from util.log import * from util.cases import * from util.sql import * import numpy as np import random class TDTestCase: updatecfgDict = {"maxTablesPerVnode":2 ,"minTablesPerVnode":2,"tableIncStepPerVnode":2 } def init(self, conn, logSql): tdLog.debug("start to execute %s" % __file__) tdSql.init(conn.cursor()) self.vnode_disbutes = None self.ts = 1537146000000 def prepare_datas_of_distribute(self, dbname="testdb"): # prepate datas for 20 tables distributed at different vgroups tdSql.execute(f"create database if not exists {dbname} keep 3650 duration 1000 vgroups 5") tdSql.execute(f" use {dbname} ") tdSql.execute( f'''create table {dbname}.stb1 (ts timestamp, c1 int, c2 bigint, c3 smallint, c4 tinyint, c5 float, c6 double, c7 bool, c8 binary(16),c9 nchar(32), c10 timestamp) tags (t0 timestamp, t1 int, t2 bigint, t3 smallint, t4 tinyint, t5 float, t6 double, t7 bool, t8 binary(16),t9 nchar(32)) ''' ) for i in range(20): tdSql.execute(f'create table {dbname}.ct{i+1} using {dbname}.stb1 tags ( now(), {1*i}, {11111*i}, {111*i}, {1*i}, {1.11*i}, {11.11*i}, {i%2}, "binary{i}", "nchar{i}" )') for i in range(9): tdSql.execute( f"insert into {dbname}.ct1 values ( now()-{i*10}s, {1*i}, {11111*i}, {111*i}, {11*i}, {1.11*i}, {11.11*i}, {i%2}, 'binary{i}', 'nchar{i}', now()+{1*i}a )" ) tdSql.execute( f"insert into {dbname}.ct4 values ( now()-{i*90}d, {1*i}, {11111*i}, {111*i}, {11*i}, {1.11*i}, {11.11*i}, {i%2}, 'binary{i}', 'nchar{i}', now()+{1*i}a )" ) for i in range(1,21): if i ==1 or i == 4: continue else: tbname = f"{dbname}.ct{i}" for j in range(9): tdSql.execute( f"insert into {tbname} values ( now()-{(i+j)*10}s, {1*(j+i)}, {11111*(j+i)}, {111*(j+i)}, {11*(j)}, {1.11*(j+i)}, {11.11*(j+i)}, {(j+i)%2}, 'binary{j}', 'nchar{j}', now()+{1*j}a )" ) tdSql.execute(f"insert into {dbname}.ct1 values (now()-45s, 0, 0, 0, 0, 0, 0, 0, 'binary0', 'nchar0', now()+8a )") tdSql.execute(f"insert into {dbname}.ct1 values (now()+10s, 9, -99999, -999, -99, -9.99, -99.99, 1, 'binary9', 'nchar9', now()+9a )") tdSql.execute(f"insert into {dbname}.ct1 values (now()+15s, 9, -99999, -999, -99, -9.99, NULL, 1, 'binary9', 'nchar9', now()+9a )") tdSql.execute(f"insert into {dbname}.ct1 values (now()+20s, 9, -99999, -999, NULL, -9.99, -99.99, 1, 'binary9', 'nchar9', now()+9a )") tdSql.execute(f"insert into {dbname}.ct4 values (now()-810d, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL ) ") tdSql.execute(f"insert into {dbname}.ct4 values (now()-400d, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL ) ") tdSql.execute(f"insert into {dbname}.ct4 values (now()+90d, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL ) ") tdLog.info(" prepare data for distributed_aggregate done! ") def check_distribute_datas(self, dbname="testdb"): # get vgroup_ids of all tdSql.query(f"show {dbname}.vgroups ") vgroups = tdSql.queryResult vnode_tables={} for vgroup_id in vgroups: vnode_tables[vgroup_id[0]]=[] # check sub_table of per vnode ,make sure sub_table has been distributed tdSql.query(f"select * from information_schema.ins_tables where db_name = '{dbname}' and table_name like 'ct%'") table_names = tdSql.queryResult tablenames = [] for table_name in table_names: vnode_tables[table_name[6]].append(table_name[0]) self.vnode_disbutes = vnode_tables count = 0 for k ,v in vnode_tables.items(): if len(v)>=2: count+=1 if count < 2: tdLog.exit(" the datas of all not satisfy sub_table has been distributed ") def distribute_agg_query(self, dbname="testdb"): # basic filter tdSql.query(f"select apercentile(c1 , 20) from {dbname}.stb1 where c1 is null") tdSql.checkRows(1) tdSql.query(f"select apercentile(c1 , 20) from {dbname}.stb1 where t1=1") tdSql.checkData(0,0,2.800000000) tdSql.query(f"select apercentile(c1+c2 ,100) from {dbname}.stb1 where c1 =1 ") tdSql.checkData(0,0,11112.000000000) tdSql.query(f"select apercentile(c1 ,10 ) from {dbname}.stb1 where tbname=\"ct2\"") tdSql.checkData(0,0,2.000000000) tdSql.query(f"select apercentile(c1,20) from {dbname}.stb1 partition by tbname") tdSql.checkRows(20) tdSql.query(f"select apercentile(c1,20) from {dbname}.stb1 where t1> 4 partition by tbname") tdSql.checkRows(15) # union all tdSql.query(f"select apercentile(c1,20) from {dbname}.stb1 union all select apercentile(c1,20) from {dbname}.stb1 ") tdSql.checkRows(2) tdSql.checkData(0,0,7.389181281) # join tdSql.execute(" create database if not exists db ") tdSql.execute(" use db ") tdSql.execute(" create stable db.st (ts timestamp , c1 int ,c2 float) tags(t1 int) ") tdSql.execute(" create table db.tb1 using db.st tags(1) ") tdSql.execute(" create table db.tb2 using db.st tags(2) ") for i in range(10): ts = i*10 + self.ts tdSql.execute(f" insert into db.tb1 values({ts},{i},{i}.0)") tdSql.execute(f" insert into db.tb2 values({ts},{i},{i}.0)") tdSql.query(f"select apercentile(tb1.c1,100), apercentile(tb2.c2,100) from db.tb1 tb1, db.tb2 tb2 where tb1.ts=tb2.ts") tdSql.checkRows(1) tdSql.checkData(0,0,9.000000000) tdSql.checkData(0,0,9.000000000) # group by tdSql.execute(f"use {dbname} ") tdSql.query(f" select max(c1),c1 from {dbname}.stb1 group by t1 ") tdSql.checkRows(20) tdSql.query(f" select max(c1),c1 from {dbname}.stb1 group by c1 ") tdSql.checkRows(30) tdSql.query(f" select max(c1),c2 from {dbname}.stb1 group by c2 ") tdSql.checkRows(31) # partition by tbname or partition by tag tdSql.query(f"select apercentile(c1 ,10)from {dbname}.stb1 partition by tbname") query_data = tdSql.queryResult # nest query for support max #tdSql.query(f"select apercentile(c2+2,10)+1 from (select max(c1) c2 from {dbname}.stb1)") #tdSql.checkData(0,0,31.000000000) tdSql.query(f"select apercentile(c1+2,10)+1 as c2 from (select ts ,c1 ,c2 from {dbname}.stb1)") tdSql.checkData(0,0,7.560701700) tdSql.query(f"select apercentile(a+2,10)+1 as c2 from (select ts ,abs(c1) a ,c2 from {dbname}.stb1)") tdSql.checkData(0,0,7.560701700) # mixup with other functions tdSql.query(f"select max(c1),count(c1),last(c2,c3),spread(c1), apercentile(c1,10) from {dbname}.stb1") tdSql.checkData(0,0,28) tdSql.checkData(0,1,184) tdSql.checkData(0,2,-99999) tdSql.checkData(0,3,-999) tdSql.checkData(0,4,28.000000000) tdSql.checkData(0,5,4.560701700) def run(self): self.prepare_datas_of_distribute() self.check_distribute_datas() self.distribute_agg_query() def stop(self): tdSql.close() tdLog.success("%s successfully executed" % __file__) tdCases.addWindows(__file__, TDTestCase()) tdCases.addLinux(__file__, TDTestCase())