ClickHouse/tests/ci/functional_test_check.py

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

390 lines
13 KiB
Python
Raw Normal View History

2021-10-27 07:21:48 +00:00
#!/usr/bin/env python3
2021-10-28 07:29:13 +00:00
2022-03-18 20:32:59 +00:00
import argparse
2021-10-27 07:21:48 +00:00
import csv
import logging
import os
import re
2022-03-18 20:32:59 +00:00
import subprocess
2021-10-27 07:21:48 +00:00
import sys
from pathlib import Path
2022-11-16 09:01:17 +00:00
from typing import List, Tuple
2021-10-27 07:21:48 +00:00
2021-11-12 11:07:54 +00:00
from build_download_helper import download_all_deb_packages
from clickhouse_helper import CiLogsCredentials
2024-02-26 17:46:15 +00:00
from docker_images_helper import DockerImage, get_docker_image, pull_image
from download_release_packages import download_last_release
2024-02-26 17:46:15 +00:00
from env_helper import REPO_COPY, REPORT_PATH, TEMP_PATH
2024-04-24 15:30:47 +00:00
from get_robot_token import get_parameter_from_ssm
from pr_info import PRInfo
from report import (
ERROR,
SUCCESS,
JobReport,
StatusType,
TestResults,
read_test_results,
FAILURE,
)
from stopwatch import Stopwatch
2021-12-03 08:33:16 +00:00
from tee_popen import TeePopen
from ci_config import CI
from ci_utils import Utils
2021-10-27 07:21:48 +00:00
2022-03-18 12:36:45 +00:00
NO_CHANGES_MSG = "Nothing to run"
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)
2023-09-22 11:16:46 +00:00
def get_additional_envs(
check_name: str, run_by_hash_num: int, run_by_hash_total: int
) -> List[str]:
result = []
2024-04-24 15:30:47 +00:00
azure_connection_string = get_parameter_from_ssm("azure_connection_string")
result.append(f"AZURE_CONNECTION_STRING='{azure_connection_string}'")
2021-11-30 11:40:19 +00:00
if "DatabaseReplicated" in check_name:
result.append("USE_DATABASE_REPLICATED=1")
2021-11-30 11:40:19 +00:00
if "DatabaseOrdinary" in check_name:
result.append("USE_DATABASE_ORDINARY=1")
2021-11-30 11:40:19 +00:00
if "wide parts enabled" in check_name:
result.append("USE_POLYMORPHIC_PARTS=1")
2023-02-03 13:34:18 +00:00
if "ParallelReplicas" in check_name:
result.append("USE_PARALLEL_REPLICAS=1")
if "s3 storage" in check_name:
result.append("USE_S3_STORAGE_FOR_MERGE_TREE=1")
result.append("RANDOMIZE_OBJECT_KEY_TYPE=1")
2023-04-12 15:18:28 +00:00
if "analyzer" in check_name:
2024-03-20 12:35:30 +00:00
result.append("USE_OLD_ANALYZER=1")
2024-04-11 14:50:18 +00:00
if "azure" in check_name:
assert "USE_S3_STORAGE_FOR_MERGE_TREE=1" not in result
result.append("USE_AZURE_STORAGE_FOR_MERGE_TREE=1")
2021-11-30 11:40:19 +00:00
if run_by_hash_total != 0:
result.append(f"RUN_BY_HASH_NUM={run_by_hash_num}")
result.append(f"RUN_BY_HASH_TOTAL={run_by_hash_total}")
return result
2021-10-27 07:21:48 +00:00
2023-09-22 11:16:46 +00:00
def get_image_name(check_name: str) -> str:
2021-10-27 07:21:48 +00:00
if "stateless" in check_name.lower():
return "clickhouse/stateless-test"
if "stateful" in check_name.lower():
return "clickhouse/stateful-test"
2024-02-26 17:46:15 +00:00
raise ValueError(f"Cannot deduce image name based on check name {check_name}")
2021-10-27 07:21:48 +00:00
2022-03-10 13:56:59 +00:00
def get_run_command(
2023-08-16 20:53:51 +00:00
check_name: str,
2023-09-22 11:16:46 +00:00
builds_path: Path,
repo_path: Path,
result_path: Path,
server_log_path: Path,
2023-08-16 20:53:51 +00:00
additional_envs: List[str],
ci_logs_args: str,
image: DockerImage,
flaky_check: bool,
tests_to_run: List[str],
) -> str:
2021-10-27 07:21:48 +00:00
additional_options = ["--hung-check"]
additional_options.append("--print-time")
2021-11-10 13:49:19 +00:00
if tests_to_run:
additional_options += tests_to_run
2021-10-27 07:21:48 +00:00
additional_options_str = (
'-e ADDITIONAL_OPTIONS="' + " ".join(additional_options) + '"'
)
2021-10-27 07:21:48 +00:00
2021-10-28 09:26:59 +00:00
envs = [
2022-08-12 07:46:36 +00:00
# a static link, don't use S3_URL or S3_DOWNLOAD
'-e S3_URL="https://s3.amazonaws.com/clickhouse-datasets"',
2021-10-28 09:26:59 +00:00
]
2021-11-10 13:49:19 +00:00
if flaky_check:
2024-07-16 08:32:10 +00:00
envs.append("-e NUM_TRIES=50")
envs.append("-e MAX_RUN_TIME=2800")
2021-11-10 13:49:19 +00:00
2021-10-27 07:21:48 +00:00
envs += [f"-e {e}" for e in additional_envs]
2021-11-10 13:49:19 +00:00
2021-10-27 07:21:48 +00:00
env_str = " ".join(envs)
2023-04-18 00:56:29 +00:00
volume_with_broken_test = (
2023-08-16 20:53:51 +00:00
f"--volume={repo_path}/tests/analyzer_tech_debt.txt:/analyzer_tech_debt.txt "
2024-03-20 12:35:30 +00:00
if "analyzer" not in check_name
2023-04-18 00:56:29 +00:00
else ""
)
2021-10-27 07:21:48 +00:00
return (
f"docker run --volume={builds_path}:/package_folder "
2024-07-09 16:39:21 +00:00
# For dmesg and sysctl
"--privileged "
2023-08-16 20:53:51 +00:00
f"{ci_logs_args}"
2023-08-06 02:38:04 +00:00
f"--volume={repo_path}/tests:/usr/share/clickhouse-test "
f"--volume={repo_path}/utils/grpc-client:/usr/share/clickhouse-utils/grpc-client "
2023-08-16 20:53:51 +00:00
f"{volume_with_broken_test}"
f"--volume={result_path}:/test_output "
f"--volume={server_log_path}:/var/log/clickhouse-server "
2024-02-25 22:49:25 +00:00
"--security-opt seccomp=unconfined " # required to issue io_uring sys-calls
2021-10-27 09:58:21 +00:00
f"--cap-add=SYS_PTRACE {env_str} {additional_options_str} {image}"
)
2021-10-27 07:21:48 +00:00
def _get_statless_tests_to_run(pr_info: PRInfo) -> List[str]:
result = set()
2021-11-10 13:49:19 +00:00
if pr_info.changed_files is None:
return []
for fpath in pr_info.changed_files:
if re.match(r"tests/queries/0_stateless/[0-9]{5}", fpath):
path_ = Path(REPO_COPY + "/" + fpath)
if not path_.exists():
logging.info("File '%s' is removed - skip", fpath)
continue
logging.info("File '%s' is changed and seems like a test", fpath)
2021-11-10 13:49:19 +00:00
fname = fpath.split("/")[3]
fname_without_ext = os.path.splitext(fname)[0]
# add '.' to the end of the test name not to run all tests with the same prefix
# e.g. we changed '00001_some_name.reference'
# and we have ['00001_some_name.sh', '00001_some_name_2.sql']
# so we want to run only '00001_some_name.sh'
2021-11-10 13:49:19 +00:00
result.add(fname_without_ext + ".")
elif "tests/queries/" in fpath:
# log suspicious changes from tests/ for debugging in case of any problems
logging.info("File '%s' is changed, but it doesn't look like a test", fpath)
2021-11-10 13:49:19 +00:00
return list(result)
2022-11-16 09:01:17 +00:00
def process_results(
ret_code: int,
2023-09-22 11:16:46 +00:00
result_directory: Path,
server_log_path: Path,
) -> Tuple[StatusType, str, TestResults, List[Path]]:
test_results = [] # type: TestResults
2021-10-27 07:21:48 +00:00
additional_files = []
2023-09-22 11:16:46 +00:00
# Just upload all files from result_directory.
# If task provides processed results, then it's responsible for content of result_directory.
if result_directory.exists():
additional_files = [p for p in result_directory.iterdir() if p.is_file()]
2021-10-27 07:21:48 +00:00
2023-09-22 11:16:46 +00:00
if server_log_path.exists():
2021-10-27 07:21:48 +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()
]
2021-10-27 07:21:48 +00:00
2021-12-12 12:09:44 +00:00
status = []
2023-09-22 11:16:46 +00:00
status_path = result_directory / "check_status.tsv"
if status_path.exists():
logging.info("Found %s", status_path.name)
2021-12-12 12:09:44 +00:00
with open(status_path, "r", encoding="utf-8") as status_file:
status = list(csv.reader(status_file, delimiter="\t"))
2021-10-28 07:29:13 +00:00
2021-10-27 07:21:48 +00:00
if len(status) != 1 or len(status[0]) != 2:
2023-09-22 11:16:46 +00:00
logging.info("Files in result folder %s", os.listdir(result_directory))
return ERROR, "Invalid check_status.tsv", test_results, additional_files
2021-10-27 07:21:48 +00:00
state, description = status[0][0], status[0][1]
if ret_code != 0:
state = ERROR
description += " (but script exited with an error)"
2021-10-27 07:21:48 +00:00
try:
2023-09-22 11:16:46 +00:00
results_path = result_directory / "test_results.tsv"
if results_path.exists():
logging.info("Found test_results.tsv")
else:
2023-09-22 11:16:46 +00:00
logging.info("Files in result folder %s", os.listdir(result_directory))
return ERROR, "Not found test_results.tsv", test_results, additional_files
test_results = read_test_results(results_path)
if len(test_results) == 0:
return ERROR, "Empty test_results.tsv", test_results, additional_files
except Exception as e:
return (
ERROR,
f"Cannot parse test_results.tsv ({e})",
test_results,
additional_files,
)
2021-10-27 07:21:48 +00:00
return state, description, test_results, additional_files # type: ignore
2021-10-27 07:21:48 +00:00
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument("check_name")
parser.add_argument(
"--validate-bugfix",
action="store_true",
help="Check that added tests failed on latest stable",
)
2022-03-18 12:36:45 +00:00
parser.add_argument(
"--report-to-file",
type=str,
default="",
help="Path to write script report to (for --validate-bugfix)",
2022-03-18 12:36:45 +00:00
)
return parser.parse_args()
def main():
2021-10-27 07:21:48 +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)
reports_path.mkdir(parents=True, exist_ok=True)
2023-09-22 11:16:46 +00:00
repo_path = Path(REPO_COPY)
2021-10-27 07:21:48 +00:00
args = parse_args()
check_name = args.check_name or os.getenv("CHECK_NAME")
assert (
check_name
), "Check name must be provided as an input arg or in CHECK_NAME env"
validate_bugfix_check = args.validate_bugfix
2021-11-10 13:49:19 +00:00
flaky_check = "flaky" in check_name.lower()
2022-03-08 15:19:05 +00:00
run_changed_tests = flaky_check or validate_bugfix_check
pr_info = PRInfo(need_changed_files=run_changed_tests)
tests_to_run = []
assert (
not validate_bugfix_check or args.report_to_file
), "JobReport file path must be provided with --validate-bugfix"
if run_changed_tests:
tests_to_run = _get_statless_tests_to_run(pr_info)
2022-03-10 16:39:18 +00:00
if "RUN_BY_HASH_NUM" in os.environ:
2022-11-16 09:01:17 +00:00
run_by_hash_num = int(os.getenv("RUN_BY_HASH_NUM", "0"))
run_by_hash_total = int(os.getenv("RUN_BY_HASH_TOTAL", "0"))
else:
run_by_hash_num = 0
run_by_hash_total = 0
2021-10-28 20:20:30 +00:00
2021-10-27 07:21:48 +00:00
image_name = get_image_name(check_name)
docker_image = pull_image(get_docker_image(image_name))
2021-10-27 07:21:48 +00:00
2023-09-22 11:16:46 +00:00
packages_path = temp_path / "packages"
packages_path.mkdir(parents=True, exist_ok=True)
2021-10-27 07:21:48 +00:00
if validate_bugfix_check:
download_last_release(packages_path, debug=True)
else:
download_all_deb_packages(check_name, reports_path, packages_path)
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)
2021-10-27 07:21:48 +00:00
2023-09-22 11:16:46 +00:00
result_path = temp_path / "result_path"
result_path.mkdir(parents=True, exist_ok=True)
2021-10-27 07:21:48 +00:00
2023-09-22 11:16:46 +00:00
run_log_path = result_path / "run.log"
2021-10-27 07:21:48 +00:00
additional_envs = get_additional_envs(
check_name, run_by_hash_num, run_by_hash_total
)
if validate_bugfix_check:
2022-03-10 12:22:24 +00:00
additional_envs.append("GLOBAL_TAGS=no-random-settings")
additional_envs.append("BUGFIX_VALIDATE_CHECK=1")
2023-09-22 11:16:46 +00:00
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
)
if (not validate_bugfix_check and not flaky_check) or tests_to_run:
run_command = get_run_command(
check_name,
packages_path,
repo_path,
result_path,
server_log_path,
additional_envs,
ci_logs_args,
docker_image,
flaky_check,
tests_to_run,
)
logging.info("Going to run func tests: %s", run_command)
with TeePopen(run_command, run_log_path) as process:
retcode = process.wait()
if retcode == 0:
logging.info("Run successfully")
else:
logging.info("Run failed")
try:
subprocess.check_call(
f"sudo chown -R ubuntu:ubuntu {temp_path}", shell=True
)
except subprocess.CalledProcessError:
logging.warning(
"Failed to change files owner in %s, ignoring it", temp_path
)
2021-11-19 14:47:04 +00:00
ci_logs_credentials.clean_ci_logs_from_credentials(run_log_path)
2021-11-12 12:36:25 +00:00
state, description, test_results, additional_logs = process_results(
retcode, result_path, server_log_path
)
2022-03-18 12:36:45 +00:00
else:
print(
"This is validate bugfix or flaky check run, but no changes test to run - skip with success"
)
state, description, test_results, additional_logs = (
SUCCESS,
"No tests to run",
[],
[],
2022-03-18 12:36:45 +00:00
)
JobReport(
description=description,
test_results=test_results,
status=state,
start_time=stopwatch.start_time_str,
duration=stopwatch.duration_seconds,
additional_files=additional_logs,
).dump(to_file=args.report_to_file if args.report_to_file else None)
2021-11-25 10:01:29 +00:00
should_block_ci = False
if state != SUCCESS:
should_block_ci = True
if state == FAILURE and CI.is_required(check_name):
failed_cnt = Utils.get_failed_tests_number(description)
print(
f"Job status is [{state}] with [{failed_cnt}] failed test cases. status description [{description}]"
)
if (
failed_cnt
and failed_cnt <= CI.MAX_TOTAL_FAILURES_PER_JOB_BEFORE_BLOCKING_CI
):
print(f"Won't block the CI workflow")
should_block_ci = False
if should_block_ci:
sys.exit(1)
if __name__ == "__main__":
main()