Search in sources :

Example 56 with SemanticDispatcher

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

the class LlapDecider method resolve.

@Override
public PhysicalContext resolve(PhysicalContext pctx) throws SemanticException {
    this.conf = pctx.getConf();
    this.mode = LlapMode.valueOf(HiveConf.getVar(conf, HiveConf.ConfVars.LLAP_EXECUTION_MODE));
    Preconditions.checkState(this.mode != null, "Unrecognized LLAP mode configuration: " + HiveConf.getVar(conf, HiveConf.ConfVars.LLAP_EXECUTION_MODE));
    LOG.info("llap mode: " + this.mode);
    if (mode == none) {
        LOG.info("LLAP disabled.");
        return pctx;
    }
    // create dispatcher and graph walker
    SemanticDispatcher disp = new LlapDecisionDispatcher(pctx, mode);
    TaskGraphWalker ogw = new TaskGraphWalker(disp);
    // get all the tasks nodes from root task
    ArrayList<Node> topNodes = new ArrayList<Node>();
    topNodes.addAll(pctx.getRootTasks());
    // begin to walk through the task tree.
    ogw.startWalking(topNodes, null);
    return pctx;
}
Also used : TaskGraphWalker(org.apache.hadoop.hive.ql.lib.TaskGraphWalker) Node(org.apache.hadoop.hive.ql.lib.Node) SemanticDispatcher(org.apache.hadoop.hive.ql.lib.SemanticDispatcher) ArrayList(java.util.ArrayList)

Example 57 with SemanticDispatcher

use of org.apache.hadoop.hive.ql.lib.SemanticDispatcher 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;
}
Also used : SemanticRule(org.apache.hadoop.hive.ql.lib.SemanticRule) DefaultRuleDispatcher(org.apache.hadoop.hive.ql.lib.DefaultRuleDispatcher) DefaultGraphWalker(org.apache.hadoop.hive.ql.lib.DefaultGraphWalker) Node(org.apache.hadoop.hive.ql.lib.Node) RuleRegExp(org.apache.hadoop.hive.ql.lib.RuleRegExp) ArrayList(java.util.ArrayList) SemanticGraphWalker(org.apache.hadoop.hive.ql.lib.SemanticGraphWalker) LinkedHashMap(java.util.LinkedHashMap) SemanticDispatcher(org.apache.hadoop.hive.ql.lib.SemanticDispatcher) SemanticNodeProcessor(org.apache.hadoop.hive.ql.lib.SemanticNodeProcessor)

Example 58 with SemanticDispatcher

use of org.apache.hadoop.hive.ql.lib.SemanticDispatcher 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);
}
Also used : SemanticRule(org.apache.hadoop.hive.ql.lib.SemanticRule) DefaultRuleDispatcher(org.apache.hadoop.hive.ql.lib.DefaultRuleDispatcher) HashMap(java.util.HashMap) LinkedHashMap(java.util.LinkedHashMap) Node(org.apache.hadoop.hive.ql.lib.Node) RuleRegExp(org.apache.hadoop.hive.ql.lib.RuleRegExp) ArrayList(java.util.ArrayList) ExprNodeGenericFuncDesc(org.apache.hadoop.hive.ql.plan.ExprNodeGenericFuncDesc) SemanticGraphWalker(org.apache.hadoop.hive.ql.lib.SemanticGraphWalker) LinkedHashMap(java.util.LinkedHashMap) ExpressionWalker(org.apache.hadoop.hive.ql.lib.ExpressionWalker) ExprNodeFieldDesc(org.apache.hadoop.hive.ql.plan.ExprNodeFieldDesc) ExprNodeColumnDesc(org.apache.hadoop.hive.ql.plan.ExprNodeColumnDesc) SemanticDispatcher(org.apache.hadoop.hive.ql.lib.SemanticDispatcher) SemanticNodeProcessor(org.apache.hadoop.hive.ql.lib.SemanticNodeProcessor)

Example 59 with SemanticDispatcher

use of org.apache.hadoop.hive.ql.lib.SemanticDispatcher 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;
}
Also used : SemanticRule(org.apache.hadoop.hive.ql.lib.SemanticRule) ForwardWalker(org.apache.hadoop.hive.ql.lib.ForwardWalker) DefaultRuleDispatcher(org.apache.hadoop.hive.ql.lib.DefaultRuleDispatcher) Node(org.apache.hadoop.hive.ql.lib.Node) RuleRegExp(org.apache.hadoop.hive.ql.lib.RuleRegExp) ArrayList(java.util.ArrayList) SemanticGraphWalker(org.apache.hadoop.hive.ql.lib.SemanticGraphWalker) LinkedHashMap(java.util.LinkedHashMap) SemanticDispatcher(org.apache.hadoop.hive.ql.lib.SemanticDispatcher) SemanticNodeProcessor(org.apache.hadoop.hive.ql.lib.SemanticNodeProcessor)

Example 60 with SemanticDispatcher

use of org.apache.hadoop.hive.ql.lib.SemanticDispatcher 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;
}
Also used : SemanticRule(org.apache.hadoop.hive.ql.lib.SemanticRule) DefaultRuleDispatcher(org.apache.hadoop.hive.ql.lib.DefaultRuleDispatcher) Node(org.apache.hadoop.hive.ql.lib.Node) RuleRegExp(org.apache.hadoop.hive.ql.lib.RuleRegExp) ArrayList(java.util.ArrayList) SemanticGraphWalker(org.apache.hadoop.hive.ql.lib.SemanticGraphWalker) Index(org.apache.hadoop.hive.ql.optimizer.lineage.LineageCtx.Index) LinkedHashMap(java.util.LinkedHashMap) SemanticDispatcher(org.apache.hadoop.hive.ql.lib.SemanticDispatcher) SemanticNodeProcessor(org.apache.hadoop.hive.ql.lib.SemanticNodeProcessor) LevelOrderWalker(org.apache.hadoop.hive.ql.lib.LevelOrderWalker)

Aggregations

SemanticDispatcher (org.apache.hadoop.hive.ql.lib.SemanticDispatcher)94 Node (org.apache.hadoop.hive.ql.lib.Node)92 ArrayList (java.util.ArrayList)84 SemanticGraphWalker (org.apache.hadoop.hive.ql.lib.SemanticGraphWalker)84 DefaultRuleDispatcher (org.apache.hadoop.hive.ql.lib.DefaultRuleDispatcher)77 SemanticNodeProcessor (org.apache.hadoop.hive.ql.lib.SemanticNodeProcessor)65 LinkedHashMap (java.util.LinkedHashMap)64 SemanticRule (org.apache.hadoop.hive.ql.lib.SemanticRule)64 RuleRegExp (org.apache.hadoop.hive.ql.lib.RuleRegExp)55 DefaultGraphWalker (org.apache.hadoop.hive.ql.lib.DefaultGraphWalker)52 HashMap (java.util.HashMap)19 SemanticException (org.apache.hadoop.hive.ql.parse.SemanticException)15 ExprNodeGenericFuncDesc (org.apache.hadoop.hive.ql.plan.ExprNodeGenericFuncDesc)14 ExprNodeColumnDesc (org.apache.hadoop.hive.ql.plan.ExprNodeColumnDesc)13 ExprNodeDesc (org.apache.hadoop.hive.ql.plan.ExprNodeDesc)13 MapJoinOperator (org.apache.hadoop.hive.ql.exec.MapJoinOperator)10 ReduceSinkOperator (org.apache.hadoop.hive.ql.exec.ReduceSinkOperator)10 ExprNodeConstantDesc (org.apache.hadoop.hive.ql.plan.ExprNodeConstantDesc)10 Test (org.junit.Test)10 List (java.util.List)9