ClickHouse/tests/ci/stress_check.py

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

228 lines
7.1 KiB
Python
Raw Normal View History

2021-11-01 10:27:46 +00:00
#!/usr/bin/env python3
import csv
import logging
import os
2024-07-15 11:30:48 +00:00
import re
2021-11-01 10:27:46 +00:00
import subprocess
import sys
from pathlib import Path
2022-11-16 09:24:10 +00:00
from typing import List, Tuple
2021-11-01 10:27:46 +00:00
2021-11-12 11:07:54 +00:00
from build_download_helper import download_all_deb_packages
2024-11-22 11:41:49 +00:00
from ci_utils import Shell
from clickhouse_helper import CiLogsCredentials
from docker_images_helper import DockerImage, get_docker_image, pull_image
from env_helper import REPO_COPY, REPORT_PATH, TEMP_PATH
2024-04-24 15:47:32 +00:00
from get_robot_token import get_parameter_from_ssm
from pr_info import PRInfo
from report import ERROR, JobReport, TestResults, read_test_results
from stopwatch import Stopwatch
2021-12-03 08:33:16 +00:00
from tee_popen import TeePopen
2021-11-19 14:47:04 +00:00
2021-11-01 10:27:46 +00:00
2024-07-15 11:30:48 +00:00
class SensitiveFormatter(logging.Formatter):
@staticmethod
def _filter(s):
return re.sub(
r"(.*)(AZURE_CONNECTION_STRING.*\')(.*)", r"\1AZURE_CONNECTION_STRING\3", s
)
def format(self, record):
original = logging.Formatter.format(self, record)
return self._filter(original)
2024-04-11 15:15:47 +00:00
def get_additional_envs(check_name: str) -> List[str]:
result = []
2024-04-24 15:32:18 +00:00
azure_connection_string = get_parameter_from_ssm("azure_connection_string")
result.append(f"AZURE_CONNECTION_STRING='{azure_connection_string}'")
2023-10-18 08:22:18 +00:00
# some cloud-specific features require feature flags enabled
# so we need this ENV to be able to disable the randomization
# of feature flags
2023-10-18 09:19:09 +00:00
result.append("RANDOMIZE_KEEPER_FEATURE_FLAGS=1")
2024-04-11 14:50:18 +00:00
if "azure" in check_name:
result.append("USE_AZURE_STORAGE_FOR_MERGE_TREE=1")
2024-07-02 12:15:59 +00:00
if "s3" in check_name:
result.append("USE_S3_STORAGE_FOR_MERGE_TREE=1")
return result
2022-03-07 13:25:06 +00:00
def get_run_command(
2023-09-22 11:16:46 +00:00
build_path: Path,
result_path: Path,
repo_tests_path: Path,
server_log_path: Path,
additional_envs: List[str],
2023-08-16 20:53:51 +00:00
ci_logs_args: str,
image: DockerImage,
2024-08-09 14:22:33 +00:00
upgrade_check: bool,
2023-08-16 20:53:51 +00:00
) -> str:
envs = [f"-e {e}" for e in additional_envs]
env_str = " ".join(envs)
2024-08-09 14:22:33 +00:00
if upgrade_check:
run_script = "/repo/tests/docker_scripts/upgrade_runner.sh"
else:
run_script = "/repo/tests/docker_scripts/stress_runner.sh"
2022-03-07 13:25:06 +00:00
cmd = (
"docker run --cap-add=SYS_PTRACE "
# For dmesg and sysctl
"--privileged "
2023-08-16 20:53:51 +00:00
# a static link, don't use S3_URL or S3_DOWNLOAD
"-e S3_URL='https://s3.amazonaws.com/clickhouse-datasets' "
f"{ci_logs_args}"
2022-03-07 13:25:06 +00:00
f"--volume={build_path}:/package_folder "
2023-08-16 20:53:51 +00:00
f"--volume={result_path}:/test_output "
2024-08-09 14:22:33 +00:00
f"--volume={repo_tests_path}/..:/repo "
f"--volume={server_log_path}:/var/log/clickhouse-server {env_str} {image} {run_script}"
2022-03-07 13:25:06 +00:00
)
2021-11-01 10:27:46 +00:00
return cmd
2022-03-07 13:25:06 +00:00
2022-11-16 09:24:10 +00:00
def process_results(
2023-09-22 11:16:46 +00:00
result_directory: Path, server_log_path: Path, run_log_path: Path
) -> Tuple[str, str, TestResults, List[Path]]:
test_results = [] # type: TestResults
2021-11-01 10:27:46 +00:00
additional_files = []
# Just upload all files from result_folder.
2022-03-07 13:25:06 +00:00
# If task provides processed results, then it's responsible for content
# of result_folder.
2023-09-22 11:16:46 +00:00
if result_directory.exists():
additional_files = [p for p in result_directory.iterdir() if p.is_file()]
2021-11-01 10:27:46 +00:00
2023-09-22 11:16:46 +00:00
if server_log_path.exists():
2022-03-07 13:25:06 +00:00
additional_files = additional_files + [
2023-09-22 11:16:46 +00:00
p for p in server_log_path.iterdir() if p.is_file()
2022-03-07 13:25:06 +00:00
]
2021-11-01 10:27:46 +00:00
2021-11-01 13:27:55 +00:00
additional_files.append(run_log_path)
2023-09-22 11:16:46 +00:00
status_path = result_directory / "check_status.tsv"
if not status_path.exists():
2022-03-07 13:25:06 +00:00
return (
"failure",
"check_status.tsv doesn't exists",
test_results,
additional_files,
)
2021-11-01 10:27:46 +00:00
logging.info("Found check_status.tsv")
2022-03-07 13:25:06 +00:00
with open(status_path, "r", encoding="utf-8") as status_file:
status = list(csv.reader(status_file, delimiter="\t"))
2021-11-01 10:27:46 +00:00
if len(status) != 1 or len(status[0]) != 2:
return ERROR, "Invalid check_status.tsv", test_results, additional_files
2021-11-01 10:27:46 +00:00
state, description = status[0][0], status[0][1]
try:
2023-09-22 11:16:46 +00:00
results_path = result_directory / "test_results.tsv"
test_results = read_test_results(results_path, True)
if len(test_results) == 0:
2024-02-26 17:46:15 +00:00
raise ValueError("Empty results")
except Exception as e:
return (
ERROR,
f"Cannot parse test_results.tsv ({e})",
test_results,
additional_files,
)
2021-11-01 10:27:46 +00:00
return state, description, test_results, additional_files
2024-08-09 14:22:33 +00:00
def run_stress_test(upgrade_check: bool = False) -> None:
2021-11-01 10:27:46 +00:00
logging.basicConfig(level=logging.INFO)
2024-07-15 11:30:48 +00:00
for handler in logging.root.handlers:
2024-07-15 11:42:00 +00:00
# pylint: disable=protected-access
handler.setFormatter(SensitiveFormatter(handler.formatter._fmt)) # type: ignore
2021-11-19 14:47:04 +00:00
stopwatch = Stopwatch()
2023-09-22 11:16:46 +00:00
temp_path = Path(TEMP_PATH)
reports_path = Path(REPORT_PATH)
2023-09-22 11:16:46 +00:00
temp_path.mkdir(parents=True, exist_ok=True)
repo_path = Path(REPO_COPY)
repo_tests_path = repo_path / "tests"
2021-11-01 10:27:46 +00:00
check_name = sys.argv[1] if len(sys.argv) > 1 else os.getenv("CHECK_NAME")
assert (
check_name
), "Check name must be provided as an input arg or in CHECK_NAME env"
2021-11-01 10:27:46 +00:00
2021-11-26 14:00:09 +00:00
pr_info = PRInfo()
2021-11-01 10:27:46 +00:00
2023-09-22 11:16:46 +00:00
packages_path = temp_path / "packages"
packages_path.mkdir(parents=True, exist_ok=True)
2021-11-01 10:27:46 +00:00
if check_name in ("amd_release", "amd_debug", "arm_release"):
# this is praktika based CI
print("Copy input *.deb artifacts")
assert Shell.check(
f"cp /tmp/praktika/input/*.deb {packages_path}", verbose=True
)
2024-11-22 11:41:49 +00:00
docker_image = pull_image(get_docker_image("clickhouse/stateful-test"))
else:
download_all_deb_packages(check_name, reports_path, packages_path)
2024-11-22 11:41:49 +00:00
docker_image = pull_image(get_docker_image("clickhouse/stress-test"))
2021-11-12 11:07:54 +00:00
2023-09-22 11:16:46 +00:00
server_log_path = temp_path / "server_log"
server_log_path.mkdir(parents=True, exist_ok=True)
2021-11-01 10:27:46 +00:00
2023-09-22 11:16:46 +00:00
result_path = temp_path / "result_path"
result_path.mkdir(parents=True, exist_ok=True)
2021-11-01 10:27:46 +00:00
2023-09-22 11:16:46 +00:00
run_log_path = temp_path / "run.log"
ci_logs_credentials = CiLogsCredentials(temp_path / "export-logs-config.sh")
2023-08-16 20:53:51 +00:00
ci_logs_args = ci_logs_credentials.get_docker_arguments(
pr_info, stopwatch.start_time_str, check_name
)
2021-11-01 10:27:46 +00:00
2024-04-11 14:50:18 +00:00
additional_envs = get_additional_envs(check_name)
2022-03-07 13:25:06 +00:00
run_command = get_run_command(
2023-08-12 20:30:18 +00:00
packages_path,
result_path,
repo_tests_path,
server_log_path,
additional_envs,
2023-08-16 20:53:51 +00:00
ci_logs_args,
2023-08-12 20:43:29 +00:00
docker_image,
2024-08-09 14:22:33 +00:00
upgrade_check,
2022-03-07 13:25:06 +00:00
)
2023-08-12 20:30:18 +00:00
logging.info("Going to run stress test: %s", run_command)
2021-11-01 10:27:46 +00:00
with TeePopen(run_command, run_log_path) as process:
2021-12-03 08:33:16 +00:00
retcode = process.wait()
if retcode == 0:
2021-12-03 08:33:16 +00:00
logging.info("Run successfully")
else:
logging.info("Run failed")
2021-11-01 10:27:46 +00:00
subprocess.check_call(f"sudo chown -R ubuntu:ubuntu {temp_path}", shell=True)
2023-09-22 11:16:46 +00:00
ci_logs_credentials.clean_ci_logs_from_credentials(run_log_path)
2021-11-01 10:27:46 +00:00
2022-03-07 13:25:06 +00:00
state, description, test_results, additional_logs = process_results(
result_path, server_log_path, run_log_path
)
2023-09-07 12:56:22 +00:00
Shell.check("pwd", verbose=True)
JobReport(
description=description,
test_results=test_results,
status=state,
start_time=stopwatch.start_time_str,
duration=stopwatch.duration_seconds,
additional_files=additional_logs,
2024-11-22 11:41:49 +00:00
).dump()
2022-03-29 12:41:47 +00:00
if state == "failure":
2022-03-29 12:41:47 +00:00
sys.exit(1)
2023-01-03 19:32:11 +00:00
if __name__ == "__main__":
2024-08-09 14:22:33 +00:00
run_stress_test()