mirror of
https://github.com/ClickHouse/ClickHouse.git
synced 2024-11-24 00:22:29 +00:00
413 lines
13 KiB
Python
413 lines
13 KiB
Python
#!/usr/bin/env python3
|
|
|
|
import argparse
|
|
import csv
|
|
import logging
|
|
import os
|
|
import re
|
|
import signal
|
|
import subprocess
|
|
import sys
|
|
from pathlib import Path
|
|
from typing import List, Optional, Tuple
|
|
|
|
from build_download_helper import download_all_deb_packages
|
|
from ci_config import CI
|
|
from ci_utils import Shell, Utils
|
|
from clickhouse_helper import CiLogsCredentials
|
|
from docker_images_helper import DockerImage, get_docker_image, pull_image
|
|
from download_release_packages import download_last_release
|
|
from env_helper import REPO_COPY, REPORT_PATH, TEMP_PATH
|
|
from get_robot_token import get_parameter_from_ssm
|
|
from pr_info import PRInfo
|
|
from report import (
|
|
ERROR,
|
|
FAILURE,
|
|
SUCCESS,
|
|
JobReport,
|
|
StatusType,
|
|
TestResult,
|
|
TestResults,
|
|
read_test_results,
|
|
)
|
|
from stopwatch import Stopwatch
|
|
from tee_popen import TeePopen
|
|
|
|
NO_CHANGES_MSG = "Nothing to run"
|
|
|
|
|
|
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)
|
|
|
|
|
|
def get_additional_envs(
|
|
check_name: str, run_by_hash_num: int, run_by_hash_total: int
|
|
) -> List[str]:
|
|
result = []
|
|
azure_connection_string = get_parameter_from_ssm("azure_connection_string")
|
|
result.append(f"AZURE_CONNECTION_STRING='{azure_connection_string}'")
|
|
if "DatabaseReplicated" in check_name:
|
|
result.append("USE_DATABASE_REPLICATED=1")
|
|
if "DatabaseOrdinary" in check_name:
|
|
result.append("USE_DATABASE_ORDINARY=1")
|
|
if "wide parts enabled" in check_name:
|
|
result.append("USE_POLYMORPHIC_PARTS=1")
|
|
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")
|
|
if "analyzer" in check_name:
|
|
result.append("USE_OLD_ANALYZER=1")
|
|
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")
|
|
|
|
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
|
|
|
|
|
|
def get_image_name(check_name: str) -> str:
|
|
if "stateless" in check_name.lower():
|
|
return "clickhouse/stateless-test"
|
|
if "stateful" in check_name.lower():
|
|
return "clickhouse/stateful-test"
|
|
raise ValueError(f"Cannot deduce image name based on check name {check_name}")
|
|
|
|
|
|
def get_run_command(
|
|
check_name: str,
|
|
builds_path: Path,
|
|
repo_path: Path,
|
|
result_path: Path,
|
|
server_log_path: Path,
|
|
additional_envs: List[str],
|
|
ci_logs_args: str,
|
|
image: DockerImage,
|
|
flaky_check: bool,
|
|
tests_to_run: List[str],
|
|
) -> str:
|
|
additional_options = ["--hung-check"]
|
|
additional_options.append("--print-time")
|
|
|
|
if tests_to_run:
|
|
additional_options += tests_to_run
|
|
|
|
additional_options_str = (
|
|
'-e ADDITIONAL_OPTIONS="' + " ".join(additional_options) + '"'
|
|
)
|
|
|
|
envs = [
|
|
# a static link, don't use S3_URL or S3_DOWNLOAD
|
|
'-e S3_URL="https://s3.amazonaws.com/clickhouse-datasets"',
|
|
]
|
|
|
|
if flaky_check:
|
|
envs.append("-e NUM_TRIES=50")
|
|
|
|
envs += [f"-e {e}" for e in additional_envs]
|
|
|
|
env_str = " ".join(envs)
|
|
|
|
if "stateful" in check_name.lower():
|
|
run_script = "/repo/tests/docker_scripts/stateful_runner.sh"
|
|
elif "stateless" in check_name.lower():
|
|
run_script = "/repo/tests/docker_scripts/stateless_runner.sh"
|
|
else:
|
|
assert False
|
|
|
|
return (
|
|
f"docker run --rm --name func-tester --volume={builds_path}:/package_folder "
|
|
# For dmesg and sysctl
|
|
"--privileged "
|
|
f"{ci_logs_args} "
|
|
f"--volume={repo_path}:/repo "
|
|
f"--volume={result_path}:/test_output "
|
|
f"--volume={server_log_path}:/var/log/clickhouse-server "
|
|
"--security-opt seccomp=unconfined " # required to issue io_uring sys-calls
|
|
f"--cap-add=SYS_PTRACE {env_str} {additional_options_str} {image} {run_script}"
|
|
)
|
|
|
|
|
|
def _get_statless_tests_to_run(pr_info: PRInfo) -> List[str]:
|
|
result = set()
|
|
|
|
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)
|
|
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'
|
|
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)
|
|
return list(result)
|
|
|
|
|
|
def process_results(
|
|
ret_code: int,
|
|
result_directory: Path,
|
|
server_log_path: Path,
|
|
) -> Tuple[StatusType, str, TestResults, List[Path]]:
|
|
test_results = [] # type: TestResults
|
|
additional_files = []
|
|
# 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()]
|
|
|
|
if server_log_path.exists():
|
|
additional_files = additional_files + [
|
|
p for p in server_log_path.iterdir() if p.is_file()
|
|
]
|
|
|
|
status = []
|
|
status_path = result_directory / "check_status.tsv"
|
|
if status_path.exists():
|
|
logging.info("Found %s", status_path.name)
|
|
with open(status_path, "r", encoding="utf-8") as status_file:
|
|
status = list(csv.reader(status_file, delimiter="\t"))
|
|
|
|
if len(status) != 1 or len(status[0]) != 2:
|
|
logging.info("Files in result folder %s", os.listdir(result_directory))
|
|
return ERROR, "Invalid check_status.tsv", test_results, additional_files
|
|
state, description = status[0][0], status[0][1]
|
|
if ret_code != 0:
|
|
state = ERROR
|
|
description = f"Job failed, exit code: {ret_code}. " + description
|
|
|
|
try:
|
|
results_path = result_directory / "test_results.tsv"
|
|
|
|
if results_path.exists():
|
|
logging.info("Found test_results.tsv")
|
|
else:
|
|
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,
|
|
)
|
|
|
|
return state, description, test_results, additional_files # type: ignore
|
|
|
|
|
|
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",
|
|
)
|
|
parser.add_argument(
|
|
"--report-to-file",
|
|
type=str,
|
|
default="",
|
|
help="Path to write script report to (for --validate-bugfix)",
|
|
)
|
|
return parser.parse_args()
|
|
|
|
|
|
test_process = None # type: Optional[TeePopen]
|
|
timeout_expired = False
|
|
|
|
|
|
def handle_sigterm(signum, _frame):
|
|
# TODO: think on how to process it without globals
|
|
print(f"WARNING: Received signal {signum}")
|
|
global timeout_expired # pylint:disable=global-statement
|
|
timeout_expired = True
|
|
Shell.check("docker exec func-tester pkill -f clickhouse-test", verbose=True)
|
|
|
|
|
|
def main():
|
|
signal.signal(signal.SIGTERM, handle_sigterm)
|
|
logging.basicConfig(level=logging.INFO)
|
|
for handler in logging.root.handlers:
|
|
# pylint: disable=protected-access
|
|
handler.setFormatter(SensitiveFormatter(handler.formatter._fmt)) # type: ignore
|
|
|
|
stopwatch = Stopwatch()
|
|
|
|
temp_path = Path(TEMP_PATH)
|
|
reports_path = Path(REPORT_PATH)
|
|
temp_path.mkdir(parents=True, exist_ok=True)
|
|
reports_path.mkdir(parents=True, exist_ok=True)
|
|
|
|
repo_path = Path(REPO_COPY)
|
|
|
|
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
|
|
|
|
flaky_check = "flaky" in check_name.lower()
|
|
|
|
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)
|
|
|
|
if "RUN_BY_HASH_NUM" in os.environ:
|
|
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
|
|
|
|
image_name = get_image_name(check_name)
|
|
|
|
docker_image = pull_image(get_docker_image(image_name))
|
|
|
|
packages_path = temp_path / "packages"
|
|
packages_path.mkdir(parents=True, exist_ok=True)
|
|
|
|
if validate_bugfix_check:
|
|
download_last_release(packages_path, debug=True)
|
|
else:
|
|
download_all_deb_packages(check_name, reports_path, packages_path)
|
|
|
|
server_log_path = temp_path / "server_log"
|
|
server_log_path.mkdir(parents=True, exist_ok=True)
|
|
|
|
result_path = temp_path / "result_path"
|
|
result_path.mkdir(parents=True, exist_ok=True)
|
|
|
|
run_log_path = result_path / "run.log"
|
|
|
|
additional_envs = get_additional_envs(
|
|
check_name, run_by_hash_num, run_by_hash_total
|
|
)
|
|
if validate_bugfix_check:
|
|
additional_envs.append("GLOBAL_TAGS=no-random-settings")
|
|
additional_envs.append("BUGFIX_VALIDATE_CHECK=1")
|
|
|
|
ci_logs_credentials = CiLogsCredentials(temp_path / "export-logs-config.sh")
|
|
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:
|
|
global test_process # pylint:disable=global-statement
|
|
test_process = process
|
|
retcode = process.wait()
|
|
if retcode == 0:
|
|
logging.info("Run successfully")
|
|
else:
|
|
logging.info("Run failed, exit code %s", retcode)
|
|
|
|
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
|
|
)
|
|
|
|
ci_logs_credentials.clean_ci_logs_from_credentials(run_log_path)
|
|
|
|
state, description, test_results, additional_logs = process_results(
|
|
retcode, result_path, server_log_path
|
|
)
|
|
if timeout_expired:
|
|
description = "Timeout expired"
|
|
state = FAILURE
|
|
test_results.insert(
|
|
0, TestResult.create_check_timeout_expired(stopwatch.duration_seconds)
|
|
)
|
|
|
|
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",
|
|
[],
|
|
[],
|
|
)
|
|
|
|
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)
|
|
|
|
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("Won't block the CI workflow")
|
|
should_block_ci = False
|
|
|
|
if should_block_ci:
|
|
sys.exit(1)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|