提交 a1f74aab 编写于 作者: C cpwu

fix case

上级 b6acf93f
# -*- coding: utf-8 -*-
# basic data type boundary
TINYINT_MAX = 127
TINYINT_MIN = -128
TINYINT_UN_MAX = 255
TINYINT_UN_MIN = 0
SMALLINT_MAX = 32767
SMALLINT_MIN = -32768
SMALLINT_UN_MAX = 65535
MALLINT_UN_MIN = 0
INT_MAX = 2147483647
INT_MIN = -2147483648
INT_UN_MAX = 4294967295
INT_UN_MIN = 0
BIGINT_MAX = 9223372036854775807
BIGINT_MIN = -9223372036854775808
BIGINT_UN_MAX = 18446744073709551615
BIGINT_UN_MIN = 0
FLOAT_MAX = 3.40E+38
FLOAT_MIN = -3.40E+38
DOUBLE_MAX = 1.7E+308
DOUBLE_MIN = -1.7E+308
# schema boundary
BINARY_LENGTH_MAX = 16374
NCAHR_LENGTH_MAX_ = 4093
DBNAME_LENGTH_MAX_ = 64
STBNAME_LENGTH_MAX = 192
STBNAME_LENGTH_MIN = 1
TBNAME_LENGTH_MAX = 192
TBNAME_LENGTH_MIN = 1
CHILD_TBNAME_LENGTH_MAX = 192
CHILD_TBNAME_LENGTH_MIN = 1
TAG_NAME_LENGTH_MAX = 64
TAG_NAME_LENGTH_MIN = 1
COL_NAME_LENGTH_MAX = 64
COL_NAME_LENGTH_MIN = 1
TAG_COUNT_MAX = 128
TAG_COUNT_MIN = 1
COL_COUNT_MAX = 4096
COL_COUNT_MIN = 2
TAG_COL_COUNT_MAX = 4096
TAG_COL_COUNT_MIN = 3
MNODE_SHM_SIZE_MAX = 2147483647
MNODE_SHM_SIZE_MIN = 6292480
MNODE_SHM_SIZE_DEFAULT = 6292480
VNODE_SHM_SIZE_MAX = 2147483647
VNODE_SHM_SIZE_MIN = 6292480
VNODE_SHM_SIZE_DEFAULT = 31458304
\ No newline at end of file
......@@ -2,11 +2,11 @@ import datetime
from dataclasses import dataclass
from typing import List, Any, Tuple
from tests.pytest.util.sql import TDSql
from util.log import *
from util.sql import *
from util.cases import *
from util.dnodes import *
from util.constant import *
PRIMARY_COL = "ts"
......@@ -21,10 +21,11 @@ TINT_UN_COL = "c_tint_un"
SINT_UN_COL = "c_sint_un"
BINT_UN_COL = "c_bint_un"
INT_UN_COL = "c_int_un"
BINARY_COL = "c_binary"
NCHAR_COL = "c_nchar"
TS_COL = "c_ts"
BINARY_COL = "c8"
NCHAR_COL = "c9"
TS_COL = "c10"
NUM_COL = [INT_COL, BINT_COL, SINT_COL, TINT_COL, FLOAT_COL, DOUBLE_COL, ]
CHAR_COL = [BINARY_COL, NCHAR_COL, ]
......@@ -53,10 +54,26 @@ class DataSet:
binary_data : List[str] = None
nchar_data : List[str] = None
def __post_init__(self):
self.ts_data = []
self.int_data = []
self.bint_data = []
self.sint_data = []
self.tint_data = []
self.int_un_data = []
self.bint_un_data = []
self.sint_un_data = []
self.tint_un_data = []
self.float_data = []
self.double_data = []
self.bool_data = []
self.binary_data = []
self.nchar_data = []
@dataclass
class SMAschema:
creation : str = "create"
creation : str = "CREATE"
index_name : str = "sma_index_1"
index_flag : str = "SMA INDEX"
operator : str = "ON"
......@@ -67,6 +84,53 @@ class SMAschema:
interval : Tuple[str] = None
sliding : str = None
other : Any = None
drop : str = "DROP"
drop_flag : str = "INDEX"
def __post_init__(self):
if isinstance(self.other, dict):
for k,v in self.other.items():
if k.lower() == "index_name" and isinstance(v, str) and not self.index_name:
self.index_name = v
del self.other[k]
if k.lower() == "index_flag" and isinstance(v, str) and not self.index_flag:
self.index_flag = v
del self.other[k]
if k.lower() == "operator" and isinstance(v, str) and not self.operator:
self.operator = v
del self.other[k]
if k.lower() == "tbname" and isinstance(v, str) and not self.tbname:
self.tbname = v
del self.other[k]
if k.lower() == "watermark" and isinstance(v, str) and not self.watermark:
self.watermark = v
del self.other[k]
if k.lower() == "maxdelay" and isinstance(v, str) and not self.maxdelay:
self.maxdelay = v
del self.other[k]
if k.lower() == "functions" and isinstance(v, tuple) and not self.func:
self.func = v
del self.other[k]
if k.lower() == "interval" and isinstance(v, tuple) and not self.interval:
self.interval = v
del self.other[k]
if k.lower() == "sliding" and isinstance(v, str) and not self.sliding:
self.sliding = v
del self.other[k]
if k.lower() == "drop_flag" and isinstance(v, str) and not self.drop_flag:
self.drop_flag = v
del self.other[k]
class TDTestCase:
......@@ -98,6 +162,17 @@ class TDTestCase:
sql += f" watermark {sma.watermark}"
if sma.maxdelay:
sql += f" maxdelay {sma.maxdelay}"
if isinstance(sma.other, dict):
for k,v in sma.other.items():
if isinstance(v,tuple) or isinstance(v, list):
sql += f" {k} ({' '.join(v)})"
else:
sql += f" {k} {v}"
if isinstance(sma.other, tuple) or isinstance(sma.other, list):
sql += " ".join(sma.other)
if isinstance(sma.other, int) or isinstance(sma.other, float) or isinstance(sma.other, str):
sql += sma.other
return sql
def sma_create_check(self, sma:SMAschema):
......@@ -129,13 +204,8 @@ class TDTestCase:
tdSql.error(self.__create_sma_index(sma))
def all_test(self):
self.test_create_databases()
self.test_create_stb()
pass
def __create_tb(self):
tdLog.printNoPrefix("==========step: create table")
......@@ -145,7 +215,7 @@ class TDTestCase:
{BINARY_COL} binary(16), {NCHAR_COL} nchar(32), {TS_COL} timestamp,
{TINT_UN_COL} tinyint unsigned, {SINT_UN_COL} smallint unsigned,
{INT_UN_COL} int unsigned, {BINT_UN_COL} bigint unsigned
) tags (t1 int)
) tags (tag1 int)
'''
create_ntb_sql = f'''create table t1(
ts timestamp, {INT_COL} int, {BINT_COL} bigint, {SINT_COL} smallint, {TINT_COL} tinyint,
......@@ -163,21 +233,6 @@ class TDTestCase:
def __data_set(self, rows):
data_set = DataSet()
# neg_data_set = DataSet()
data_set.ts_data = []
data_set.int_data = []
data_set.bint_data = []
data_set.sint_data = []
data_set.tint_data = []
data_set.int_un_data = []
data_set.bint_un_data = []
data_set.sint_un_data = []
data_set.tint_un_data = []
data_set.float_data = []
data_set.double_data = []
data_set.bool_data = []
data_set.binary_data = []
data_set.nchar_data = []
for i in range(rows):
data_set.ts_data.append(NOW + 1 * (rows - i))
......@@ -248,11 +303,11 @@ class TDTestCase:
def run(self):
sma1 = SMAschema(func=("min(c1)","max(c2)"))
sql1 = self.create_sma_index(sma1)
sql1 = self.__create_sma_index(sma1)
print("================")
print(sql1)
# a = DataSet()
return
# return
self.rows = 10
tdLog.printNoPrefix("==========step0:all check")
......@@ -262,7 +317,7 @@ class TDTestCase:
tdSql.prepare()
self.__create_tb()
self.__insert_data()
# return
return
tdLog.printNoPrefix("==========step2:create table in rollup database")
tdSql.execute("create database db3 retentions 1s:4m,2s:8m,3s:12m")
......
......@@ -238,7 +238,7 @@ class TDTestCase:
ts timestamp, {INT_COL} int, {BINT_COL} bigint, {SINT_COL} smallint, {TINT_COL} tinyint,
{FLOAT_COL} float, {DOUBLE_COL} double, {BOOL_COL} bool,
{BINARY_COL} binary(16), {NCHAR_COL} nchar(32), {TS_COL} timestamp
) tags (t1 int)
) tags (tag1 int)
'''
create_ntb_sql = f'''create table t1(
ts timestamp, {INT_COL} int, {BINT_COL} bigint, {SINT_COL} smallint, {TINT_COL} tinyint,
......
......@@ -23,6 +23,9 @@ CHAR_COL = [ BINARY_COL, NCHAR_COL, ]
BOOLEAN_COL = [ BOOL_COL, ]
TS_TYPE_COL = [ TS_COL, ]
ERR_POS = 0
CURRENT_POS = 1
LENS = 6
class TDTestCase:
......@@ -76,6 +79,7 @@ class TDTestCase:
if pos < 1:
tdSql.error(f"select substr( {condition}, {pos}, {lens}) , {condition} from {tbname} ")
break
tdSql.query(f"select substr( {condition}, {pos}, {lens}) , {condition} from {tbname} ")
for j in range(tdSql.queryRows):
......@@ -127,7 +131,7 @@ class TDTestCase:
tdLog.printNoPrefix("==========current sql condition check , must return query ok==========")
tbname = ["ct1", "ct2", "ct4", "t1", "stb1"]
for tb in tbname:
self.__substr_check(tb, 1, 6)
self.__substr_check(tb, CURRENT_POS, LENS)
tdLog.printNoPrefix(f"==========current sql condition check in {tb} over==========")
def __test_error(self):
......@@ -137,7 +141,7 @@ class TDTestCase:
for tb in tbname:
for errsql in self.__substr_err_check(tb):
tdSql.error(sql=errsql)
self.__substr_check(tb, 0, 6)
self.__substr_check(tb, ERR_POS, LENS)
tdLog.printNoPrefix(f"==========err sql condition check in {tb} over==========")
......@@ -154,7 +158,7 @@ class TDTestCase:
ts timestamp, {INT_COL} int, {BINT_COL} bigint, {SINT_COL} smallint, {TINT_COL} tinyint,
{FLOAT_COL} float, {DOUBLE_COL} double, {BOOL_COL} bool,
{BINARY_COL} binary(16), {NCHAR_COL} nchar(32), {TS_COL} timestamp
) tags (t1 int)
) tags (tag1 int)
'''
create_ntb_sql = f'''create table t1(
ts timestamp, {INT_COL} int, {BINT_COL} bigint, {SINT_COL} smallint, {TINT_COL} tinyint,
......
......@@ -73,7 +73,7 @@ class TDTestCase:
f"select sum(ceil( {un_num_col} )) from {tbanme} ",
)
)
sqls.extend( f"select sum( {un_num_col} + {un_num_col_2} ) from {tbanme} " for un_num_col_2 in UN_NUM_COL )
# sqls.extend( f"select sum( {un_num_col} + {un_num_col_2} ) from {tbanme} " for un_num_col_2 in UN_NUM_COL )
sqls.extend( f"select sum( {num_col} + {ts_col} ) from {tbanme} " for num_col in NUM_COL for ts_col in TS_TYPE_COL)
sqls.extend(
......
......@@ -35,6 +35,7 @@ python3 ./test.py -f 2-query/lower.py
python3 ./test.py -f 2-query/join.py
python3 ./test.py -f 2-query/join2.py
python3 ./test.py -f 2-query/cast.py
python3 ./test.py -f 2-query/substr.py
python3 ./test.py -f 2-query/union.py
python3 ./test.py -f 2-query/union1.py
python3 ./test.py -f 2-query/concat.py
......@@ -99,11 +100,11 @@ python3 ./test.py -f 2-query/function_stateduration.py
python3 ./test.py -f 2-query/statecount.py
python3 ./test.py -f 2-query/tail.py
python3 ./test.py -f 6-cluster/5dnode1mnode.py
python3 ./test.py -f 6-cluster/5dnode2mnode.py
python3 ./test.py -f 6-cluster/5dnode1mnode.py
python3 ./test.py -f 6-cluster/5dnode2mnode.py
#python3 ./test.py -f 6-cluster/5dnode3mnodeStop.py
python3 ./test.py -f 6-cluster/5dnode3mnodeDrop.py
# BUG python3 ./test.py -f 6-cluster/5dnode3mnodeStopInsert.py
# BUG python3 ./test.py -f 6-cluster/5dnode3mnodeStopInsert.py
python3 ./test.py -f 7-tmq/basic5.py
python3 ./test.py -f 7-tmq/subscribeDb.py
......
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