import os import subprocess import time import docker class PartitionManager: """Allows introducing failures in the network between docker containers. Can act as a context manager: with PartitionManager() as pm: pm.partition_instances(instance1, instance2) ... # At exit all partitions are removed automatically. """ def __init__(self): self._iptables_rules = [] self._netem_delayed_instances = [] _NetworkManager.get() def drop_instance_zk_connections(self, instance, action='DROP'): self._check_instance(instance) self._add_rule({'source': instance.ip_address, 'destination_port': 2181, 'action': action}) self._add_rule({'destination': instance.ip_address, 'source_port': 2181, 'action': action}) def restore_instance_zk_connections(self, instance, action='DROP'): self._check_instance(instance) self._delete_rule({'source': instance.ip_address, 'destination_port': 2181, 'action': action}) self._delete_rule({'destination': instance.ip_address, 'source_port': 2181, 'action': action}) def partition_instances(self, left, right, port=None, action='DROP'): self._check_instance(left) self._check_instance(right) def create_rule(src, dst): rule = {'source': src.ip_address, 'destination': dst.ip_address, 'action': action} if port is not None: rule['destination_port'] = port return rule self._add_rule(create_rule(left, right)) self._add_rule(create_rule(right, left)) def add_network_delay(self, instance, delay_ms): self._add_tc_netem_delay(instance, delay_ms) def heal_all(self): while self._iptables_rules: rule = self._iptables_rules.pop() _NetworkManager.get().delete_iptables_rule(**rule) while self._netem_delayed_instances: instance = self._netem_delayed_instances.pop() instance.exec_in_container(["bash", "-c", "tc qdisc del dev eth0 root netem"], user="root") def pop_rules(self): res = self._iptables_rules[:] self.heal_all() return res def push_rules(self, rules): for rule in rules: self._add_rule(rule) @staticmethod def _check_instance(instance): if instance.ip_address is None: raise Exception('Instance + ' + instance.name + ' is not launched!') def _add_rule(self, rule): _NetworkManager.get().add_iptables_rule(**rule) self._iptables_rules.append(rule) def _delete_rule(self, rule): _NetworkManager.get().delete_iptables_rule(**rule) self._iptables_rules.remove(rule) def _add_tc_netem_delay(self, instance, delay_ms): instance.exec_in_container(["bash", "-c", "tc qdisc add dev eth0 root netem delay {}ms".format(delay_ms)], user="root") self._netem_delayed_instances.append(instance) def __enter__(self): return self def __exit__(self, exc_type, exc_val, exc_tb): self.heal_all() def __del__(self): self.heal_all() class PartitionManagerDisabler: def __init__(self, manager): self.manager = manager self.rules = self.manager.pop_rules() def __enter__(self): return self def __exit__(self, exc_type, exc_val, exc_tb): self.manager.push_rules(self.rules) class _NetworkManager: """Execute commands inside a container with access to network settings. We need to call iptables to create partitions, but we want to avoid sudo. The way to circumvent this restriction is to run iptables in a container with network=host. The container is long-running and periodically renewed - this is an optimization to avoid the overhead of container creation on each call. Source of the idea: https://github.com/worstcase/blockade/blob/master/blockade/host.py """ # Singleton instance. _instance = None @classmethod def get(cls, **kwargs): if cls._instance is None: cls._instance = cls(**kwargs) return cls._instance def add_iptables_rule(self, **kwargs): cmd = ['iptables', '-I', 'DOCKER-USER', '1'] cmd.extend(self._iptables_cmd_suffix(**kwargs)) self._exec_run(cmd, privileged=True) def delete_iptables_rule(self, **kwargs): cmd = ['iptables', '-D', 'DOCKER-USER'] cmd.extend(self._iptables_cmd_suffix(**kwargs)) self._exec_run(cmd, privileged=True) @staticmethod def _iptables_cmd_suffix( source=None, destination=None, source_port=None, destination_port=None, action=None, probability=None): ret = [] if probability is not None: ret.extend(['-m', 'statistic', '--mode', 'random', '--probability', str(probability)]) ret.extend(['-p', 'tcp']) if source is not None: ret.extend(['-s', source]) if destination is not None: ret.extend(['-d', destination]) if source_port is not None: ret.extend(['--sport', str(source_port)]) if destination_port is not None: ret.extend(['--dport', str(destination_port)]) if action is not None: ret.extend(['-j'] + action.split()) return ret def __init__( self, container_expire_timeout=50, container_exit_timeout=60): self.container_expire_timeout = container_expire_timeout self.container_exit_timeout = container_exit_timeout self._docker_client = docker.from_env(version=os.environ.get("DOCKER_API_VERSION")) self._container = None self._ensure_container() def _ensure_container(self): if self._container is None or self._container_expire_time <= time.time(): if self._container is not None: try: self._container.remove(force=True) except docker.errors.NotFound: pass # for some reason docker api may hang if image doesn't exist, so we download it # before running for i in range(5): try: subprocess.check_call("docker pull yandex/clickhouse-integration-helper", shell=True) break except: time.sleep(i) else: raise Exception("Cannot pull yandex/clickhouse-integration-helper image") self._container = self._docker_client.containers.run('yandex/clickhouse-integration-helper', auto_remove=True, command=('sleep %s' % self.container_exit_timeout), detach=True, network_mode='host') container_id = self._container.id self._container_expire_time = time.time() + self.container_expire_timeout return self._container def _exec_run(self, cmd, **kwargs): container = self._ensure_container() handle = self._docker_client.api.exec_create(container.id, cmd, **kwargs) output = self._docker_client.api.exec_start(handle).decode('utf8') exit_code = self._docker_client.api.exec_inspect(handle)['ExitCode'] if exit_code != 0: print(output) raise subprocess.CalledProcessError(exit_code, cmd) return output