未验证 提交 e0097168 编写于 作者: H Hui Li 提交者: GitHub

Merge pull request #7112 from taosdata/test/td-5314

[TD-5314]<test>: finish schemaless test
......@@ -379,6 +379,7 @@ python3 test.py -f alter/alter_cacheLastRow.py
python3 ./test.py -f query/querySession.py
python3 test.py -f alter/alter_create_exception.py
python3 ./test.py -f insert/flushwhiledrop.py
python3 ./test.py -f insert/schemalessInsert.py
#======================p4-end===============
python3 test.py -f tools/taosdemoAllTest/pytest.py
......
###################################################################
# Copyright (c) 2021 by TAOS Technologies, Inc.
# All rights reserved.
#
# This file is proprietary and confidential to TAOS Technologies.
# No part of this file may be reproduced, stored, transmitted,
# disclosed or used in any form or by any means other than as
# expressly provided by the written permission from Jianhui Tao
#
###################################################################
# -*- coding: utf-8 -*-
import random
import string
import time
from copy import deepcopy
import numpy as np
from util.log import *
from util.cases import *
from util.sql import *
import threading
class TDTestCase:
def init(self, conn, logSql):
tdLog.debug("start to execute %s" % __file__)
tdSql.init(conn.cursor(), logSql)
self._conn = conn
def createDb(self, name="test", db_update_tag=0):
if db_update_tag == 0:
tdSql.execute(f"drop database if exists {name}")
tdSql.execute(f"create database if not exists {name} precision 'us'")
else:
tdSql.execute(f"drop database if exists {name}")
tdSql.execute(f"create database if not exists {name} precision 'us' update 1")
tdSql.execute(f'use {name}')
def getLongName(self, len, mode = "mixed"):
"""
generate long name
mode could be numbers/letters/mixed
"""
if mode is "numbers":
chars = ''.join(random.choice(string.digits) for i in range(len))
elif mode is "letters":
chars = ''.join(random.choice(string.ascii_letters.lower()) for i in range(len))
else:
chars = ''.join(random.choice(string.ascii_letters.lower() + string.digits) for i in range(len))
return chars
def timeTrans(self, time_value):
if time_value.endswith("ns"):
ts = int(''.join(list(filter(str.isdigit, time_value))))/1000000000
elif time_value.endswith("us") or time_value.isdigit() and int(time_value) != 0:
ts = int(''.join(list(filter(str.isdigit, time_value))))/1000000
elif time_value.endswith("ms"):
ts = int(''.join(list(filter(str.isdigit, time_value))))/1000
elif time_value.endswith("s") and list(time_value)[-1] not in "num":
ts = int(''.join(list(filter(str.isdigit, time_value))))/1
elif int(time_value) == 0:
ts = time.time()
else:
print("input ts maybe not right format")
ulsec = repr(ts).split('.')[1][:6]
if len(ulsec) < 6 and int(ulsec) != 0:
ulsec = int(ulsec) * (10 ** (6 - len(ulsec)))
elif int(ulsec) == 0:
ulsec *= 6
# * follow two rows added for tsCheckCase
td_ts = time.strftime("%Y-%m-%d %H:%M:%S", time.localtime(ts))
return td_ts
#td_ts = time.strftime("%Y-%m-%d %H:%M:%S", time.localtime(ts))
td_ts = time.strftime("%Y-%m-%d %H:%M:%S.{}".format(ulsec), time.localtime(ts))
return td_ts
#return repr(datetime.datetime.strptime(td_ts, "%Y-%m-%d %H:%M:%S.%f"))
def dateToTs(self, datetime_input):
return int(time.mktime(time.strptime(datetime_input, "%Y-%m-%d %H:%M:%S.%f")))
def getTdTypeValue(self, value):
if value.endswith("i8"):
td_type = "TINYINT"
td_tag_value = ''.join(list(value)[:-2])
elif value.endswith("i16"):
td_type = "SMALLINT"
td_tag_value = ''.join(list(value)[:-3])
elif value.endswith("i32"):
td_type = "INT"
td_tag_value = ''.join(list(value)[:-3])
elif value.endswith("i64"):
td_type = "BIGINT"
td_tag_value = ''.join(list(value)[:-3])
elif value.endswith("u64"):
td_type = "BIGINT UNSIGNED"
td_tag_value = ''.join(list(value)[:-3])
elif value.endswith("f32"):
td_type = "FLOAT"
td_tag_value = ''.join(list(value)[:-3])
td_tag_value = '{}'.format(np.float32(td_tag_value))
elif value.endswith("f64"):
td_type = "DOUBLE"
td_tag_value = ''.join(list(value)[:-3])
elif value.startswith('L"'):
td_type = "NCHAR"
td_tag_value = ''.join(list(value)[2:-1])
elif value.startswith('"') and value.endswith('"'):
td_type = "BINARY"
td_tag_value = ''.join(list(value)[1:-1])
elif value.lower() == "t" or value == "true" or value == "True":
td_type = "BOOL"
td_tag_value = "True"
elif value.lower() == "f" or value == "false" or value == "False":
td_type = "BOOL"
td_tag_value = "False"
else:
td_type = "FLOAT"
td_tag_value = value
return td_type, td_tag_value
def typeTrans(self, type_list):
type_num_list = []
for tp in type_list:
if tp.upper() == "TIMESTAMP":
type_num_list.append(9)
elif tp.upper() == "BOOL":
type_num_list.append(1)
elif tp.upper() == "TINYINT":
type_num_list.append(2)
elif tp.upper() == "SMALLINT":
type_num_list.append(3)
elif tp.upper() == "INT":
type_num_list.append(4)
elif tp.upper() == "BIGINT":
type_num_list.append(5)
elif tp.upper() == "FLOAT":
type_num_list.append(6)
elif tp.upper() == "DOUBLE":
type_num_list.append(7)
elif tp.upper() == "BINARY":
type_num_list.append(8)
elif tp.upper() == "NCHAR":
type_num_list.append(10)
elif tp.upper() == "BIGINT UNSIGNED":
type_num_list.append(14)
return type_num_list
def inputHandle(self, input_sql):
input_sql_split_list = input_sql.split(" ")
stb_tag_list = input_sql_split_list[0].split(',')
stb_col_list = input_sql_split_list[1].split(',')
ts_value = self.timeTrans(input_sql_split_list[2])
stb_name = stb_tag_list[0]
stb_tag_list.pop(0)
tag_name_list = []
tag_value_list = []
td_tag_value_list = []
td_tag_type_list = []
col_name_list = []
col_value_list = []
td_col_value_list = []
td_col_type_list = []
for elm in stb_tag_list:
if "id=" in elm.lower():
tb_name = elm.split('=')[1]
else:
tag_name_list.append(elm.split("=")[0])
tag_value_list.append(elm.split("=")[1])
tb_name = ""
td_tag_value_list.append(self.getTdTypeValue(elm.split("=")[1])[1])
td_tag_type_list.append(self.getTdTypeValue(elm.split("=")[1])[0])
for elm in stb_col_list:
col_name_list.append(elm.split("=")[0])
col_value_list.append(elm.split("=")[1])
td_col_value_list.append(self.getTdTypeValue(elm.split("=")[1])[1])
td_col_type_list.append(self.getTdTypeValue(elm.split("=")[1])[0])
final_field_list = []
final_field_list.extend(col_name_list)
final_field_list.extend(tag_name_list)
final_type_list = []
final_type_list.append("TIMESTAMP")
final_type_list.extend(td_col_type_list)
final_type_list.extend(td_tag_type_list)
final_type_list = self.typeTrans(final_type_list)
final_value_list = []
final_value_list.append(ts_value)
final_value_list.extend(td_col_value_list)
final_value_list.extend(td_tag_value_list)
return final_value_list, final_field_list, final_type_list, stb_name, tb_name
def genFullTypeSql(self, stb_name="", tb_name="", t0="", t1="127i8", t2="32767i16", t3="2147483647i32",
t4="9223372036854775807i64", t5="11.12345f32", t6="22.123456789f64", t7="\"binaryTagValue\"",
t8="L\"ncharTagValue\"", c0="", c1="127i8", c2="32767i16", c3="2147483647i32",
c4="9223372036854775807i64", c5="11.12345f32", c6="22.123456789f64", c7="\"binaryColValue\"",
c8="L\"ncharColValue\"", c9="7u64", ts="1626006833639000000ns",
id_noexist_tag=None, id_change_tag=None, id_upper_tag=None, id_double_tag=None,
ct_add_tag=None, ct_am_tag=None, ct_ma_tag=None, ct_min_tag=None):
if stb_name == "":
stb_name = self.getLongName(len=6, mode="letters")
if tb_name == "":
tb_name = f'{stb_name}_{random.randint(0, 65535)}_{random.randint(0, 65535)}'
if t0 == "":
t0 = random.choice(["f", "F", "false", "False", "t", "T", "true", "True"])
if c0 == "":
c0 = random.choice(["f", "F", "false", "False", "t", "T", "true", "True"])
#sql_seq = f'{stb_name},id=\"{tb_name}\",t0={t0},t1=127i8,t2=32767i16,t3=125.22f64,t4=11.321f32,t5=11.12345f32,t6=22.123456789f64,t7=\"binaryTagValue\",t8=L\"ncharTagValue\" c0={bool_value},c1=127i8,c2=32767i16,c3=2147483647i32,c4=9223372036854775807i64,c5=11.12345f32,c6=22.123456789f64,c7=\"binaryValue\",c8=L\"ncharValue\" 1626006833639000000ns'
if id_upper_tag is not None:
id = "ID"
else:
id = "id"
sql_seq = f'{stb_name},{id}=\"{tb_name}\",t0={t0},t1={t1},t2={t2},t3={t3},t4={t4},t5={t5},t6={t6},t7={t7},t8={t8} c0={c0},c1={c1},c2={c2},c3={c3},c4={c4},c5={c5},c6={c6},c7={c7},c8={c8},c9={c9} {ts}'
if id_noexist_tag is not None:
sql_seq = f'{stb_name},t0={t0},t1={t1},t2={t2},t3={t3},t4={t4},t5={t5},t6={t6},t7={t7},t8={t8} c0={c0},c1={c1},c2={c2},c3={c3},c4={c4},c5={c5},c6={c6},c7={c7},c8={c8},c9={c9} {ts}'
if ct_add_tag is not None:
sql_seq = f'{stb_name},t0={t0},t1={t1},t2={t2},t3={t3},t4={t4},t5={t5},t6={t6},t7={t7},t8={t8},t9={t8} c0={c0},c1={c1},c2={c2},c3={c3},c4={c4},c5={c5},c6={c6},c7={c7},c8={c8},c9={c9} {ts}'
if id_change_tag is not None:
sql_seq = f'{stb_name},t0={t0},t1={t1},{id}=\"{tb_name}\",t2={t2},t3={t3},t4={t4},t5={t5},t6={t6},t7={t7},t8={t8} c0={c0},c1={c1},c2={c2},c3={c3},c4={c4},c5={c5},c6={c6},c7={c7},c8={c8},c9={c9} {ts}'
if id_double_tag is not None:
sql_seq = f'{stb_name},{id}=\"{tb_name}_1\",t0={t0},t1={t1},{id}=\"{tb_name}_2\",t2={t2},t3={t3},t4={t4},t5={t5},t6={t6},t7={t7},t8={t8} c0={c0},c1={c1},c2={c2},c3={c3},c4={c4},c5={c5},c6={c6},c7={c7},c8={c8},c9={c9} {ts}'
if ct_add_tag is not None:
sql_seq = f'{stb_name},{id}=\"{tb_name}\",t0={t0},t1={t1},t2={t2},t3={t3},t4={t4},t5={t5},t6={t6},t7={t7},t8={t8},t11={t1},t10={t8} c0={c0},c1={c1},c2={c2},c3={c3},c4={c4},c5={c5},c6={c6},c7={c7},c8={c8},c9={c9},c11={c8},c10={t0} {ts}'
if ct_am_tag is not None:
sql_seq = f'{stb_name},{id}=\"{tb_name}\",t0={t0},t1={t1},t2={t2},t3={t3},t4={t4},t5={t5},t6={t6} c0={c0},c1={c1},c2={c2},c3={c3},c4={c4},c5={c5},c6={c6},c7={c7},c8={c8},c9={c9},c11={c8},c10={t0} {ts}'
if id_noexist_tag is not None:
sql_seq = f'{stb_name},t0={t0},t1={t1},t2={t2},t3={t3},t4={t4},t5={t5},t6={t6} c0={c0},c1={c1},c2={c2},c3={c3},c4={c4},c5={c5},c6={c6},c7={c7},c8={c8},c9={c9},c11={c8},c10={t0} {ts}'
if ct_ma_tag is not None:
sql_seq = f'{stb_name},{id}=\"{tb_name}\",t0={t0},t1={t1},t2={t2},t3={t3},t4={t4},t5={t5},t6={t6},t7={t7},t8={t8},t11={t1},t10={t8} c0={c0},c1={c1},c2={c2},c3={c3},c4={c4},c5={c5},c6={c6} {ts}'
if id_noexist_tag is not None:
sql_seq = f'{stb_name},t0={t0},t1={t1},t2={t2},t3={t3},t4={t4},t5={t5},t6={t6},t7={t7},t8={t8},t11={t1},t10={t8} c0={c0},c1={c1},c2={c2},c3={c3},c4={c4},c5={c5},c6={c6} {ts}'
if ct_min_tag is not None:
sql_seq = f'{stb_name},{id}=\"{tb_name}\",t0={t0},t1={t1},t2={t2},t3={t3},t4={t4},t5={t5},t6={t6} c0={c0},c1={c1},c2={c2},c3={c3},c4={c4},c5={c5},c6={c6} {ts}'
return sql_seq, stb_name
def genMulTagColStr(self, genType, count):
"""
genType must be tag/col
"""
tag_str = ""
col_str = ""
if genType == "tag":
for i in range(0, count):
if i < (count-1):
tag_str += f't{i}=f,'
else:
tag_str += f't{i}=f '
return tag_str
if genType == "col":
for i in range(0, count):
if i < (count-1):
col_str += f'c{i}=t,'
else:
col_str += f'c{i}=t '
return col_str
def genLongSql(self, tag_count, col_count):
stb_name = self.getLongName(7, mode="letters")
tb_name = f'{stb_name}_1'
tag_str = self.genMulTagColStr("tag", tag_count)
col_str = self.genMulTagColStr("col", col_count)
ts = "1626006833640000000ns"
long_sql = stb_name + ',' + f'id=\"{tb_name}\"' + ',' + tag_str + col_str + ts
return long_sql, stb_name
def getNoIdTbName(self, stb_name):
query_sql = f"select tbname from {stb_name}"
tb_name = self.resHandle(query_sql, True)[0][0]
return tb_name
def resHandle(self, query_sql, query_tag):
tdSql.execute('reset query cache')
row_info = tdSql.query(query_sql, query_tag)
col_info = tdSql.getColNameList(query_sql, query_tag)
res_row_list = []
sub_list = []
for row_mem in row_info:
for i in row_mem:
sub_list.append(str(i))
res_row_list.append(sub_list)
res_field_list_without_ts = col_info[0][1:]
res_type_list = col_info[1]
return res_row_list, res_field_list_without_ts, res_type_list
def resCmp(self, input_sql, stb_name, query_sql="select * from", condition="", ts=None, id=True, none_check_tag=None):
expect_list = self.inputHandle(input_sql)
self._conn.insertLines([input_sql])
query_sql = f"{query_sql} {stb_name} {condition}"
res_row_list, res_field_list_without_ts, res_type_list = self.resHandle(query_sql, True)
if ts == 0:
res_ts = self.dateToTs(res_row_list[0][0])
current_time = time.time()
if current_time - res_ts < 60:
tdSql.checkEqual(res_row_list[0][1:], expect_list[0][1:])
else:
print("timeout")
tdSql.checkEqual(res_row_list[0], expect_list[0])
else:
if none_check_tag is not None:
none_index_list = [i for i,x in enumerate(res_row_list[0]) if x=="None"]
none_index_list.reverse()
for j in none_index_list:
res_row_list[0].pop(j)
expect_list[0].pop(j)
tdSql.checkEqual(res_row_list[0], expect_list[0])
tdSql.checkEqual(res_field_list_without_ts, expect_list[1])
tdSql.checkEqual(res_type_list, expect_list[2])
def cleanStb(self):
query_sql = "show stables"
res_row_list = tdSql.query(query_sql, True)
stb_list = map(lambda x: x[0], res_row_list)
for stb in stb_list:
tdSql.execute(f'drop table if exists {stb}')
def initCheckCase(self):
"""
normal tags and cols, one for every elm
"""
self.cleanStb()
input_sql, stb_name = self.genFullTypeSql()
self.resCmp(input_sql, stb_name)
def boolTypeCheckCase(self):
"""
check all normal type
"""
self.cleanStb()
full_type_list = ["f", "F", "false", "False", "t", "T", "true", "True"]
for t_type in full_type_list:
input_sql, stb_name = self.genFullTypeSql(c0=t_type, t0=t_type)
self.resCmp(input_sql, stb_name)
def symbolsCheckCase(self):
"""
check symbols = `~!@#$%^&*()_-+={[}]\|:;'\",<.>/?
"""
'''
please test :
binary_symbols = '\"abcd`~!@#$%^&*()_-{[}]|:;<.>?lfjal"\'\'"\"'
'''
self.cleanStb()
binary_symbols = '\"abcd`~!@#$%^&*()_-{[}]|:;<.>?lfjal"\"'
nchar_symbols = f'L{binary_symbols}'
input_sql, stb_name = self.genFullTypeSql(c7=binary_symbols, c8=nchar_symbols, t7=binary_symbols, t8=nchar_symbols)
self.resCmp(input_sql, stb_name)
def tsCheckCase(self):
"""
test ts list --> ["1626006833639000000ns", "1626006833639019us", "1626006833640ms", "1626006834s", "1626006822639022"]
# ! us级时间戳都为0时,数据库中查询显示,但python接口拿到的结果不显示 .000000的情况请确认,目前修改时间处理代码可以通过
"""
self.cleanStb()
ts_list = ["1626006833639000000ns", "1626006833639019us", "1626006833640ms", "1626006834s", "1626006822639022", 0]
for ts in ts_list:
input_sql, stb_name = self.genFullTypeSql(ts=ts)
self.resCmp(input_sql, stb_name, ts=ts)
def idSeqCheckCase(self):
"""
check id.index in tags
eg: t0=**,id=**,t1=**
"""
self.cleanStb()
input_sql, stb_name = self.genFullTypeSql(id_change_tag=True)
self.resCmp(input_sql, stb_name)
def idUpperCheckCase(self):
"""
check id param
eg: id and ID
"""
self.cleanStb()
input_sql, stb_name = self.genFullTypeSql(id_upper_tag=True)
self.resCmp(input_sql, stb_name)
input_sql, stb_name = self.genFullTypeSql(id_change_tag=True, id_upper_tag=True)
self.resCmp(input_sql, stb_name)
def noIdCheckCase(self):
"""
id not exist
"""
self.cleanStb()
input_sql, stb_name = self.genFullTypeSql(id_noexist_tag=True)
self.resCmp(input_sql, stb_name)
query_sql = f"select tbname from {stb_name}"
res_row_list = self.resHandle(query_sql, True)[0]
if len(res_row_list[0][0]) > 0:
tdSql.checkColNameList(res_row_list, res_row_list)
else:
tdSql.checkColNameList(res_row_list, "please check noIdCheckCase")
def maxColTagCheckCase(self):
"""
max tag count is 128
max col count is ??
"""
for input_sql in [self.genLongSql(128, 1)[0], self.genLongSql(1, 4094)[0]]:
self.cleanStb()
code = self._conn.insertLines([input_sql])
tdSql.checkEqual(code, 0)
for input_sql in [self.genLongSql(129, 1)[0], self.genLongSql(1, 4095)[0]]:
self.cleanStb()
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
def idIllegalNameCheckCase(self):
"""
test illegal id name
mix "`~!@#$¥%^&*()-+={}|[]、「」【】\:;《》<>?"
"""
self.cleanStb()
rstr = list("`~!@#$¥%^&*()-+={}|[]、「」【】\:;《》<>?")
for i in rstr:
input_sql = self.genFullTypeSql(tb_name=f"\"aaa{i}bbb\"")[0]
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
def idStartWithNumCheckCase(self):
"""
id is start with num
"""
self.cleanStb()
input_sql = self.genFullTypeSql(tb_name=f"\"1aaabbb\"")[0]
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
def nowTsCheckCase(self):
"""
check now unsupported
"""
self.cleanStb()
input_sql = self.genFullTypeSql(ts="now")[0]
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
def dateFormatTsCheckCase(self):
"""
check date format ts unsupported
"""
self.cleanStb()
input_sql = self.genFullTypeSql(ts="2021-07-21\ 19:01:46.920")[0]
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
def illegalTsCheckCase(self):
"""
check ts format like 16260068336390us19
"""
self.cleanStb()
input_sql = self.genFullTypeSql(ts="16260068336390us19")[0]
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
def tagValueLengthCheckCase(self):
"""
check full type tag value limit
"""
self.cleanStb()
# i8
for t1 in ["-127i8", "127i8"]:
input_sql, stb_name = self.genFullTypeSql(t1=t1)
self.resCmp(input_sql, stb_name)
for t1 in ["-128i8", "128i8"]:
input_sql = self.genFullTypeSql(t1=t1)[0]
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
#i16
for t2 in ["-32767i16", "32767i16"]:
input_sql, stb_name = self.genFullTypeSql(t2=t2)
self.resCmp(input_sql, stb_name)
for t2 in ["-32768i16", "32768i16"]:
input_sql = self.genFullTypeSql(t2=t2)[0]
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
#i32
for t3 in ["-2147483647i32", "2147483647i32"]:
input_sql, stb_name = self.genFullTypeSql(t3=t3)
self.resCmp(input_sql, stb_name)
for t3 in ["-2147483648i32", "2147483648i32"]:
input_sql = self.genFullTypeSql(t3=t3)[0]
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
#i64
for t4 in ["-9223372036854775807i64", "9223372036854775807i64"]:
input_sql, stb_name = self.genFullTypeSql(t4=t4)
self.resCmp(input_sql, stb_name)
for t4 in ["-9223372036854775808i64", "9223372036854775808i64"]:
input_sql = self.genFullTypeSql(t4=t4)[0]
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
# f32
for t5 in [f"{-3.4028234663852885981170418348451692544*(10**38)}f32", f"{3.4028234663852885981170418348451692544*(10**38)}f32"]:
input_sql, stb_name = self.genFullTypeSql(t5=t5)
self.resCmp(input_sql, stb_name)
# * limit set to 4028234664*(10**38)
for t5 in [f"{-3.4028234664*(10**38)}f32", f"{3.4028234664*(10**38)}f32"]:
input_sql = self.genFullTypeSql(t5=t5)[0]
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
# f64
for t6 in [f'{-1.79769*(10**308)}f64', f'{-1.79769*(10**308)}f64']:
input_sql, stb_name = self.genFullTypeSql(t6=t6)
self.resCmp(input_sql, stb_name)
# * limit set to 1.797693134862316*(10**308)
for c6 in [f'{-1.797693134862316*(10**308)}f64', f'{-1.797693134862316*(10**308)}f64']:
input_sql = self.genFullTypeSql(c6=c6)[0]
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
# binary
stb_name = self.getLongName(7, "letters")
input_sql = f'{stb_name},t0=t,t1="{self.getLongName(16374, "letters")}" c0=f 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
tdSql.checkEqual(code, 0)
input_sql = f'{stb_name},t0=t,t1="{self.getLongName(16375, "letters")}" c0=f 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
# nchar
# * legal nchar could not be larger than 16374/4
stb_name = self.getLongName(7, "letters")
input_sql = f'{stb_name},t0=t,t1=L"{self.getLongName(4093, "letters")}" c0=f 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
tdSql.checkEqual(code, 0)
input_sql = f'{stb_name},t0=t,t1=L"{self.getLongName(4094, "letters")}" c0=f 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
def colValueLengthCheckCase(self):
"""
check full type col value limit
"""
self.cleanStb()
# i8
for c1 in ["-127i8", "127i8"]:
input_sql, stb_name = self.genFullTypeSql(c1=c1)
self.resCmp(input_sql, stb_name)
for c1 in ["-128i8", "128i8"]:
input_sql = self.genFullTypeSql(c1=c1)[0]
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
# i16
for c2 in ["-32767i16"]:
input_sql, stb_name = self.genFullTypeSql(c2=c2)
self.resCmp(input_sql, stb_name)
for c2 in ["-32768i16", "32768i16"]:
input_sql = self.genFullTypeSql(c2=c2)[0]
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
# i32
for c3 in ["-2147483647i32"]:
input_sql, stb_name = self.genFullTypeSql(c3=c3)
self.resCmp(input_sql, stb_name)
for c3 in ["-2147483648i32", "2147483648i32"]:
input_sql = self.genFullTypeSql(c3=c3)[0]
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
# i64
for c4 in ["-9223372036854775807i64"]:
input_sql, stb_name = self.genFullTypeSql(c4=c4)
self.resCmp(input_sql, stb_name)
for c4 in ["-9223372036854775808i64", "9223372036854775808i64"]:
input_sql = self.genFullTypeSql(c4=c4)[0]
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
# f32
for c5 in [f"{-3.4028234663852885981170418348451692544*(10**38)}f32", f"{3.4028234663852885981170418348451692544*(10**38)}f32"]:
input_sql, stb_name = self.genFullTypeSql(c5=c5)
self.resCmp(input_sql, stb_name)
# * limit set to 4028234664*(10**38)
for c5 in [f"{-3.4028234664*(10**38)}f32", f"{3.4028234664*(10**38)}f32"]:
input_sql = self.genFullTypeSql(c5=c5)[0]
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
# f64
for c6 in [f'{-1.79769313486231570814527423731704356798070567525844996598917476803157260780*(10**308)}f64', f'{-1.79769313486231570814527423731704356798070567525844996598917476803157260780*(10**308)}f64']:
input_sql, stb_name = self.genFullTypeSql(c6=c6)
self.resCmp(input_sql, stb_name)
# * limit set to 1.797693134862316*(10**308)
for c6 in [f'{-1.797693134862316*(10**308)}f64', f'{-1.797693134862316*(10**308)}f64']:
input_sql = self.genFullTypeSql(c6=c6)[0]
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
# # binary
stb_name = self.getLongName(7, "letters")
input_sql = f'{stb_name},t0=t c0=f,c1="{self.getLongName(16374, "letters")}" 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
tdSql.checkEqual(code, 0)
input_sql = f'{stb_name},t0=t c0=f,c1="{self.getLongName(16375, "letters")}" 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
# nchar
# * legal nchar could not be larger than 16374/4
stb_name = self.getLongName(7, "letters")
input_sql = f'{stb_name},t0=t c0=f,c1=L"{self.getLongName(4093, "letters")}" 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
tdSql.checkEqual(code, 0)
input_sql = f'{stb_name},t0=t c0=f,c1=L"{self.getLongName(4094, "letters")}" 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
def tagColIllegalValueCheckCase(self):
"""
test illegal tag col value
"""
self.cleanStb()
# bool
for i in ["TrUe", "tRue", "trUe", "truE", "FalsE", "fAlse", "faLse", "falSe", "falsE"]:
input_sql1 = self.genFullTypeSql(t0=i)[0]
code = self._conn.insertLines([input_sql1])
tdSql.checkNotEqual(code, 0)
input_sql2 = self.genFullTypeSql(c0=i)[0]
code = self._conn.insertLines([input_sql2])
tdSql.checkNotEqual(code, 0)
# i8 i16 i32 i64 f32 f64
for input_sql in [
self.genFullTypeSql(t1="1s2i8")[0],
self.genFullTypeSql(t2="1s2i16")[0],
self.genFullTypeSql(t3="1s2i32")[0],
self.genFullTypeSql(t4="1s2i64")[0],
self.genFullTypeSql(t5="11.1s45f32")[0],
self.genFullTypeSql(t6="11.1s45f64")[0],
self.genFullTypeSql(c1="1s2i8")[0],
self.genFullTypeSql(c2="1s2i16")[0],
self.genFullTypeSql(c3="1s2i32")[0],
self.genFullTypeSql(c4="1s2i64")[0],
self.genFullTypeSql(c5="11.1s45f32")[0],
self.genFullTypeSql(c6="11.1s45f64")[0],
self.genFullTypeSql(c9="1s1u64")[0]
]:
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
# check binary and nchar blank
stb_name = self.getLongName(7, "letters")
input_sql1 = f'{stb_name},t0=t c0=f,c1="abc aaa" 1626006833639000000ns'
input_sql2 = f'{stb_name},t0=t c0=f,c1=L"abc aaa" 1626006833639000000ns'
input_sql3 = f'{stb_name},t0=t,t1="abc aaa" c0=f 1626006833639000000ns'
input_sql4 = f'{stb_name},t0=t,t1=L"abc aaa" c0=f 1626006833639000000ns'
for input_sql in [input_sql1, input_sql2, input_sql3, input_sql4]:
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
# check accepted binary and nchar symbols
# # * ~!@#$¥%^&*()-+={}|[]、「」:;
for symbol in list('~!@#$¥%^&*()-+={}|[]、「」:;'):
input_sql1 = f'{stb_name},t0=t c0=f,c1="abc{symbol}aaa" 1626006833639000000ns'
input_sql2 = f'{stb_name},t0=t,t1="abc{symbol}aaa" c0=f 1626006833639000000ns'
code = self._conn.insertLines([input_sql1])
tdSql.checkEqual(code, 0)
code = self._conn.insertLines([input_sql2])
tdSql.checkEqual(code, 0)
def duplicateIdTagColInsertCheckCase(self):
"""
check duplicate Id Tag Col
"""
self.cleanStb()
input_sql_id = self.genFullTypeSql(id_double_tag=True)[0]
code = self._conn.insertLines([input_sql_id])
tdSql.checkNotEqual(code, 0)
input_sql = self.genFullTypeSql()[0]
input_sql_tag = input_sql.replace("t5", "t6")
code = self._conn.insertLines([input_sql_tag])
tdSql.checkNotEqual(code, 0)
input_sql = self.genFullTypeSql()[0]
input_sql_col = input_sql.replace("c5", "c6")
code = self._conn.insertLines([input_sql_col])
tdSql.checkNotEqual(code, 0)
input_sql = self.genFullTypeSql()[0]
input_sql_col = input_sql.replace("c5", "C6")
code = self._conn.insertLines([input_sql_col])
tdSql.checkNotEqual(code, 0)
##### stb exist #####
def noIdStbExistCheckCase(self):
"""
case no id when stb exist
"""
self.cleanStb()
input_sql, stb_name = self.genFullTypeSql(t0="f", c0="f")
self.resCmp(input_sql, stb_name)
input_sql, stb_name = self.genFullTypeSql(stb_name=stb_name, id_noexist_tag=True, t0="f", c0="f")
self.resCmp(input_sql, stb_name, condition='where tbname like "t_%"')
tdSql.query(f"select * from {stb_name}")
tdSql.checkRows(2)
# TODO cover other case
def duplicateInsertExistCheckCase(self):
"""
check duplicate insert when stb exist
"""
self.cleanStb()
input_sql, stb_name = self.genFullTypeSql()
self.resCmp(input_sql, stb_name)
code = self._conn.insertLines([input_sql])
tdSql.checkEqual(code, 0)
self.resCmp(input_sql, stb_name)
def tagColBinaryNcharLengthCheckCase(self):
"""
check length increase
"""
self.cleanStb()
input_sql, stb_name = self.genFullTypeSql()
self.resCmp(input_sql, stb_name)
tb_name = self.getLongName(5, "letters")
input_sql, stb_name = self.genFullTypeSql(stb_name=stb_name, tb_name=tb_name,t7="\"binaryTagValuebinaryTagValue\"", t8="L\"ncharTagValuencharTagValue\"", c7="\"binaryTagValuebinaryTagValue\"", c8="L\"ncharTagValuencharTagValue\"")
self.resCmp(input_sql, stb_name, condition=f'where tbname like "{tb_name}"')
def tagColAddDupIDCheckCase(self):
"""
check column and tag count add, stb and tb duplicate
* tag: alter table ...
* col: when update==0 and ts is same, unchange
* so this case tag&&value will be added,
* col is added without value when update==0
* col is added with value when update==1
"""
self.cleanStb()
tb_name = self.getLongName(7, "letters")
for db_update_tag in [0, 1]:
if db_update_tag == 1 :
self.createDb("test_update", db_update_tag=db_update_tag)
input_sql, stb_name = self.genFullTypeSql(tb_name=tb_name, t0="f", c0="f")
self.resCmp(input_sql, stb_name)
self.genFullTypeSql(stb_name=stb_name, tb_name=tb_name, t0="f", c0="f", ct_add_tag=True)
if db_update_tag == 1 :
self.resCmp(input_sql, stb_name, condition=f'where tbname like "{tb_name}"')
else:
self.resCmp(input_sql, stb_name, condition=f'where tbname like "{tb_name}"', none_check_tag=True)
def tagColAddCheckCase(self):
"""
check column and tag count add
"""
self.cleanStb()
tb_name = self.getLongName(7, "letters")
input_sql, stb_name = self.genFullTypeSql(tb_name=tb_name, t0="f", c0="f")
self.resCmp(input_sql, stb_name)
tb_name_1 = self.getLongName(7, "letters")
input_sql, stb_name = self.genFullTypeSql(stb_name=stb_name, tb_name=tb_name_1, t0="f", c0="f", ct_add_tag=True)
self.resCmp(input_sql, stb_name, condition=f'where tbname like "{tb_name_1}"')
res_row_list = self.resHandle(f"select c10,c11,t10,t11 from {tb_name}", True)[0]
tdSql.checkEqual(res_row_list[0], ['None', 'None', 'None', 'None'])
self.resCmp(input_sql, stb_name, condition=f'where tbname like "{tb_name}"', none_check_tag=True)
def tagMd5Check(self):
"""
condition: stb not change
insert two table, keep tag unchange, change col
"""
self.cleanStb()
input_sql, stb_name = self.genFullTypeSql(t0="f", c0="f", id_noexist_tag=True)
self.resCmp(input_sql, stb_name)
tb_name1 = self.getNoIdTbName(stb_name)
input_sql, stb_name = self.genFullTypeSql(stb_name=stb_name, t0="f", c0="f", id_noexist_tag=True)
self.resCmp(input_sql, stb_name)
tb_name2 = self.getNoIdTbName(stb_name)
tdSql.query(f"select * from {stb_name}")
tdSql.checkRows(1)
tdSql.checkEqual(tb_name1, tb_name2)
input_sql, stb_name = self.genFullTypeSql(stb_name=stb_name, t0="f", c0="f", id_noexist_tag=True, ct_add_tag=True)
self._conn.insertLines([input_sql])
tb_name3 = self.getNoIdTbName(stb_name)
tdSql.query(f"select * from {stb_name}")
tdSql.checkRows(2)
tdSql.checkNotEqual(tb_name1, tb_name3)
# * tag binary max is 16384, col+ts binary max 49151
def tagColBinaryMaxLengthCheckCase(self):
"""
every binary and nchar must be length+2
"""
self.cleanStb()
stb_name = self.getLongName(7, "letters")
tb_name = f'{stb_name}_1'
input_sql = f'{stb_name},id="{tb_name}",t0=t c0=f 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
# * every binary and nchar must be length+2, so here is two tag, max length could not larger than 16384-2*2
input_sql = f'{stb_name},t0=t,t1="{self.getLongName(16374, "letters")}",t2="{self.getLongName(5, "letters")}" c0=f 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
tdSql.checkEqual(code, 0)
tdSql.query(f"select * from {stb_name}")
tdSql.checkRows(2)
input_sql = f'{stb_name},t0=t,t1="{self.getLongName(16374, "letters")}",t2="{self.getLongName(6, "letters")}" c0=f 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
tdSql.query(f"select * from {stb_name}")
tdSql.checkRows(2)
# # * check col,col+ts max in describe ---> 16143
input_sql = f'{stb_name},t0=t c0=f,c1="{self.getLongName(16374, "letters")}",c2="{self.getLongName(16374, "letters")}",c3="{self.getLongName(16374, "letters")}",c4="{self.getLongName(12, "letters")}" 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
tdSql.checkEqual(code, 0)
tdSql.query(f"select * from {stb_name}")
tdSql.checkRows(3)
input_sql = f'{stb_name},t0=t c0=f,c1="{self.getLongName(16374, "letters")}",c2="{self.getLongName(16374, "letters")}",c3="{self.getLongName(16374, "letters")}",c4="{self.getLongName(13, "letters")}" 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
tdSql.query(f"select * from {stb_name}")
tdSql.checkRows(3)
# * tag nchar max is 16374/4, col+ts nchar max 49151
def tagColNcharMaxLengthCheckCase(self):
"""
check nchar length limit
"""
self.cleanStb()
stb_name = self.getLongName(7, "letters")
tb_name = f'{stb_name}_1'
input_sql = f'{stb_name},id="{tb_name}",t0=t c0=f 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
# * legal nchar could not be larger than 16374/4
input_sql = f'{stb_name},t0=t,t1=L"{self.getLongName(4093, "letters")}",t2=L"{self.getLongName(1, "letters")}" c0=f 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
tdSql.checkEqual(code, 0)
tdSql.query(f"select * from {stb_name}")
tdSql.checkRows(2)
input_sql = f'{stb_name},t0=t,t1=L"{self.getLongName(4093, "letters")}",t2=L"{self.getLongName(2, "letters")}" c0=f 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
tdSql.query(f"select * from {stb_name}")
tdSql.checkRows(2)
input_sql = f'{stb_name},t0=t c0=f,c1=L"{self.getLongName(4093, "letters")}",c2=L"{self.getLongName(4093, "letters")}",c3=L"{self.getLongName(4093, "letters")}",c4=L"{self.getLongName(4, "letters")}" 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
tdSql.checkEqual(code, 0)
tdSql.query(f"select * from {stb_name}")
tdSql.checkRows(3)
input_sql = f'{stb_name},t0=t c0=f,c1=L"{self.getLongName(4093, "letters")}",c2=L"{self.getLongName(4093, "letters")}",c3=L"{self.getLongName(4093, "letters")}",c4=L"{self.getLongName(5, "letters")}" 1626006833639000000ns'
code = self._conn.insertLines([input_sql])
tdSql.checkNotEqual(code, 0)
tdSql.query(f"select * from {stb_name}")
tdSql.checkRows(3)
def batchInsertCheckCase(self):
"""
test batch insert
"""
self.cleanStb()
stb_name = self.getLongName(8, "letters")
tdSql.execute(f'create stable {stb_name}(ts timestamp, f int) tags(t1 bigint)')
lines = ["st123456,t1=3i64,t2=4f64,t3=\"t3\" c1=3i64,c3=L\"passit\",c2=false,c4=4f64 1626006833639000000ns",
"st123456,t1=4i64,t3=\"t4\",t2=5f64,t4=5f64 c1=3i64,c3=L\"passitagin\",c2=true,c4=5f64,c5=5f64 1626006833640000000ns",
f"{stb_name},t2=5f64,t3=L\"ste\" c1=true,c2=4i64,c3=\"iam\" 1626056811823316532ns",
"stf567890,t1=4i64,t3=\"t4\",t2=5f64,t4=5f64 c1=3i64,c3=L\"passitagin\",c2=true,c4=5f64,c5=5f64,c6=7u64 1626006933640000000ns",
"st123456,t1=4i64,t2=5f64,t3=\"t4\" c1=3i64,c3=L\"passitagain\",c2=true,c4=5f64 1626006833642000000ns",
f"{stb_name},t2=5f64,t3=L\"ste2\" c3=\"iamszhou\",c4=false 1626056811843316532ns",
f"{stb_name},t2=5f64,t3=L\"ste2\" c3=\"iamszhou\",c4=false,c5=32i8,c6=64i16,c7=32i32,c8=88.88f32 1626056812843316532ns",
"st123456,t1=4i64,t3=\"t4\",t2=5f64,t4=5f64 c1=3i64,c3=L\"passitagin\",c2=true,c4=5f64,c5=5f64,c6=7u64 1626006933640000000ns",
"st123456,t1=4i64,t3=\"t4\",t2=5f64,t4=5f64 c1=3i64,c3=L\"passitagin_stf\",c2=false,c5=5f64,c6=7u64 1626006933641000000ns"
]
code = self._conn.insertLines(lines)
tdSql.checkEqual(code, 0)
def multiInsertCheckCase(self, count):
"""
test multi insert
"""
self.cleanStb()
sql_list = []
stb_name = self.getLongName(8, "letters")
tdSql.execute(f'create stable {stb_name}(ts timestamp, f int) tags(t1 bigint)')
for i in range(count):
input_sql = self.genFullTypeSql(stb_name=stb_name, t7=f'"{self.getLongName(8, "letters")}"', c7=f'"{self.getLongName(8, "letters")}"', id_noexist_tag=True)[0]
sql_list.append(input_sql)
code = self._conn.insertLines(sql_list)
tdSql.checkEqual(code, 0)
def batchErrorInsertCheckCase(self):
"""
test batch error insert
"""
self.cleanStb()
stb_name = self.getLongName(8, "letters")
lines = ["st123456,t1=3i64,t2=4f64,t3=\"t3\" c1=3i64,c3=L\"passit\",c2=false,c4=4f64 1626006833639000000ns",
f"{stb_name},t2=5f64,t3=L\"ste\" c1=tRue,c2=4i64,c3=\"iam\" 1626056811823316532ns"]
code = self._conn.insertLines(lines)
tdSql.checkNotEqual(code, 0)
def genSqlList(self, count=5, stb_name="", tb_name=""):
"""
stb --> supertable
tb --> table
ts --> timestamp, same default
col --> column, same default
tag --> tag, same default
d --> different
s --> same
a --> add
m --> minus
"""
d_stb_d_tb_list = list()
s_stb_s_tb_list = list()
s_stb_s_tb_a_col_a_tag_list = list()
s_stb_s_tb_m_col_m_tag_list = list()
s_stb_d_tb_list = list()
s_stb_d_tb_a_col_m_tag_list = list()
s_stb_d_tb_a_tag_m_col_list = list()
s_stb_s_tb_d_ts_list = list()
s_stb_s_tb_d_ts_a_col_m_tag_list = list()
s_stb_s_tb_d_ts_a_tag_m_col_list = list()
s_stb_d_tb_d_ts_list = list()
s_stb_d_tb_d_ts_a_col_m_tag_list = list()
s_stb_d_tb_d_ts_a_tag_m_col_list = list()
for i in range(count):
d_stb_d_tb_list.append(self.genFullTypeSql(t0="f", c0="f"))
s_stb_s_tb_list.append(self.genFullTypeSql(stb_name=stb_name, tb_name=tb_name, t7=f'"{self.getLongName(8, "letters")}"', c7=f'"{self.getLongName(8, "letters")}"'))
s_stb_s_tb_a_col_a_tag_list.append(self.genFullTypeSql(stb_name=stb_name, tb_name=tb_name, t7=f'"{self.getLongName(8, "letters")}"', c7=f'"{self.getLongName(8, "letters")}"', ct_add_tag=True))
s_stb_s_tb_m_col_m_tag_list.append(self.genFullTypeSql(stb_name=stb_name, tb_name=tb_name, t7=f'"{self.getLongName(8, "letters")}"', c7=f'"{self.getLongName(8, "letters")}"', ct_min_tag=True))
s_stb_d_tb_list.append(self.genFullTypeSql(stb_name=stb_name, t7=f'"{self.getLongName(8, "letters")}"', c7=f'"{self.getLongName(8, "letters")}"', id_noexist_tag=True))
s_stb_d_tb_a_col_m_tag_list.append(self.genFullTypeSql(stb_name=stb_name, t7=f'"{self.getLongName(8, "letters")}"', c7=f'"{self.getLongName(8, "letters")}"', id_noexist_tag=True, ct_am_tag=True))
s_stb_d_tb_a_tag_m_col_list.append(self.genFullTypeSql(stb_name=stb_name, t7=f'"{self.getLongName(8, "letters")}"', c7=f'"{self.getLongName(8, "letters")}"', id_noexist_tag=True, ct_ma_tag=True))
s_stb_s_tb_d_ts_list.append(self.genFullTypeSql(stb_name=stb_name, tb_name=tb_name, t7=f'"{self.getLongName(8, "letters")}"', c7=f'"{self.getLongName(8, "letters")}"', ts=0))
s_stb_s_tb_d_ts_a_col_m_tag_list.append(self.genFullTypeSql(stb_name=stb_name, tb_name=tb_name, t7=f'"{self.getLongName(8, "letters")}"', c7=f'"{self.getLongName(8, "letters")}"', ts=0, ct_am_tag=True))
s_stb_s_tb_d_ts_a_tag_m_col_list.append(self.genFullTypeSql(stb_name=stb_name, tb_name=tb_name, t7=f'"{self.getLongName(8, "letters")}"', c7=f'"{self.getLongName(8, "letters")}"', ts=0, ct_ma_tag=True))
s_stb_d_tb_d_ts_list.append(self.genFullTypeSql(stb_name=stb_name, t7=f'"{self.getLongName(8, "letters")}"', c7=f'"{self.getLongName(8, "letters")}"', id_noexist_tag=True, ts=0))
s_stb_d_tb_d_ts_a_col_m_tag_list.append(self.genFullTypeSql(stb_name=stb_name, t7=f'"{self.getLongName(8, "letters")}"', c7=f'"{self.getLongName(8, "letters")}"', id_noexist_tag=True, ts=0, ct_am_tag=True))
s_stb_d_tb_d_ts_a_tag_m_col_list.append(self.genFullTypeSql(stb_name=stb_name, t7=f'"{self.getLongName(8, "letters")}"', c7=f'"{self.getLongName(8, "letters")}"', id_noexist_tag=True, ts=0, ct_ma_tag=True))
return d_stb_d_tb_list, s_stb_s_tb_list, s_stb_s_tb_a_col_a_tag_list, s_stb_s_tb_m_col_m_tag_list, \
s_stb_d_tb_list, s_stb_d_tb_a_col_m_tag_list, s_stb_d_tb_a_tag_m_col_list, s_stb_s_tb_d_ts_list, \
s_stb_s_tb_d_ts_a_col_m_tag_list, s_stb_s_tb_d_ts_a_tag_m_col_list, s_stb_d_tb_d_ts_list, \
s_stb_d_tb_d_ts_a_col_m_tag_list, s_stb_d_tb_d_ts_a_tag_m_col_list
def genMultiThreadSeq(self, sql_list):
tlist = list()
for insert_sql in sql_list:
t = threading.Thread(target=self._conn.insertLines,args=([insert_sql[0]],))
tlist.append(t)
return tlist
def multiThreadRun(self, tlist):
for t in tlist:
t.start()
for t in tlist:
t.join()
def stbInsertMultiThreadCheckCase(self):
"""
thread input different stb
"""
self.cleanStb()
input_sql = self.genSqlList()[0]
self.multiThreadRun(self.genMultiThreadSeq(input_sql))
tdSql.query(f"show tables;")
tdSql.checkRows(5)
def sStbStbDdataInsertMultiThreadCheckCase(self):
"""
thread input same stb tb, different data, result keep first data
"""
self.cleanStb()
tb_name = self.getLongName(7, "letters")
input_sql, stb_name = self.genFullTypeSql(tb_name=tb_name)
self.resCmp(input_sql, stb_name)
s_stb_s_tb_list = self.genSqlList(stb_name=stb_name, tb_name=tb_name)[1]
self.multiThreadRun(self.genMultiThreadSeq(s_stb_s_tb_list))
tdSql.query(f"show tables;")
tdSql.checkRows(1)
expected_tb_name = self.getNoIdTbName(stb_name)[0]
tdSql.checkEqual(tb_name, expected_tb_name)
tdSql.query(f"select * from {stb_name};")
tdSql.checkRows(1)
def sStbStbDdataAtcInsertMultiThreadCheckCase(self):
"""
thread input same stb tb, different data, add columes and tags, result keep first data
"""
self.cleanStb()
tb_name = self.getLongName(7, "letters")
input_sql, stb_name = self.genFullTypeSql(tb_name=tb_name)
self.resCmp(input_sql, stb_name)
s_stb_s_tb_a_col_a_tag_list = self.genSqlList(stb_name=stb_name, tb_name=tb_name)[2]
self.multiThreadRun(self.genMultiThreadSeq(s_stb_s_tb_a_col_a_tag_list))
tdSql.query(f"show tables;")
tdSql.checkRows(1)
expected_tb_name = self.getNoIdTbName(stb_name)[0]
tdSql.checkEqual(tb_name, expected_tb_name)
tdSql.query(f"select * from {stb_name};")
tdSql.checkRows(1)
def sStbStbDdataMtcInsertMultiThreadCheckCase(self):
"""
thread input same stb tb, different data, minus columes and tags, result keep first data
"""
self.cleanStb()
tb_name = self.getLongName(7, "letters")
input_sql, stb_name = self.genFullTypeSql(tb_name=tb_name)
self.resCmp(input_sql, stb_name)
s_stb_s_tb_m_col_m_tag_list = self.genSqlList(stb_name=stb_name, tb_name=tb_name)[3]
self.multiThreadRun(self.genMultiThreadSeq(s_stb_s_tb_m_col_m_tag_list))
tdSql.query(f"show tables;")
tdSql.checkRows(1)
expected_tb_name = self.getNoIdTbName(stb_name)[0]
tdSql.checkEqual(tb_name, expected_tb_name)
tdSql.query(f"select * from {stb_name};")
tdSql.checkRows(1)
def sStbDtbDdataInsertMultiThreadCheckCase(self):
"""
thread input same stb, different tb, different data
"""
self.cleanStb()
input_sql, stb_name = self.genFullTypeSql()
self.resCmp(input_sql, stb_name)
s_stb_d_tb_list = self.genSqlList(stb_name=stb_name)[4]
self.multiThreadRun(self.genMultiThreadSeq(s_stb_d_tb_list))
tdSql.query(f"show tables;")
tdSql.checkRows(6)
def sStbDtbDdataAcMtInsertMultiThreadCheckCase(self):
"""
#! concurrency conflict
"""
"""
thread input same stb, different tb, different data, add col, mul tag
"""
self.cleanStb()
input_sql, stb_name = self.genFullTypeSql()
self.resCmp(input_sql, stb_name)
s_stb_d_tb_a_col_m_tag_list = self.genSqlList(stb_name=stb_name)[5]
self.multiThreadRun(self.genMultiThreadSeq(s_stb_d_tb_a_col_m_tag_list))
tdSql.query(f"show tables;")
tdSql.checkRows(6)
def sStbDtbDdataAtMcInsertMultiThreadCheckCase(self):
"""
#! concurrency conflict
"""
"""
thread input same stb, different tb, different data, add tag, mul col
"""
self.cleanStb()
input_sql, stb_name = self.genFullTypeSql()
self.resCmp(input_sql, stb_name)
s_stb_d_tb_a_tag_m_col_list = self.genSqlList(stb_name=stb_name)[6]
self.multiThreadRun(self.genMultiThreadSeq(s_stb_d_tb_a_tag_m_col_list))
tdSql.query(f"show tables;")
tdSql.checkRows(6)
def sStbStbDdataDtsInsertMultiThreadCheckCase(self):
"""
thread input same stb tb, different ts
"""
self.cleanStb()
tb_name = self.getLongName(7, "letters")
input_sql, stb_name = self.genFullTypeSql(tb_name=tb_name)
self.resCmp(input_sql, stb_name)
s_stb_s_tb_d_ts_list = self.genSqlList(stb_name=stb_name, tb_name=tb_name)[7]
self.multiThreadRun(self.genMultiThreadSeq(s_stb_s_tb_d_ts_list))
tdSql.query(f"show tables;")
tdSql.checkRows(1)
tdSql.query(f"select * from {stb_name}")
tdSql.checkRows(6)
def sStbStbDdataDtsAcMtInsertMultiThreadCheckCase(self):
"""
thread input same stb tb, different ts, add col, mul tag
"""
self.cleanStb()
tb_name = self.getLongName(7, "letters")
input_sql, stb_name = self.genFullTypeSql(tb_name=tb_name)
self.resCmp(input_sql, stb_name)
s_stb_s_tb_d_ts_a_col_m_tag_list = self.genSqlList(stb_name=stb_name, tb_name=tb_name)[8]
self.multiThreadRun(self.genMultiThreadSeq(s_stb_s_tb_d_ts_a_col_m_tag_list))
tdSql.query(f"show tables;")
tdSql.checkRows(1)
tdSql.query(f"select * from {stb_name}")
tdSql.checkRows(6)
tdSql.query(f"select * from {stb_name} where t8 is not NULL")
tdSql.checkRows(6)
tdSql.query(f"select * from {tb_name} where c11 is not NULL;")
tdSql.checkRows(5)
def sStbStbDdataDtsAtMcInsertMultiThreadCheckCase(self):
"""
thread input same stb tb, different ts, add tag, mul col
"""
self.cleanStb()
tb_name = self.getLongName(7, "letters")
input_sql, stb_name = self.genFullTypeSql(tb_name=tb_name)
self.resCmp(input_sql, stb_name)
s_stb_s_tb_d_ts_a_tag_m_col_list = self.genSqlList(stb_name=stb_name, tb_name=tb_name)[9]
self.multiThreadRun(self.genMultiThreadSeq(s_stb_s_tb_d_ts_a_tag_m_col_list))
tdSql.query(f"show tables;")
tdSql.checkRows(1)
tdSql.query(f"select * from {stb_name}")
tdSql.checkRows(6)
for c in ["c7", "c8", "c9"]:
tdSql.query(f"select * from {stb_name} where {c} is NULL")
tdSql.checkRows(5)
for t in ["t10", "t11"]:
tdSql.query(f"select * from {stb_name} where {t} is not NULL;")
tdSql.checkRows(6)
def sStbDtbDdataDtsInsertMultiThreadCheckCase(self):
"""
thread input same stb, different tb, data, ts
"""
self.cleanStb()
input_sql, stb_name = self.genFullTypeSql()
self.resCmp(input_sql, stb_name)
s_stb_d_tb_d_ts_list = self.genSqlList(stb_name=stb_name)[10]
self.multiThreadRun(self.genMultiThreadSeq(s_stb_d_tb_d_ts_list))
tdSql.query(f"show tables;")
tdSql.checkRows(6)
def sStbDtbDdataDtsAcMtInsertMultiThreadCheckCase(self):
"""
# ! concurrency conflict
"""
"""
thread input same stb, different tb, data, ts, add col, mul tag
"""
self.cleanStb()
input_sql, stb_name = self.genFullTypeSql()
self.resCmp(input_sql, stb_name)
s_stb_d_tb_d_ts_a_col_m_tag_list = self.genSqlList(stb_name=stb_name)[11]
self.multiThreadRun(self.genMultiThreadSeq(s_stb_d_tb_d_ts_a_col_m_tag_list))
tdSql.query(f"show tables;")
tdSql.checkRows(6)
def test(self):
input_sql1 = "rfasta,id=\"rfasta_1\",t0=true,t1=127i8,t2=32767i16,t3=2147483647i32,t4=9223372036854775807i64,t5=11.12345f32,t6=22.123456789f64,t7=\"ddzhiksj\",t8=L\"ncharTagValue\" c0=True,c1=127i8,c2=32767i16,c3=2147483647i32,c4=9223372036854775807i64,c5=11.12345f32,c6=22.123456789f64,c7=\"bnhwlgvj\",c8=L\"ncharTagValue\",c9=7u64 1626006933640000000ns"
input_sql2 = "rfasta,id=\"rfasta_1\",t0=true,t1=127i8,t2=32767i16,t3=2147483647i32,t4=9223372036854775807i64,t5=11.12345f32,t6=22.123456789f64 c0=True,c1=127i8,c2=32767i16,c3=2147483647i32,c4=9223372036854775807i64,c5=11.12345f32,c6=22.123456789f64 1626006933640000000ns"
code = self._conn.insertLines([input_sql1])
code = self._conn.insertLines([input_sql2])
print(code)
# self._conn.insertLines([input_sql2])
# input_sql3 = f'abcd,id="cc¥Ec",t0=True,t1=127i8,t2=32767i16,t3=2147483647i32,t4=9223372036854775807i64,t5=11.12345f32,t6=22.123456789f64,t7="ndsfdrum",t8=L"ncharTagValue" c0=f,c1=127i8,c2=32767i16,c3=2147483647i32,c4=9223372036854775807i64,c5=11.12345f32,c6=22.123456789f64,c7="igwoehkm",c8=L"ncharColValue",c9=7u64 0'
# print(input_sql3)
# input_sql4 = 'hmemeb,id="kilrcrldgf",t0=F,t1=127i8,t2=32767i16,t3=2147483647i32,t4=9223372036854775807i64,t5=11.12345f32,t6=22.123456789f64,t7="fysodjql",t8=L"ncharTagValue" c0=True,c1=127i8,c2=32767i16,c3=2147483647i32,c4=9223372036854775807i64,c5=11.12345f32,c6=22.123456789f64,c7="waszbfvc",c8=L"ncharColValue",c9=7u64 0'
# code = self._conn.insertLines([input_sql3])
# print(code)
# self._conn.insertLines([input_sql4])
def runAll(self):
self.initCheckCase()
self.boolTypeCheckCase()
self.symbolsCheckCase()
self.tsCheckCase()
self.idSeqCheckCase()
self.idUpperCheckCase()
self.noIdCheckCase()
self.maxColTagCheckCase()
self.idIllegalNameCheckCase()
self.idStartWithNumCheckCase()
self.nowTsCheckCase()
self.dateFormatTsCheckCase()
self.illegalTsCheckCase()
self.tagValueLengthCheckCase()
self.colValueLengthCheckCase()
self.tagColIllegalValueCheckCase()
self.duplicateIdTagColInsertCheckCase()
self.noIdStbExistCheckCase()
self.duplicateInsertExistCheckCase()
self.tagColBinaryNcharLengthCheckCase()
self.tagColAddDupIDCheckCase()
self.tagColAddCheckCase()
self.tagMd5Check()
self.tagColBinaryMaxLengthCheckCase()
# self.tagColNcharMaxLengthCheckCase()
self.batchInsertCheckCase()
self.multiInsertCheckCase(1000)
self.batchErrorInsertCheckCase()
# MultiThreads
self.stbInsertMultiThreadCheckCase()
self.sStbStbDdataInsertMultiThreadCheckCase()
self.sStbStbDdataAtcInsertMultiThreadCheckCase()
self.sStbStbDdataMtcInsertMultiThreadCheckCase()
self.sStbDtbDdataInsertMultiThreadCheckCase()
# # ! concurrency conflict
# self.sStbDtbDdataAcMtInsertMultiThreadCheckCase()
# self.sStbDtbDdataAtMcInsertMultiThreadCheckCase()
self.sStbStbDdataDtsInsertMultiThreadCheckCase()
# # ! concurrency conflict
# self.sStbStbDdataDtsAcMtInsertMultiThreadCheckCase()
# self.sStbStbDdataDtsAtMcInsertMultiThreadCheckCase()
self.sStbDtbDdataDtsInsertMultiThreadCheckCase()
# ! concurrency conflict
# self.sStbDtbDdataDtsAcMtInsertMultiThreadCheckCase()
def run(self):
print("running {}".format(__file__))
self.createDb()
self.runAll()
# self.tagColIllegalValueCheckCase()
# self.test()
def stop(self):
tdSql.close()
tdLog.success("%s successfully executed" % __file__)
tdCases.addWindows(__file__, TDTestCase())
tdCases.addLinux(__file__, TDTestCase())
......@@ -65,7 +65,7 @@ class TDSql:
self.queryResult = None
tdLog.info("sql:%s, expect error occured" % (sql))
def query(self, sql):
def query(self, sql, row_tag=None):
self.sql = sql
try:
self.cursor.execute(sql)
......@@ -77,21 +77,27 @@ class TDSql:
args = (caller.filename, caller.lineno, sql, repr(e))
tdLog.notice("%s(%d) failed: sql:%s, %s" % args)
raise Exception(repr(e))
if row_tag:
return self.queryResult
return self.queryRows
def getColNameList(self, sql):
def getColNameList(self, sql, col_tag=None):
self.sql = sql
try:
col_name_list = []
col_type_list = []
self.cursor.execute(sql)
self.queryCols = self.cursor.description
for query_col in self.queryCols:
col_name_list.append(query_col[0])
col_type_list.append(query_col[1])
except Exception as e:
caller = inspect.getframeinfo(inspect.stack()[1][0])
args = (caller.filename, caller.lineno, sql, repr(e))
tdLog.notice("%s(%d) failed: sql:%s, %s" % args)
raise Exception(repr(e))
if col_tag:
return col_name_list, col_type_list
return col_name_list
def waitedQuery(self, sql, expectRows, timeout):
......@@ -245,6 +251,22 @@ class TDSql:
args = (caller.filename, caller.lineno, self.sql, col_name_list, expect_col_name_list)
tdLog.exit("%s(%d) failed: sql:%s, col_name_list:%s != expect_col_name_list:%s" % args)
def checkEqual(self, elm, expect_elm):
if elm == expect_elm:
tdLog.info("sql:%s, elm:%s == expect_elm:%s" % (self.sql, elm, expect_elm))
else:
caller = inspect.getframeinfo(inspect.stack()[1][0])
args = (caller.filename, caller.lineno, self.sql, elm, expect_elm)
tdLog.exit("%s(%d) failed: sql:%s, elm:%s != expect_elm:%s" % args)
def checkNotEqual(self, elm, expect_elm):
if elm != expect_elm:
tdLog.info("sql:%s, elm:%s != expect_elm:%s" % (self.sql, elm, expect_elm))
else:
caller = inspect.getframeinfo(inspect.stack()[1][0])
args = (caller.filename, caller.lineno, self.sql, elm, expect_elm)
tdLog.exit("%s(%d) failed: sql:%s, elm:%s == expect_elm:%s" % args)
def taosdStatus(self, state):
tdLog.sleep(5)
pstate = 0
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册