Examples with JoinOperator org.apache.hadoop.hive.ql.exec.JoinOperator used on opensource projects

Example 21 with JoinOperator

use of org.apache.hadoop.hive.ql.exec.JoinOperator in project hive by apache.

the class MapJoinProcessor method convertMapJoin.

/**
   * convert a regular join to a a map-side join.
   *
   * @param opParseCtxMap
   * @param op
   *          join operator
   * @param joinTree
   *          qb join tree
   * @param mapJoinPos
   *          position of the source to be read as part of map-reduce framework. All other sources
   *          are cached in memory
   * @param noCheckOuterJoin
   * @param validateMapJoinTree
   */
public MapJoinOperator convertMapJoin(HiveConf conf, JoinOperator op, boolean leftInputJoin, String[] baseSrc, List<String> mapAliases, int mapJoinPos, boolean noCheckOuterJoin, boolean validateMapJoinTree) throws SemanticException {
    // outer join cannot be performed on a table which is being cached
    JoinDesc desc = op.getConf();
    JoinCondDesc[] condns = desc.getConds();
    if (!noCheckOuterJoin) {
        if (checkMapJoin(mapJoinPos, condns) < 0) {
            throw new SemanticException(ErrorMsg.NO_OUTER_MAPJOIN.getMsg());
        }
    }
    // Walk over all the sources (which are guaranteed to be reduce sink
    // operators).
    // The join outputs a concatenation of all the inputs.
    List<Operator<? extends OperatorDesc>> parentOps = op.getParentOperators();
    List<Operator<? extends OperatorDesc>> newParentOps = new ArrayList<Operator<? extends OperatorDesc>>();
    List<Operator<? extends OperatorDesc>> oldReduceSinkParentOps = new ArrayList<Operator<? extends OperatorDesc>>();
    // found a source which is not to be stored in memory
    if (leftInputJoin) {
        // assert mapJoinPos == 0;
        Operator<? extends OperatorDesc> parentOp = parentOps.get(0);
        assert parentOp.getParentOperators().size() == 1;
        Operator<? extends OperatorDesc> grandParentOp = parentOp.getParentOperators().get(0);
        oldReduceSinkParentOps.add(parentOp);
        newParentOps.add(grandParentOp);
    }
    byte pos = 0;
    // Remove parent reduce-sink operators
    for (String src : baseSrc) {
        if (src != null) {
            Operator<? extends OperatorDesc> parentOp = parentOps.get(pos);
            assert parentOp.getParentOperators().size() == 1;
            Operator<? extends OperatorDesc> grandParentOp = parentOp.getParentOperators().get(0);
            oldReduceSinkParentOps.add(parentOp);
            newParentOps.add(grandParentOp);
        }
        pos++;
    }
    // create the map-join operator
    MapJoinOperator mapJoinOp = convertJoinOpMapJoinOp(conf, op, leftInputJoin, baseSrc, mapAliases, mapJoinPos, noCheckOuterJoin);
    // remove old parents
    for (pos = 0; pos < newParentOps.size(); pos++) {
        newParentOps.get(pos).replaceChild(oldReduceSinkParentOps.get(pos), mapJoinOp);
    }
    mapJoinOp.getParentOperators().removeAll(oldReduceSinkParentOps);
    mapJoinOp.setParentOperators(newParentOps);
    // make sure only map-joins can be performed.
    if (validateMapJoinTree) {
        validateMapJoinTypes(mapJoinOp);
    }
    return mapJoinOp;
}

Example 22 with JoinOperator

use of org.apache.hadoop.hive.ql.exec.JoinOperator in project hive by apache.

the class SparkSortMergeJoinOptimizer method process.

@Override
public Object process(Node nd, Stack<Node> stack, NodeProcessorCtx procCtx, Object... nodeOutputs) throws SemanticException {
    JoinOperator joinOp = (JoinOperator) nd;
    HiveConf conf = ((OptimizeSparkProcContext) procCtx).getParseContext().getConf();
    if (!conf.getBoolVar(HiveConf.ConfVars.HIVE_AUTO_SORTMERGE_JOIN)) {
        return null;
    }
    SortBucketJoinProcCtx smbJoinContext = new SortBucketJoinProcCtx(conf);
    boolean convert = canConvertJoinToSMBJoin(joinOp, smbJoinContext, pGraphContext, stack);
    if (convert) {
        return convertJoinToSMBJoinAndReturn(joinOp, smbJoinContext);
    }
    return null;
}

Example 23 with JoinOperator

use of org.apache.hadoop.hive.ql.exec.JoinOperator in project hive by apache.

the class HiveOpConverter method genJoin.

private static JoinOperator genJoin(RelNode join, ExprNodeDesc[][] joinExpressions, List<List<ExprNodeDesc>> filterExpressions, List<Operator<?>> children, String[] baseSrc, String tabAlias) throws SemanticException {
    // 1. Extract join type
    JoinCondDesc[] joinCondns;
    boolean semiJoin;
    boolean noOuterJoin;
    if (join instanceof HiveMultiJoin) {
        HiveMultiJoin hmj = (HiveMultiJoin) join;
        joinCondns = new JoinCondDesc[hmj.getJoinInputs().size()];
        for (int i = 0; i < hmj.getJoinInputs().size(); i++) {
            joinCondns[i] = new JoinCondDesc(new JoinCond(hmj.getJoinInputs().get(i).left, hmj.getJoinInputs().get(i).right, transformJoinType(hmj.getJoinTypes().get(i))));
        }
        semiJoin = false;
        noOuterJoin = !hmj.isOuterJoin();
    } else {
        joinCondns = new JoinCondDesc[1];
        semiJoin = join instanceof SemiJoin;
        JoinType joinType;
        if (semiJoin) {
            joinType = JoinType.LEFTSEMI;
        } else {
            joinType = extractJoinType((Join) join);
        }
        joinCondns[0] = new JoinCondDesc(new JoinCond(0, 1, joinType));
        noOuterJoin = joinType != JoinType.FULLOUTER && joinType != JoinType.LEFTOUTER && joinType != JoinType.RIGHTOUTER;
    }
    // 2. We create the join aux structures
    ArrayList<ColumnInfo> outputColumns = new ArrayList<ColumnInfo>();
    ArrayList<String> outputColumnNames = new ArrayList<String>(join.getRowType().getFieldNames());
    Operator<?>[] childOps = new Operator[children.size()];
    Map<String, Byte> reversedExprs = new HashMap<String, Byte>();
    Map<Byte, List<ExprNodeDesc>> exprMap = new HashMap<Byte, List<ExprNodeDesc>>();
    Map<Byte, List<ExprNodeDesc>> filters = new HashMap<Byte, List<ExprNodeDesc>>();
    Map<String, ExprNodeDesc> colExprMap = new HashMap<String, ExprNodeDesc>();
    HashMap<Integer, Set<String>> posToAliasMap = new HashMap<Integer, Set<String>>();
    int outputPos = 0;
    for (int pos = 0; pos < children.size(); pos++) {
        // 2.1. Backtracking from RS
        ReduceSinkOperator inputRS = (ReduceSinkOperator) children.get(pos);
        if (inputRS.getNumParent() != 1) {
            throw new SemanticException("RS should have single parent");
        }
        Operator<?> parent = inputRS.getParentOperators().get(0);
        ReduceSinkDesc rsDesc = inputRS.getConf();
        int[] index = inputRS.getValueIndex();
        Byte tag = (byte) rsDesc.getTag();
        // 2.1.1. If semijoin...
        if (semiJoin && pos != 0) {
            exprMap.put(tag, new ArrayList<ExprNodeDesc>());
            childOps[pos] = inputRS;
            continue;
        }
        posToAliasMap.put(pos, new HashSet<String>(inputRS.getSchema().getTableNames()));
        List<String> keyColNames = rsDesc.getOutputKeyColumnNames();
        List<String> valColNames = rsDesc.getOutputValueColumnNames();
        Map<String, ExprNodeDesc> descriptors = buildBacktrackFromReduceSinkForJoin(outputPos, outputColumnNames, keyColNames, valColNames, index, parent, baseSrc[pos]);
        List<ColumnInfo> parentColumns = parent.getSchema().getSignature();
        for (int i = 0; i < index.length; i++) {
            ColumnInfo info = new ColumnInfo(parentColumns.get(i));
            info.setInternalName(outputColumnNames.get(outputPos));
            info.setTabAlias(tabAlias);
            outputColumns.add(info);
            reversedExprs.put(outputColumnNames.get(outputPos), tag);
            outputPos++;
        }
        exprMap.put(tag, new ArrayList<ExprNodeDesc>(descriptors.values()));
        colExprMap.putAll(descriptors);
        childOps[pos] = inputRS;
    }
    // 3. We populate the filters and filterMap structure needed in the join descriptor
    List<List<ExprNodeDesc>> filtersPerInput = Lists.newArrayList();
    int[][] filterMap = new int[children.size()][];
    for (int i = 0; i < children.size(); i++) {
        filtersPerInput.add(new ArrayList<ExprNodeDesc>());
    }
    // 3. We populate the filters structure
    for (int i = 0; i < filterExpressions.size(); i++) {
        int leftPos = joinCondns[i].getLeft();
        int rightPos = joinCondns[i].getRight();
        for (ExprNodeDesc expr : filterExpressions.get(i)) {
            // We need to update the exprNode, as currently
            // they refer to columns in the output of the join;
            // they should refer to the columns output by the RS
            int inputPos = updateExprNode(expr, reversedExprs, colExprMap);
            if (inputPos == -1) {
                inputPos = leftPos;
            }
            filtersPerInput.get(inputPos).add(expr);
            if (joinCondns[i].getType() == JoinDesc.FULL_OUTER_JOIN || joinCondns[i].getType() == JoinDesc.LEFT_OUTER_JOIN || joinCondns[i].getType() == JoinDesc.RIGHT_OUTER_JOIN) {
                if (inputPos == leftPos) {
                    updateFilterMap(filterMap, leftPos, rightPos);
                } else {
                    updateFilterMap(filterMap, rightPos, leftPos);
                }
            }
        }
    }
    for (int pos = 0; pos < children.size(); pos++) {
        ReduceSinkOperator inputRS = (ReduceSinkOperator) children.get(pos);
        ReduceSinkDesc rsDesc = inputRS.getConf();
        Byte tag = (byte) rsDesc.getTag();
        filters.put(tag, filtersPerInput.get(pos));
    }
    // 4. We create the join operator with its descriptor
    JoinDesc desc = new JoinDesc(exprMap, outputColumnNames, noOuterJoin, joinCondns, filters, joinExpressions);
    desc.setReversedExprs(reversedExprs);
    desc.setFilterMap(filterMap);
    JoinOperator joinOp = (JoinOperator) OperatorFactory.getAndMakeChild(childOps[0].getCompilationOpContext(), desc, new RowSchema(outputColumns), childOps);
    joinOp.setColumnExprMap(colExprMap);
    joinOp.setPosToAliasMap(posToAliasMap);
    joinOp.getConf().setBaseSrc(baseSrc);
    if (LOG.isDebugEnabled()) {
        LOG.debug("Generated " + joinOp + " with row schema: [" + joinOp.getSchema() + "]");
    }
    return joinOp;
}

Example 24 with JoinOperator

use of org.apache.hadoop.hive.ql.exec.JoinOperator in project hive by apache.

the class AbstractSMBJoinProc method canConvertJoinToBucketMapJoin.

// Can the join operator be converted to a bucket map-merge join operator ?
@SuppressWarnings("unchecked")
protected boolean canConvertJoinToBucketMapJoin(JoinOperator joinOp, SortBucketJoinProcCtx context) throws SemanticException {
    // This has already been inspected and rejected
    if (context.getRejectedJoinOps().contains(joinOp)) {
        return false;
    }
    if (!this.pGraphContext.getJoinOps().contains(joinOp)) {
        return false;
    }
    Class<? extends BigTableSelectorForAutoSMJ> bigTableMatcherClass = null;
    try {
        String selector = HiveConf.getVar(pGraphContext.getConf(), HiveConf.ConfVars.HIVE_AUTO_SORTMERGE_JOIN_BIGTABLE_SELECTOR);
        bigTableMatcherClass = JavaUtils.loadClass(selector);
    } catch (ClassNotFoundException e) {
        throw new SemanticException(e.getMessage());
    }
    BigTableSelectorForAutoSMJ bigTableMatcher = ReflectionUtils.newInstance(bigTableMatcherClass, null);
    JoinDesc joinDesc = joinOp.getConf();
    JoinCondDesc[] joinCondns = joinDesc.getConds();
    Set<Integer> joinCandidates = MapJoinProcessor.getBigTableCandidates(joinCondns);
    if (joinCandidates.isEmpty()) {
        // of any type. So return false.
        return false;
    }
    int bigTablePosition = bigTableMatcher.getBigTablePosition(pGraphContext, joinOp, joinCandidates);
    if (bigTablePosition < 0) {
        // contains aliases from sub-query
        return false;
    }
    context.setBigTablePosition(bigTablePosition);
    String joinAlias = bigTablePosition == 0 ? joinOp.getConf().getLeftAlias() : joinOp.getConf().getRightAliases()[bigTablePosition - 1];
    joinAlias = QB.getAppendedAliasFromId(joinOp.getConf().getId(), joinAlias);
    Map<Byte, List<ExprNodeDesc>> keyExprMap = new HashMap<Byte, List<ExprNodeDesc>>();
    List<Operator<? extends OperatorDesc>> parentOps = joinOp.getParentOperators();
    // get the join keys from parent ReduceSink operators
    for (Operator<? extends OperatorDesc> parentOp : parentOps) {
        ReduceSinkDesc rsconf = ((ReduceSinkOperator) parentOp).getConf();
        Byte tag = (byte) rsconf.getTag();
        List<ExprNodeDesc> keys = rsconf.getKeyCols();
        keyExprMap.put(tag, keys);
    }
    context.setKeyExprMap(keyExprMap);
    // Make a deep copy of the aliases so that they are not changed in the context
    String[] joinSrcs = joinOp.getConf().getBaseSrc();
    String[] srcs = new String[joinSrcs.length];
    for (int srcPos = 0; srcPos < joinSrcs.length; srcPos++) {
        joinSrcs[srcPos] = QB.getAppendedAliasFromId(joinOp.getConf().getId(), joinSrcs[srcPos]);
        srcs[srcPos] = new String(joinSrcs[srcPos]);
    }
    // table matcher.
    return checkConvertBucketMapJoin(context, joinOp.getConf().getAliasToOpInfo(), keyExprMap, joinAlias, Arrays.asList(srcs));
}

Example 25 with JoinOperator

use of org.apache.hadoop.hive.ql.exec.JoinOperator in project hive by apache.

the class SemanticAnalyzer method analyzeInternal.

void analyzeInternal(ASTNode ast, PlannerContext plannerCtx) throws SemanticException {
    // 1. Generate Resolved Parse tree from syntax tree
    LOG.info("Starting Semantic Analysis");
    //change the location of position alias process here
    processPositionAlias(ast);
    if (!genResolvedParseTree(ast, plannerCtx)) {
        return;
    }
    // 2. Gen OP Tree from resolved Parse Tree
    Operator sinkOp = genOPTree(ast, plannerCtx);
    if (!unparseTranslator.isEnabled() && tableMask.isEnabled()) {
        // Here we rewrite the * and also the masking table
        ASTNode tree = rewriteASTWithMaskAndFilter(ast);
        if (tree != ast) {
            ctx.setSkipTableMasking(true);
            init(true);
            //change the location of position alias process here
            processPositionAlias(tree);
            genResolvedParseTree(tree, plannerCtx);
            if (this instanceof CalcitePlanner) {
                ((CalcitePlanner) this).resetCalciteConfiguration();
            }
            sinkOp = genOPTree(tree, plannerCtx);
        }
    }
    // 3. Deduce Resultset Schema
    if (createVwDesc != null && !this.ctx.isCboSucceeded()) {
        resultSchema = convertRowSchemaToViewSchema(opParseCtx.get(sinkOp).getRowResolver());
    } else {
        // succeeds.
        if (resultSchema == null) {
            resultSchema = convertRowSchemaToResultSetSchema(opParseCtx.get(sinkOp).getRowResolver(), HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVE_RESULTSET_USE_UNIQUE_COLUMN_NAMES));
        }
    }
    // 4. Generate Parse Context for Optimizer & Physical compiler
    copyInfoToQueryProperties(queryProperties);
    ParseContext pCtx = new ParseContext(queryState, opToPartPruner, opToPartList, topOps, new HashSet<JoinOperator>(joinContext.keySet()), new HashSet<SMBMapJoinOperator>(smbMapJoinContext.keySet()), loadTableWork, loadFileWork, columnStatsAutoGatherContexts, ctx, idToTableNameMap, destTableId, uCtx, listMapJoinOpsNoReducer, prunedPartitions, tabNameToTabObject, opToSamplePruner, globalLimitCtx, nameToSplitSample, inputs, rootTasks, opToPartToSkewedPruner, viewAliasToInput, reduceSinkOperatorsAddedByEnforceBucketingSorting, analyzeRewrite, tableDesc, createVwDesc, queryProperties, viewProjectToTableSchema, acidFileSinks);
    // 5. Take care of view creation
    if (createVwDesc != null) {
        if (ctx.getExplainAnalyze() == AnalyzeState.RUNNING) {
            return;
        }
        if (!ctx.isCboSucceeded()) {
            saveViewDefinition();
        }
        // validate the create view statement at this point, the createVwDesc gets
        // all the information for semanticcheck
        validateCreateView();
        if (!createVwDesc.isMaterialized()) {
            // Since we're only creating a view (not executing it), we don't need to
            // optimize or translate the plan (and in fact, those procedures can
            // interfere with the view creation). So skip the rest of this method.
            ctx.setResDir(null);
            ctx.setResFile(null);
            try {
                PlanUtils.addInputsForView(pCtx);
            } catch (HiveException e) {
                throw new SemanticException(e);
            }
            // Generate lineage info for create view statements
            // if LineageLogger hook is configured.
            // Add the transformation that computes the lineage information.
            Set<String> postExecHooks = Sets.newHashSet(Splitter.on(",").trimResults().omitEmptyStrings().split(Strings.nullToEmpty(HiveConf.getVar(conf, HiveConf.ConfVars.POSTEXECHOOKS))));
            if (postExecHooks.contains("org.apache.hadoop.hive.ql.hooks.PostExecutePrinter") || postExecHooks.contains("org.apache.hadoop.hive.ql.hooks.LineageLogger") || postExecHooks.contains("org.apache.atlas.hive.hook.HiveHook")) {
                ArrayList<Transform> transformations = new ArrayList<Transform>();
                transformations.add(new HiveOpConverterPostProc());
                transformations.add(new Generator());
                for (Transform t : transformations) {
                    pCtx = t.transform(pCtx);
                }
                // we just use view name as location.
                SessionState.get().getLineageState().mapDirToOp(new Path(createVwDesc.getViewName()), sinkOp);
            }
            return;
        }
    }
    // 6. Generate table access stats if required
    if (HiveConf.getBoolVar(this.conf, HiveConf.ConfVars.HIVE_STATS_COLLECT_TABLEKEYS)) {
        TableAccessAnalyzer tableAccessAnalyzer = new TableAccessAnalyzer(pCtx);
        setTableAccessInfo(tableAccessAnalyzer.analyzeTableAccess());
    }
    // 7. Perform Logical optimization
    if (LOG.isDebugEnabled()) {
        LOG.debug("Before logical optimization\n" + Operator.toString(pCtx.getTopOps().values()));
    }
    Optimizer optm = new Optimizer();
    optm.setPctx(pCtx);
    optm.initialize(conf);
    pCtx = optm.optimize();
    if (pCtx.getColumnAccessInfo() != null) {
        // set ColumnAccessInfo for view column authorization
        setColumnAccessInfo(pCtx.getColumnAccessInfo());
    }
    FetchTask origFetchTask = pCtx.getFetchTask();
    if (LOG.isDebugEnabled()) {
        LOG.debug("After logical optimization\n" + Operator.toString(pCtx.getTopOps().values()));
    }
    // 8. Generate column access stats if required - wait until column pruning
    // takes place during optimization
    boolean isColumnInfoNeedForAuth = SessionState.get().isAuthorizationModeV2() && HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVE_AUTHORIZATION_ENABLED);
    if (isColumnInfoNeedForAuth || HiveConf.getBoolVar(this.conf, HiveConf.ConfVars.HIVE_STATS_COLLECT_SCANCOLS)) {
        ColumnAccessAnalyzer columnAccessAnalyzer = new ColumnAccessAnalyzer(pCtx);
        // view column access info is carried by this.getColumnAccessInfo().
        setColumnAccessInfo(columnAccessAnalyzer.analyzeColumnAccess(this.getColumnAccessInfo()));
    }
    // TEZ..)
    if (!ctx.getExplainLogical()) {
        TaskCompiler compiler = TaskCompilerFactory.getCompiler(conf, pCtx);
        compiler.init(queryState, console, db);
        compiler.compile(pCtx, rootTasks, inputs, outputs);
        fetchTask = pCtx.getFetchTask();
    }
    LOG.info("Completed plan generation");
    // 10. put accessed columns to readEntity
    if (HiveConf.getBoolVar(this.conf, HiveConf.ConfVars.HIVE_STATS_COLLECT_SCANCOLS)) {
        putAccessedColumnsToReadEntity(inputs, columnAccessInfo);
    }
    // 11. if desired check we're not going over partition scan limits
    if (!ctx.isExplainSkipExecution()) {
        enforceScanLimits(pCtx, origFetchTask);
    }
    return;
}