ClickHouse/tests/integration/test_globs_in_filepath/test.py

134 lines
6.4 KiB
Python
Raw Normal View History

import pytest
from helpers.cluster import ClickHouseCluster
cluster = ClickHouseCluster(__file__)
node = cluster.add_instance('node')
2019-08-10 16:00:01 +00:00
path_to_userfiles_from_defaut_config = "/var/lib/clickhouse/user_files/" # should be the same as in config file
@pytest.fixture(scope="module")
def start_cluster():
try:
cluster.start()
yield cluster
except Exception as ex:
print(ex)
raise ex
finally:
cluster.shutdown()
2019-08-10 16:00:01 +00:00
def test_strange_filenames(start_cluster):
# 2 rows data
some_data = "\t111.222\nData\t333.444"
node.exec_in_container(['bash', '-c', 'mkdir {}strange_names/'.format(path_to_userfiles_from_defaut_config)], privileged=True, user='root')
files = ["p.o.i.n.t.s",
"b}{ra{ces",
"b}.o{t.h"]
# filename inside testing data for debug simplicity
for filename in files:
node.exec_in_container(['bash', '-c', 'echo "{}{}" > {}strange_names/{}'.format(filename, some_data, path_to_userfiles_from_defaut_config, filename)], privileged=True, user='root')
test_requests = [("p.o.??n.t.s", "2"),
("p.o.*t.s", "2"),
("b}{r?{ces", "2"),
("b}*ces", "2"),
2019-08-27 15:20:31 +00:00
("b}.?{t.h", "2")]
2019-08-10 16:00:01 +00:00
for pattern, value in test_requests:
assert node.query('''
select count(*) from file('strange_names/{}', 'TSV', 'text String, number Float64')
'''.format(pattern)) == '{}\n'.format(value)
assert node.query('''
select count(*) from file('{}strange_names/{}', 'TSV', 'text String, number Float64')
'''.format(path_to_userfiles_from_defaut_config, pattern)) == '{}\n'.format(value)
def test_linear_structure(start_cluster):
# 2 rows data
some_data = "\t123.456\nData\t789.012"
files = ["file1", "file2", "file3", "file4", "file5",
"file000", "file111", "file222", "file333", "file444",
"a_file", "b_file", "c_file", "d_file", "e_file",
"a_data", "b_data", "c_data", "d_data", "e_data"]
# filename inside testing data for debug simplicity
for filename in files:
node.exec_in_container(['bash', '-c', 'echo "{}{}" > {}{}'.format(filename, some_data, path_to_userfiles_from_defaut_config, filename)], privileged=True, user='root')
test_requests = [("file{0..9}", "10"),
("file?", "10"),
("nothing*", "0"),
2019-08-10 16:00:01 +00:00
("file{0..9}{0..9}{0..9}", "10"),
("file{000..999}", "10"),
2019-08-10 16:00:01 +00:00
("file???", "10"),
("file*", "20"),
("a_{file,data}", "4"),
("?_{file,data}", "20"),
2019-08-27 15:20:31 +00:00
("{a,b,c,d,e}_{file,data}", "20"),
("{a,b,c,d,e}?{file,data}", "20"),
2019-08-10 16:00:01 +00:00
("*", "40")]
for pattern, value in test_requests:
assert node.query('''
select count(*) from file('{}', 'TSV', 'text String, number Float64')
'''.format(pattern)) == '{}\n'.format(value)
assert node.query('''
select count(*) from file('{}{}', 'TSV', 'text String, number Float64')
'''.format(path_to_userfiles_from_defaut_config, pattern)) == '{}\n'.format(value)
def test_deep_structure(start_cluster):
# 2 rows data
some_data = "\t135.791\nData\t246.802"
dirs = ["directory1/", "directory2/", "some_more_dir/", "we/",
"directory1/big_dir/",
"directory1/dir1/", "directory1/dir2/", "directory1/dir3/",
"directory2/dir1/", "directory2/dir2/", "directory2/one_more_dir/",
"some_more_dir/yet_another_dir/",
"we/need/", "we/need/to/", "we/need/to/go/", "we/need/to/go/deeper/"]
for dir in dirs:
2019-08-10 16:00:01 +00:00
node.exec_in_container(['bash', '-c', 'mkdir {}{}'.format(path_to_userfiles_from_defaut_config, dir)], privileged=True, user='root')
# all directories appeared in files must be listed in dirs
2019-08-10 16:00:01 +00:00
files = []
for i in range(10):
for j in range(10):
for k in range(10):
files.append("directory1/big_dir/file" + str(i) + str(j) + str(k))
2019-08-10 16:00:01 +00:00
for dir in dirs:
files.append(dir+"file")
# filename inside testing data for debug simplicity
for filename in files:
2019-08-10 16:00:01 +00:00
node.exec_in_container(['bash', '-c', 'echo "{}{}" > {}{}'.format(filename, some_data, path_to_userfiles_from_defaut_config, filename)], privileged=True, user='root')
2019-08-10 16:00:01 +00:00
test_requests = [ ("directory{1..5}/big_dir/*", "2002"), ("directory{0..6}/big_dir/*{0..9}{0..9}{0..9}", "2000"),
("?", "0"),
("directory{0..5}/dir{1..3}/file", "10"), ("directory{0..5}/dir?/file", "10"),
2019-08-27 15:20:31 +00:00
("we/need/to/go/deeper/file", "2"), ("*/*/*/*/*/*", "2"), ("we/need/??/go/deeper/*?*?*?*?*", "2")]
for pattern, value in test_requests:
assert node.query('''
select count(*) from file('{}', 'TSV', 'text String, number Float64')
2019-08-10 16:00:01 +00:00
'''.format(pattern)) == '{}\n'.format(value)
assert node.query('''
select count(*) from file('{}{}', 'TSV', 'text String, number Float64')
2019-11-13 18:35:35 +00:00
'''.format(path_to_userfiles_from_defaut_config, pattern)) == '{}\n'.format(value)
def test_table_function_and_virtual_columns(start_cluster):
2019-11-13 18:35:35 +00:00
node.exec_in_container(['bash', '-c', 'mkdir -p {}some/path/to/'.format(path_to_userfiles_from_defaut_config)])
node.exec_in_container(['bash', '-c', 'touch {}some/path/to/data.CSV'.format(path_to_userfiles_from_defaut_config)])
node.query("insert into table function file('some/path/to/data.CSV', CSV, 'n UInt8, s String') select number, concat('str_', toString(number)) from numbers(100000)")
assert node.query("select count() from file('some/path/to/data.CSV', CSV, 'n UInt8, s String')").rstrip() == '100000'
node.query("insert into table function file('nonexist.csv', 'CSV', 'val1 UInt32') values (1)")
assert node.query("select * from file('nonexist.csv', 'CSV', 'val1 UInt32')").rstrip()== '1'
assert "nonexist.csv" in node.query("select _path from file('nonexis?.csv', 'CSV', 'val1 UInt32')").rstrip()
assert "nonexist.csv" in node.query("select _path from file('nonexist.csv', 'CSV', 'val1 UInt32')").rstrip()
assert "nonexist.csv" == node.query("select _file from file('nonexis?.csv', 'CSV', 'val1 UInt32')").rstrip()
assert "nonexist.csv" == node.query("select _file from file('nonexist.csv', 'CSV', 'val1 UInt32')").rstrip()