Search in sources :

Example 11 with GraphWalker

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

the class StatsOptimizer method transform.

@Override
public ParseContext transform(ParseContext pctx) throws SemanticException {
    if (pctx.getFetchTask() != null || !pctx.getQueryProperties().isQuery() || pctx.getQueryProperties().isAnalyzeRewrite() || pctx.getQueryProperties().isCTAS() || pctx.getLoadFileWork().size() > 1 || !pctx.getLoadTableWork().isEmpty() || // tables is being sampled and we can not optimize.
    !pctx.getNameToSplitSample().isEmpty()) {
        return pctx;
    }
    String TS = TableScanOperator.getOperatorName() + "%";
    String GBY = GroupByOperator.getOperatorName() + "%";
    String RS = ReduceSinkOperator.getOperatorName() + "%";
    String SEL = SelectOperator.getOperatorName() + "%";
    String FS = FileSinkOperator.getOperatorName() + "%";
    Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
    opRules.put(new RuleRegExp("R1", TS + SEL + GBY + RS + GBY + SEL + FS), new MetaDataProcessor(pctx));
    opRules.put(new RuleRegExp("R2", TS + SEL + GBY + RS + GBY + FS), new MetaDataProcessor(pctx));
    NodeProcessorCtx soProcCtx = new StatsOptimizerProcContext();
    Dispatcher disp = new DefaultRuleDispatcher(null, opRules, soProcCtx);
    GraphWalker ogw = new DefaultGraphWalker(disp);
    ArrayList<Node> topNodes = new ArrayList<Node>();
    topNodes.addAll(pctx.getTopOps().values());
    ogw.startWalking(topNodes, null);
    return pctx;
}
Also used : NodeProcessor(org.apache.hadoop.hive.ql.lib.NodeProcessor) 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) Dispatcher(org.apache.hadoop.hive.ql.lib.Dispatcher) DefaultRuleDispatcher(org.apache.hadoop.hive.ql.lib.DefaultRuleDispatcher) LinkedHashMap(java.util.LinkedHashMap) NodeProcessorCtx(org.apache.hadoop.hive.ql.lib.NodeProcessorCtx) Rule(org.apache.hadoop.hive.ql.lib.Rule) GraphWalker(org.apache.hadoop.hive.ql.lib.GraphWalker) DefaultGraphWalker(org.apache.hadoop.hive.ql.lib.DefaultGraphWalker)

Example 12 with GraphWalker

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

the class SkewJoinOptimizer 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 {
    Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
    opRules.put(new RuleRegExp("R1", "TS%.*RS%JOIN%"), getSkewJoinProc(pctx));
    SkewJoinOptProcCtx skewJoinOptProcCtx = new SkewJoinOptProcCtx(pctx);
    // The dispatcher fires the processor corresponding to the closest matching
    // rule and passes the context along
    Dispatcher disp = new DefaultRuleDispatcher(null, opRules, skewJoinOptProcCtx);
    GraphWalker 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);
    return pctx;
}
Also used : NodeProcessor(org.apache.hadoop.hive.ql.lib.NodeProcessor) 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) Dispatcher(org.apache.hadoop.hive.ql.lib.Dispatcher) DefaultRuleDispatcher(org.apache.hadoop.hive.ql.lib.DefaultRuleDispatcher) LinkedHashMap(java.util.LinkedHashMap) Rule(org.apache.hadoop.hive.ql.lib.Rule) GraphWalker(org.apache.hadoop.hive.ql.lib.GraphWalker) DefaultGraphWalker(org.apache.hadoop.hive.ql.lib.DefaultGraphWalker)

Example 13 with GraphWalker

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

the class SortedMergeBucketMapJoinOptimizer method getListOfRejectedJoins.

private void getListOfRejectedJoins(ParseContext pctx, SortBucketJoinProcCtx smbJoinContext) throws SemanticException {
    // Go through all joins - it should only contain selects and filters between
    // tablescan and join operators.
    Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
    opRules.put(new RuleRegExp("R1", JoinOperator.getOperatorName() + "%"), getCheckCandidateJoin());
    // The dispatcher fires the processor corresponding to the closest matching
    // rule and passes the context along
    Dispatcher disp = new DefaultRuleDispatcher(getDefaultProc(), opRules, smbJoinContext);
    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);
}
Also used : NodeProcessor(org.apache.hadoop.hive.ql.lib.NodeProcessor) 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) Rule(org.apache.hadoop.hive.ql.lib.Rule) Dispatcher(org.apache.hadoop.hive.ql.lib.Dispatcher) DefaultRuleDispatcher(org.apache.hadoop.hive.ql.lib.DefaultRuleDispatcher) GraphWalker(org.apache.hadoop.hive.ql.lib.GraphWalker) DefaultGraphWalker(org.apache.hadoop.hive.ql.lib.DefaultGraphWalker) LinkedHashMap(java.util.LinkedHashMap)

Example 14 with GraphWalker

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

the class SamplePruner 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 {
    // create a the context for walking operators
    SamplePrunerCtx samplePrunerCtx = new SamplePrunerCtx(pctx.getOpToSamplePruner());
    Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
    opRules.put(new RuleRegExp("R1", "(" + TableScanOperator.getOperatorName() + "%" + FilterOperator.getOperatorName() + "%" + FilterOperator.getOperatorName() + "%|" + TableScanOperator.getOperatorName() + "%" + FilterOperator.getOperatorName() + "%)"), getFilterProc());
    // The dispatcher fires the processor corresponding to the closest matching
    // rule and passes the context along
    Dispatcher disp = new DefaultRuleDispatcher(getDefaultProc(), opRules, samplePrunerCtx);
    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);
    return pctx;
}
Also used : NodeProcessor(org.apache.hadoop.hive.ql.lib.NodeProcessor) 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) Dispatcher(org.apache.hadoop.hive.ql.lib.Dispatcher) DefaultRuleDispatcher(org.apache.hadoop.hive.ql.lib.DefaultRuleDispatcher) LinkedHashMap(java.util.LinkedHashMap) Rule(org.apache.hadoop.hive.ql.lib.Rule) DefaultGraphWalker(org.apache.hadoop.hive.ql.lib.DefaultGraphWalker) GraphWalker(org.apache.hadoop.hive.ql.lib.GraphWalker)

Example 15 with GraphWalker

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

the class ReduceSinkDeDuplication method transform.

@Override
public ParseContext transform(ParseContext pctx) throws SemanticException {
    pGraphContext = pctx;
    // generate pruned column list for all relevant operators
    ReduceSinkDeduplicateProcCtx cppCtx = new ReduceSinkDeduplicateProcCtx(pGraphContext);
    // for auto convert map-joins, it not safe to dedup in here (todo)
    boolean mergeJoins = !pctx.getConf().getBoolVar(HIVECONVERTJOIN) && !pctx.getConf().getBoolVar(HIVECONVERTJOINNOCONDITIONALTASK) && !pctx.getConf().getBoolVar(ConfVars.HIVE_CONVERT_JOIN_BUCKET_MAPJOIN_TEZ) && !pctx.getConf().getBoolVar(ConfVars.HIVEDYNAMICPARTITIONHASHJOIN);
    // If multiple rules can be matched with same cost, last rule will be choosen as a processor
    // see DefaultRuleDispatcher#dispatch()
    Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
    opRules.put(new RuleRegExp("R1", RS + "%.*%" + RS + "%"), ReduceSinkDeduplicateProcFactory.getReducerReducerProc());
    opRules.put(new RuleRegExp("R2", RS + "%" + GBY + "%.*%" + RS + "%"), ReduceSinkDeduplicateProcFactory.getGroupbyReducerProc());
    if (mergeJoins) {
        opRules.put(new RuleRegExp("R3", JOIN + "%.*%" + RS + "%"), ReduceSinkDeduplicateProcFactory.getJoinReducerProc());
    }
    // TODO RS+JOIN
    // The dispatcher fires the processor corresponding to the closest matching
    // rule and passes the context along
    Dispatcher disp = new DefaultRuleDispatcher(ReduceSinkDeduplicateProcFactory.getDefaultProc(), opRules, cppCtx);
    GraphWalker ogw = new DefaultGraphWalker(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 : NodeProcessor(org.apache.hadoop.hive.ql.lib.NodeProcessor) 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) Dispatcher(org.apache.hadoop.hive.ql.lib.Dispatcher) DefaultRuleDispatcher(org.apache.hadoop.hive.ql.lib.DefaultRuleDispatcher) LinkedHashMap(java.util.LinkedHashMap) Rule(org.apache.hadoop.hive.ql.lib.Rule) DefaultGraphWalker(org.apache.hadoop.hive.ql.lib.DefaultGraphWalker) GraphWalker(org.apache.hadoop.hive.ql.lib.GraphWalker)

Aggregations

GraphWalker (org.apache.hadoop.hive.ql.lib.GraphWalker)70 Node (org.apache.hadoop.hive.ql.lib.Node)70 Dispatcher (org.apache.hadoop.hive.ql.lib.Dispatcher)68 ArrayList (java.util.ArrayList)67 DefaultRuleDispatcher (org.apache.hadoop.hive.ql.lib.DefaultRuleDispatcher)65 NodeProcessor (org.apache.hadoop.hive.ql.lib.NodeProcessor)57 Rule (org.apache.hadoop.hive.ql.lib.Rule)57 LinkedHashMap (java.util.LinkedHashMap)56 DefaultGraphWalker (org.apache.hadoop.hive.ql.lib.DefaultGraphWalker)56 RuleRegExp (org.apache.hadoop.hive.ql.lib.RuleRegExp)49 HashMap (java.util.HashMap)16 SemanticException (org.apache.hadoop.hive.ql.parse.SemanticException)12 ExprNodeColumnDesc (org.apache.hadoop.hive.ql.plan.ExprNodeColumnDesc)11 ExprNodeDesc (org.apache.hadoop.hive.ql.plan.ExprNodeDesc)11 ExprNodeGenericFuncDesc (org.apache.hadoop.hive.ql.plan.ExprNodeGenericFuncDesc)11 TypeRule (org.apache.hadoop.hive.ql.lib.TypeRule)10 ExprNodeConstantDesc (org.apache.hadoop.hive.ql.plan.ExprNodeConstantDesc)7 GenericUDFOPEqualOrGreaterThan (org.apache.hadoop.hive.ql.udf.generic.GenericUDFOPEqualOrGreaterThan)7 Test (org.junit.Test)7 List (java.util.List)6