ClickHouse/tests/integration/test_dictionaries_complex_key_cache_string/test.py

52 lines
2.5 KiB
Python

import pytest
import os
from helpers.cluster import ClickHouseCluster
@pytest.fixture(scope="function")
def cluster(request):
SCRIPT_DIR = os.path.dirname(os.path.realpath(__file__))
cluster = ClickHouseCluster(__file__)
try:
if request.param == "memory":
node = cluster.add_instance('node', main_configs=['configs/enable_dictionaries.xml', 'configs/dictionaries/complex_key_cache_string.xml'])
if request.param == "ssd":
node = cluster.add_instance('node', main_configs=['configs/enable_dictionaries.xml', 'configs/dictionaries/ssd_complex_key_cache_string.xml'])
cluster.start()
node.query("create table radars_table (radar_id String, radar_ip String, client_id String) engine=MergeTree() order by radar_id")
yield cluster
finally:
cluster.shutdown()
@pytest.mark.parametrize("cluster", ["memory", "ssd"], indirect=True)
def test_memory_consumption(cluster):
node = cluster.instances['node']
node.query("insert into radars_table select toString(rand() % 5000), '{0}', '{0}' from numbers(1000)".format('w' * 8))
node.query("insert into radars_table select toString(rand() % 5000), '{0}', '{0}' from numbers(1000)".format('x' * 16))
node.query("insert into radars_table select toString(rand() % 5000), '{0}', '{0}' from numbers(1000)".format('y' * 32))
node.query("insert into radars_table select toString(rand() % 5000), '{0}', '{0}' from numbers(1000)".format('z' * 64))
# Fill dictionary
node.query("select dictGetString('radars', 'client_id', tuple(toString(number))) from numbers(0, 5000)")
allocated_first = int(node.query("select bytes_allocated from system.dictionaries where name = 'radars'").strip())
alloc_array = []
for i in xrange(5):
node.query("select dictGetString('radars', 'client_id', tuple(toString(number))) from numbers(0, 5000)")
allocated = int(node.query("select bytes_allocated from system.dictionaries where name = 'radars'").strip())
alloc_array.append(allocated)
# size doesn't grow
assert all(allocated_first >= a for a in alloc_array)
for i in xrange(5):
node.query("select dictGetString('radars', 'client_id', tuple(toString(number))) from numbers(0, 5000)")
allocated = int(node.query("select bytes_allocated from system.dictionaries where name = 'radars'").strip())
alloc_array.append(allocated)
# size doesn't grow
assert all(allocated_first >= a for a in alloc_array)