ClickHouse/tests/integration/test_old_versions/test.py

74 lines
3.5 KiB
Python
Raw Normal View History

import time
import os
import pytest
from helpers.cluster import ClickHouseCluster
from multiprocessing.dummy import Pool
from helpers.client import QueryRuntimeException, QueryTimeoutExceedException
from helpers.test_tools import assert_eq_with_retry
cluster = ClickHouseCluster(__file__)
node18_14 = cluster.add_instance('node18_14', image='yandex/clickhouse-server:18.14.19', with_installed_binary=True, config_dir="configs")
node19_1 = cluster.add_instance('node19_1', image='yandex/clickhouse-server:19.1.16', with_installed_binary=True, config_dir="configs")
node19_4 = cluster.add_instance('node19_4', image='yandex/clickhouse-server:19.4.5.35', with_installed_binary=True, config_dir="configs")
node19_8 = cluster.add_instance('node19_8', image='yandex/clickhouse-server:19.8.3.8', with_installed_binary=True, config_dir="configs")
node19_11 = cluster.add_instance('node19_11', image='yandex/clickhouse-server:19.11.13.74', with_installed_binary=True, config_dir="configs")
node19_13 = cluster.add_instance('node19_13', image='yandex/clickhouse-server:19.13.7.57', with_installed_binary=True, config_dir="configs")
node19_16 = cluster.add_instance('node19_16', image='yandex/clickhouse-server:19.16.2.2', with_installed_binary=True, config_dir="configs")
old_nodes = [node18_14, node19_1, node19_4, node19_8, node19_11, node19_13, node19_16]
new_node = cluster.add_instance('node_new')
def query_from_one_node_to_another(client_node, server_node, query):
client_node.exec_in_container(["bash", "-c", "/usr/bin/clickhouse client --host {} --query {!r}".format(server_node.name, query)])
@pytest.fixture(scope="module")
def setup_nodes():
try:
cluster.start()
for n in old_nodes + [new_node]:
n.query('''CREATE TABLE test_table (id UInt32, value UInt64) ENGINE = MergeTree() ORDER BY tuple()''')
for n in old_nodes:
n.query('''CREATE TABLE dist_table AS test_table ENGINE = Distributed('test_cluster', 'default', 'test_table')''')
yield cluster
finally:
cluster.shutdown()
def test_client_is_older_than_server(setup_nodes):
server = new_node
for i, client in enumerate(old_nodes):
query_from_one_node_to_another(client, server, "INSERT INTO test_table VALUES (1, {})".format(i))
for client in old_nodes:
query_from_one_node_to_another(client, server, "SELECT COUNT() FROM test_table")
assert server.query("SELECT COUNT() FROM test_table WHERE id=1") == str(len(old_nodes)) + "\n"
def test_server_is_older_than_client(setup_nodes):
client = new_node
for i, server in enumerate(old_nodes):
query_from_one_node_to_another(client, server, "INSERT INTO test_table VALUES (2, {})".format(i))
for server in old_nodes:
query_from_one_node_to_another(client, server, "SELECT COUNT() FROM test_table")
for server in old_nodes:
assert server.query("SELECT COUNT() FROM test_table WHERE id=2") == "1\n"
def test_distributed_query_initiator_is_older_than_shard(setup_nodes):
distributed_query_initiator_old_nodes = [node18_14, node19_13, node19_16]
shard = new_node
for i, initiator in enumerate(distributed_query_initiator_old_nodes):
initiator.query("INSERT INTO dist_table VALUES (3, {})".format(i))
assert_eq_with_retry(shard, "SELECT COUNT() FROM test_table WHERE id=3", str(len(distributed_query_initiator_old_nodes)))
assert_eq_with_retry(initiator, "SELECT COUNT() FROM dist_table WHERE id=3", str(len(distributed_query_initiator_old_nodes)))