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

Example 56 with SemanticGraphWalker

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

the class PcrExprProcFactory method walkExprTree.

/**
 * Remove partition conditions when necessary from the the expression tree.
 *
 * @param tabAlias
 *          the table alias
 * @param parts
 *          the list of all pruned partitions for the table
 * @param vcs
 *          virtual columns referenced
 * @param pred
 *          expression tree of the target filter operator
 * @return the node information of the root expression
 * @throws SemanticException
 */
public static NodeInfoWrapper walkExprTree(String tabAlias, ArrayList<Partition> parts, List<VirtualColumn> vcs, ExprNodeDesc pred) throws SemanticException {
    // Create the walker, the rules dispatcher and the context.
    PcrExprProcCtx pprCtx = new PcrExprProcCtx(tabAlias, parts, vcs);
    Map<SemanticRule, SemanticNodeProcessor> exprRules = new LinkedHashMap<SemanticRule, SemanticNodeProcessor>();
    exprRules.put(new RuleRegExp("R1", ExprNodeColumnDesc.class.getName() + "%"), getColumnProcessor());
    exprRules.put(new RuleRegExp("R2", ExprNodeFieldDesc.class.getName() + "%"), getFieldProcessor());
    exprRules.put(new RuleRegExp("R5", ExprNodeGenericFuncDesc.class.getName() + "%"), getGenericFuncProcessor());
    // The dispatcher fires the processor corresponding to the closest matching
    // rule and passes the context along
    SemanticDispatcher disp = new DefaultRuleDispatcher(getDefaultExprProcessor(), exprRules, pprCtx);
    SemanticGraphWalker egw = new ExpressionWalker(disp);
    List<Node> startNodes = new ArrayList<Node>();
    startNodes.add(pred);
    HashMap<Node, Object> outputMap = new HashMap<Node, Object>();
    egw.startWalking(startNodes, outputMap);
    // Return the wrapper of the root node
    return (NodeInfoWrapper) outputMap.get(pred);
}

Example 57 with SemanticGraphWalker

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

the class ReduceSinkJoinDeDuplication method transform.

@Override
public ParseContext transform(ParseContext pctx) throws SemanticException {
    pGraphContext = pctx;
    ReduceSinkJoinDeDuplicateProcCtx cppCtx = new ReduceSinkJoinDeDuplicateProcCtx(pGraphContext);
    Map<SemanticRule, SemanticNodeProcessor> opRules = new LinkedHashMap<SemanticRule, SemanticNodeProcessor>();
    opRules.put(new RuleRegExp("R1", ReduceSinkOperator.getOperatorName() + "%"), ReduceSinkJoinDeDuplicateProcFactory.getReducerMapJoinProc());
    SemanticDispatcher disp = new DefaultRuleDispatcher(ReduceSinkJoinDeDuplicateProcFactory.getDefaultProc(), opRules, cppCtx);
    SemanticGraphWalker ogw = new ForwardWalker(disp);
    // Create a list of topop nodes
    ArrayList<Node> topNodes = new ArrayList<Node>();
    topNodes.addAll(pGraphContext.getTopOps().values());
    ogw.startWalking(topNodes, null);
    return pGraphContext;
}

Example 58 with SemanticGraphWalker

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

the class Generator method transform.

/* (non-Javadoc)
   * @see org.apache.hadoop.hive.ql.optimizer.Transform#transform(org.apache.hadoop.hive.ql.parse.ParseContext)
   */
@Override
public ParseContext transform(ParseContext pctx) throws SemanticException {
    if (hooks != null && hooks.contains(ATLAS_HOOK_CLASSNAME)) {
        // Atlas would be interested in lineage information for insert,load,create etc.
        if (!pctx.getQueryProperties().isCTAS() && !pctx.getQueryProperties().isMaterializedView() && pctx.getQueryProperties().isQuery() && pctx.getCreateTable() == null && pctx.getCreateViewDesc() == null && (pctx.getLoadTableWork() == null || pctx.getLoadTableWork().isEmpty())) {
            LOG.debug("Not evaluating lineage");
            return pctx;
        }
    }
    Index index = pctx.getQueryState().getLineageState().getIndex();
    if (index == null) {
        index = new Index();
    }
    long sTime = System.currentTimeMillis();
    // Create the lineage context
    LineageCtx lCtx = new LineageCtx(pctx, index);
    Map<SemanticRule, SemanticNodeProcessor> opRules = new LinkedHashMap<SemanticRule, SemanticNodeProcessor>();
    opRules.put(new RuleRegExp("R1", TableScanOperator.getOperatorName() + "%"), OpProcFactory.getTSProc());
    opRules.put(new RuleRegExp("R2", ScriptOperator.getOperatorName() + "%"), OpProcFactory.getTransformProc());
    opRules.put(new RuleRegExp("R3", UDTFOperator.getOperatorName() + "%"), OpProcFactory.getTransformProc());
    opRules.put(new RuleRegExp("R4", SelectOperator.getOperatorName() + "%"), OpProcFactory.getSelProc());
    opRules.put(new RuleRegExp("R5", GroupByOperator.getOperatorName() + "%"), OpProcFactory.getGroupByProc());
    opRules.put(new RuleRegExp("R6", UnionOperator.getOperatorName() + "%"), OpProcFactory.getUnionProc());
    opRules.put(new RuleRegExp("R7", CommonJoinOperator.getOperatorName() + "%|" + MapJoinOperator.getOperatorName() + "%"), OpProcFactory.getJoinProc());
    opRules.put(new RuleRegExp("R8", ReduceSinkOperator.getOperatorName() + "%"), OpProcFactory.getReduceSinkProc());
    opRules.put(new RuleRegExp("R9", LateralViewJoinOperator.getOperatorName() + "%"), OpProcFactory.getLateralViewJoinProc());
    opRules.put(new RuleRegExp("R10", PTFOperator.getOperatorName() + "%"), OpProcFactory.getTransformProc());
    opRules.put(new RuleRegExp("R11", FilterOperator.getOperatorName() + "%"), OpProcFactory.getFilterProc());
    // The dispatcher fires the processor corresponding to the closest matching rule and passes the context along
    SemanticDispatcher disp = new DefaultRuleDispatcher(OpProcFactory.getDefaultProc(), opRules, lCtx);
    SemanticGraphWalker ogw = new LevelOrderWalker(disp, 2);
    // Create a list of topop nodes
    ArrayList<Node> topNodes = new ArrayList<Node>();
    topNodes.addAll(pctx.getTopOps().values());
    ogw.startWalking(topNodes, null);
    LOG.debug("Time taken for lineage transform={}", (System.currentTimeMillis() - sTime));
    return pctx;
}

Example 59 with SemanticGraphWalker

use of org.apache.hadoop.hive.ql.lib.SemanticGraphWalker 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<SemanticRule, SemanticNodeProcessor> opRules = new LinkedHashMap<SemanticRule, SemanticNodeProcessor>();
    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 SemanticNodeProcessor() {

        @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 SemanticNodeProcessor() {

        @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
    SemanticDispatcher disp = new DefaultRuleDispatcher(null, opRules, procCtx);
    SemanticGraphWalker ogw = new GenSparkWorkWalker(disp, procCtx);
    ogw.startWalking(topNodes, null);
}

Example 60 with SemanticGraphWalker

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

the class OperatorHealthCheckerHook method walkTree.

private void walkTree(Collection<Node> rootOps) throws SemanticException {
    if (rootOps.isEmpty()) {
        return;
    }
    SemanticDispatcher disp = new DefaultRuleDispatcher(new OperatorHealthCheckerProcessor(), new HashMap<>(), null);
    SemanticGraphWalker ogw = new DefaultGraphWalker(disp);
    HashMap<Node, Object> nodeOutput = new HashMap<Node, Object>();
    ogw.startWalking(rootOps, nodeOutput);
}