mirror of
https://github.com/ClickHouse/ClickHouse.git
synced 2024-11-24 08:32:02 +00:00
581 lines
21 KiB
Python
581 lines
21 KiB
Python
import os
|
|
|
|
import pytest
|
|
from helpers.cluster import ClickHouseCluster
|
|
from pyhdfs import HdfsClient
|
|
|
|
cluster = ClickHouseCluster(__file__)
|
|
node1 = cluster.add_instance("node1", with_hdfs=True)
|
|
|
|
|
|
@pytest.fixture(scope="module")
|
|
def started_cluster():
|
|
try:
|
|
cluster.start()
|
|
yield cluster
|
|
finally:
|
|
cluster.shutdown()
|
|
|
|
|
|
def test_read_write_storage(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
node1.query("drop table if exists SimpleHDFSStorage SYNC")
|
|
node1.query(
|
|
"create table SimpleHDFSStorage (id UInt32, name String, weight Float64) ENGINE = HDFS('hdfs://hdfs1:9000/simple_storage', 'TSV')"
|
|
)
|
|
node1.query("insert into SimpleHDFSStorage values (1, 'Mark', 72.53)")
|
|
assert hdfs_api.read_data("/simple_storage") == "1\tMark\t72.53\n"
|
|
assert node1.query("select * from SimpleHDFSStorage") == "1\tMark\t72.53\n"
|
|
|
|
|
|
def test_read_write_storage_with_globs(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
node1.query(
|
|
"create table HDFSStorageWithRange (id UInt32, name String, weight Float64) ENGINE = HDFS('hdfs://hdfs1:9000/storage{1..5}', 'TSV')"
|
|
)
|
|
node1.query(
|
|
"create table HDFSStorageWithEnum (id UInt32, name String, weight Float64) ENGINE = HDFS('hdfs://hdfs1:9000/storage{1,2,3,4,5}', 'TSV')"
|
|
)
|
|
node1.query(
|
|
"create table HDFSStorageWithQuestionMark (id UInt32, name String, weight Float64) ENGINE = HDFS('hdfs://hdfs1:9000/storage?', 'TSV')"
|
|
)
|
|
node1.query(
|
|
"create table HDFSStorageWithAsterisk (id UInt32, name String, weight Float64) ENGINE = HDFS('hdfs://hdfs1:9000/storage*', 'TSV')"
|
|
)
|
|
|
|
for i in ["1", "2", "3"]:
|
|
hdfs_api.write_data("/storage" + i, i + "\tMark\t72.53\n")
|
|
assert hdfs_api.read_data("/storage" + i) == i + "\tMark\t72.53\n"
|
|
|
|
assert node1.query("select count(*) from HDFSStorageWithRange") == "3\n"
|
|
assert node1.query("select count(*) from HDFSStorageWithEnum") == "3\n"
|
|
assert node1.query("select count(*) from HDFSStorageWithQuestionMark") == "3\n"
|
|
assert node1.query("select count(*) from HDFSStorageWithAsterisk") == "3\n"
|
|
|
|
try:
|
|
node1.query("insert into HDFSStorageWithEnum values (1, 'NEW', 4.2)")
|
|
assert False, "Exception have to be thrown"
|
|
except Exception as ex:
|
|
print(ex)
|
|
assert "in readonly mode" in str(ex)
|
|
|
|
try:
|
|
node1.query("insert into HDFSStorageWithQuestionMark values (1, 'NEW', 4.2)")
|
|
assert False, "Exception have to be thrown"
|
|
except Exception as ex:
|
|
print(ex)
|
|
assert "in readonly mode" in str(ex)
|
|
|
|
try:
|
|
node1.query("insert into HDFSStorageWithAsterisk values (1, 'NEW', 4.2)")
|
|
assert False, "Exception have to be thrown"
|
|
except Exception as ex:
|
|
print(ex)
|
|
assert "in readonly mode" in str(ex)
|
|
|
|
|
|
def test_read_write_table(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
data = "1\tSerialize\t555.222\n2\tData\t777.333\n"
|
|
hdfs_api.write_data("/simple_table_function", data)
|
|
|
|
assert hdfs_api.read_data("/simple_table_function") == data
|
|
|
|
assert (
|
|
node1.query(
|
|
"select * from hdfs('hdfs://hdfs1:9000/simple_table_function', 'TSV', 'id UInt64, text String, number Float64')"
|
|
)
|
|
== data
|
|
)
|
|
|
|
|
|
def test_write_table(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
node1.query(
|
|
"create table OtherHDFSStorage (id UInt32, name String, weight Float64) ENGINE = HDFS('hdfs://hdfs1:9000/other_storage', 'TSV')"
|
|
)
|
|
node1.query(
|
|
"insert into OtherHDFSStorage values (10, 'tomas', 55.55), (11, 'jack', 32.54)"
|
|
)
|
|
|
|
result = "10\ttomas\t55.55\n11\tjack\t32.54\n"
|
|
assert hdfs_api.read_data("/other_storage") == result
|
|
assert node1.query("select * from OtherHDFSStorage order by id") == result
|
|
|
|
|
|
def test_bad_hdfs_uri(started_cluster):
|
|
try:
|
|
node1.query(
|
|
"create table BadStorage1 (id UInt32, name String, weight Float64) ENGINE = HDFS('hads:hgsdfs100500:9000/other_storage', 'TSV')"
|
|
)
|
|
except Exception as ex:
|
|
print(ex)
|
|
assert "Bad hdfs url" in str(ex)
|
|
try:
|
|
node1.query(
|
|
"create table BadStorage2 (id UInt32, name String, weight Float64) ENGINE = HDFS('hdfs://hdfs100500:9000/other_storage', 'TSV')"
|
|
)
|
|
except Exception as ex:
|
|
print(ex)
|
|
assert "Unable to create builder to connect to HDFS" in str(ex)
|
|
|
|
try:
|
|
node1.query(
|
|
"create table BadStorage3 (id UInt32, name String, weight Float64) ENGINE = HDFS('hdfs://hdfs1:9000/<>', 'TSV')"
|
|
)
|
|
except Exception as ex:
|
|
print(ex)
|
|
assert "Unable to open HDFS file" in str(ex)
|
|
|
|
|
|
@pytest.mark.timeout(800)
|
|
def test_globs_in_read_table(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
some_data = "1\tSerialize\t555.222\n2\tData\t777.333\n"
|
|
globs_dir = "/dir_for_test_with_globs/"
|
|
files = [
|
|
"dir1/dir_dir/file1",
|
|
"dir2/file2",
|
|
"simple_table_function",
|
|
"dir/file",
|
|
"some_dir/dir1/file",
|
|
"some_dir/dir2/file",
|
|
"some_dir/file",
|
|
"table1_function",
|
|
"table2_function",
|
|
"table3_function",
|
|
]
|
|
for filename in files:
|
|
hdfs_api.write_data(globs_dir + filename, some_data)
|
|
|
|
test_requests = [
|
|
("dir{1..5}/dir_dir/file1", 1, 1),
|
|
("*_table_functio?", 1, 1),
|
|
("dir/fil?", 1, 1),
|
|
("table{3..8}_function", 1, 1),
|
|
("table{2..8}_function", 2, 2),
|
|
("dir/*", 1, 1),
|
|
("dir/*?*?*?*?*", 1, 1),
|
|
("dir/*?*?*?*?*?*", 0, 0),
|
|
("some_dir/*/file", 2, 1),
|
|
("some_dir/dir?/*", 2, 1),
|
|
("*/*/*", 3, 2),
|
|
("?", 0, 0),
|
|
]
|
|
|
|
for pattern, paths_amount, files_amount in test_requests:
|
|
inside_table_func = (
|
|
"'hdfs://hdfs1:9000"
|
|
+ globs_dir
|
|
+ pattern
|
|
+ "', 'TSV', 'id UInt64, text String, number Float64'"
|
|
)
|
|
print("inside_table_func ", inside_table_func)
|
|
assert (
|
|
node1.query("select * from hdfs(" + inside_table_func + ")")
|
|
== paths_amount * some_data
|
|
)
|
|
assert node1.query(
|
|
"select count(distinct _path) from hdfs(" + inside_table_func + ")"
|
|
).rstrip() == str(paths_amount)
|
|
assert node1.query(
|
|
"select count(distinct _file) from hdfs(" + inside_table_func + ")"
|
|
).rstrip() == str(files_amount)
|
|
|
|
|
|
def test_read_write_gzip_table(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
data = "1\tHello Jessica\t555.222\n2\tI rolled a joint\t777.333\n"
|
|
hdfs_api.write_gzip_data("/simple_table_function.gz", data)
|
|
|
|
assert hdfs_api.read_gzip_data("/simple_table_function.gz") == data
|
|
|
|
assert (
|
|
node1.query(
|
|
"select * from hdfs('hdfs://hdfs1:9000/simple_table_function.gz', 'TSV', 'id UInt64, text String, number Float64')"
|
|
)
|
|
== data
|
|
)
|
|
|
|
|
|
def test_read_write_gzip_table_with_parameter_gzip(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
data = "1\tHello Jessica\t555.222\n2\tI rolled a joint\t777.333\n"
|
|
hdfs_api.write_gzip_data("/simple_table_function", data)
|
|
|
|
assert hdfs_api.read_gzip_data("/simple_table_function") == data
|
|
|
|
assert (
|
|
node1.query(
|
|
"select * from hdfs('hdfs://hdfs1:9000/simple_table_function', 'TSV', 'id UInt64, text String, number Float64', 'gzip')"
|
|
)
|
|
== data
|
|
)
|
|
|
|
|
|
def test_read_write_table_with_parameter_none(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
data = "1\tHello Jessica\t555.222\n2\tI rolled a joint\t777.333\n"
|
|
hdfs_api.write_data("/simple_table_function.gz", data)
|
|
|
|
assert hdfs_api.read_data("/simple_table_function.gz") == data
|
|
|
|
assert (
|
|
node1.query(
|
|
"select * from hdfs('hdfs://hdfs1:9000/simple_table_function.gz', 'TSV', 'id UInt64, text String, number Float64', 'none')"
|
|
)
|
|
== data
|
|
)
|
|
|
|
|
|
def test_read_write_gzip_table_with_parameter_auto_gz(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
data = "1\tHello Jessica\t555.222\n2\tI rolled a joint\t777.333\n"
|
|
hdfs_api.write_gzip_data("/simple_table_function.gz", data)
|
|
|
|
assert hdfs_api.read_gzip_data("/simple_table_function.gz") == data
|
|
|
|
assert (
|
|
node1.query(
|
|
"select * from hdfs('hdfs://hdfs1:9000/simple_table_function.gz', 'TSV', 'id UInt64, text String, number Float64', 'auto')"
|
|
)
|
|
== data
|
|
)
|
|
|
|
|
|
def test_write_gz_storage(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
node1.query(
|
|
"create table GZHDFSStorage (id UInt32, name String, weight Float64) ENGINE = HDFS('hdfs://hdfs1:9000/storage.gz', 'TSV')"
|
|
)
|
|
node1.query("insert into GZHDFSStorage values (1, 'Mark', 72.53)")
|
|
assert hdfs_api.read_gzip_data("/storage.gz") == "1\tMark\t72.53\n"
|
|
assert node1.query("select * from GZHDFSStorage") == "1\tMark\t72.53\n"
|
|
|
|
|
|
def test_write_gzip_storage(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
node1.query(
|
|
"create table GZIPHDFSStorage (id UInt32, name String, weight Float64) ENGINE = HDFS('hdfs://hdfs1:9000/gzip_storage', 'TSV', 'gzip')"
|
|
)
|
|
node1.query("insert into GZIPHDFSStorage values (1, 'Mark', 72.53)")
|
|
assert hdfs_api.read_gzip_data("/gzip_storage") == "1\tMark\t72.53\n"
|
|
assert node1.query("select * from GZIPHDFSStorage") == "1\tMark\t72.53\n"
|
|
|
|
|
|
def test_virtual_columns(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
node1.query(
|
|
"create table virtual_cols (id UInt32) ENGINE = HDFS('hdfs://hdfs1:9000/file*', 'TSV')"
|
|
)
|
|
hdfs_api.write_data("/file1", "1\n")
|
|
hdfs_api.write_data("/file2", "2\n")
|
|
hdfs_api.write_data("/file3", "3\n")
|
|
expected = "1\tfile1\thdfs://hdfs1:9000//file1\n2\tfile2\thdfs://hdfs1:9000//file2\n3\tfile3\thdfs://hdfs1:9000//file3\n"
|
|
assert (
|
|
node1.query(
|
|
"select id, _file as file_name, _path as file_path from virtual_cols order by id"
|
|
)
|
|
== expected
|
|
)
|
|
|
|
|
|
def test_read_files_with_spaces(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
fs = HdfsClient(hosts=started_cluster.hdfs_ip)
|
|
dir = "/test_spaces"
|
|
exists = fs.exists(dir)
|
|
if exists:
|
|
fs.delete(dir, recursive=True)
|
|
fs.mkdirs(dir)
|
|
|
|
hdfs_api.write_data(f"{dir}/test test test 1.txt", "1\n")
|
|
hdfs_api.write_data(f"{dir}/test test test 2.txt", "2\n")
|
|
hdfs_api.write_data(f"{dir}/test test test 3.txt", "3\n")
|
|
|
|
node1.query(
|
|
f"create table test (id UInt32) ENGINE = HDFS('hdfs://hdfs1:9000/{dir}/test*', 'TSV')"
|
|
)
|
|
assert node1.query("select * from test order by id") == "1\n2\n3\n"
|
|
fs.delete(dir, recursive=True)
|
|
|
|
|
|
def test_truncate_table(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
node1.query(
|
|
"create table test_truncate (id UInt32, name String, weight Float64) ENGINE = HDFS('hdfs://hdfs1:9000/tr', 'TSV')"
|
|
)
|
|
node1.query("insert into test_truncate values (1, 'Mark', 72.53)")
|
|
assert hdfs_api.read_data("/tr") == "1\tMark\t72.53\n"
|
|
assert node1.query("select * from test_truncate") == "1\tMark\t72.53\n"
|
|
node1.query("truncate table test_truncate")
|
|
assert node1.query("select * from test_truncate") == ""
|
|
node1.query("drop table test_truncate")
|
|
|
|
|
|
def test_partition_by(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
table_format = "column1 UInt32, column2 UInt32, column3 UInt32"
|
|
file_name = "test_{_partition_id}"
|
|
partition_by = "column3"
|
|
values = "(1, 2, 3), (3, 2, 1), (1, 3, 2)"
|
|
table_function = f"hdfs('hdfs://hdfs1:9000/{file_name}', 'TSV', '{table_format}')"
|
|
|
|
node1.query(
|
|
f"insert into table function {table_function} PARTITION BY {partition_by} values {values}"
|
|
)
|
|
result = node1.query(
|
|
f"select * from hdfs('hdfs://hdfs1:9000/test_1', 'TSV', '{table_format}')"
|
|
)
|
|
assert result.strip() == "3\t2\t1"
|
|
result = node1.query(
|
|
f"select * from hdfs('hdfs://hdfs1:9000/test_2', 'TSV', '{table_format}')"
|
|
)
|
|
assert result.strip() == "1\t3\t2"
|
|
result = node1.query(
|
|
f"select * from hdfs('hdfs://hdfs1:9000/test_3', 'TSV', '{table_format}')"
|
|
)
|
|
assert result.strip() == "1\t2\t3"
|
|
|
|
file_name = "test2_{_partition_id}"
|
|
node1.query(
|
|
f"create table p(column1 UInt32, column2 UInt32, column3 UInt32) engine = HDFS('hdfs://hdfs1:9000/{file_name}', 'TSV') partition by column3"
|
|
)
|
|
node1.query(f"insert into p values {values}")
|
|
result = node1.query(
|
|
f"select * from hdfs('hdfs://hdfs1:9000/test2_1', 'TSV', '{table_format}')"
|
|
)
|
|
assert result.strip() == "3\t2\t1"
|
|
result = node1.query(
|
|
f"select * from hdfs('hdfs://hdfs1:9000/test2_2', 'TSV', '{table_format}')"
|
|
)
|
|
assert result.strip() == "1\t3\t2"
|
|
result = node1.query(
|
|
f"select * from hdfs('hdfs://hdfs1:9000/test2_3', 'TSV', '{table_format}')"
|
|
)
|
|
assert result.strip() == "1\t2\t3"
|
|
|
|
|
|
def test_seekable_formats(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
table_function = (
|
|
f"hdfs('hdfs://hdfs1:9000/parquet', 'Parquet', 'a Int32, b String')"
|
|
)
|
|
node1.query(
|
|
f"insert into table function {table_function} SELECT number, randomString(100) FROM numbers(5000000)"
|
|
)
|
|
|
|
result = node1.query(f"SELECT count() FROM {table_function}")
|
|
assert int(result) == 5000000
|
|
|
|
table_function = f"hdfs('hdfs://hdfs1:9000/orc', 'ORC', 'a Int32, b String')"
|
|
node1.query(
|
|
f"insert into table function {table_function} SELECT number, randomString(100) FROM numbers(5000000)"
|
|
)
|
|
result = node1.query(f"SELECT count() FROM {table_function}")
|
|
assert int(result) == 5000000
|
|
|
|
|
|
def test_read_table_with_default(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
data = "n\n100\n"
|
|
hdfs_api.write_data("/simple_table_function", data)
|
|
assert hdfs_api.read_data("/simple_table_function") == data
|
|
|
|
output = "n\tm\n100\t200\n"
|
|
assert (
|
|
node1.query(
|
|
"select * from hdfs('hdfs://hdfs1:9000/simple_table_function', 'TSVWithNames', 'n UInt32, m UInt32 DEFAULT n * 2') FORMAT TSVWithNames"
|
|
)
|
|
== output
|
|
)
|
|
|
|
|
|
def test_schema_inference(started_cluster):
|
|
node1.query(
|
|
f"insert into table function hdfs('hdfs://hdfs1:9000/native', 'Native', 'a Int32, b String') SELECT number, randomString(100) FROM numbers(5000000)"
|
|
)
|
|
|
|
result = node1.query(f"desc hdfs('hdfs://hdfs1:9000/native', 'Native')")
|
|
assert result == "a\tInt32\t\t\t\t\t\nb\tString\t\t\t\t\t\n"
|
|
|
|
result = node1.query(
|
|
f"select count(*) from hdfs('hdfs://hdfs1:9000/native', 'Native')"
|
|
)
|
|
assert int(result) == 5000000
|
|
|
|
node1.query(
|
|
f"create table schema_inference engine=HDFS('hdfs://hdfs1:9000/native', 'Native')"
|
|
)
|
|
result = node1.query(f"desc schema_inference")
|
|
assert result == "a\tInt32\t\t\t\t\t\nb\tString\t\t\t\t\t\n"
|
|
|
|
result = node1.query(f"select count(*) from schema_inference")
|
|
assert int(result) == 5000000
|
|
|
|
|
|
def test_hdfsCluster(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
fs = HdfsClient(hosts=started_cluster.hdfs_ip)
|
|
dir = "/test_hdfsCluster"
|
|
exists = fs.exists(dir)
|
|
if exists:
|
|
fs.delete(dir, recursive=True)
|
|
fs.mkdirs(dir)
|
|
hdfs_api.write_data("/test_hdfsCluster/file1", "1\n")
|
|
hdfs_api.write_data("/test_hdfsCluster/file2", "2\n")
|
|
hdfs_api.write_data("/test_hdfsCluster/file3", "3\n")
|
|
|
|
actual = node1.query(
|
|
"select id, _file as file_name, _path as file_path from hdfs('hdfs://hdfs1:9000/test_hdfsCluster/file*', 'TSV', 'id UInt32') order by id"
|
|
)
|
|
expected = "1\tfile1\thdfs://hdfs1:9000/test_hdfsCluster/file1\n2\tfile2\thdfs://hdfs1:9000/test_hdfsCluster/file2\n3\tfile3\thdfs://hdfs1:9000/test_hdfsCluster/file3\n"
|
|
assert actual == expected
|
|
|
|
actual = node1.query(
|
|
"select id, _file as file_name, _path as file_path from hdfsCluster('test_cluster_two_shards', 'hdfs://hdfs1:9000/test_hdfsCluster/file*', 'TSV', 'id UInt32') order by id"
|
|
)
|
|
expected = "1\tfile1\thdfs://hdfs1:9000/test_hdfsCluster/file1\n2\tfile2\thdfs://hdfs1:9000/test_hdfsCluster/file2\n3\tfile3\thdfs://hdfs1:9000/test_hdfsCluster/file3\n"
|
|
assert actual == expected
|
|
fs.delete(dir, recursive=True)
|
|
|
|
|
|
def test_hdfs_directory_not_exist(started_cluster):
|
|
ddl = "create table HDFSStorageWithNotExistDir (id UInt32, name String, weight Float64) ENGINE = HDFS('hdfs://hdfs1:9000/data/not_eixst', 'TSV')"
|
|
node1.query(ddl)
|
|
assert "" == node1.query("select * from HDFSStorageWithNotExistDir")
|
|
|
|
|
|
def test_overwrite(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
table_function = f"hdfs('hdfs://hdfs1:9000/data', 'Parquet', 'a Int32, b String')"
|
|
node1.query(f"create table test_overwrite as {table_function}")
|
|
node1.query(
|
|
f"insert into test_overwrite select number, randomString(100) from numbers(5)"
|
|
)
|
|
node1.query_and_get_error(
|
|
f"insert into test_overwrite select number, randomString(100) FROM numbers(10)"
|
|
)
|
|
node1.query(
|
|
f"insert into test_overwrite select number, randomString(100) from numbers(10) settings hdfs_truncate_on_insert=1"
|
|
)
|
|
|
|
result = node1.query(f"select count() from test_overwrite")
|
|
assert int(result) == 10
|
|
|
|
|
|
def test_multiple_inserts(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
table_function = f"hdfs('hdfs://hdfs1:9000/data_multiple_inserts', 'Parquet', 'a Int32, b String')"
|
|
node1.query(f"create table test_multiple_inserts as {table_function}")
|
|
node1.query(
|
|
f"insert into test_multiple_inserts select number, randomString(100) from numbers(10)"
|
|
)
|
|
node1.query(
|
|
f"insert into test_multiple_inserts select number, randomString(100) from numbers(20) settings hdfs_create_new_file_on_insert=1"
|
|
)
|
|
node1.query(
|
|
f"insert into test_multiple_inserts select number, randomString(100) from numbers(30) settings hdfs_create_new_file_on_insert=1"
|
|
)
|
|
|
|
result = node1.query(f"select count() from test_multiple_inserts")
|
|
assert int(result) == 60
|
|
|
|
result = node1.query(f"drop table test_multiple_inserts")
|
|
|
|
table_function = f"hdfs('hdfs://hdfs1:9000/data_multiple_inserts.gz', 'Parquet', 'a Int32, b String')"
|
|
node1.query(f"create table test_multiple_inserts as {table_function}")
|
|
node1.query(
|
|
f"insert into test_multiple_inserts select number, randomString(100) FROM numbers(10)"
|
|
)
|
|
node1.query(
|
|
f"insert into test_multiple_inserts select number, randomString(100) FROM numbers(20) settings hdfs_create_new_file_on_insert=1"
|
|
)
|
|
node1.query(
|
|
f"insert into test_multiple_inserts select number, randomString(100) FROM numbers(30) settings hdfs_create_new_file_on_insert=1"
|
|
)
|
|
|
|
result = node1.query(f"select count() from test_multiple_inserts")
|
|
assert int(result) == 60
|
|
|
|
|
|
def test_format_detection(started_cluster):
|
|
node1.query(
|
|
f"create table arrow_table (x UInt64) engine=HDFS('hdfs://hdfs1:9000/data.arrow')"
|
|
)
|
|
node1.query(f"insert into arrow_table select 1")
|
|
result = node1.query(f"select * from hdfs('hdfs://hdfs1:9000/data.arrow')")
|
|
assert int(result) == 1
|
|
|
|
|
|
def test_schema_inference_with_globs(started_cluster):
|
|
node1.query(
|
|
f"insert into table function hdfs('hdfs://hdfs1:9000/data1.jsoncompacteachrow', 'JSONCompactEachRow', 'x Nullable(UInt32)') select NULL"
|
|
)
|
|
node1.query(
|
|
f"insert into table function hdfs('hdfs://hdfs1:9000/data2.jsoncompacteachrow', 'JSONCompactEachRow', 'x Nullable(UInt32)') select 0"
|
|
)
|
|
|
|
result = node1.query(f"desc hdfs('hdfs://hdfs1:9000/data*.jsoncompacteachrow')")
|
|
assert result.strip() == "c1\tNullable(Float64)"
|
|
|
|
result = node1.query(
|
|
f"select * from hdfs('hdfs://hdfs1:9000/data*.jsoncompacteachrow')"
|
|
)
|
|
assert sorted(result.split()) == ["0", "\\N"]
|
|
|
|
|
|
def test_insert_select_schema_inference(started_cluster):
|
|
node1.query(
|
|
f"insert into table function hdfs('hdfs://hdfs1:9000/test.native.zst') select toUInt64(1) as x"
|
|
)
|
|
|
|
result = node1.query(f"desc hdfs('hdfs://hdfs1:9000/test.native.zst')")
|
|
assert result.strip() == "x\tUInt64"
|
|
|
|
result = node1.query(f"select * from hdfs('hdfs://hdfs1:9000/test.native.zst')")
|
|
assert int(result) == 1
|
|
|
|
|
|
def test_virtual_columns_2(started_cluster):
|
|
hdfs_api = started_cluster.hdfs_api
|
|
|
|
table_function = (
|
|
f"hdfs('hdfs://hdfs1:9000/parquet_2', 'Parquet', 'a Int32, b String')"
|
|
)
|
|
node1.query(f"insert into table function {table_function} SELECT 1, 'kek'")
|
|
|
|
result = node1.query(f"SELECT _path FROM {table_function}")
|
|
assert result.strip() == "hdfs://hdfs1:9000/parquet_2"
|
|
|
|
table_function = (
|
|
f"hdfs('hdfs://hdfs1:9000/parquet_3', 'Parquet', 'a Int32, _path String')"
|
|
)
|
|
node1.query(f"insert into table function {table_function} SELECT 1, 'kek'")
|
|
|
|
result = node1.query(f"SELECT _path FROM {table_function}")
|
|
assert result.strip() == "kek"
|
|
|
|
|
|
if __name__ == "__main__":
|
|
cluster.start()
|
|
input("Cluster created, press any key to destroy...")
|
|
cluster.shutdown()
|