Calcite RelNode和RexNode 介绍

SQL 数据处理流程

RelNode

RelNode 是关系表达式，主要描述处理数据的关系，比如何sort ，join,投影，过滤，扫描

LogicalTableScan,扫描整表

// SQL 表示
select *
from
test_user
 
 // 关系表达式生成
 RelNode root = builder.scan("test_user").build();
 
 //实际生成的逻辑计划
 LogicalTableScan(table=[[test_user]])
1
2
3
4
5
6
7
8
9
10

LogicalSort 排序

// SQL 表示
select id,name
from test_user
group  by id
 
 // 关系表达式生成
 RelNode root =
        builder
                .scan("test_user").sort(0)
                .build();
                
 //实际生成的逻辑计划               
LogicalSort(sort0=[$0], dir0=[ASC])
  LogicalTableScan(table=[[test_user]])             
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

LogicalProject 投影

// SQL 表示
select id
from test_user

 // 关系表达式生成
RelNode root = builder.scan("test_user").project(builder.field("id")).build();

 //实际生成的逻辑计划
LogicalProject(id=[$0])
  LogicalTableScan(table=[[test_user]])
1
2
3
4
5
6
7
8
9
10

LogicalFilter 过滤

// SQL 表示
select * from test_user where id=10

// 关系表达式生成
RelNode root = builder.scan("test_user").filter(builder.equals(builder.field(0),builder.literal(10))).build();

//实际生成的逻辑计划
LogicalFilter(condition=[=($0, 10)])
  LogicalTableScan(table=[[test_user]])
1
2
3
4
5
6
7
8
9

LogicalJoin 双表join

// SQL 表示
select * from test_user join test_address 
on  test_user.id=test_address.id

// 关系表达式生成
RelNode root = builder.scan("test_user")        
.scan("test_address").join(JoinRelType.INNER, builder.field("id"))    
    .build();

//实际生成的逻辑计划
LogicalJoin(condition=[$0], joinType=[inner])
  LogicalTableScan(table=[[test_user]])
  LogicalTableScan(table=[[test_address]])
1
2
3
4
5
6
7
8
9
10
11
12
13

LogicalMinus 差集

// 关系表达式生成
RelNode root = builder.scan("test_user")
        .scan("test_address").minus(true)
        .build();
   //实际生成的逻辑计划     
 LogicalMinus(all=[true])
   LogicalTableScan(table=[[test_user]])  
   LogicalTableScan(table=[[test_address]])
LogicalMinus  交集

// 关系表达式生成
RelNode root = builder.scan("test_user")
        .scan("test_address").intersect(true)
        .build();
   //实际生成的逻辑计划     
LogicalIntersect(all=[true])
  LogicalTableScan(table=[[test_user]])
  LogicalTableScan(table=[[test_address]])
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18

LogicalUnion union 并集

// SQL 表示
select id,name from test_user
union all 
select id,address as name from
test_address

// 关系表达式生成
RelNode root = builder.scan("test_user")
        .scan("test_address").union(true)
        .build();
  //实际生成的逻辑计划 
LogicalUnion(all=[true]) 
 LogicalTableScan(table=[[test_user]]) 
 LogicalTableScan(table=[[test_address]])
        
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

LogicalAggregate 分组操作

// SQL 表示
select 
id,max(name)
from
test_user
group by id

// 关系表达式生成
RelNode root = builder.scan("test_user").aggregate(builder.groupKey("id"), builder.max(builder.field("name")))
        .build();
        
     //实际生成的逻辑计划   
LogicalAggregate(group=[{0}], agg#0=[MAX($1)])
  LogicalTableScan(table=[[test_user]])
1
2
3
4
5
6
7
8
9
10
11
12
13
14

ReXNode

RexNode 主要表示的对数据行的处理

RexLiteral 常量表达式

//SQL 表示
select * from user where id=1
1 就是常量
builder.literal(10)

RelNode root = builder.scan("test_user").filter(builder.equals(builder.field("id"),builder.literal(10)))
        .build();
        
LogicalFilter(condition=[=($0, 10)])
  LogicalTableScan(table=[[test_user]])
1
2
3
4
5
6
7
8
9
10

RexInputRef 输入引用符

//SQL 表示
select id from test_user

RelNode root = builder.scan("test_user").project(builder.field(0))
        .build();

//这里builder.filed(0) 返回的就是第一个元素
//逻辑计划
LogicalProject(id=[$0])
  LogicalTableScan(table=[[test_user]])
  
1
2
3
4
5
6
7
8
9
10
11

RexCall 多个操作符构建的表达式

select id,case name='ming' then null else name
end  from test_user
比如这里的
case name='ming' then null else name
end 

RexSubQuery 子查询


select * from test_user where id existes  （10,40)

RelNode root = builder.scan("test_user").filter(builder.call(SqlStdOperatorTable.EXISTS, builder.field(0), builder.literal(10), builder.literal(40)))
        .build();

LogicalFilter(condition=[EXISTS($0, 10, 40)])
  LogicalTableScan(table=[[test_user]])
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

参考代码

public static Frameworks.ConfigBuilder config() {

    SchemaPlus schemaPlus = Frameworks.createRootSchema(true);
    schemaPlus.add("test_user", new AbstractTable() {
        @Override
        public RelDataType getRowType(RelDataTypeFactory typeFactory) {

            List dataTypeList = new ArrayList<>();
            RelDataType integerDataType = typeFactory.createTypeWithNullability(typeFactory.createSqlType(SqlTypeName.BIGINT), true);

            RelDataType stringDataType = typeFactory.createTypeWithNullability(typeFactory.createSqlType(SqlTypeName.VARCHAR), true);

            dataTypeList.add(integerDataType);
            dataTypeList.add(stringDataType);


            List fieldNames = new ArrayList<>();
            fieldNames.add("id");
            fieldNames.add("name");

            return typeFactory.createStructType(dataTypeList, fieldNames);
        }
    });


    schemaPlus.add("test_address", new AbstractTable() {
        @Override
        public RelDataType getRowType(RelDataTypeFactory typeFactory) {

            List dataTypeList = new ArrayList<>();
            RelDataType integerDataType = typeFactory.createTypeWithNullability(typeFactory.createSqlType(SqlTypeName.BIGINT), true);

            RelDataType stringDataType = typeFactory.createTypeWithNullability(typeFactory.createSqlType(SqlTypeName.VARCHAR), true);

            dataTypeList.add(integerDataType);
            dataTypeList.add(stringDataType);


            List fieldNames = new ArrayList<>();
            fieldNames.add("id");
            fieldNames.add("address");

            return typeFactory.createStructType(dataTypeList, fieldNames);
        }
    });


    return Frameworks.newConfigBuilder()
            .parserConfig(SqlParser.Config.DEFAULT)
            .defaultSchema(schemaPlus)
            .traitDefs((List) null)
            .programs(Programs.heuristicJoinOrder(Programs.RULE_SET, true, 2));
}


public static void main(String[] args) {
    RelBuilder builder = RelBuilder.create(config().build());
    RelNode root = builder.scan("test_user")
            .build();

    System.out.println(RelOptUtil.toString(root));
}
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62