mirror of
https://github.com/ClickHouse/ClickHouse.git
synced 2024-11-18 13:42:02 +00:00
b75963d370
This PR formats all the `*.py` files found under the `tests/integration` folder. It also reorders the imports and cleans up a bunch of unused imports. The formatting also takes care of other things like wrapping lines and fixing spaces and indents such that the tests look more readable.
80 lines
2.2 KiB
Python
80 lines
2.2 KiB
Python
import io
|
|
import logging
|
|
|
|
import avro.schema
|
|
import pytest
|
|
from confluent.schemaregistry.serializers import MessageSerializer
|
|
from helpers.cluster import ClickHouseCluster, ClickHouseInstance
|
|
|
|
logging.getLogger().setLevel(logging.INFO)
|
|
logging.getLogger().addHandler(logging.StreamHandler())
|
|
|
|
|
|
@pytest.fixture(scope="module")
|
|
def cluster():
|
|
try:
|
|
cluster = ClickHouseCluster(__file__)
|
|
cluster.add_instance("dummy", with_kafka=True)
|
|
logging.info("Starting cluster...")
|
|
cluster.start()
|
|
logging.info("Cluster started")
|
|
|
|
yield cluster
|
|
finally:
|
|
cluster.shutdown()
|
|
|
|
|
|
def run_query(instance, query, data=None, settings=None):
|
|
# type: (ClickHouseInstance, str, object, dict) -> str
|
|
|
|
logging.info("Running query '{}'...".format(query))
|
|
# use http to force parsing on server
|
|
if not data:
|
|
data = " " # make POST request
|
|
result = instance.http_query(query, data=data, params=settings)
|
|
logging.info("Query finished")
|
|
|
|
return result
|
|
|
|
|
|
def test_select(cluster):
|
|
# type: (ClickHouseCluster) -> None
|
|
|
|
schema_registry_client = cluster.schema_registry_client
|
|
serializer = MessageSerializer(schema_registry_client)
|
|
|
|
schema = avro.schema.make_avsc_object({
|
|
'name': 'test_record',
|
|
'type': 'record',
|
|
'fields': [
|
|
{
|
|
'name': 'value',
|
|
'type': 'long'
|
|
}
|
|
]
|
|
})
|
|
|
|
buf = io.BytesIO()
|
|
for x in range(0, 3):
|
|
message = serializer.encode_record_with_schema(
|
|
'test_subject', schema, {'value': x}
|
|
)
|
|
buf.write(message)
|
|
data = buf.getvalue()
|
|
|
|
instance = cluster.instances["dummy"] # type: ClickHouseInstance
|
|
schema_registry_url = "http://{}:{}".format(
|
|
cluster.schema_registry_host,
|
|
cluster.schema_registry_port
|
|
)
|
|
|
|
run_query(instance, "create table avro_data(value Int64) engine = Memory()")
|
|
settings = {'format_avro_schema_registry_url': schema_registry_url}
|
|
run_query(instance, "insert into avro_data format AvroConfluent", data, settings)
|
|
stdout = run_query(instance, "select * from avro_data")
|
|
assert list(map(str.split, stdout.splitlines())) == [
|
|
["0"],
|
|
["1"],
|
|
["2"],
|
|
]
|