# Copyright (c) 2023 Egor Tensin # This file is part of the "cimple" project. # For details, see https://github.com/egor-tensin/cimple. # Distributed under the MIT License. import logging import multiprocessing as mp import re import pytest from lib.logging import child_logging_thread, configure_logging_in_child from lib.process import LoggingEvent from lib.tests import my_parametrize def test_sigsegv(sigsegv): ec, output = sigsegv.try_run() assert ec == -11 assert output == 'Started the test program.\n' class LoggingEventRunComplete(LoggingEvent): def __init__(self, target): self.counter = 0 self.target = target self.re = re.compile(r'run \d+ as finished') super().__init__(timeout=150) def log_line_matches(self, line): return bool(self.re.search(line)) def set(self): self.counter += 1 if self.counter == self.target: super().set() def client_runner_process(log_queue, client, runs_per_process, repo): with configure_logging_in_child(log_queue): logging.info('Executing %s clients', runs_per_process) for i in range(runs_per_process): client.run('run', repo.path, 'HEAD') def _test_repo_internal(env, repo, numof_processes, runs_per_process): numof_runs = numof_processes * runs_per_process event = LoggingEventRunComplete(numof_runs) # Count the number of times the server receives the "run complete" message. env.server.logger.add_event(event) with child_logging_thread() as log_queue: ctx = mp.get_context('spawn') args = (log_queue, env.client, runs_per_process, repo) processes = [ctx.Process(target=client_runner_process, args=args) for i in range(numof_processes)] for proc in processes: proc.start() event.wait() for proc in processes: proc.join() repo.run_files_are_present(numof_runs) runs = env.db.get_all_runs() assert numof_runs == len(runs) for id, status, ec, output, url, rev in runs: assert status == 'finished', f'Invalid status for run {id}: {status}' assert repo.run_exit_code_matches(ec), f"Exit code doesn't match: {ec}" assert repo.run_output_matches(output), f"Output doesn't match: {output}" @my_parametrize('runs_per_client', [1, 5]) @my_parametrize('numof_clients', [1, 5]) def test_repo(env, test_repo, numof_clients, runs_per_client): _test_repo_internal(env, test_repo, numof_clients, runs_per_client) @pytest.mark.stress @my_parametrize('numof_clients,runs_per_client', [ (10, 50), (1, 2000), (4, 500), ]) def test_repo_stress(env, stress_test_repo, numof_clients, runs_per_client): _test_repo_internal(env, stress_test_repo, numof_clients, runs_per_client) @pytest.mark.flame_graph def test_repo_flame_graph(env, profiler, flame_graph_repo): _test_repo_internal(env, flame_graph_repo, 4, 500)