Examples of PlanOptimizer

edu.brown.optimizer.PlanOptimizer
@author pavlo @author sw47
org.apache.pig.experimental.plan.optimizer.PlanOptimizer
The core class of the optimizer. The basic design of this class is that it is provided a list of RuleSets. RuleSets represent all of the optimizer rules that can be run together. The rules in the RuleSet will be run repeatedly until either no rule in the RuleSet passes check and calls transform or until maxIter iterations (default 500) has been made over the RuleSet. Then the next RuleSet will be moved to. Once finished, a given RuleSet is never returned to. Each rule is has two parts: a pattern and and associated transformer. Transformers have two important functions: check(), and transform(). The pattern describes a pattern of node types that the optimizer will look to match. If that match is found anywhere in the plan, then check() will be called. check() allows the rule to look more in depth at the matched pattern and decide whether the rule should be run or not. For example, one might design a rule to push filters above join that would look for the pattern filter(join) (meaning a filter followed by a join). But only certain types of filters can be pushed. The check() function would need to decide whether the filter that it found was pushable or not. If check() returns true, the rule is said to have matched, and transform() is then called. This function is responsible for making changes in the logical plan. Once transform is complete PlanPatcher.patchUp will be called to do any necessary cleanup in the plan, such as resetting schemas, etc.
org.apache.pig.newplan.optimizer.PlanOptimizer
The core class of the optimizer. The basic design of this class is that it is provided a list of RuleSets. RuleSets represent all of the optimizer rules that can be run together. The rules in the RuleSet will be run repeatedly until either no rule in the RuleSet passes check and calls transform or until maxIter iterations (default 500) has been made over the RuleSet. Then the next RuleSet will be moved to. Once finished, a given RuleSet is never returned to. Each rule is has two parts: a pattern and and associated transformer. Transformers have two important functions: check(), and transform(). The pattern describes a pattern of node types that the optimizer will look to match. If that match is found anywhere in the plan, then check() will be called. check() allows the rule to look more in depth at the matched pattern and decide whether the rule should be run or not. For example, one might design a rule to push filters above join that would look for the pattern filter(join) (meaning a filter followed by a join). But only certain types of filters can be pushed. The check() function would need to decide whether the filter that it found was pushable or not. If check() returns true, the rule is said to have matched, and transform() is then called. This function is responsible for making changes in the logical plan. Once transform is complete PlanPatcher.patchUp will be called to do any necessary cleanup in the plan, such as resetting schemas, etc.

Examples of org.apache.pig.newplan.optimizer.PlanOptimizer

        LogicalPlanTester lpt = new LogicalPlanTester(pc);
        lpt.buildPlan("b = filter (load 'd.txt' as (id:int, v1, v2)) by ((v1 is not null) AND (id == 1) AND (1 == 1));");
        org.apache.pig.impl.logicalLayer.LogicalPlan plan = lpt.buildPlan("store b into 'empty';");
        LogicalPlan newLogicalPlan = migratePlan(plan);


        PlanOptimizer optimizer = new MyPlanOptimizer(newLogicalPlan, 10);
        optimizer.optimize();


        lpt = new LogicalPlanTester(pc);
        lpt.buildPlan("b = filter (load 'd.txt' as (id:int, v1, v2)) by ((v1 is not null) AND (id == 1));");
        plan = lpt.buildPlan("store b into 'empty';");
        LogicalPlan expected = migratePlan(plan);


        assertTrue(expected.isEqual(newLogicalPlan));


        // boolean constant elimination: OR
        lpt = new LogicalPlanTester(pc);
        lpt.buildPlan("b = filter (load 'd.txt' as (id:int, v1, v2)) by (((v1 is not null) AND (id == 1)) OR (1 == 0));");
        plan = lpt.buildPlan("store b into 'empty';");
        newLogicalPlan = migratePlan(plan);


        optimizer = new MyPlanOptimizer(newLogicalPlan, 10);
        optimizer.optimize();


        lpt = new LogicalPlanTester(pc);
        lpt.buildPlan("b = filter (load 'd.txt' as (id:int, v1, v2)) by ((v1 is not null) AND (id == 1));");
        plan = lpt.buildPlan("store b into 'empty';");
        expected = migratePlan(plan);


        assertTrue(expected.isEqual(newLogicalPlan));
        
        // the mirror case of the above
        lpt = new LogicalPlanTester(pc);
        lpt.buildPlan("b = filter (load 'd.txt' as (id:int, v1, v2)) by ((1 == 0) OR ((v1 is not null) AND (id == 1)));");
        plan = lpt.buildPlan("store b into 'empty';");
        newLogicalPlan = migratePlan(plan);


        optimizer = new MyPlanOptimizer(newLogicalPlan, 10);
        optimizer.optimize();


        lpt = new LogicalPlanTester(pc);
        lpt.buildPlan("b = filter (load 'd.txt' as (id:int, v1, v2)) by ((v1 is not null) AND (id == 1));");
        plan = lpt.buildPlan("store b into 'empty';");
        expected = migratePlan(plan);

View Full Code Here

Examples of org.apache.pig.newplan.optimizer.PlanOptimizer

        lpt.buildPlan("b = filter (load 'd.txt' as (a:chararray, b:long, c:map[], d:chararray, e:chararray)) by a == 'v' and b == 117L and c#'p1' == 'h' and c#'p2' == 'to' and ((d is not null and d != '') or (e is not null and e != ''));"); 


        org.apache.pig.impl.logicalLayer.LogicalPlan plan = lpt.buildPlan("store b into 'empty';");
        LogicalPlan newLogicalPlan = migratePlan(plan);


        PlanOptimizer optimizer = new MyPlanOptimizer(newLogicalPlan, 10);
        optimizer.optimize();


        lpt = new LogicalPlanTester(pc);
        lpt.buildPlan("b = filter (load 'd.txt' as (a:chararray, b:long, c:map[], d:chararray, e:chararray)) by a == 'v' and b == 117L and c#'p1' == 'h' and c#'p2' == 'to' and ((d is not null and d != '') or (e is not null and e != ''));"); 
        plan = lpt.buildPlan("store b into 'empty';");
        LogicalPlan expected = migratePlan(plan);


        assertTrue(expected.isEqual(newLogicalPlan));


        // mirror of the above
        lpt.buildPlan("b = filter (load 'd.txt' as (a:chararray, b:long, c:map[], d:chararray, e:chararray)) by ((d is not null and d != '') or (e is not null and e != '')) and a == 'v' and b == 117L and c#'p1' == 'h' and c#'p2' == 'to';"); 


        plan = lpt.buildPlan("store b into 'empty';");
        newLogicalPlan = migratePlan(plan);


        optimizer = new MyPlanOptimizer(newLogicalPlan, 10);
        optimizer.optimize();


        lpt = new LogicalPlanTester(pc);
        lpt.buildPlan("b = filter (load 'd.txt' as (a:chararray, b:long, c:map[], d:chararray, e:chararray)) by ((d is not null and d != '') or (e is not null and e != '')) and a == 'v' and b == 117L and c#'p1' == 'h' and c#'p2' == 'to';"); 
        plan = lpt.buildPlan("store b into 'empty';");
        expected = migratePlan(plan);

View Full Code Here

Examples of org.apache.pig.newplan.optimizer.PlanOptimizer

        lpt.buildPlan(query); 


        org.apache.pig.impl.logicalLayer.LogicalPlan plan = lpt.buildPlan("store b into 'empty';");
        LogicalPlan newLogicalPlan = migratePlan(plan);


        PlanOptimizer optimizer = new MyPlanOptimizer(newLogicalPlan, 10);
        optimizer.optimize();


        lpt = new LogicalPlanTester(pc);
        lpt.buildPlan(query); 
        plan = lpt.buildPlan("store b into 'empty';");
        LogicalPlan expected = migratePlan(plan);

View Full Code Here

Examples of org.apache.pig.newplan.optimizer.PlanOptimizer

        lpt.buildPlan(query); 


        org.apache.pig.impl.logicalLayer.LogicalPlan plan = lpt.buildPlan("store b into 'empty';");
        LogicalPlan newLogicalPlan = migratePlan(plan);


        PlanOptimizer optimizer = new MyPlanOptimizer(newLogicalPlan, 10);
        optimizer.optimize();


        lpt = new LogicalPlanTester(pc);
        lpt.buildPlan(query); 
        plan = lpt.buildPlan("store b into 'empty';");
        LogicalPlan expected = migratePlan(plan);

View Full Code Here

Examples of org.apache.pig.newplan.optimizer.PlanOptimizer

        int forEachCount1 = getForEachOperatorCount( newLogicalPlan );
        int outputExprCount1 = getOutputExprCount( newLogicalPlan );
        LOForEach foreach1 = getForEachOperator( newLogicalPlan );
        Assert.assertTrue( foreach1.getAlias().equals( "C" ) );
               
        PlanOptimizer optimizer = new MyPlanOptimizer( newLogicalPlan, 3 );
        optimizer.optimize();
        
        int forEachCount2 = getForEachOperatorCount( newLogicalPlan );
        Assert.assertEquals( 1, forEachCount1 - forEachCount2 );
        int outputExprCount2 = getOutputExprCount( newLogicalPlan );
        Assert.assertTrue( outputExprCount1 == outputExprCount2 );

View Full Code Here

Examples of org.apache.pig.newplan.optimizer.PlanOptimizer

        int forEachCount1 = getForEachOperatorCount( newLogicalPlan );
        int outputExprCount1 = getOutputExprCount( newLogicalPlan );
        LOForEach foreach1 = getForEachOperator( newLogicalPlan );
        Assert.assertTrue( foreach1.getAlias().equals( "C" ) );
               
        PlanOptimizer optimizer = new MyPlanOptimizer( newLogicalPlan, 3 );
        optimizer.optimize();
        
        int forEachCount2 = getForEachOperatorCount( newLogicalPlan );
        // The number of FOREACHes didn't change because one is genereated because of type cast and
        // one is reduced because of the merge.
        Assert.assertEquals( 0, forEachCount1 - forEachCount2 );

View Full Code Here

Examples of org.apache.pig.newplan.optimizer.PlanOptimizer

        Operator store = newLogicalPlan.getSinks().get(0);
        int forEachCount1 = getForEachOperatorCount( newLogicalPlan );
        LOForEach foreach1 = (LOForEach)newLogicalPlan.getPredecessors(store).get(0);
        Assert.assertTrue( foreach1.getAlias().equals( "D" ) );
               
        PlanOptimizer optimizer = new MyPlanOptimizer( newLogicalPlan, 3 );
        optimizer.optimize();
        
        int forEachCount2 = getForEachOperatorCount( newLogicalPlan );
        // The number of FOREACHes didn't change because one is genereated because of type cast and
        // one is reduced because of the merge.
        Assert.assertEquals( 1, forEachCount1 - forEachCount2 );

View Full Code Here

Examples of org.apache.pig.newplan.optimizer.PlanOptimizer

        org.apache.pig.impl.logicalLayer.LogicalPlan plan = lpt.buildPlan( "store C into 'empty';" );  
        LogicalPlan newLogicalPlan = migratePlan( plan );
        
        int forEachCount1 = getForEachOperatorCount( newLogicalPlan );
               
        PlanOptimizer optimizer = new MyPlanOptimizer( newLogicalPlan, 3 );
        optimizer.optimize();
        int forEachCount2 = getForEachOperatorCount( newLogicalPlan );
        
        // Actually MergeForEach optimization is happening here. A new foreach will be inserted after A because
        // of typ casting. The inserted one and the one in B can be merged due to this optimization. However, 
        // the plan cannot be further optimized because C has inner plan.

View Full Code Here

Examples of org.apache.pig.newplan.optimizer.PlanOptimizer

        org.apache.pig.impl.logicalLayer.LogicalPlan plan = lpt.buildPlan( "store C into 'empty';" );  
        LogicalPlan newLogicalPlan = migratePlan( plan );
        
        int forEachCount1 = getForEachOperatorCount( newLogicalPlan );
               
        PlanOptimizer optimizer = new MyPlanOptimizer( newLogicalPlan, 3 );
        optimizer.optimize();
        
        int forEachCount2 = getForEachOperatorCount( newLogicalPlan );
        Assert.assertEquals( 2, forEachCount1 );
        Assert.assertEquals( 2, forEachCount2 );
    }

View Full Code Here

Examples of org.apache.pig.newplan.optimizer.PlanOptimizer

        LogicalPlan newLogicalPlan = migratePlan( plan );


        int forEachCount1 = getForEachOperatorCount( newLogicalPlan );
        int outputExprCount1 = getOutputExprCount( newLogicalPlan );
               
        PlanOptimizer optimizer = new MyPlanOptimizer( newLogicalPlan, 3 );
        optimizer.optimize();
        
        int forEachCount2 = getForEachOperatorCount( newLogicalPlan );
        Assert.assertEquals( 1, forEachCount1 - forEachCount2 );
        int outputExprCount2 = getOutputExprCount( newLogicalPlan );
        Assert.assertTrue( outputExprCount1 == outputExprCount2 );

View Full Code Here

0 1 2 3 4 5

TOP

All source code are property of their respective owners. Java is a trademark of Sun Microsystems, Inc and owned by ORACLE Inc. Contact coftware#gmail.com.