Beispiel #1
0
def run_job_on_worker(worker_info, call_ids_range, job_payload):
    """
    Install all the Lithops dependencies into the worker.
    Runs the job
    """
    instance_name, ip_address, instance_id = worker_info
    logger.info('Going to setup {}, IP address {}'.format(
        instance_name, ip_address))

    ssh_client = SSHClient(ip_address, STANDALONE_SSH_CREDNTIALS)
    wait_instance_ready(ssh_client)

    # upload zip lithops package
    logger.info('Uploading lithops files to VM instance {}'.format(ip_address))
    ssh_client.upload_local_file('/opt/lithops/lithops_standalone.zip',
                                 '/tmp/lithops_standalone.zip')
    logger.info(
        'Executing lithops installation process on VM instance {}'.format(
            ip_address))

    vm_data = {
        'instance_name': instance_name,
        'ip_address': ip_address,
        'instance_id': instance_id
    }

    script = get_worker_setup_script(STANDALONE_CONFIG, vm_data)
    ssh_client.run_remote_command(script, run_async=True)
    ssh_client.close()

    # Wait until the proxy is ready
    wait_proxy_ready(ip_address)

    dbr = job_payload['data_byte_ranges']
    job_payload['call_ids'] = call_ids_range
    job_payload['data_byte_ranges'] = [
        dbr[int(call_id)] for call_id in call_ids_range
    ]

    url = "http://{}:{}/run".format(ip_address, STANDALONE_SERVICE_PORT)
    r = requests.post(url, data=json.dumps(job_payload))
    response = r.json()

    if 'activationId' in response:
        logger.info('Calls {} invoked. Activation ID: {}'.format(
            ', '.join(call_ids_range), response['activationId']))
    else:
        logger.error('calls {} failed invocation: {}'.format(
            ', '.join(call_ids_range), response['error']))
Beispiel #2
0
class StandaloneHandler:
    """
    A StandaloneHandler object is used by invokers and other components to access
    underlying standalone backend without exposing the implementation details.
    """
    def __init__(self, standalone_config):
        self.config = standalone_config
        self.backend_name = self.config['backend']
        self.runtime = self.config['runtime']
        self.is_lithops_worker = is_lithops_worker()

        self.start_timeout = self.config.get('start_timeout', 300)

        self.auto_dismantle = self.config.get('auto_dismantle')
        self.hard_dismantle_timeout = self.config.get('hard_dismantle_timeout')
        self.soft_dismantle_timeout = self.config.get('soft_dismantle_timeout')

        try:
            module_location = 'lithops.standalone.backends.{}'.format(
                self.backend_name)
            sb_module = importlib.import_module(module_location)
            StandaloneBackend = getattr(sb_module, 'StandaloneBackend')
            self.backend = StandaloneBackend(self.config[self.backend_name])

        except Exception as e:
            logger.error("There was an error trying to create the "
                         "{} standalone backend".format(self.backend_name))
            raise e

        self.log_monitors = {}

        self.ssh_credentials = self.backend.get_ssh_credentials()
        self.ip_address = self.backend.get_ip_address()

        from lithops.util.ssh_client import SSHClient
        self.ssh_client = SSHClient(self.ssh_credentials)

        logger.debug("Standalone handler created successfully")

    def _is_backend_ready(self):
        """
        Checks if the VM instance is ready to receive ssh connections
        """
        try:
            self.ssh_client.run_remote_command(self.ip_address,
                                               'id',
                                               timeout=2)
        except Exception:
            return False
        return True

    def _wait_backend_ready(self):
        """
        Waits until the VM instance is ready to receive ssh connections
        """
        logger.debug('Waiting VM instance to become ready')

        start = time.time()
        while (time.time() - start < self.start_timeout):
            if self._is_backend_ready():
                return True
            time.sleep(1)

        self.dismantle()
        raise Exception('VM readiness probe expired. Check your VM')

    def _start_backend(self):
        if not self._is_backend_ready():
            # The VM instance is stopped
            init_time = time.time()
            self.backend.start()
            self._wait_backend_ready()
            total_start_time = round(time.time() - init_time, 2)
            logger.info(
                'VM instance ready in {} seconds'.format(total_start_time))

    def _is_proxy_ready(self):
        """
        Checks if the proxy is ready to receive http connections
        """
        try:
            if self.is_lithops_worker:
                url = "http://{}:{}/ping".format('127.0.0.1',
                                                 PROXY_SERVICE_PORT)
                r = requests.get(url, timeout=1, verify=True)
                if r.status_code == 200:
                    return True
                return False
            else:
                cmd = 'curl -X GET http://127.0.0.1:8080/ping'
                out = self.ssh_client.run_remote_command(self.ip_address,
                                                         cmd,
                                                         timeout=2)
                data = json.loads(out)
                if data['response'] == 'pong':
                    return True
        except Exception:
            return False

    def _wait_proxy_ready(self):
        """
        Waits until the proxy is ready to receive http connections
        """
        logger.info('Waiting Lithops proxy to become ready')

        start = time.time()
        while (time.time() - start < self.start_timeout):
            if self._is_proxy_ready():
                return True
            time.sleep(1)

        self.dismantle()
        raise Exception('Proxy readiness probe expired. Check your VM')

    def _start_log_monitor(self, executor_id, job_id):
        """
        Starts a process that polls the remote log into a local file
        """

        job_key = create_job_key(executor_id, job_id)

        def log_monitor():
            os.makedirs(LOGS_DIR, exist_ok=True)
            log_file = os.path.join(LOGS_DIR, job_key + '.log')
            fdout_0 = open(log_file, 'wb')
            fdout_1 = open(FN_LOG_FILE, 'ab')

            ssh_client = self.ssh_client.create_client(self.ip_address)
            cmd = 'tail -n +1 -F /tmp/lithops/logs/{}.log'.format(job_key)
            stdin, stdout, stderr = ssh_client.exec_command(cmd)
            channel = stdout.channel
            stdin.close()
            channel.shutdown_write()

            data = None
            while not channel.closed:
                try:
                    readq, _, _ = select.select([channel], [], [], 10)
                    if readq and readq[0].recv_ready():
                        data = channel.recv(len(readq[0].in_buffer))
                        fdout_0.write(data)
                        fdout_0.flush()
                        fdout_1.write(data)
                        fdout_1.flush()
                    else:
                        if data:
                            cmd = 'ls /tmp/lithops/jobs/{}.done'.format(
                                job_key)
                            _, out, _ = ssh_client.exec_command(cmd)
                            if out.read().decode().strip():
                                break
                        time.sleep(0.5)
                except Exception:
                    pass

        if not self.is_lithops_worker:
            Thread(target=log_monitor, daemon=True).start()
            logger.debug('ExecutorID {} | JobID {} - Remote log monitor '
                         'started'.format(executor_id, job_id))

    def run_job(self, job_payload):
        """
        Run the job description against the selected environment
        """
        executor_id = job_payload['executor_id']
        job_id = job_payload['job_id']
        job_key = create_job_key(executor_id, job_id)
        log_file = os.path.join(LOGS_DIR, job_key + '.log')

        if not self._is_proxy_ready():
            # The VM instance is stopped
            init_time = time.time()
            self.backend.start()
            self._wait_proxy_ready()
            total_start_time = round(time.time() - init_time, 2)
            logger.info(
                'VM instance ready in {} seconds'.format(total_start_time))

        self._start_log_monitor(executor_id, job_id)

        logger.info('ExecutorID {} | JobID {} - Running job'.format(
            executor_id, job_id))
        logger.info("View execution logs at {}".format(log_file))

        if self.is_lithops_worker:
            url = "http://{}:{}/run".format('127.0.0.1', PROXY_SERVICE_PORT)
            r = requests.post(url, data=json.dumps(job_payload), verify=True)
            response = r.json()
        else:
            cmd = ('curl -X POST http://127.0.0.1:8080/run -d {} '
                   '-H \'Content-Type: application/json\''.format(
                       shlex.quote(json.dumps(job_payload))))
            out = self.ssh_client.run_remote_command(self.ip_address, cmd)
            response = json.loads(out)

        return response['activationId']

    def create_runtime(self, runtime):
        """
        Installs the proxy and extracts the runtime metadata and
        preinstalled modules
        """
        self._start_backend()
        self._setup_proxy()
        self._wait_proxy_ready()

        logger.debug('Extracting runtime metadata information')
        payload = {'runtime': runtime}

        if self.is_lithops_worker:
            url = "http://{}:{}/preinstalls".format('127.0.0.1',
                                                    PROXY_SERVICE_PORT)
            r = requests.get(url, data=json.dumps(payload), verify=True)
            runtime_meta = r.json()
        else:
            cmd = ('curl http://127.0.0.1:8080/preinstalls -d {} '
                   '-H \'Content-Type: application/json\' -X GET'.format(
                       shlex.quote(json.dumps(payload))))
            out = self.ssh_client.run_remote_command(self.ip_address, cmd)
            runtime_meta = json.loads(out)

        return runtime_meta

    def get_runtime_key(self, runtime_name):
        """
        Wrapper method that returns a formated string that represents the
        runtime key. Each backend has its own runtime key format. Used to
        store modules preinstalls into the storage
        """
        return self.backend.get_runtime_key(runtime_name)

    def dismantle(self):
        """
        Stop VM instance
        """
        self.backend.stop()

    def init(self):
        """
        Start the VM instance and initialize runtime
        """
        self._start_backend()

        # Not sure if mandatory, but sleep several seconds to let proxy server start
        time.sleep(2)

        # if proxy not started, install it
        if not self._is_proxy_ready():
            self._setup_proxy()

        self._wait_proxy_ready()

    def clean(self):
        pass

    def clear(self):
        pass

    def _setup_proxy(self):
        logger.debug('Installing Lithops proxy in the VM instance')
        logger.debug(
            'Be patient, installation process can take up to 3 minutes '
            'if this is the first time you use the VM instance')

        service_file = '/etc/systemd/system/{}'.format(PROXY_SERVICE_NAME)
        self.ssh_client.upload_data_to_file(self.ip_address,
                                            PROXY_SERVICE_FILE, service_file)

        cmd = 'rm -R {}; mkdir -p {}; '.format(REMOTE_INSTALL_DIR,
                                               REMOTE_INSTALL_DIR)
        cmd += 'systemctl daemon-reload; systemctl stop {}; '.format(
            PROXY_SERVICE_NAME)
        self.ssh_client.run_remote_command(self.ip_address, cmd)

        config_file = os.path.join(REMOTE_INSTALL_DIR, 'config')
        self.ssh_client.upload_data_to_file(self.ip_address,
                                            json.dumps(self.config),
                                            config_file)

        src_proxy = os.path.join(os.path.dirname(__file__), 'proxy.py')
        create_handler_zip(FH_ZIP_LOCATION, src_proxy)
        self.ssh_client.upload_local_file(self.ip_address, FH_ZIP_LOCATION,
                                          '/tmp/lithops_standalone.zip')
        os.remove(FH_ZIP_LOCATION)

        # Install dependenices
        cmd = 'mkdir -p /tmp/lithops; '
        cmd += 'apt-get update >> /tmp/lithops/proxy.log; '
        cmd += 'apt-get install unzip python3-pip -y >> /tmp/lithops/proxy.log; '
        cmd += 'pip3 install flask gevent pika==0.13.1 >> /tmp/lithops/proxy.log; '
        cmd += 'unzip -o /tmp/lithops_standalone.zip -d {} > /dev/null 2>&1; '.format(
            REMOTE_INSTALL_DIR)
        cmd += 'rm /tmp/lithops_standalone.zip; '
        cmd += 'chmod 644 {}; '.format(service_file)
        # Start proxy service
        cmd += 'systemctl daemon-reload; '
        cmd += 'systemctl stop {}; '.format(PROXY_SERVICE_NAME)
        cmd += 'systemctl enable {}; '.format(PROXY_SERVICE_NAME)
        cmd += 'systemctl start {}; '.format(PROXY_SERVICE_NAME)
        self.ssh_client.run_remote_command(self.ip_address,
                                           cmd,
                                           background=True)