ClickHouse/docker/test/stress/stress

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
from multiprocessing import cpu_count
from subprocess import Popen, call, STDOUT
import os
import sys
import shutil
import argparse
import logging
import time


def get_skip_list_cmd(path):
    with open(path, 'r') as f:
        for line in f:
            if '--use-skip-list' in line:
                return '--use-skip-list'
    return ''


def get_options(i):
    options = ""
    if 0 < i:
        options += " --order=random"

    if i % 3 == 1:
        options += " --db-engine=Ordinary"

    if i % 3 == 2:
        options += ''' --db-engine="Replicated('/test/db/test_{}', 's1', 'r1')"'''.format(i)

    # If database name is not specified, new database is created for each functional test.
    # Run some threads with one database for all tests.
    if i % 2 == 1:
        options += " --database=test_{}".format(i)

    if i == 13:
        options += " --client-option='memory_tracker_fault_probability=0.00001'"

    return options


def run_func_test(cmd, output_prefix, num_processes, skip_tests_option, global_time_limit):
    skip_list_opt = get_skip_list_cmd(cmd)

    global_time_limit_option = ''
    if global_time_limit:
        global_time_limit_option = "--global_time_limit={}".format(global_time_limit)

    output_paths = [os.path.join(output_prefix, "stress_test_run_{}.txt".format(i)) for i in range(num_processes)]
    pipes = []
    for i in range(0, len(output_paths)):
        f = open(output_paths[i], 'w')
        full_command = "{} {} {} {} {}".format(cmd, skip_list_opt, get_options(i), global_time_limit_option, skip_tests_option)
        logging.info("Run func tests '%s'", full_command)
        p = Popen(full_command, shell=True, stdout=f, stderr=f)
        pipes.append(p)
        time.sleep(0.5)
    return pipes

def prepare_for_hung_check():
    # FIXME this function should not exist, but...

    # We attach gdb to clickhouse-server before running tests
    # to print stacktraces of all crashes even if clickhouse cannot print it for some reason.
    # However, it obstruct checking for hung queries.
    logging.info("Will terminate gdb (if any)")
    call("kill -TERM $(pidof gdb)", shell=True, stderr=STDOUT)

    # Some tests set too low memory limit for default user and forget to reset in back.
    # It may cause SYSTEM queries to fail, let's disable memory limit.
    call("clickhouse client --max_memory_usage_for_user=0 -q 'SELECT 1 FORMAT Null'", shell=True, stderr=STDOUT)

    # Some tests execute SYSTEM STOP MERGES or similar queries.
    # It may cause some ALTERs to hang.
    # Possibly we should fix tests and forbid to use such queries without specifying table.
    call("clickhouse client -q 'SYSTEM START MERGES'", shell=True, stderr=STDOUT)
    call("clickhouse client -q 'SYSTEM START DISTRIBUTED SENDS'", shell=True, stderr=STDOUT)
    call("clickhouse client -q 'SYSTEM START TTL MERGES'", shell=True, stderr=STDOUT)
    call("clickhouse client -q 'SYSTEM START MOVES'", shell=True, stderr=STDOUT)
    call("clickhouse client -q 'SYSTEM START FETCHES'", shell=True, stderr=STDOUT)
    call("clickhouse client -q 'SYSTEM START REPLICATED SENDS'", shell=True, stderr=STDOUT)
    call("clickhouse client -q 'SYSTEM START REPLICATION QUEUES'", shell=True, stderr=STDOUT)

    # Issue #21004, live views are experimental, so let's just suppress it
    call("""clickhouse client -q "KILL QUERY WHERE upper(query) LIKE 'WATCH %'" """, shell=True, stderr=STDOUT)

    # Wait for last queries to finish if any, not longer than 120 seconds
    call("""clickhouse client -q "select sleepEachRow((
            select maxOrDefault(120 - elapsed) + 1 from system.processes where query not like '%from system.processes%' and elapsed < 120
            ) / 120) from numbers(120) format Null" """, shell=True, stderr=STDOUT)

if __name__ == "__main__":
    logging.basicConfig(level=logging.INFO, format='%(asctime)s %(message)s')
    parser = argparse.ArgumentParser(description="ClickHouse script for running stresstest")
    parser.add_argument("--test-cmd", default='/usr/bin/clickhouse-test')
    parser.add_argument("--skip-func-tests", default='')
    parser.add_argument("--client-cmd", default='clickhouse-client')
    parser.add_argument("--server-log-folder", default='/var/log/clickhouse-server')
    parser.add_argument("--output-folder")
    parser.add_argument("--global-time-limit", type=int, default=3600)
    parser.add_argument("--num-parallel", default=cpu_count())
    parser.add_argument('--hung-check', action='store_true', default=False)

    args = parser.parse_args()
    func_pipes = []
    func_pipes = run_func_test(args.test_cmd, args.output_folder, args.num_parallel, args.skip_func_tests, args.global_time_limit)

    logging.info("Will wait functests to finish")
    while True:
        retcodes = []
        for p in func_pipes:
            if p.poll() is not None:
                retcodes.append(p.returncode)
        if len(retcodes) == len(func_pipes):
            break
        logging.info("Finished %s from %s processes", len(retcodes), len(func_pipes))
        time.sleep(5)

    logging.info("All processes finished")
    if args.hung_check:
        prepare_for_hung_check()
        logging.info("Checking if some queries hung")
        cmd = "{} {} {}".format(args.test_cmd, "--hung-check", "00001_select_1")
        res = call(cmd, shell=True, stderr=STDOUT)
        hung_check_status = "No queries hung\tOK\n"
        if res != 0:
            logging.info("Hung check failed with exit code {}".format(res))
            hung_check_status = "Hung check failed\tFAIL\n"
        open(os.path.join(args.output_folder, "test_results.tsv"), 'w+').write(hung_check_status)

    logging.info("Stress test finished")