更新时间:2024-10-28 GMT+08:00
EXPLAIN
语法
EXPLAIN [ ( option [, ...] ) ] statement
其中选项可以是以下选项之一:
FORMAT { TEXT | GRAPHVIZ | JSON }
TYPE { LOGICAL | DISTRIBUTED | VALIDATE | IO}
描述
显示一条语句的逻辑的或者分布式的执行计划,也可以用于校验一条SQL语句,或者是分析IO。
参数TYPE DISTRIBUTED用于显示分片后的计划(fragmented plan)。每一个fragment都会被一个或者多个节点执行。Fragments separation表示数据在两个节点之间进行交换。Fragment type表示一个fragment如何被执行以及数据在不同fragment之间怎样分布。
示例
- LOGICAL:
CREATE TABLE testTable (regionkey int, name varchar); EXPLAIN SELECT regionkey, count(*) FROM testTable GROUP BY 1; Query Plan ------------------------------------------------------------------------------------------------------------------------------------- Output[regionkey, _col1] │ Layout: [regionkey:integer, count:bigint] │ Estimates: {rows: ? (?), cpu: ?, memory: ?, network: ?} │ _col1 := count └─ RemoteExchange[GATHER] │ Layout: [regionkey:integer, count:bigint] │ Estimates: {rows: ? (?), cpu: ?, memory: ?, network: ?} └─ Project[] │ Layout: [regionkey:integer, count:bigint] │ Estimates: {rows: ? (?), cpu: ?, memory: ?, network: ?} └─ Aggregate(FINAL)[regionkey][$hashvalue] │ Layout: [regionkey:integer, $hashvalue:bigint, count:bigint] │ Estimates: {rows: ? (?), cpu: ?, memory: ?, network: ?} │ count := count("count_8") └─ LocalExchange[HASH][$hashvalue] ("regionkey") │ Layout: [regionkey:integer, count_8:bigint, $hashvalue:bigint] │ Estimates: {rows: ? (?), cpu: ?, memory: ?, network: ?} └─ RemoteExchange[REPARTITION][$hashvalue_9] │ Layout: [regionkey:integer, count_8:bigint, $hashvalue_9:bigint] │ Estimates: {rows: ? (?), cpu: ?, memory: ?, network: ?} └─ Aggregate(PARTIAL)[regionkey][$hashvalue_10] │ Layout: [regionkey:integer, $hashvalue_10:bigint, count_8:bigint] │ count_8 := count(*) └─ ScanProject[table = hive:default:testtable] Layout: [regionkey:integer, $hashvalue_10:bigint] Estimates: {rows: 0 (0B), cpu: 0, memory: 0B, network: 0B}/{rows: 0 (0B), cpu: 0, memory: 0B, network: 0B} $hashvalue_10 := "combine_hash"(bigint '0', COALESCE("$operator$hash_code"("regionkey"), 0)) regionkey := regionkey:int:0:REGULAR
- DISTRIBUTED:
EXPLAIN (type DISTRIBUTED) SELECT regionkey, count(*) FROM testTable GROUP BY 1; Query Plan ----------------------------------------------------------------------------------------------------------------------- Fragment 0 [SINGLE] Output layout: [regionkey, count] Output partitioning: SINGLE [] Stage Execution Strategy: UNGROUPED_EXECUTION Output[regionkey, _col1] │ Layout: [regionkey:integer, count:bigint] │ Estimates: {rows: ? (?), cpu: ?, memory: ?, network: ?} │ _col1 := count └─ RemoteSource[1] Layout: [regionkey:integer, count:bigint] Fragment 1 [HASH] Output layout: [regionkey, count] Output partitioning: SINGLE [] Stage Execution Strategy: UNGROUPED_EXECUTION Project[] │ Layout: [regionkey:integer, count:bigint] │ Estimates: {rows: ? (?), cpu: ?, memory: ?, network: ?} └─ Aggregate(FINAL)[regionkey][$hashvalue] │ Layout: [regionkey:integer, $hashvalue:bigint, count:bigint] │ Estimates: {rows: ? (?), cpu: ?, memory: ?, network: ?} │ count := count("count_8") └─ LocalExchange[HASH][$hashvalue] ("regionkey") │ Layout: [regionkey:integer, count_8:bigint, $hashvalue:bigint] │ Estimates: {rows: ? (?), cpu: ?, memory: ?, network: ?} └─ RemoteSource[2] Layout: [regionkey:integer, count_8:bigint, $hashvalue_9:bigint] Fragment 2 [SOURCE] Output layout: [regionkey, count_8, $hashvalue_10] Output partitioning: HASH [regionkey][$hashvalue_10] Stage Execution Strategy: UNGROUPED_EXECUTION Aggregate(PARTIAL)[regionkey][$hashvalue_10] │ Layout: [regionkey:integer, $hashvalue_10:bigint, count_8:bigint] │ count_8 := count(*) └─ ScanProject[table = hive:default:testtable, grouped = false] Layout: [regionkey:integer, $hashvalue_10:bigint] Estimates: {rows: 0 (0B), cpu: 0, memory: 0B, network: 0B}/{rows: 0 (0B), cpu: 0, memory: 0B, network: 0B} $hashvalue_10 := "combine_hash"(bigint '0', COALESCE("$operator$hash_code"("regionkey"), 0)) regionkey := regionkey:int:0:REGULAR
- VALIDATE:
EXPLAIN (TYPE VALIDATE) SELECT id, count(*) FROM testTable GROUP BY 1; Valid ------- true
- IO:
EXPLAIN (TYPE IO, FORMAT JSON) SELECT regionkey , count(*) FROM testTable GROUP BY 1; Query Plan --------------------------------- { "inputTableColumnInfos" : [ { "table" : { "catalog" : "hive", "schemaTable" : { "schema" : "default", "table" : "testtable" } }, "columnConstraints" : [ ] } ] }
父主题: HetuEngine辅助命令语法