ClickHouse/tests/integration/test_random_inserts/test.py

160 lines
5.5 KiB
Python
Raw Normal View History

import time
import os
import threading
import random
from contextlib import contextmanager
import pytest
from helpers.cluster import ClickHouseCluster
from helpers.network import PartitionManager
from helpers.test_tools import TSV
from helpers.client import CommandRequest
cluster = ClickHouseCluster(__file__)
node1 = cluster.add_instance('node1', config_dir='configs', with_zookeeper=True, macros={"layer": 0, "shard": 0, "replica": 1})
node2 = cluster.add_instance('node2', config_dir='configs', with_zookeeper=True, macros={"layer": 0, "shard": 0, "replica": 2})
nodes = [node1, node2]
@pytest.fixture(scope="module")
def started_cluster():
try:
cluster.start()
yield cluster
finally:
pass
cluster.shutdown()
def test_random_inserts(started_cluster):
# Duration of the test, reduce it if don't want to wait
DURATION_SECONDS = 10# * 60
node1.query("""
CREATE TABLE simple ON CLUSTER test_cluster (date Date, i UInt32, s String)
ENGINE = ReplicatedMergeTree('/clickhouse/tables/{shard}/simple', '{replica}', date, i, 8192)""")
with PartitionManager() as pm_random_drops:
for sacrifice in nodes:
pass # This test doesn't work with partition problems still
#pm_random_drops._add_rule({'probability': 0.01, 'destination': sacrifice.ip_address, 'source_port': 2181, 'action': 'REJECT --reject-with tcp-reset'})
#pm_random_drops._add_rule({'probability': 0.01, 'source': sacrifice.ip_address, 'destination_port': 2181, 'action': 'REJECT --reject-with tcp-reset'})
min_timestamp = int(time.time())
max_timestamp = min_timestamp + DURATION_SECONDS
num_timestamps = max_timestamp - min_timestamp + 1
bash_script = os.path.join(os.path.dirname(__file__), "test.sh")
inserters = []
for node in nodes:
2018-09-07 11:51:51 +00:00
cmd = ['/bin/bash', bash_script, node.ip_address, str(min_timestamp), str(max_timestamp), str(cluster.get_client_cmd())]
inserters.append(CommandRequest(cmd, timeout=DURATION_SECONDS * 2, stdin=''))
print node.name, node.ip_address
for inserter in inserters:
inserter.get_answer()
answer="{}\t{}\t{}\t{}\n".format(num_timestamps, num_timestamps, min_timestamp, max_timestamp)
for node in nodes:
2018-09-07 11:51:51 +00:00
res = node.query_with_retry("SELECT count(), uniqExact(i), min(i), max(i) FROM simple", check_callback=lambda res: TSV(res) == TSV(answer))
assert TSV(res) == TSV(answer), node.name + " : " + node.query("SELECT groupArray(_part), i, count() AS c FROM simple GROUP BY i ORDER BY c DESC LIMIT 1")
node1.query("""DROP TABLE simple ON CLUSTER test_cluster""")
class Runner:
def __init__(self):
self.mtx = threading.Lock()
self.total_inserted = 0
self.inserted_vals = set()
self.inserted_payloads = set()
self.stop_ev = threading.Event()
def do_insert(self, thread_num):
self.stop_ev.wait(random.random())
year = 2000
month = '01'
day = str(thread_num + 1).zfill(2)
x = 1
while not self.stop_ev.is_set():
payload = """
{year}-{month}-{day} {x1}
{year}-{month}-{day} {x2}
""".format(year=year, month=month, day=day, x1=x, x2=(x + 1)).strip()
try:
random.choice(nodes).query("INSERT INTO repl_test FORMAT TSV", payload)
# print 'thread {}: insert {}, {}'.format(thread_num, i, i + 1)
self.mtx.acquire()
if payload not in self.inserted_payloads:
self.inserted_payloads.add(payload)
self.inserted_vals.add(x)
self.inserted_vals.add(x + 1)
self.total_inserted += 2 * x + 1
self.mtx.release()
except Exception, e:
print 'Exception:', e
x += 2
self.stop_ev.wait(0.1 + random.random() / 10)
def test_insert_multithreaded(started_cluster):
DURATION_SECONDS = 50
for node in nodes:
node.query("DROP TABLE IF EXISTS repl_test")
for node in nodes:
node.query("CREATE TABLE repl_test(d Date, x UInt32) ENGINE ReplicatedMergeTree('/clickhouse/tables/test/repl_test', '{replica}') ORDER BY x PARTITION BY toYYYYMM(d)")
runner = Runner()
threads = []
for thread_num in range(5):
threads.append(threading.Thread(target=runner.do_insert, args=(thread_num, )))
for t in threads:
t.start()
time.sleep(DURATION_SECONDS)
runner.stop_ev.set()
for t in threads:
t.join()
# Sanity check: at least something was inserted
assert runner.total_inserted > 0
2018-07-05 13:29:22 +00:00
all_replicated = False
for i in range(100): # wait for replication 50 seconds max
time.sleep(0.5)
def get_delay(node):
return int(node.query("SELECT absolute_delay FROM system.replicas WHERE table = 'repl_test'").rstrip())
if all([get_delay(n) == 0 for n in nodes]):
2018-07-05 13:29:22 +00:00
all_replicated = True
break
2018-07-05 13:29:22 +00:00
assert all_replicated
2020-03-02 15:00:34 +00:00
# Now we can be sure that all replicated fetches started, but they may not
# be finished yet so we additionaly sync replicas, to be sure, that we have
# all data on both replicas
for node in nodes:
node.query("SYSTEM SYNC REPLICA repl_test", timeout=10)
2018-07-05 13:29:22 +00:00
actual_inserted = []
for i, node in enumerate(nodes):
actual_inserted.append(int(node.query("SELECT sum(x) FROM repl_test").rstrip()))
assert actual_inserted[i] == runner.total_inserted