Source code for airflow.contrib.hooks.spark_submit_hook

# -*- coding: utf-8 -*-
#
# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements.  See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership.  The ASF licenses this file
# to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance
# with the License.  You may obtain a copy of the License at
#
#   http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, either express or implied.  See the License for the
# specific language governing permissions and limitations
# under the License.
#
import os
import subprocess
import re
import time

from airflow.hooks.base_hook import BaseHook
from airflow.exceptions import AirflowException
from airflow.utils.log.logging_mixin import LoggingMixin
from airflow.contrib.kubernetes import kube_client


[docs]class SparkSubmitHook(BaseHook, LoggingMixin): """ This hook is a wrapper around the spark-submit binary to kick off a spark-submit job. It requires that the "spark-submit" binary is in the PATH or the spark_home to be supplied. :param conf: Arbitrary Spark configuration properties :type conf: dict :param conn_id: The connection id as configured in Airflow administration. When an invalid connection_id is supplied, it will default to yarn. :type conn_id: str :param files: Upload additional files to the executor running the job, separated by a comma. Files will be placed in the working directory of each executor. For example, serialized objects. :type files: str :param py_files: Additional python files used by the job, can be .zip, .egg or .py. :type py_files: str :param: archives: Archives that spark should unzip (and possibly tag with #ALIAS) into the application working directory. :param driver_class_path: Additional, driver-specific, classpath settings. :type driver_class_path: str :param jars: Submit additional jars to upload and place them in executor classpath. :type jars: str :param java_class: the main class of the Java application :type java_class: str :param packages: Comma-separated list of maven coordinates of jars to include on the driver and executor classpaths :type packages: str :param exclude_packages: Comma-separated list of maven coordinates of jars to exclude while resolving the dependencies provided in 'packages' :type exclude_packages: str :param repositories: Comma-separated list of additional remote repositories to search for the maven coordinates given with 'packages' :type repositories: str :param total_executor_cores: (Standalone & Mesos only) Total cores for all executors (Default: all the available cores on the worker) :type total_executor_cores: int :param executor_cores: (Standalone, YARN and Kubernetes only) Number of cores per executor (Default: 2) :type executor_cores: int :param executor_memory: Memory per executor (e.g. 1000M, 2G) (Default: 1G) :type executor_memory: str :param driver_memory: Memory allocated to the driver (e.g. 1000M, 2G) (Default: 1G) :type driver_memory: str :param keytab: Full path to the file that contains the keytab :type keytab: str :param principal: The name of the kerberos principal used for keytab :type principal: str :param name: Name of the job (default airflow-spark) :type name: str :param num_executors: Number of executors to launch :type num_executors: int :param application_args: Arguments for the application being submitted :type application_args: list :param env_vars: Environment variables for spark-submit. It supports yarn and k8s mode too. :type env_vars: dict :param verbose: Whether to pass the verbose flag to spark-submit process for debugging :type verbose: bool :param spark_binary: The command to use for spark submit. Some distros may use spark2-submit. :type spark_binary: str """ def __init__(self, conf=None, conn_id='spark_default', files=None, py_files=None, archives=None, driver_class_path=None, jars=None, java_class=None, packages=None, exclude_packages=None, repositories=None, total_executor_cores=None, executor_cores=None, executor_memory=None, driver_memory=None, keytab=None, principal=None, name='default-name', num_executors=None, application_args=None, env_vars=None, verbose=False, spark_binary="spark-submit"): self._conf = conf self._conn_id = conn_id self._files = files self._py_files = py_files self._archives = archives self._driver_class_path = driver_class_path self._jars = jars self._java_class = java_class self._packages = packages self._exclude_packages = exclude_packages self._repositories = repositories self._total_executor_cores = total_executor_cores self._executor_cores = executor_cores self._executor_memory = executor_memory self._driver_memory = driver_memory self._keytab = keytab self._principal = principal self._name = name self._num_executors = num_executors self._application_args = application_args self._env_vars = env_vars self._verbose = verbose self._submit_sp = None self._yarn_application_id = None self._kubernetes_driver_pod = None self._spark_binary = spark_binary self._connection = self._resolve_connection() self._is_yarn = 'yarn' in self._connection['master'] self._is_kubernetes = 'k8s' in self._connection['master'] if self._is_kubernetes and kube_client is None: raise RuntimeError( "{} specified by kubernetes dependencies are not installed!".format( self._connection['master'])) self._should_track_driver_status = self._resolve_should_track_driver_status() self._driver_id = None self._driver_status = None self._spark_exit_code = None
[docs] def _resolve_should_track_driver_status(self): """ Determines whether or not this hook should poll the spark driver status through subsequent spark-submit status requests after the initial spark-submit request :return: if the driver status should be tracked """ return ('spark://' in self._connection['master'] and self._connection['deploy_mode'] == 'cluster')
[docs] def _resolve_connection(self): # Build from connection master or default to yarn if not available conn_data = {'master': 'yarn', 'queue': None, 'deploy_mode': None, 'spark_home': None, 'spark_binary': self._spark_binary, 'namespace': 'default'} try: # Master can be local, yarn, spark://HOST:PORT, mesos://HOST:PORT and # k8s://https://<HOST>:<PORT> conn = self.get_connection(self._conn_id) if conn.port: conn_data['master'] = "{}:{}".format(conn.host, conn.port) else: conn_data['master'] = conn.host # Determine optional yarn queue from the extra field extra = conn.extra_dejson conn_data['queue'] = extra.get('queue', None) conn_data['deploy_mode'] = extra.get('deploy-mode', None) conn_data['spark_home'] = extra.get('spark-home', None) conn_data['spark_binary'] = extra.get('spark-binary', "spark-submit") conn_data['namespace'] = extra.get('namespace', 'default') except AirflowException: self.log.debug( "Could not load connection string %s, defaulting to %s", self._conn_id, conn_data['master'] ) return conn_data
[docs] def get_conn(self): pass
[docs] def _get_spark_binary_path(self): # If the spark_home is passed then build the spark-submit executable path using # the spark_home; otherwise assume that spark-submit is present in the path to # the executing user if self._connection['spark_home']: connection_cmd = [os.path.join(self._connection['spark_home'], 'bin', self._connection['spark_binary'])] else: connection_cmd = [self._connection['spark_binary']] return connection_cmd
[docs] def _build_spark_submit_command(self, application): """ Construct the spark-submit command to execute. :param application: command to append to the spark-submit command :type application: str :return: full command to be executed """ connection_cmd = self._get_spark_binary_path() # The url ot the spark master connection_cmd += ["--master", self._connection['master']] if self._conf: for key in self._conf: connection_cmd += ["--conf", "{}={}".format(key, str(self._conf[key]))] if self._env_vars and (self._is_kubernetes or self._is_yarn): if self._is_yarn: tmpl = "spark.yarn.appMasterEnv.{}={}" else: tmpl = "spark.kubernetes.driverEnv.{}={}" for key in self._env_vars: connection_cmd += [ "--conf", tmpl.format(key, str(self._env_vars[key]))] elif self._env_vars and self._connection['deploy_mode'] != "cluster": self._env = self._env_vars # Do it on Popen of the process elif self._env_vars and self._connection['deploy_mode'] == "cluster": raise AirflowException( "SparkSubmitHook env_vars is not supported in standalone-cluster mode.") if self._is_kubernetes: connection_cmd += ["--conf", "spark.kubernetes.namespace={}".format( self._connection['namespace'])] if self._files: connection_cmd += ["--files", self._files] if self._py_files: connection_cmd += ["--py-files", self._py_files] if self._archives: connection_cmd += ["--archives", self._archives] if self._driver_class_path: connection_cmd += ["--driver-class-path", self._driver_class_path] if self._jars: connection_cmd += ["--jars", self._jars] if self._packages: connection_cmd += ["--packages", self._packages] if self._exclude_packages: connection_cmd += ["--exclude-packages", self._exclude_packages] if self._repositories: connection_cmd += ["--repositories", self._repositories] if self._num_executors: connection_cmd += ["--num-executors", str(self._num_executors)] if self._total_executor_cores: connection_cmd += ["--total-executor-cores", str(self._total_executor_cores)] if self._executor_cores: connection_cmd += ["--executor-cores", str(self._executor_cores)] if self._executor_memory: connection_cmd += ["--executor-memory", self._executor_memory] if self._driver_memory: connection_cmd += ["--driver-memory", self._driver_memory] if self._keytab: connection_cmd += ["--keytab", self._keytab] if self._principal: connection_cmd += ["--principal", self._principal] if self._name: connection_cmd += ["--name", self._name] if self._java_class: connection_cmd += ["--class", self._java_class] if self._verbose: connection_cmd += ["--verbose"] if self._connection['queue']: connection_cmd += ["--queue", self._connection['queue']] if self._connection['deploy_mode']: connection_cmd += ["--deploy-mode", self._connection['deploy_mode']] # The actual script to execute connection_cmd += [application] # Append any application arguments if self._application_args: connection_cmd += self._application_args self.log.info("Spark-Submit cmd: %s", connection_cmd) return connection_cmd
[docs] def _build_track_driver_status_command(self): """ Construct the command to poll the driver status. :return: full command to be executed """ connection_cmd = self._get_spark_binary_path() # The url ot the spark master connection_cmd += ["--master", self._connection['master']] # The driver id so we can poll for its status if self._driver_id: connection_cmd += ["--status", self._driver_id] else: raise AirflowException( "Invalid status: attempted to poll driver " + "status but no driver id is known. Giving up.") self.log.debug("Poll driver status cmd: %s", connection_cmd) return connection_cmd
[docs] def submit(self, application="", **kwargs): """ Remote Popen to execute the spark-submit job :param application: Submitted application, jar or py file :type application: str :param kwargs: extra arguments to Popen (see subprocess.Popen) """ spark_submit_cmd = self._build_spark_submit_command(application) if hasattr(self, '_env'): env = os.environ.copy() env.update(self._env) kwargs["env"] = env self._submit_sp = subprocess.Popen(spark_submit_cmd, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, bufsize=-1, universal_newlines=True, **kwargs) self._process_spark_submit_log(iter(self._submit_sp.stdout.readline, '')) returncode = self._submit_sp.wait() # Check spark-submit return code. In Kubernetes mode, also check the value # of exit code in the log, as it may differ. if returncode or (self._is_kubernetes and self._spark_exit_code != 0): raise AirflowException( "Cannot execute: {}. Error code is: {}.".format( spark_submit_cmd, returncode ) ) self.log.debug("Should track driver: {}".format(self._should_track_driver_status)) # We want the Airflow job to wait until the Spark driver is finished if self._should_track_driver_status: if self._driver_id is None: raise AirflowException( "No driver id is known: something went wrong when executing " + "the spark submit command" ) # We start with the SUBMITTED status as initial status self._driver_status = "SUBMITTED" # Start tracking the driver status (blocking function) self._start_driver_status_tracking() if self._driver_status != "FINISHED": raise AirflowException( "ERROR : Driver {} badly exited with status {}" .format(self._driver_id, self._driver_status)
)
[docs] def _process_spark_submit_log(self, itr): """ Processes the log files and extracts useful information out of it. If the deploy-mode is 'client', log the output of the submit command as those are the output logs of the Spark worker directly. Remark: If the driver needs to be tracked for its status, the log-level of the spark deploy needs to be at least INFO (log4j.logger.org.apache.spark.deploy=INFO) :param itr: An iterator which iterates over the input of the subprocess """ # Consume the iterator for line in itr: line = line.strip() # If we run yarn cluster mode, we want to extract the application id from # the logs so we can kill the application when we stop it unexpectedly if self._is_yarn and self._connection['deploy_mode'] == 'cluster': match = re.search('(application[0-9_]+)', line) if match: self._yarn_application_id = match.groups()[0] self.log.info("Identified spark driver id: %s", self._yarn_application_id) # If we run Kubernetes cluster mode, we want to extract the driver pod id # from the logs so we can kill the application when we stop it unexpectedly elif self._is_kubernetes: match = re.search(r'\s*pod name: ((.+?)-([a-z0-9]+)-driver)', line) if match: self._kubernetes_driver_pod = match.groups()[0] self.log.info("Identified spark driver pod: %s", self._kubernetes_driver_pod) # Store the Spark Exit code match_exit_code = re.search(r'\s*Exit code: (\d+)', line) if match_exit_code: self._spark_exit_code = int(match_exit_code.groups()[0]) # if we run in standalone cluster mode and we want to track the driver status # we need to extract the driver id from the logs. This allows us to poll for # the status using the driver id. Also, we can kill the driver when needed. elif self._should_track_driver_status and not self._driver_id: match_driver_id = re.search(r'(driver-[0-9\-]+)', line) if match_driver_id: self._driver_id = match_driver_id.groups()[0] self.log.info("identified spark driver id: {}" .format(self._driver_id)) else: self.log.info(line) self.log.debug("spark submit log: {}".format(line))
[docs] def _process_spark_status_log(self, itr): """ parses the logs of the spark driver status query process :param itr: An iterator which iterates over the input of the subprocess """ # Consume the iterator for line in itr: line = line.strip() # Check if the log line is about the driver status and extract the status. if "driverState" in line: self._driver_status = line.split(' : ')[1] \ .replace(',', '').replace('\"', '').strip() self.log.debug("spark driver status log: {}".format(line))
[docs] def _start_driver_status_tracking(self): """ Polls the driver based on self._driver_id to get the status. Finish successfully when the status is FINISHED. Finish failed when the status is ERROR/UNKNOWN/KILLED/FAILED. Possible status: SUBMITTED Submitted but not yet scheduled on a worker RUNNING Has been allocated to a worker to run FINISHED Previously ran and exited cleanly RELAUNCHING Exited non-zero or due to worker failure, but has not yet started running again UNKNOWN The status of the driver is temporarily not known due to master failure recovery KILLED A user manually killed this driver FAILED The driver exited non-zero and was not supervised ERROR Unable to run or restart due to an unrecoverable error (e.g. missing jar file) """ # When your Spark Standalone cluster is not performing well # due to misconfiguration or heavy loads. # it is possible that the polling request will timeout. # Therefore we use a simple retry mechanism. missed_job_status_reports = 0 max_missed_job_status_reports = 10 # Keep polling as long as the driver is processing while self._driver_status not in ["FINISHED", "UNKNOWN", "KILLED", "FAILED", "ERROR"]: # Sleep for 1 second as we do not want to spam the cluster time.sleep(1) self.log.debug("polling status of spark driver with id {}" .format(self._driver_id)) poll_drive_status_cmd = self._build_track_driver_status_command() status_process = subprocess.Popen(poll_drive_status_cmd, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, bufsize=-1, universal_newlines=True) self._process_spark_status_log(iter(status_process.stdout.readline, '')) returncode = status_process.wait() if returncode: if missed_job_status_reports < max_missed_job_status_reports: missed_job_status_reports = missed_job_status_reports + 1 else: raise AirflowException( "Failed to poll for the driver status {} times: returncode = {}" .format(max_missed_job_status_reports, returncode)
)
[docs] def _build_spark_driver_kill_command(self): """ Construct the spark-submit command to kill a driver. :return: full command to kill a driver """ # If the spark_home is passed then build the spark-submit executable path using # the spark_home; otherwise assume that spark-submit is present in the path to # the executing user if self._connection['spark_home']: connection_cmd = [os.path.join(self._connection['spark_home'], 'bin', self._connection['spark_binary'])] else: connection_cmd = [self._connection['spark_binary']] # The url ot the spark master connection_cmd += ["--master", self._connection['master']] # The actual kill command connection_cmd += ["--kill", self._driver_id] self.log.debug("Spark-Kill cmd: %s", connection_cmd) return connection_cmd
[docs] def on_kill(self): self.log.debug("Kill Command is being called") if self._should_track_driver_status: if self._driver_id: self.log.info('Killing driver {} on cluster' .format(self._driver_id)) kill_cmd = self._build_spark_driver_kill_command() driver_kill = subprocess.Popen(kill_cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE) self.log.info("Spark driver {} killed with return code: {}" .format(self._driver_id, driver_kill.wait())) if self._submit_sp and self._submit_sp.poll() is None: self.log.info('Sending kill signal to %s', self._connection['spark_binary']) self._submit_sp.kill() if self._yarn_application_id: self.log.info('Killing application {} on YARN' .format(self._yarn_application_id)) kill_cmd = "yarn application -kill {}" \ .format(self._yarn_application_id).split() yarn_kill = subprocess.Popen(kill_cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE) self.log.info("YARN killed with return code: %s", yarn_kill.wait()) if self._kubernetes_driver_pod: self.log.info('Killing pod %s on Kubernetes', self._kubernetes_driver_pod) # Currently only instantiate Kubernetes client for killing a spark pod. try: client = kube_client.get_kube_client() api_response = client.delete_namespaced_pod( self._kubernetes_driver_pod, self._connection['namespace'], body=client.V1DeleteOptions(), pretty=True) self.log.info("Spark on K8s killed with response: %s", api_response) except kube_client.ApiException as e: self.log.info("Exception when attempting to kill Spark on K8s:") self.log.exception(e)