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

Example 41 with DefaultRuleDispatcher

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

the class MapJoinProcessor method transform.

/**
   * Transform the query tree. For each join, check if it is a map-side join (user specified). If
   * yes, convert it to a map-side join.
   *
   * @param pactx
   *          current parse context
   */
@Override
public ParseContext transform(ParseContext pactx) throws SemanticException {
    List<MapJoinOperator> listMapJoinOps = new ArrayList<MapJoinOperator>();
    // traverse all the joins and convert them if necessary
    if (pactx.getJoinOps() != null) {
        Set<JoinOperator> joinMap = new HashSet<JoinOperator>();
        Set<MapJoinOperator> mapJoinMap = pactx.getMapJoinOps();
        if (mapJoinMap == null) {
            mapJoinMap = new HashSet<MapJoinOperator>();
            pactx.setMapJoinOps(mapJoinMap);
        }
        Iterator<JoinOperator> joinCtxIter = pactx.getJoinOps().iterator();
        while (joinCtxIter.hasNext()) {
            JoinOperator joinOp = joinCtxIter.next();
            int mapJoinPos = mapSideJoin(joinOp);
            if (mapJoinPos >= 0) {
                MapJoinOperator mapJoinOp = generateMapJoinOperator(pactx, joinOp, mapJoinPos);
                listMapJoinOps.add(mapJoinOp);
                mapJoinOp.getConf().setQBJoinTreeProps(joinOp.getConf());
                mapJoinMap.add(mapJoinOp);
            } else {
                joinOp.getConf().setQBJoinTreeProps(joinOp.getConf());
                joinMap.add(joinOp);
            }
        }
        // store the new joinContext
        pactx.setJoinOps(joinMap);
    }
    // Go over the list and find if a reducer is not needed
    List<AbstractMapJoinOperator<? extends MapJoinDesc>> listMapJoinOpsNoRed = new ArrayList<AbstractMapJoinOperator<? extends MapJoinDesc>>();
    // create a walker which walks the tree in a DFS manner while maintaining
    // the operator stack.
    // The dispatcher generates the plan from the operator tree
    Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
    opRules.put(new RuleRegExp("R0", MapJoinOperator.getOperatorName() + "%"), getCurrentMapJoin());
    opRules.put(new RuleRegExp("R1", MapJoinOperator.getOperatorName() + "%.*" + FileSinkOperator.getOperatorName() + "%"), getMapJoinFS());
    opRules.put(new RuleRegExp("R2", MapJoinOperator.getOperatorName() + "%.*" + ReduceSinkOperator.getOperatorName() + "%"), getMapJoinDefault());
    opRules.put(new RuleRegExp("R4", MapJoinOperator.getOperatorName() + "%.*" + UnionOperator.getOperatorName() + "%"), getMapJoinDefault());
    // The dispatcher fires the processor corresponding to the closest matching
    // rule and passes the context along
    Dispatcher disp = new DefaultRuleDispatcher(getDefault(), opRules, new MapJoinWalkerCtx(listMapJoinOpsNoRed, pactx));
    GraphWalker ogw = new GenMapRedWalker(disp);
    ArrayList<Node> topNodes = new ArrayList<Node>();
    topNodes.addAll(listMapJoinOps);
    ogw.startWalking(topNodes, null);
    pactx.setListMapJoinOpsNoReducer(listMapJoinOpsNoRed);
    return pactx;
}

Example 42 with DefaultRuleDispatcher

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

the class PartitionColumnsSeparator method transform.

@Override
public ParseContext transform(ParseContext pctx) throws SemanticException {
    // 1. Trigger transformation
    Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
    opRules.put(new RuleRegExp("R1", FilterOperator.getOperatorName() + "%"), new StructInTransformer());
    Dispatcher disp = new DefaultRuleDispatcher(null, opRules, null);
    GraphWalker ogw = new ForwardWalker(disp);
    List<Node> topNodes = new ArrayList<Node>();
    topNodes.addAll(pctx.getTopOps().values());
    ogw.startWalking(topNodes, null);
    return pctx;
}

Example 43 with DefaultRuleDispatcher

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

the class PrunerUtils method walkOperatorTree.

/**
   * Walk operator tree for pruner generation.
   *
   * @param pctx
   * @param opWalkerCtx
   * @param filterProc
   * @param defaultProc
   * @throws SemanticException
   */
public static void walkOperatorTree(ParseContext pctx, NodeProcessorCtx opWalkerCtx, NodeProcessor filterProc, NodeProcessor defaultProc) throws SemanticException {
    Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
    // Build regular expression for operator rule.
    // "(TS%FIL%)|(TS%FIL%FIL%)"
    String tsOprName = TableScanOperator.getOperatorName();
    String filtOprName = FilterOperator.getOperatorName();
    opRules.put(new RuleExactMatch("R1", new String[] { tsOprName, filtOprName, filtOprName }), filterProc);
    opRules.put(new RuleExactMatch("R2", new String[] { tsOprName, filtOprName }), filterProc);
    // The dispatcher fires the processor corresponding to the closest matching
    // rule and passes the context along
    Dispatcher disp = new DefaultRuleDispatcher(defaultProc, opRules, opWalkerCtx);
    GraphWalker ogw = new DefaultGraphWalker(disp);
    // Create a list of topop nodes
    ArrayList<Node> topNodes = new ArrayList<Node>();
    topNodes.addAll(pctx.getTopOps().values());
    ogw.startWalking(topNodes, null);
}

Example 44 with DefaultRuleDispatcher

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

the class ExprWalkerProcFactory method extractPushdownPreds.

/**
   * Extracts pushdown predicates from the given list of predicate expression.
   *
   * @param opContext
   *          operator context used for resolving column references
   * @param op
   *          operator of the predicates being processed
   * @param preds
   * @return The expression walker information
   * @throws SemanticException
   */
public static ExprWalkerInfo extractPushdownPreds(OpWalkerInfo opContext, Operator<? extends OperatorDesc> op, List<ExprNodeDesc> preds) throws SemanticException {
    // Create the walker, the rules dispatcher and the context.
    ExprWalkerInfo exprContext = new ExprWalkerInfo(op);
    // create a walker which walks the tree in a DFS manner while maintaining
    // the operator stack. The dispatcher
    // generates the plan from the operator tree
    Map<Rule, NodeProcessor> exprRules = new LinkedHashMap<Rule, NodeProcessor>();
    exprRules.put(new TypeRule(ExprNodeColumnDesc.class), getColumnProcessor());
    exprRules.put(new TypeRule(ExprNodeFieldDesc.class), getFieldProcessor());
    exprRules.put(new TypeRule(ExprNodeGenericFuncDesc.class), getGenericFuncProcessor());
    // The dispatcher fires the processor corresponding to the closest matching
    // rule and passes the context along
    Dispatcher disp = new DefaultRuleDispatcher(getDefaultExprProcessor(), exprRules, exprContext);
    GraphWalker egw = new DefaultGraphWalker(disp);
    List<Node> startNodes = new ArrayList<Node>();
    List<ExprNodeDesc> clonedPreds = new ArrayList<ExprNodeDesc>();
    for (ExprNodeDesc node : preds) {
        ExprNodeDesc clone = node.clone();
        clonedPreds.add(clone);
        exprContext.getNewToOldExprMap().put(clone, node);
    }
    startNodes.addAll(clonedPreds);
    egw.startWalking(startNodes, null);
    HiveConf conf = opContext.getParseContext().getConf();
    // check the root expression for final candidates
    for (ExprNodeDesc pred : clonedPreds) {
        extractFinalCandidates(pred, exprContext, conf);
    }
    return exprContext;
}

Example 45 with DefaultRuleDispatcher

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

the class SparkCompiler method generateTaskTreeHelper.

private void generateTaskTreeHelper(GenSparkProcContext procCtx, List<Node> topNodes) throws SemanticException {
    // create a walker which walks the tree in a DFS manner while maintaining
    // the operator stack. The dispatcher generates the plan from the operator tree
    Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
    GenSparkWork genSparkWork = new GenSparkWork(GenSparkUtils.getUtils());
    opRules.put(new RuleRegExp("Split Work - ReduceSink", ReduceSinkOperator.getOperatorName() + "%"), genSparkWork);
    opRules.put(new RuleRegExp("Split Work - SparkPartitionPruningSink", SparkPartitionPruningSinkOperator.getOperatorName() + "%"), genSparkWork);
    opRules.put(new TypeRule(MapJoinOperator.class), new SparkReduceSinkMapJoinProc());
    opRules.put(new RuleRegExp("Split Work + Move/Merge - FileSink", FileSinkOperator.getOperatorName() + "%"), new CompositeProcessor(new SparkFileSinkProcessor(), genSparkWork));
    opRules.put(new RuleRegExp("Handle Analyze Command", TableScanOperator.getOperatorName() + "%"), new SparkProcessAnalyzeTable(GenSparkUtils.getUtils()));
    opRules.put(new RuleRegExp("Remember union", UnionOperator.getOperatorName() + "%"), new NodeProcessor() {

        @Override
        public Object process(Node n, Stack<Node> s, NodeProcessorCtx procCtx, Object... os) throws SemanticException {
            GenSparkProcContext context = (GenSparkProcContext) procCtx;
            UnionOperator union = (UnionOperator) n;
            // simply need to remember that we've seen a union.
            context.currentUnionOperators.add(union);
            return null;
        }
    });
    /**
     *  SMB join case:   (Big)   (Small)  (Small)
     *                     TS       TS       TS
     *                      \       |       /
     *                       \      DS     DS
     *                         \   |    /
     *                         SMBJoinOP
     *
     * Some of the other processors are expecting only one traversal beyond SMBJoinOp.
     * We need to traverse from the big-table path only, and stop traversing on the
     * small-table path once we reach SMBJoinOp.
     * Also add some SMB join information to the context, so we can properly annotate
     * the MapWork later on.
     */
    opRules.put(new TypeRule(SMBMapJoinOperator.class), new NodeProcessor() {

        @Override
        public Object process(Node currNode, Stack<Node> stack, NodeProcessorCtx procCtx, Object... os) throws SemanticException {
            GenSparkProcContext context = (GenSparkProcContext) procCtx;
            SMBMapJoinOperator currSmbNode = (SMBMapJoinOperator) currNode;
            SparkSMBMapJoinInfo smbMapJoinCtx = context.smbMapJoinCtxMap.get(currSmbNode);
            if (smbMapJoinCtx == null) {
                smbMapJoinCtx = new SparkSMBMapJoinInfo();
                context.smbMapJoinCtxMap.put(currSmbNode, smbMapJoinCtx);
            }
            for (Node stackNode : stack) {
                if (stackNode instanceof DummyStoreOperator) {
                    //If coming from small-table side, do some book-keeping, and skip traversal.
                    smbMapJoinCtx.smallTableRootOps.add(context.currentRootOperator);
                    return true;
                }
            }
            //If coming from big-table side, do some book-keeping, and continue traversal
            smbMapJoinCtx.bigTableRootOp = context.currentRootOperator;
            return false;
        }
    });
    // The dispatcher fires the processor corresponding to the closest matching
    // rule and passes the context along
    Dispatcher disp = new DefaultRuleDispatcher(null, opRules, procCtx);
    GraphWalker ogw = new GenSparkWorkWalker(disp, procCtx);
    ogw.startWalking(topNodes, null);
}