mirror of
https://github.com/ClickHouse/ClickHouse.git
synced 2024-11-22 23:52:03 +00:00
310 lines
9.9 KiB
Python
310 lines
9.9 KiB
Python
import random
|
|
import string
|
|
|
|
import pytest
|
|
from helpers.cluster import ClickHouseCluster
|
|
|
|
cluster = ClickHouseCluster(__file__)
|
|
|
|
node1 = cluster.add_instance(
|
|
"node1",
|
|
main_configs=["configs/zstd_compression_by_default.xml"],
|
|
user_configs=["configs/allow_suspicious_codecs.xml"],
|
|
)
|
|
node2 = cluster.add_instance(
|
|
"node2",
|
|
main_configs=["configs/lz4hc_compression_by_default.xml"],
|
|
user_configs=["configs/allow_suspicious_codecs.xml"],
|
|
)
|
|
node3 = cluster.add_instance(
|
|
"node3",
|
|
main_configs=["configs/custom_compression_by_default.xml"],
|
|
user_configs=["configs/allow_suspicious_codecs.xml"],
|
|
)
|
|
node4 = cluster.add_instance(
|
|
"node4",
|
|
user_configs=[
|
|
"configs/enable_uncompressed_cache.xml",
|
|
"configs/allow_suspicious_codecs.xml",
|
|
],
|
|
)
|
|
node5 = cluster.add_instance(
|
|
"node5",
|
|
main_configs=["configs/zstd_compression_by_default.xml"],
|
|
user_configs=[
|
|
"configs/enable_uncompressed_cache.xml",
|
|
"configs/allow_suspicious_codecs.xml",
|
|
],
|
|
)
|
|
node6 = cluster.add_instance(
|
|
"node6",
|
|
main_configs=["configs/deflateqpl_compression_by_default.xml"],
|
|
user_configs=[
|
|
"configs/allow_suspicious_codecs.xml",
|
|
"configs/enable_deflateqpl_codec.xml",
|
|
],
|
|
)
|
|
node7 = cluster.add_instance(
|
|
"node7",
|
|
main_configs=["configs/allow_experimental_codecs.xml"],
|
|
user_configs=["configs/allow_suspicious_codecs.xml"],
|
|
)
|
|
|
|
|
|
@pytest.fixture(scope="module")
|
|
def start_cluster():
|
|
try:
|
|
cluster.start()
|
|
|
|
yield cluster
|
|
finally:
|
|
cluster.shutdown()
|
|
|
|
|
|
def test_preconfigured_default_codec(start_cluster):
|
|
for node in [node1, node2]:
|
|
node.query(
|
|
"""
|
|
CREATE TABLE compression_codec_multiple_with_key (
|
|
somedate Date CODEC(ZSTD, ZSTD, ZSTD(12), LZ4HC(12)),
|
|
id UInt64 CODEC(LZ4, ZSTD, NONE, LZ4HC),
|
|
data String CODEC(ZSTD(2), LZ4HC, NONE, LZ4, LZ4),
|
|
somecolumn Float64
|
|
) ENGINE = MergeTree() PARTITION BY somedate ORDER BY id SETTINGS index_granularity = 2;
|
|
"""
|
|
)
|
|
node.query(
|
|
"INSERT INTO compression_codec_multiple_with_key VALUES(toDate('2018-10-12'), 100000, 'hello', 88.88), (toDate('2018-10-12'), 100002, 'world', 99.99), (toDate('2018-10-12'), 1111, '!', 777.777)"
|
|
)
|
|
assert (
|
|
node.query(
|
|
"SELECT COUNT(*) FROM compression_codec_multiple_with_key WHERE id % 2 == 0"
|
|
)
|
|
== "2\n"
|
|
)
|
|
assert (
|
|
node.query(
|
|
"SELECT DISTINCT somecolumn FROM compression_codec_multiple_with_key ORDER BY id"
|
|
)
|
|
== "777.777\n88.88\n99.99\n"
|
|
)
|
|
assert (
|
|
node.query(
|
|
"SELECT data FROM compression_codec_multiple_with_key WHERE id >= 1112 AND somedate = toDate('2018-10-12') AND somecolumn <= 100"
|
|
)
|
|
== "hello\nworld\n"
|
|
)
|
|
|
|
node.query(
|
|
"INSERT INTO compression_codec_multiple_with_key SELECT toDate('2018-10-12'), number, toString(number), 1.0 FROM system.numbers LIMIT 10000"
|
|
)
|
|
|
|
assert (
|
|
node.query(
|
|
"SELECT COUNT(id) FROM compression_codec_multiple_with_key WHERE id % 10 == 0"
|
|
)
|
|
== "1001\n"
|
|
)
|
|
assert (
|
|
node.query(
|
|
"SELECT SUM(somecolumn) FROM compression_codec_multiple_with_key"
|
|
)
|
|
== str(777.777 + 88.88 + 99.99 + 1.0 * 10000) + "\n"
|
|
)
|
|
assert (
|
|
node.query(
|
|
"SELECT count(*) FROM compression_codec_multiple_with_key GROUP BY somedate"
|
|
)
|
|
== "10003\n"
|
|
)
|
|
|
|
|
|
def test_preconfigured_custom_codec(start_cluster):
|
|
node3.query(
|
|
"""
|
|
CREATE TABLE compression_codec_multiple_with_key (
|
|
somedate Date CODEC(ZSTD, ZSTD, ZSTD(12), LZ4HC(12)),
|
|
id UInt64 CODEC(LZ4, ZSTD, NONE, LZ4HC),
|
|
data String,
|
|
somecolumn Float64 CODEC(ZSTD(2), LZ4HC, NONE, NONE, NONE, LZ4HC(5))
|
|
) ENGINE = MergeTree() PARTITION BY somedate ORDER BY id SETTINGS index_granularity = 2;
|
|
"""
|
|
)
|
|
|
|
node3.query(
|
|
"INSERT INTO compression_codec_multiple_with_key VALUES(toDate('2018-10-12'), 100000, 'hello', 88.88), (toDate('2018-10-12'), 100002, 'world', 99.99), (toDate('2018-10-12'), 1111, '!', 777.777)"
|
|
)
|
|
assert (
|
|
node3.query(
|
|
"SELECT COUNT(*) FROM compression_codec_multiple_with_key WHERE id % 2 == 0"
|
|
)
|
|
== "2\n"
|
|
)
|
|
assert (
|
|
node3.query(
|
|
"SELECT DISTINCT somecolumn FROM compression_codec_multiple_with_key ORDER BY id"
|
|
)
|
|
== "777.777\n88.88\n99.99\n"
|
|
)
|
|
assert (
|
|
node3.query(
|
|
"SELECT data FROM compression_codec_multiple_with_key WHERE id >= 1112 AND somedate = toDate('2018-10-12') AND somecolumn <= 100"
|
|
)
|
|
== "hello\nworld\n"
|
|
)
|
|
|
|
node3.query(
|
|
"INSERT INTO compression_codec_multiple_with_key VALUES(toDate('2018-10-12'), 100000, '{}', 88.88)".format(
|
|
"".join(
|
|
random.choice(string.ascii_uppercase + string.digits)
|
|
for _ in range(10000)
|
|
)
|
|
)
|
|
)
|
|
|
|
node3.query("OPTIMIZE TABLE compression_codec_multiple_with_key FINAL")
|
|
assert (
|
|
node3.query(
|
|
"SELECT max(length(data)) from compression_codec_multiple_with_key GROUP BY data ORDER BY max(length(data)) DESC LIMIT 1"
|
|
)
|
|
== "10000\n"
|
|
)
|
|
|
|
for i in range(10):
|
|
node3.query(
|
|
"INSERT INTO compression_codec_multiple_with_key VALUES(toDate('2018-10-12'), {}, '{}', 88.88)".format(
|
|
i,
|
|
"".join(
|
|
random.choice(string.ascii_uppercase + string.digits)
|
|
for _ in range(10000)
|
|
),
|
|
)
|
|
)
|
|
|
|
node3.query("OPTIMIZE TABLE compression_codec_multiple_with_key FINAL")
|
|
|
|
assert (
|
|
node3.query(
|
|
"SELECT COUNT(*) from compression_codec_multiple_with_key WHERE length(data) = 10000"
|
|
)
|
|
== "11\n"
|
|
)
|
|
|
|
|
|
def test_uncompressed_cache_custom_codec(start_cluster):
|
|
node4.query(
|
|
"""
|
|
CREATE TABLE compression_codec_multiple_with_key (
|
|
somedate Date CODEC(ZSTD, ZSTD, ZSTD(12), LZ4HC(12)),
|
|
id UInt64 CODEC(LZ4, ZSTD, NONE, LZ4HC),
|
|
data String,
|
|
somecolumn Float64 CODEC(ZSTD(2), LZ4HC, NONE, NONE, NONE, LZ4HC(5))
|
|
) ENGINE = MergeTree() PARTITION BY somedate ORDER BY id SETTINGS index_granularity = 2;
|
|
"""
|
|
)
|
|
|
|
node4.query(
|
|
"INSERT INTO compression_codec_multiple_with_key VALUES(toDate('2018-10-12'), 100000, '{}', 88.88)".format(
|
|
"".join(
|
|
random.choice(string.ascii_uppercase + string.digits)
|
|
for _ in range(10000)
|
|
)
|
|
)
|
|
)
|
|
|
|
# two equal requests one by one, to get into UncompressedCache for the first block
|
|
assert (
|
|
node4.query(
|
|
"SELECT max(length(data)) from compression_codec_multiple_with_key GROUP BY data ORDER BY max(length(data)) DESC LIMIT 1"
|
|
)
|
|
== "10000\n"
|
|
)
|
|
|
|
assert (
|
|
node4.query(
|
|
"SELECT max(length(data)) from compression_codec_multiple_with_key GROUP BY data ORDER BY max(length(data)) DESC LIMIT 1"
|
|
)
|
|
== "10000\n"
|
|
)
|
|
|
|
|
|
def test_uncompressed_cache_plus_zstd_codec(start_cluster):
|
|
node5.query(
|
|
"""
|
|
CREATE TABLE compression_codec_multiple_with_key (
|
|
somedate Date CODEC(ZSTD, ZSTD, ZSTD(12), LZ4HC(12)),
|
|
id UInt64 CODEC(LZ4, ZSTD, NONE, LZ4HC),
|
|
data String,
|
|
somecolumn Float64 CODEC(ZSTD(2), LZ4HC, NONE, NONE, NONE, LZ4HC(5))
|
|
) ENGINE = MergeTree() PARTITION BY somedate ORDER BY id SETTINGS index_granularity = 2;
|
|
"""
|
|
)
|
|
|
|
node5.query(
|
|
"INSERT INTO compression_codec_multiple_with_key VALUES(toDate('2018-10-12'), 100000, '{}', 88.88)".format(
|
|
"a" * 10000
|
|
)
|
|
)
|
|
|
|
assert (
|
|
node5.query(
|
|
"SELECT max(length(data)) from compression_codec_multiple_with_key GROUP BY data ORDER BY max(length(data)) DESC LIMIT 1"
|
|
)
|
|
== "10000\n"
|
|
)
|
|
|
|
|
|
def test_preconfigured_deflateqpl_codec(start_cluster):
|
|
node6.query(
|
|
"""
|
|
CREATE TABLE compression_codec_multiple_with_key (
|
|
somedate Date CODEC(ZSTD, ZSTD, ZSTD(12), LZ4HC(12), DEFLATE_QPL),
|
|
id UInt64 CODEC(LZ4, ZSTD, NONE, LZ4HC, DEFLATE_QPL),
|
|
data String CODEC(ZSTD(2), LZ4HC, NONE, LZ4, LZ4, DEFLATE_QPL),
|
|
somecolumn Float64
|
|
) ENGINE = MergeTree() PARTITION BY somedate ORDER BY id SETTINGS index_granularity = 2;
|
|
"""
|
|
)
|
|
node6.query(
|
|
"INSERT INTO compression_codec_multiple_with_key VALUES(toDate('2018-10-12'), 100000, 'hello', 88.88), (toDate('2018-10-12'), 100002, 'world', 99.99), (toDate('2018-10-12'), 1111, '!', 777.777)"
|
|
)
|
|
assert (
|
|
node6.query(
|
|
"SELECT COUNT(*) FROM compression_codec_multiple_with_key WHERE id % 2 == 0"
|
|
)
|
|
== "2\n"
|
|
)
|
|
assert (
|
|
node6.query(
|
|
"SELECT DISTINCT somecolumn FROM compression_codec_multiple_with_key ORDER BY id"
|
|
)
|
|
== "777.777\n88.88\n99.99\n"
|
|
)
|
|
assert (
|
|
node6.query(
|
|
"SELECT data FROM compression_codec_multiple_with_key WHERE id >= 1112 AND somedate = toDate('2018-10-12') AND somecolumn <= 100"
|
|
)
|
|
== "hello\nworld\n"
|
|
)
|
|
|
|
node6.query(
|
|
"INSERT INTO compression_codec_multiple_with_key SELECT toDate('2018-10-12'), number, toString(number), 1.0 FROM system.numbers LIMIT 10000"
|
|
)
|
|
|
|
assert (
|
|
node6.query(
|
|
"SELECT COUNT(id) FROM compression_codec_multiple_with_key WHERE id % 10 == 0"
|
|
)
|
|
== "1001\n"
|
|
)
|
|
assert (
|
|
node6.query("SELECT SUM(somecolumn) FROM compression_codec_multiple_with_key")
|
|
== str(777.777 + 88.88 + 99.99 + 1.0 * 10000) + "\n"
|
|
)
|
|
assert (
|
|
node6.query(
|
|
"SELECT count(*) FROM compression_codec_multiple_with_key GROUP BY somedate"
|
|
)
|
|
== "10003\n"
|
|
)
|