mirror of
https://github.com/ClickHouse/ClickHouse.git
synced 2024-11-10 01:25:21 +00:00
54 lines
1.6 KiB
Python
54 lines
1.6 KiB
Python
import pytest
|
|
|
|
from helpers.cluster import ClickHouseCluster
|
|
from helpers.client import QueryRuntimeException
|
|
|
|
cluster = ClickHouseCluster(__file__)
|
|
|
|
initiator = cluster.add_instance(
|
|
"initiator", main_configs=["configs/remote_servers.xml"], with_zookeeper=True
|
|
)
|
|
|
|
|
|
@pytest.fixture(scope="module")
|
|
def start_cluster():
|
|
try:
|
|
cluster.start()
|
|
yield cluster
|
|
finally:
|
|
cluster.shutdown()
|
|
|
|
|
|
def create_tables(cluster, table_name):
|
|
initiator.query(f"DROP TABLE IF EXISTS {table_name} SYNC")
|
|
|
|
initiator.query(
|
|
f"CREATE TABLE IF NOT EXISTS {table_name} (key Int64, value String) Engine=ReplicatedMergeTree('/test_parallel_replicas/shard1/{table_name}', 'r1') ORDER BY (key)"
|
|
)
|
|
|
|
# populate data
|
|
initiator.query(
|
|
f"INSERT INTO {table_name} SELECT number % 4, number FROM numbers(1000)"
|
|
)
|
|
|
|
|
|
@pytest.mark.parametrize("skip_unavailable_shards", [1, 0])
|
|
@pytest.mark.parametrize("max_parallel_replicas", [2, 3, 100])
|
|
def test_skip_all_replicas(
|
|
start_cluster, skip_unavailable_shards, max_parallel_replicas
|
|
):
|
|
cluster_name = "test_1_shard_3_unavaliable_replicas"
|
|
table_name = "tt"
|
|
create_tables(cluster_name, table_name)
|
|
|
|
with pytest.raises(QueryRuntimeException):
|
|
initiator.query(
|
|
f"SELECT key, count() FROM {table_name} GROUP BY key ORDER BY key",
|
|
settings={
|
|
"allow_experimental_parallel_reading_from_replicas": 2,
|
|
"max_parallel_replicas": max_parallel_replicas,
|
|
"cluster_for_parallel_replicas": cluster_name,
|
|
"skip_unavailable_shards": skip_unavailable_shards,
|
|
},
|
|
)
|