Calcite执行计划的四种展示格式

• 前言
• 格式
• • Text/String
  • XML
  • JSON
  • DOT
• 属性级别

我们知道Calcite会将SQL为RelNode构成的执行计划，为了分析执行计划，经常会使用RelOptUtil.toString()将其转为字符串。其实，除了这种字符串之外还有更多的格式，可以展示执行计划：XML、JSON、DOT。

在SqlExplainFormat中可以看到这四种格式，其中TEXT其实和String一样

在RelOptUtilTest.java中构建一个单测，利用已有的relBuilder可以快速构建一个RelNode

RelNode aggregate =  relBuilder.scan("DEPT")
   .aggregate(relBuilder.groupKey(),
       relBuilder.aggregateCall(SqlStdOperatorTable.COUNT))
   .limit(0, 10)
   .build();

    String strPlan = RelOptUtil.toString(aggregate);
    System.out.println(strPlan);

    String textPlan = RelOptUtil.dumpPlan("TEXT:",
        aggregate,
        SqlExplainFormat.TEXT,
        SqlExplainLevel.ALL_ATTRIBUTES);
    System.out.println(textPlan);

打印的结果有些差别，是因为SqlExplainLevel.ALL_ATTRIBUTES选择的是所有属性；toString只是EXPPLAN_ATTRIBUTES级别

LogicalSort(fetch=[10])
  LogicalAggregate(group=[{}], agg#0=[COUNT()])
    LogicalTableScan(table=[[scott, DEPT]])


TEXT:
LogicalSort(fetch=[10]): rowcount = 1.0, cumulative cost = {6.125 rows, 21.0 cpu, 0.0 io}, id = 4
  LogicalAggregate(group=[{}], agg#0=[COUNT()]): rowcount = 1.0, cumulative cost = {5.125 rows, 5.0 cpu, 0.0 io}, id = 3
    LogicalTableScan(table=[[scott, DEPT]]): rowcount = 4.0, cumulative cost = {4.0 rows, 5.0 cpu, 0.0 io}, id = 2

    String xmlPlan = RelOptUtil.dumpPlan("xml:",
        aggregate,
        SqlExplainFormat.XML,
        SqlExplainLevel.ALL_ATTRIBUTES);
    System.out.println(xmlPlan);

XML格式的，直观上看起来还不如Text了

	
		10
	
	
		
			
				{}
			
			
				COUNT()
			
			
				
					
						[scott, DEPT]
					
					
				
			
		
	

String jsonPlan = RelOptUtil.dumpPlan("json:",
    aggregate,
    SqlExplainFormat.JSON,
    SqlExplainLevel.ALL_ATTRIBUTES);
System.out.println(jsonPlan);

看起来也不是很方便，但是json格式便于传输

{
  "rels": [
    {
      "id": "0",
      "relOp": "LogicalTableScan",
      "table": [
        "scott",
        "DEPT"
      ],
      "inputs": []
    },
    {
      "id": "1",
      "relOp": "LogicalAggregate",
      "group": [],
      "aggs": [
        {
          "agg": {
            "name": "COUNT",
            "kind": "COUNT",
            "syntax": "FUNCTION_STAR"
          },
          "type": {
            "type": "BIGINT",
            "nullable": false
          },
          "distinct": false,
          "operands": [],
          "name": null
        }
      ]
    },
    {
      "id": "2",
      "relOp": "LogicalSort",
      "collation": [],
      "fetch": {
        "literal": 10,
        "type": {
          "type": "INTEGER",
          "nullable": false
        }
      }
    }
  ]
}

String dotPlan = RelOptUtil.dumpPlan("dot:",
    aggregate,
    SqlExplainFormat.DOT,
    SqlExplainLevel.ALL_ATTRIBUTES);
System.out.println(dotPlan);

这个主要是用于绘图，而不是直接观察计划

digraph {
"LogicalAggregatengroup = {}nagg#0 = COUNT()nrowcount = 1.0 cumulnative cost = {5.125 nrows, 5.0 cpu, 0.0 ino} id = 3" -> "LogicalSortnfetch = 10nrowcount = 1.0 cumulnative cost = {6.125 nrows, 21.0 cpu, 0.0 nio} id = 4" [label="0"]
"LogicalTableScanntable = [scott, DEPTn]nrowcount = 4.0 cumulnative cost = {4.0 ronws, 5.0 cpu, 0.0 io}n id = 2" -> "LogicalAggregatengroup = {}nagg#0 = COUNT()nrowcount = 1.0 cumulnative cost = {5.125 nrows, 5.0 cpu, 0.0 ino} id = 3" [label="0"]
}

将plan保存plan.dot文件，然后使用graphviz的命令dot -Tsvg -o plan.svg plan.dot生成svg图片

也可以通过-T指定其他格式，如-Tpng

用浏览器可以直接打开生成的svg图

刚刚提到SqlExplainLevel导致了TEXT和String的不一样，其实一共有5种级别

用上文的RelNode感受一下，这些plan的不同吧

noAttributes:
LogicalSort
  LogicalAggregate
    LogicalTableScan

expPlanAttributes:
LogicalSort(fetch=[10])
  LogicalAggregate(group=[{}], agg#0=[COUNT()])
    LogicalTableScan(table=[[scott, DEPT]])

digestAttributes:
LogicalSort(fetch=[10])
  LogicalAggregate(group=[{}], agg#0=[COUNT()])
    LogicalTableScan(table=[[scott, DEPT]])

nonCostAttributes:
LogicalSort(fetch=[10]), id = 4
  LogicalAggregate(group=[{}], agg#0=[COUNT()]), id = 3
    LogicalTableScan(table=[[scott, DEPT]]), id = 2

allAttributes:
LogicalSort(fetch=[10]): rowcount = 1.0, cumulative cost = {6.125 rows, 21.0 cpu, 0.0 io}, id = 4
  LogicalAggregate(group=[{}], agg#0=[COUNT()]): rowcount = 1.0, cumulative cost = {5.125 rows, 5.0 cpu, 0.0 io}, id = 3
    LogicalTableScan(table=[[scott, DEPT]]): rowcount = 4.0, cumulative cost = {4.0 rows, 5.0 cpu, 0.0 io}, id = 2