Examples with SemanticGraphWalker org.apache.hadoop.hive.ql.lib.SemanticGraphWalker used on opensource projects

Example 51 with SemanticGraphWalker

use of org.apache.hadoop.hive.ql.lib.SemanticGraphWalker in project hive by apache.

the class TestAccumuloRangeGenerator method testRangeOverStringIndexedField.

@Test
public void testRangeOverStringIndexedField() throws Exception {
    // age >= '10'
    ExprNodeDesc column = new ExprNodeColumnDesc(TypeInfoFactory.stringTypeInfo, "age", null, false);
    ExprNodeDesc constant = new ExprNodeConstantDesc(TypeInfoFactory.stringTypeInfo, "10");
    List<ExprNodeDesc> children = Lists.newArrayList();
    children.add(column);
    children.add(constant);
    ExprNodeDesc node = new ExprNodeGenericFuncDesc(TypeInfoFactory.stringTypeInfo, new GenericUDFOPEqualOrGreaterThan(), children);
    assertNotNull(node);
    // age <= '50'
    ExprNodeDesc column2 = new ExprNodeColumnDesc(TypeInfoFactory.stringTypeInfo, "age", null, false);
    ExprNodeDesc constant2 = new ExprNodeConstantDesc(TypeInfoFactory.stringTypeInfo, "50");
    List<ExprNodeDesc> children2 = Lists.newArrayList();
    children2.add(column2);
    children2.add(constant2);
    ExprNodeDesc node2 = new ExprNodeGenericFuncDesc(TypeInfoFactory.stringTypeInfo, new GenericUDFOPEqualOrLessThan(), children2);
    assertNotNull(node2);
    // And UDF
    List<ExprNodeDesc> bothFilters = Lists.newArrayList();
    bothFilters.add(node);
    bothFilters.add(node2);
    ExprNodeGenericFuncDesc both = new ExprNodeGenericFuncDesc(TypeInfoFactory.stringTypeInfo, new GenericUDFOPAnd(), bothFilters);
    AccumuloRangeGenerator rangeGenerator = new AccumuloRangeGenerator(conf, handler, rowIdMapping, "rid");
    rangeGenerator.setIndexScanner(TestAccumuloDefaultIndexScanner.buildMockHandler(10));
    SemanticDispatcher disp = new DefaultRuleDispatcher(rangeGenerator, Collections.<SemanticRule, SemanticNodeProcessor>emptyMap(), null);
    SemanticGraphWalker ogw = new DefaultGraphWalker(disp);
    ArrayList<Node> topNodes = new ArrayList<Node>();
    topNodes.add(both);
    HashMap<Node, Object> nodeOutput = new HashMap<Node, Object>();
    try {
        ogw.startWalking(topNodes, nodeOutput);
    } catch (SemanticException ex) {
        throw new RuntimeException(ex);
    }
    // Filters are using an index which should match 3 rows
    Object result = nodeOutput.get(both);
    if (result instanceof List) {
        List results = (List) result;
        Assert.assertEquals(3, results.size());
        Assert.assertTrue("does not contain row1", results.contains(new Range("row1")));
        Assert.assertTrue("does not contain row2", results.contains(new Range("row2")));
        Assert.assertTrue("does not contain row3", results.contains(new Range("row3")));
    } else {
        Assert.fail("Results not a list");
    }
}

Example 52 with SemanticGraphWalker

use of org.apache.hadoop.hive.ql.lib.SemanticGraphWalker in project hive by apache.

the class HiveOpConverterPostProc method transform.

@Override
public ParseContext transform(ParseContext pctx) throws SemanticException {
    // 0. We check the conditions to apply this transformation,
    // if we do not meet them we bail out
    final boolean cboEnabled = HiveConf.getBoolVar(pctx.getConf(), HiveConf.ConfVars.HIVE_CBO_ENABLED);
    final boolean returnPathEnabled = HiveConf.getBoolVar(pctx.getConf(), HiveConf.ConfVars.HIVE_CBO_RETPATH_HIVEOP);
    final boolean cboSucceeded = pctx.getContext().isCboSucceeded();
    if (!(cboEnabled && returnPathEnabled && cboSucceeded)) {
        return pctx;
    }
    // 1. Initialize aux data structures
    this.pctx = pctx;
    this.aliasToOpInfo = new HashMap<String, Operator<? extends OperatorDesc>>();
    // 2. Trigger transformation
    Map<SemanticRule, SemanticNodeProcessor> opRules = new LinkedHashMap<SemanticRule, SemanticNodeProcessor>();
    opRules.put(new RuleRegExp("R1", JoinOperator.getOperatorName() + "%"), new JoinAnnotate());
    opRules.put(new RuleRegExp("R2", TableScanOperator.getOperatorName() + "%"), new TableScanAnnotate());
    SemanticDispatcher disp = new DefaultRuleDispatcher(null, opRules, null);
    SemanticGraphWalker ogw = new ForwardWalker(disp);
    List<Node> topNodes = new ArrayList<Node>();
    topNodes.addAll(pctx.getTopOps().values());
    ogw.startWalking(topNodes, null);
    return pctx;
}

Example 53 with SemanticGraphWalker

use of org.apache.hadoop.hive.ql.lib.SemanticGraphWalker in project hive by apache.

the class AnnotateRunTimeStatsOptimizer method resolve.

@Override
public PhysicalContext resolve(PhysicalContext pctx) throws SemanticException {
    Map<SemanticRule, SemanticNodeProcessor> opRules = new LinkedHashMap<SemanticRule, SemanticNodeProcessor>();
    SemanticDispatcher disp = new AnnotateRunTimeStatsDispatcher(pctx, opRules);
    SemanticGraphWalker ogw = new DefaultGraphWalker(disp);
    ArrayList<Node> topNodes = new ArrayList<Node>();
    topNodes.addAll(pctx.getRootTasks());
    ogw.startWalking(topNodes, null);
    return pctx;
}

Example 54 with SemanticGraphWalker

use of org.apache.hadoop.hive.ql.lib.SemanticGraphWalker in project hive by apache.

the class BucketingSortingInferenceOptimizer method inferBucketingSorting.

/**
 * For each map reduce task, if it has a reducer, infer whether or not the final output of the
 * reducer is bucketed and/or sorted
 *
 * @param mapRedTasks
 * @throws SemanticException
 */
private void inferBucketingSorting(List<ExecDriver> mapRedTasks) throws SemanticException {
    for (ExecDriver mapRedTask : mapRedTasks) {
        // of the outputs of intermediate map reduce jobs.
        if (!mapRedTask.getWork().isFinalMapRed()) {
            continue;
        }
        if (mapRedTask.getWork().getReduceWork() == null) {
            continue;
        }
        Operator<? extends OperatorDesc> reducer = mapRedTask.getWork().getReduceWork().getReducer();
        // uses sampling, which means it's not bucketed
        boolean disableBucketing = mapRedTask.getWork().getMapWork().getSamplingType() > 0;
        BucketingSortingCtx bCtx = new BucketingSortingCtx(disableBucketing);
        // RuleRegExp rules are used to match operators anywhere in the tree
        // RuleExactMatch rules are used to specify exactly what the tree should look like
        // In particular, this guarantees that the first operator is the reducer
        // (and its parent(s) are ReduceSinkOperators) since it begins walking the tree from
        // the reducer.
        Map<SemanticRule, SemanticNodeProcessor> opRules = new LinkedHashMap<SemanticRule, SemanticNodeProcessor>();
        opRules.put(new RuleRegExp("R1", SelectOperator.getOperatorName() + "%"), BucketingSortingOpProcFactory.getSelProc());
        // Matches only GroupByOperators which are reducers, rather than map group by operators,
        // or multi group by optimization specific operators
        opRules.put(new RuleExactMatch("R2", new String[] { GroupByOperator.getOperatorName() }), BucketingSortingOpProcFactory.getGroupByProc());
        // Matches only JoinOperators which are reducers, rather than map joins, SMB map joins, etc.
        opRules.put(new RuleExactMatch("R3", new String[] { JoinOperator.getOperatorName() }), BucketingSortingOpProcFactory.getJoinProc());
        opRules.put(new RuleRegExp("R5", FileSinkOperator.getOperatorName() + "%"), BucketingSortingOpProcFactory.getFileSinkProc());
        opRules.put(new RuleRegExp("R7", FilterOperator.getOperatorName() + "%"), BucketingSortingOpProcFactory.getFilterProc());
        opRules.put(new RuleRegExp("R8", LimitOperator.getOperatorName() + "%"), BucketingSortingOpProcFactory.getLimitProc());
        opRules.put(new RuleRegExp("R9", LateralViewForwardOperator.getOperatorName() + "%"), BucketingSortingOpProcFactory.getLateralViewForwardProc());
        opRules.put(new RuleRegExp("R10", LateralViewJoinOperator.getOperatorName() + "%"), BucketingSortingOpProcFactory.getLateralViewJoinProc());
        // Matches only ForwardOperators which are preceded by some other operator in the tree,
        // in particular it can't be a reducer (and hence cannot be one of the ForwardOperators
        // added by the multi group by optimization)
        opRules.put(new RuleRegExp("R11", ".+" + ForwardOperator.getOperatorName() + "%"), BucketingSortingOpProcFactory.getForwardProc());
        // Matches only ForwardOperators which are reducers and are followed by GroupByOperators
        // (specific to the multi group by optimization)
        opRules.put(new RuleExactMatch("R12", new String[] { ForwardOperator.getOperatorName(), GroupByOperator.getOperatorName() }), BucketingSortingOpProcFactory.getMultiGroupByProc());
        // The dispatcher fires the processor corresponding to the closest matching rule and passes
        // the context along
        SemanticDispatcher disp = new DefaultRuleDispatcher(BucketingSortingOpProcFactory.getDefaultProc(), opRules, bCtx);
        SemanticGraphWalker ogw = new PreOrderWalker(disp);
        // Create a list of topop nodes
        ArrayList<Node> topNodes = new ArrayList<Node>();
        topNodes.add(reducer);
        ogw.startWalking(topNodes, null);
        mapRedTask.getWork().getMapWork().getBucketedColsByDirectory().putAll(bCtx.getBucketedColsByDirectory());
        mapRedTask.getWork().getMapWork().getSortedColsByDirectory().putAll(bCtx.getSortedColsByDirectory());
    }
}

Example 55 with SemanticGraphWalker

use of org.apache.hadoop.hive.ql.lib.SemanticGraphWalker in project hive by apache.

the class CorrelationOptimizer method transform.

/**
 * Detect correlations and transform the query tree.
 *
 * @param pctx
 *          current parse context
 * @throws SemanticException
 */
@Override
public ParseContext transform(ParseContext pctx) throws SemanticException {
    pCtx = pctx;
    if (HiveConf.getBoolVar(pCtx.getConf(), HiveConf.ConfVars.HIVECONVERTJOIN)) {
        findPossibleAutoConvertedJoinOperators();
    }
    // detect correlations
    CorrelationNodeProcCtx corrCtx = new CorrelationNodeProcCtx(pCtx);
    Map<SemanticRule, SemanticNodeProcessor> opRules = new LinkedHashMap<SemanticRule, SemanticNodeProcessor>();
    opRules.put(new RuleRegExp("R1", ReduceSinkOperator.getOperatorName() + "%"), new CorrelationNodeProc());
    SemanticDispatcher disp = new DefaultRuleDispatcher(getDefaultProc(), opRules, corrCtx);
    SemanticGraphWalker ogw = new DefaultGraphWalker(disp);
    // Create a list of topOp nodes
    List<Node> topNodes = new ArrayList<Node>();
    topNodes.addAll(pCtx.getTopOps().values());
    ogw.startWalking(topNodes, null);
    // We have finished tree walking (correlation detection).
    // We will first see if we need to abort (the operator tree has not been changed).
    // If not, we will start to transform the operator tree.
    abort = corrCtx.isAbort();
    if (abort) {
        LOG.info("Abort. Reasons are ...");
        for (String reason : corrCtx.getAbortReasons()) {
            LOG.info("-- " + reason);
        }
    } else {
        // transform the operator tree
        LOG.info("Begain query plan transformation based on intra-query correlations. " + corrCtx.getCorrelations().size() + " correlation(s) to be applied");
        for (IntraQueryCorrelation correlation : corrCtx.getCorrelations()) {
            QueryPlanTreeTransformation.applyCorrelation(pCtx, corrCtx, correlation);
        }
    }
    return pCtx;
}