mirror of
https://github.com/ClickHouse/ClickHouse.git
synced 2024-11-30 11:32:03 +00:00
106 lines
3.1 KiB
Python
106 lines
3.1 KiB
Python
|
import re
|
||
|
import time
|
||
|
|
||
|
import pytest
|
||
|
import requests
|
||
|
from helpers.cluster import ClickHouseCluster
|
||
|
from helpers.client import QueryRuntimeException
|
||
|
|
||
|
cluster = ClickHouseCluster(__file__, zookeeper_config_path="configs/zookeeper.xml")
|
||
|
|
||
|
main_configs = [
|
||
|
"configs/remote_servers.xml",
|
||
|
]
|
||
|
|
||
|
node1 = cluster.add_instance(
|
||
|
"node1",
|
||
|
main_configs=main_configs + ["configs/drop_if_empty_check.xml"],
|
||
|
macros={"replica": "node1", "shard": "shard1"},
|
||
|
with_zookeeper=True,
|
||
|
)
|
||
|
|
||
|
node2 = cluster.add_instance(
|
||
|
"node2",
|
||
|
main_configs=main_configs + ["configs/drop_if_empty_check.xml"],
|
||
|
macros={"replica": "node2", "shard": "shard1"},
|
||
|
with_zookeeper=True,
|
||
|
)
|
||
|
|
||
|
|
||
|
@pytest.fixture(scope="module")
|
||
|
def start_cluster():
|
||
|
try:
|
||
|
cluster.start()
|
||
|
yield cluster
|
||
|
finally:
|
||
|
cluster.shutdown()
|
||
|
|
||
|
|
||
|
def test_drop_if_empty(start_cluster):
|
||
|
settings = {
|
||
|
"allow_experimental_database_replicated": 1,
|
||
|
}
|
||
|
node1.query(
|
||
|
"CREATE DATABASE replicateddb "
|
||
|
"ENGINE = Replicated('/clickhouse/databases/replicateddb', 'shard1', 'node1')",
|
||
|
settings=settings,
|
||
|
)
|
||
|
node2.query(
|
||
|
"CREATE DATABASE replicateddb "
|
||
|
"ENGINE = Replicated('/clickhouse/databases/replicateddb', 'shard1', 'node2')",
|
||
|
settings=settings,
|
||
|
)
|
||
|
node1.query(
|
||
|
"CREATE TABLE default.tbl ON CLUSTER 'cluster' ("
|
||
|
"x UInt64"
|
||
|
") ENGINE=ReplicatedMergeTree('/clickhouse/tables/tbl/', '{replica}')"
|
||
|
"ORDER BY x"
|
||
|
)
|
||
|
node1.query(
|
||
|
"CREATE TABLE replicateddb.tbl2 (" "x UInt64" ") ENGINE=MergeTree " "ORDER BY x"
|
||
|
)
|
||
|
|
||
|
assert 1 == int(
|
||
|
node2.query("SELECT count() FROM system.tables WHERE name = 'tbl';")
|
||
|
)
|
||
|
assert 1 == int(
|
||
|
node2.query("SELECT count() FROM system.databases WHERE name = 'replicateddb';")
|
||
|
)
|
||
|
assert 1 == int(
|
||
|
node2.query("SELECT count() FROM system.tables WHERE name = 'tbl2';")
|
||
|
)
|
||
|
|
||
|
node2.query("SYSTEM STOP MERGES;")
|
||
|
node2.query("SYSTEM STOP FETCHES;")
|
||
|
node2.query("SYSTEM STOP REPLICATION QUEUES;")
|
||
|
|
||
|
node1.query("INSERT INTO default.tbl SELECT * FROM system.numbers_mt LIMIT 10000;")
|
||
|
node1.query(
|
||
|
"INSERT INTO replicateddb.tbl2 SELECT * FROM system.numbers_mt LIMIT 10000;"
|
||
|
)
|
||
|
|
||
|
assert 0 == int(node2.query("SELECT count() FROM default.tbl;"))
|
||
|
assert 0 == int(node2.query("SELECT count() FROM replicateddb.tbl2;"))
|
||
|
|
||
|
node2.query("DROP TABLE IF EMPTY default.tbl ON CLUSTER 'cluster';")
|
||
|
node2.query("DROP TABLE IF EMPTY replicateddb.tbl2;")
|
||
|
|
||
|
assert 0 == int(
|
||
|
node1.query("SELECT count() FROM system.tables WHERE name = 'tbl';")
|
||
|
)
|
||
|
assert 0 == int(
|
||
|
node2.query("SELECT count() FROM system.tables WHERE name = 'tbl';")
|
||
|
)
|
||
|
assert 0 == int(
|
||
|
node1.query("SELECT count() FROM system.tables WHERE name = 'tbl2';")
|
||
|
)
|
||
|
assert 0 == int(
|
||
|
node2.query("SELECT count() FROM system.tables WHERE name = 'tbl2';")
|
||
|
)
|
||
|
|
||
|
with pytest.raises(
|
||
|
QueryRuntimeException,
|
||
|
match="DB::Exception: DROP IF EMPTY is not implemented for databases.",
|
||
|
):
|
||
|
node2.query("DROP DATABASE IF EMPTY replicateddb;")
|