvTask.add(tsk);
}
boolean oneLoadFileForCtas = true;
for (LoadFileDesc lfd : loadFileWork) {
if (pCtx.getQueryProperties().isCTAS() || pCtx.getQueryProperties().isMaterializedView()) {
if (!oneLoadFileForCtas) {
// should not have more than 1 load file for CTAS.
throw new SemanticException("One query is not expected to contain multiple CTAS loads statements");
}
setLoadFileLocation(pCtx, lfd);
oneLoadFileForCtas = false;
}
mvTask.add(TaskFactory.get(new MoveWork(null, null, null, lfd, false)));
}
}
generateTaskTree(rootTasks, pCtx, mvTask, inputs, outputs);
// For each task, set the key descriptor for the reducer
for (Task<? extends Serializable> rootTask : rootTasks) {
GenMapRedUtils.setKeyAndValueDescForTaskTree(rootTask);
}
// to be used, please do so
for (Task<? extends Serializable> rootTask : rootTasks) {
setInputFormat(rootTask);
}
optimizeTaskPlan(rootTasks, pCtx, ctx);
/*
* If the query was the result of analyze table column compute statistics rewrite, create
* a column stats task instead of a fetch task to persist stats to the metastore.
* As per HIVE-15903, we will also collect table stats when user computes column stats.
* That means, if isCStats || !pCtx.getColumnStatsAutoGatherContexts().isEmpty()
* We need to collect table stats
* if isCStats, we need to include a basic stats task
* else it is ColumnStatsAutoGather, which should have a move task with a stats task already.
*/
if (isCStats || !pCtx.getColumnStatsAutoGatherContexts().isEmpty()) {
// map from tablename to task (ColumnStatsTask which includes a BasicStatsTask)
Map<String, StatsTask> map = new LinkedHashMap<>();
if (isCStats) {
if (rootTasks == null || rootTasks.size() != 1 || pCtx.getTopOps() == null || pCtx.getTopOps().size() != 1) {
throw new SemanticException("Can not find correct root task!");
}
try {
Task<? extends Serializable> root = rootTasks.iterator().next();
StatsTask tsk = (StatsTask) genTableStats(pCtx, pCtx.getTopOps().values().iterator().next(), root, outputs);
root.addDependentTask(tsk);
map.put(extractTableFullName(tsk), tsk);
} catch (HiveException e) {
throw new SemanticException(e);
}
genColumnStatsTask(pCtx.getAnalyzeRewrite(), loadFileWork, map, outerQueryLimit, 0);
} else {
Set<Task<? extends Serializable>> leafTasks = new LinkedHashSet<Task<? extends Serializable>>();
getLeafTasks(rootTasks, leafTasks);
List<Task<? extends Serializable>> nonStatsLeafTasks = new ArrayList<>();
for (Task<? extends Serializable> tsk : leafTasks) {
// map table name to the correct ColumnStatsTask
if (tsk instanceof StatsTask) {
map.put(extractTableFullName((StatsTask) tsk), (StatsTask) tsk);
} else {
nonStatsLeafTasks.add(tsk);
}
}
// add cStatsTask as a dependent of all the nonStatsLeafTasks
for (Task<? extends Serializable> tsk : nonStatsLeafTasks) {
for (Task<? extends Serializable> cStatsTask : map.values()) {
tsk.addDependentTask(cStatsTask);
}
}
for (ColumnStatsAutoGatherContext columnStatsAutoGatherContext : pCtx.getColumnStatsAutoGatherContexts()) {
if (!columnStatsAutoGatherContext.isInsertInto()) {
genColumnStatsTask(columnStatsAutoGatherContext.getAnalyzeRewrite(), columnStatsAutoGatherContext.getLoadFileWork(), map, outerQueryLimit, 0);
} else {
int numBitVector;
try {
numBitVector = HiveStatsUtils.getNumBitVectorsForNDVEstimation(conf);
} catch (Exception e) {
throw new SemanticException(e.getMessage());
}
genColumnStatsTask(columnStatsAutoGatherContext.getAnalyzeRewrite(), columnStatsAutoGatherContext.getLoadFileWork(), map, outerQueryLimit, numBitVector);
}
}
}
}
decideExecMode(rootTasks, ctx, globalLimitCtx);
if (pCtx.getQueryProperties().isCTAS() && !pCtx.getCreateTable().isMaterialization()) {
// generate a DDL task and make it a dependent task of the leaf
CreateTableDesc crtTblDesc = pCtx.getCreateTable();
crtTblDesc.validate(conf);
Task<? extends Serializable> crtTblTask = TaskFactory.get(new DDLWork(inputs, outputs, crtTblDesc));
patchUpAfterCTASorMaterializedView(rootTasks, outputs, crtTblTask);
} else if (pCtx.getQueryProperties().isMaterializedView()) {
// generate a DDL task and make it a dependent task of the leaf
CreateViewDesc viewDesc = pCtx.getCreateViewDesc();
Task<? extends Serializable> crtViewTask = TaskFactory.get(new DDLWork(inputs, outputs, viewDesc));
patchUpAfterCTASorMaterializedView(rootTasks, outputs, crtViewTask);
} else if (pCtx.getMaterializedViewUpdateDesc() != null) {
// If there is a materialized view update desc, we create introduce it at the end
// of the tree.
MaterializedViewDesc materializedViewDesc = pCtx.getMaterializedViewUpdateDesc();
Set<Task<? extends Serializable>> leafTasks = new LinkedHashSet<Task<? extends Serializable>>();
getLeafTasks(rootTasks, leafTasks);
Task<? extends Serializable> materializedViewTask = TaskFactory.get(materializedViewDesc, conf);
for (Task<? extends Serializable> task : leafTasks) {
task.addDependentTask(materializedViewTask);
}
}
if (globalLimitCtx.isEnable() && pCtx.getFetchTask() != null) {
LOG.info("set least row check for FetchTask: " + globalLimitCtx.getGlobalLimit());
pCtx.getFetchTask().getWork().setLeastNumRows(globalLimitCtx.getGlobalLimit());
}
if (globalLimitCtx.isEnable() && globalLimitCtx.getLastReduceLimitDesc() != null) {
LOG.info("set least row check for LimitDesc: " + globalLimitCtx.getGlobalLimit());
globalLimitCtx.getLastReduceLimitDesc().setLeastRows(globalLimitCtx.getGlobalLimit());
}
Interner<TableDesc> interner = Interners.newStrongInterner();
for (Task<? extends Serializable> rootTask : rootTasks) {
GenMapRedUtils.internTableDesc(rootTask, interner);
GenMapRedUtils.deriveFinalExplainAttributes(rootTask, pCtx.getConf());
}
}
Aggregations