Skip to content

Latest commit

 

History

History
894 lines (679 loc) · 43.4 KB

README.md

File metadata and controls

894 lines (679 loc) · 43.4 KB
Raw

Build Status Coverage

Join the chat at https://gitter.im/spark-jobserver/spark-jobserver

spark-jobserver provides a RESTful interface for submitting and managing Apache Spark jobs, jars, and job contexts. This repo contains the complete Spark job server project, including unit tests and deploy scripts. It was originally started at Ooyala, but this is now the main development repo.

Other useful links: Troubleshooting Tips, Yarn tips, Mesos tips.

Also see Chinese docs / 中文.

Table of Contents generated with DocToc

Users

(Please add yourself to this list!)

Spark Job Server is now included in Datastax Enterprise 4.8!

Features

  • "Spark as a Service": Simple REST interface (including HTTPS) for all aspects of job, context management
  • Support for Spark SQL, Hive, Streaming Contexts/jobs and custom job contexts! See Contexts.
  • Python, Scala, and Java (see TestJob.java) support
  • LDAP Auth support via Apache Shiro integration
  • Separate JVM per SparkContext for isolation (EXPERIMENTAL)
  • Supports sub-second low-latency jobs via long-running job contexts
  • Start and stop job contexts for RDD sharing and low-latency jobs; change resources on restart
  • Kill running jobs via stop context and delete job
  • Separate jar uploading step for faster job startup
  • Asynchronous and synchronous job API. Synchronous API is great for low latency jobs!
  • Works with Standalone Spark as well as Mesos and yarn-client
  • Job and jar info is persisted via a pluggable DAO interface
  • Named Objects (such as RDDs or DataFrames) to cache and retrieve RDDs or DataFrames by name, improving object sharing and reuse among jobs.
  • Supports Scala 2.10 and 2.11

Version Information

Version Spark Version
0.3.1 0.9.1
0.4.0 1.0.2
0.4.1 1.1.0
0.5.0 1.2.0
0.5.1 1.3.0
0.5.2 1.3.1
0.6.0 1.4.1
0.6.1 1.5.2
0.6.2 1.6.1
0.7.0 1.6.2
0.8.0-SNAPSHOT 2.1.0

For release notes, look in the notes/ directory.

If you need non-released jars, please visit Jitpack - they provide non-release jar builds for any Git repo. :)

Getting Started with Spark Job Server

The easiest way to get started is to try the Docker container which prepackages a Spark distribution with the job server and lets you start and deploy it.

Alternatives:

  • Build and run Job Server in local development mode within SBT. NOTE: This does NOT work for YARN, and in fact is only recommended with spark.master set to local[*]. Please deploy if you want to try with YARN or other real cluster.
  • Deploy job server to a cluster. There are two alternatives (see the deployment section):
    • server_deploy.sh deploys job server to a directory on a remote host.
    • server_package.sh deploys job server to a local directory, from which you can deploy the directory, or create a .tar.gz for Mesos or YARN deployment.
  • EC2 Deploy scripts - follow the instructions in EC2 to spin up a Spark cluster with job server and an example application.
  • EMR Deploy instruction - follow the instruction in EMR

NOTE: Spark Job Server can optionally run SparkContexts in their own, forked JVM process when the config option spark.jobserver.context-per-jvm is set to true. This option does not currently work for SBT/local dev mode. See Deployment section for more info.

Development mode

The example walk-through below shows you how to use the job server with an included example job, by running the job server in local development mode in SBT. This is not an example of usage in production.

You need to have SBT installed.

To set the current version, do something like this:

export VER=`sbt version | tail -1 | cut -f2`

From SBT shell, simply type "reStart". This uses a default configuration file. An optional argument is a path to an alternative config file. You can also specify JVM parameters after "---". Including all the options looks like this:

job-server-extras/reStart /path/to/my.conf --- -Xmx8g

Note that reStart (SBT Revolver) forks the job server in a separate process. If you make a code change, simply type reStart again at the SBT shell prompt, it will compile your changes and restart the jobserver. It enables very fast turnaround cycles.

NOTE2: You cannot do sbt reStart from the OS shell. SBT will start job server and immediately kill it.

For example jobs see the job-server-tests/ project / folder.

When you use reStart, the log file goes to job-server/job-server-local.log. There is also an environment variable EXTRA_JAR for adding a jar to the classpath.

WordCountExample walk-through

Package Jar - Send to Server

First, to package the test jar containing the WordCountExample: sbt job-server-tests/package. Then go ahead and start the job server using the instructions above.

Let's upload the jar:

curl --data-binary @job-server-tests/target/scala-2.10/job-server-tests-$VER.jar localhost:8090/jars/test
OK⏎

Ad-hoc Mode - Single, Unrelated Jobs (Transient Context)

The above jar is uploaded as app test. Next, let's start an ad-hoc word count job, meaning that the job server will create its own SparkContext, and return a job ID for subsequent querying:

curl -d "input.string = a b c a b see" "localhost:8090/jobs?appName=test&classPath=spark.jobserver.WordCountExample"
{
  "duration": "Job not done yet",
  "classPath": "spark.jobserver.WordCountExample",
  "startTime": "2016-06-19T16:27:12.196+05:30",
  "context": "b7ea0eb5-spark.jobserver.WordCountExample",
  "status": "STARTED",
  "jobId": "5453779a-f004-45fc-a11d-a39dae0f9bf4"
}⏎

NOTE: If you want to feed in a text file config and POST using curl, you want the --data-binary option, otherwise curl will munge your line separator chars. Like:

curl --data-binary @my-job-config.json "localhost:8090/jobs?appNam=..."

NOTE2: If you want to send in UTF-8 chars, make sure you pass in a proper header to CURL for the encoding, otherwise it may assume an encoding which is not what you expect.

From this point, you could asynchronously query the status and results:

curl localhost:8090/jobs/5453779a-f004-45fc-a11d-a39dae0f9bf4
{
  "duration": "6.341 secs",
  "classPath": "spark.jobserver.WordCountExample",
  "startTime": "2015-10-16T03:17:03.127Z",
  "context": "b7ea0eb5-spark.jobserver.WordCountExample",
  "result": {
    "a": 2,
    "b": 2,
    "c": 1,
    "see": 1
  },
  "status": "FINISHED",
  "jobId": "5453779a-f004-45fc-a11d-a39dae0f9bf4"
}⏎

Note that you could append &sync=true when you POST to /jobs to get the results back in one request, but for real clusters and most jobs this may be too slow.

You can also append &timeout=XX to extend the request timeout for sync=true requests.

Persistent Context Mode - Faster & Required for Related Jobs

Another way of running this job is in a pre-created context. Start a new context:

curl -d "" "localhost:8090/contexts/test-context?num-cpu-cores=4&memory-per-node=512m"
OK⏎

You can verify that the context has been created:

curl localhost:8090/contexts
["test-context"]⏎

Now let's run the job in the context and get the results back right away:

curl -d "input.string = a b c a b see" "localhost:8090/jobs?appName=test&classPath=spark.jobserver.WordCountExample&context=test-context&sync=true"
{
  "result": {
    "a": 2,
    "b": 2,
    "c": 1,
    "see": 1
  }
}⏎

Note the addition of context= and sync=true.

Debug mode

Spark job server is started using SBT Revolver (which forks a new JVM), so debugging directly in an IDE is not feasible. To enable debugging, the Spark job server should be started from the SBT shell with the following Java options :

job-server-extras/reStart /absolute/path/to/your/dev.conf --- -Xdebug -Xrunjdwp:transport=dt_socket,address=15000,server=y,suspend=y

The above command starts a remote debugging server on port 15000. The Spark job server is not started until a debugging client (Intellij, Eclipse, telnet, ...) connects to the exposed port.

In your IDE you just have to start a Remote debugging debug job and use the above defined port. Once the client connects to the debugging server the Spark job server is started and you can start adding breakpoints and debugging requests.

Note that you might need to adjust some server parameters to avoid short Spary/Akka/Spark timeouts, in your dev.conf add the following values :

spark {
  jobserver {
    # Dev debug timeouts
    context-creation-timeout = 1000000 s
    yarn-context-creation-timeout = 1000000 s
    default-sync-timeout = 1000000 s
  }

  context-settings {
    # Dev debug timeout
    context-init-timeout = 1000000 s
  }
}
spray.can.server {
      # Debug timeouts
      idle-timeout = infinite
      request-timeout = infinite
}

Additionally, you might have to increase the Akka Timeouts by adding the following query parameter timeout=1000000 in your HTTP requests :

curl -d "input.string = a b c a b see" "localhost:8090/jobs?appName=test&classPath=spark.jobserver.WordCountExample&sync=true&timeout=100000"

Create a Job Server Project

Creating a project from scratch using giter8 template

There is a giter8 template available at https://github.com/spark-jobserver/spark-jobserver.g8

$ sbt new spark-jobserver/spark-jobserver.g8

Answer the questions to generate a project structure for you. This contains Word Count example spark job using both old API and new one.

$ cd /path/to/project/directory
$ sbt package

Now you could remove example application and start adding your one.

Creating a project manually assuming that you already have sbt project structure

In your build.sbt, add this to use the job server jar:

    resolvers += "Job Server Bintray" at "https://dl.bintray.com/spark-jobserver/maven"

    libraryDependencies += "spark.jobserver" %% "job-server-api" % "0.7.0" % "provided"

If a SQL or Hive job/context is desired, you also want to pull in job-server-extras:

libraryDependencies += "spark.jobserver" %% "job-server-extras" % "0.7.0" % "provided"

For most use cases it's better to have the dependencies be "provided" because you don't want SBT assembly to include the whole job server jar.

To create a job that can be submitted through the job server, the job must implement the SparkJob trait. Your job will look like:

object SampleJob extends SparkJob {
    override def runJob(sc: SparkContext, jobConfig: Config): Any = ???
    override def validate(sc: SparkContext, config: Config): SparkJobValidation = ???
}
  • runJob contains the implementation of the Job. The SparkContext is managed by the JobServer and will be provided to the job through this method. This relieves the developer from the boiler-plate configuration management that comes with the creation of a Spark job and allows the Job Server to manage and re-use contexts.
  • validate allows for an initial validation of the context and any provided configuration. If the context and configuration are OK to run the job, returning spark.jobserver.SparkJobValid will let the job execute, otherwise returning spark.jobserver.SparkJobInvalid(reason) prevents the job from running and provides means to convey the reason of failure. In this case, the call immediately returns an HTTP/1.1 400 Bad Request status code. validate helps you preventing running jobs that will eventually fail due to missing or wrong configuration and save both time and resources.

NEW SparkJob API

Note: As of version 0.7.0, a new SparkJob API that is significantly better than the old SparkJob API will take over. Existing jobs should continue to compile against the old spark.jobserver.SparkJob API, but this will be deprecated in the future. Note that jobs before 0.7.0 will need to be recompiled, older jobs may not work with the current SJS example. The new API looks like this:

object WordCountExampleNewApi extends NewSparkJob {
  type JobData = Seq[String]
  type JobOutput = collection.Map[String, Long]

  def runJob(sc: SparkContext, runtime: JobEnvironment, data: JobData): JobOutput =
    sc.parallelize(data).countByValue

  def validate(sc: SparkContext, runtime: JobEnvironment, config: Config):
    JobData Or Every[ValidationProblem] = {
    Try(config.getString("input.string").split(" ").toSeq)
      .map(words => Good(words))
      .getOrElse(Bad(One(SingleProblem("No input.string param"))))
  }
}

It is much more type safe, separates context configuration, job ID, named objects, and other environment variables into a separate JobEnvironment input, and allows the validation method to return specific data for the runJob method. See the WordCountExample and LongPiJob for examples.

Let's try running our sample job with an invalid configuration:

curl -i -d "bad.input=abc" "localhost:8090/jobs?appName=test&classPath=spark.jobserver.WordCountExample"

HTTP/1.1 400 Bad Request
Server: spray-can/1.2.0
Date: Tue, 10 Jun 2014 22:07:18 GMT
Content-Type: application/json; charset=UTF-8
Content-Length: 929

{
  "status": "VALIDATION FAILED",
  "result": {
    "message": "No input.string config param",
    "errorClass": "java.lang.Throwable",
    "stack": ["spark.jobserver.JobManagerActor$$anonfun$spark$jobserver$JobManagerActor$$getJobFuture$4.apply(JobManagerActor.scala:212)",
    "scala.concurrent.impl.Future$PromiseCompletingRunnable.liftedTree1$1(Future.scala:24)",
    "scala.concurrent.impl.Future$PromiseCompletingRunnable.run(Future.scala:24)",
    "akka.dispatch.TaskInvocation.run(AbstractDispatcher.scala:42)",
    "akka.dispatch.ForkJoinExecutorConfigurator$AkkaForkJoinTask.exec(AbstractDispatcher.scala:386)",
    "scala.concurrent.forkjoin.ForkJoinTask.doExec(ForkJoinTask.java:260)",
    "scala.concurrent.forkjoin.ForkJoinPool$WorkQueue.runTask(ForkJoinPool.java:1339)",
    "scala.concurrent.forkjoin.ForkJoinPool.runWorker(ForkJoinPool.java:1979)",
    "scala.concurrent.forkjoin.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:107)"]
  }
}

NEW SparkJob API with Spark v2.1

Deploying Spark JobServer with Spark v2.x cluster, you can create a SparkSession context which enables Spark-SQL and Hive support

curl -i -d "" 'http://localhost:8090/contexts/sql-context-1?num-cpu-cores=2&memory-per-node=512M&context-factory=spark.jobserver.context.SessionContextFactory'

Spark JobServer application shall extend from the SparkSessionJob to use the spark.jobserver.context.SessionContextFactory, here is an example

import com.typesafe.config.Config
import org.apache.spark.sql.SparkSession
import org.scalactic._
import spark.jobserver.SparkSessionJob
import spark.jobserver.api.{JobEnvironment, SingleProblem, ValidationProblem}

import scala.util.Try

object WordCountExampleSparkSession extends SparkSessionJob {
  type JobData = Seq[String]
  type JobOutput = collection.Map[String, Long]

  override def runJob(sparkSession: SparkSession, runtime: JobEnvironment, data: JobData): JobOutput =
    sparkSession.sparkContext.parallelize(data).countByValue

  override def validate(sparkSession: SparkSession, runtime: JobEnvironment, config: Config):
  JobData Or Every[ValidationProblem] = {
    Try(config.getString("input.string").split(" ").toSeq)
      .map(words => Good(words))
      .getOrElse(Bad(One(SingleProblem("No input.string param"))))
  }
}

Dependency jars

You have a couple options to package and upload dependency jars.

  • The easiest is to use something like sbt-assembly to produce a fat jar. Be sure to mark the Spark and job-server dependencies as "provided" so it won't blow up the jar size. This works well if the number of dependencies is not large.
  • When the dependencies are sizeable and/or you don't want to load them with every different job, you can package the dependencies separately and use one of several options:
    • Use the dependent-jar-uris context configuration param. Then the jar gets loaded for every job.
    • The dependent-jar-uris can also be used in job configuration param when submitting a job. On an ad-hoc context this has the same effect as dependent-jar-uris context configuration param. On a persistent context the jars will be loaded for the current job and then for every job that will be executed on the persistent context. 
      curl -d "" "localhost:8090/contexts/test-context?num-cpu-cores=4&memory-per-node=512m"
OK⏎

      curl "localhost:8090/jobs?appName=test&classPath=spark.jobserver.WordCountExample&context=test-context&sync=true" -d '{
    dependent-jar-uris = ["file:///myjars/deps01.jar", "file:///myjars/deps02.jar"],
    input.string = "a b c a b see"
}'
      The jars /myjars/deps01.jar & /myjars/deps02.jar (present only on the SJS node) will be loaded and made available for the Spark driver & executors.
    • Use the --package option with Maven coordinates with server_start.sh.
    • Put the extra jars in the SPARK_CLASSPATH

Named Objects

Using Named RDDs

Initially, the job server only supported Named RDDs. For backwards compatibility and convenience, the following is still supported even though it is now possible to use the more generic Named Object support described in the next section.

Named RDDs are a way to easily share RDDs among jobs. Using this facility, computed RDDs can be cached with a given name and later on retrieved. To use this feature, the SparkJob needs to mixin NamedRddSupport:

object SampleNamedRDDJob  extends SparkJob with NamedRddSupport {
    override def runJob(sc:SparkContext, jobConfig: Config): Any = ???
    override def validate(sc:SparkContext, config: Config): SparkJobValidation = ???
}

Then in the implementation of the job, RDDs can be stored with a given name:

this.namedRdds.update("french_dictionary", frenchDictionaryRDD)

Other job running in the same context can retrieve and use this RDD later on:

val rdd = this.namedRdds.get[(String, String)]("french_dictionary").get

(note the explicit type provided to get. This will allow to cast the retrieved RDD that otherwise is of type RDD[_])

For jobs that depends on a named RDDs it's a good practice to check for the existence of the NamedRDD in the validate method as explained earlier:

def validate(sc:SparkContext, config: Config): SparkJobValidation = {
  ...
  val rdd = this.namedRdds.get[(Long, scala.Seq[String])]("dictionary")
  if (rdd.isDefined) SparkJobValid else SparkJobInvalid(s"Missing named RDD [dictionary]")
}

Using Named Objects

Named Objects are a way to easily share RDDs, DataFrames or other objects among jobs. Using this facility, computed objects can be cached with a given name and later on retrieved. To use this feature, the SparkJob needs to mixin NamedObjectSupport. It is also necessary to define implicit persisters for each desired type of named objects. For convencience, we have provided implementations for RDD persistence and for DataFrame persistence (defined in job-server-extras):

object SampleNamedObjectJob  extends SparkJob with NamedObjectSupport {

  implicit def rddPersister[T] : NamedObjectPersister[NamedRDD[T]] = new RDDPersister[T]
  implicit val dataFramePersister = new DataFramePersister

    override def runJob(sc:SparkContext, jobConfig: Config): Any = ???
    override def validate(sc:SparkContext, config: Config): SparkJobValidation = ???
}

Then in the implementation of the job, RDDs can be stored with a given name:

this.namedObjects.update("rdd:french_dictionary", NamedRDD(frenchDictionaryRDD, forceComputation = false, storageLevel = StorageLevel.NONE))

DataFrames can be stored like so:

this.namedObjects.update("df:some df", NamedDataFrame(frenchDictionaryDF, forceComputation = false, storageLevel = StorageLevel.NONE))

It is advisable to use different name prefixes for different types of objects to avoid confusion.

Another job running in the same context can retrieve and use these objects later on:

val NamedRDD(frenchDictionaryRDD, _ ,_) = namedObjects.get[NamedRDD[(String, String)]]("rdd:french_dictionary").get

val NamedDataFrame(frenchDictionaryDF, _, _) = namedObjects.get[NamedDataFrame]("df:some df").get

(Note the explicit type provided to get. This will allow to cast the retrieved RDD/DataFrame object to the proper result type.)

For jobs that depends on a named objects it's a good practice to check for the existence of the NamedObject in the validate method as explained earlier:

def validate(sc:SparkContext, config: Config): SparkJobValidation = {
  ...
  val obj = this.namedObjects.get("dictionary")
  if (obj.isDefined) SparkJobValid else SparkJobInvalid(s"Missing named object [dictionary]")
}

HTTPS / SSL Configuration

To activate ssl communication, set these flags in your application.conf file (Section 'spray.can.server'):

  ssl-encryption = on
  # absolute path to keystore file
  keystore = "/some/path/sjs.jks"
  keystorePW = "changeit"

You will need a keystore that contains the server certificate. The bare minimum is achieved with this command which creates a self-signed certificate:

 keytool -genkey -keyalg RSA -alias jobserver -keystore ~/sjs.jks -storepass changeit -validity 360 -keysize 2048

You may place the keystore anywhere. Here is an example of a simple curl command that utilizes ssl:

curl -k https://localhost:8090/contexts

The -k flag tells curl to "Allow connections to SSL sites without certs". Export your server certificate and import it into the client's truststore to fully utilize ssl security.

Authentication

Authentication uses the Apache Shiro framework. Authentication is activated by setting this flag (Section 'shiro'):

authentication = on
# absolute path to shiro config file, including file name
config.path = "/some/path/shiro.ini"

Shiro-specific configuration options should be placed into a file named 'shiro.ini' in the directory as specified by the config option 'config.path'. Here is an example that configures LDAP with user group verification:

# use this for basic ldap authorization, without group checking
# activeDirectoryRealm = org.apache.shiro.realm.ldap.JndiLdapRealm
# use this for checking group membership of users based on the 'member' attribute of the groups:
activeDirectoryRealm = spark.jobserver.auth.LdapGroupRealm
# search base for ldap groups (only relevant for LdapGroupRealm):
activeDirectoryRealm.contextFactory.environment[ldap.searchBase] = dc=xxx,dc=org
# allowed groups (only relevant for LdapGroupRealm):
activeDirectoryRealm.contextFactory.environment[ldap.allowedGroups] = "cn=group1,ou=groups", "cn=group2,ou=groups"
activeDirectoryRealm.contextFactory.environment[java.naming.security.credentials] = password
activeDirectoryRealm.contextFactory.url = ldap://localhost:389
activeDirectoryRealm.userDnTemplate = cn={0},ou=people,dc=xxx,dc=org

cacheManager = org.apache.shiro.cache.MemoryConstrainedCacheManager

securityManager.cacheManager = $cacheManager

Make sure to edit the url, credentials, userDnTemplate, ldap.allowedGroups and ldap.searchBase settings in accordance with your local setup.

Here is an example of a simple curl command that authenticates a user and uses ssl (you may want to use -H to hide the credentials, this is just a simple example to get you started):

curl -k --basic --user 'user:pw' https://localhost:8090/contexts

Deployment

Manual steps

  1. Copy config/local.sh.template to <environment>.sh and edit as appropriate. NOTE: be sure to set SPARK_VERSION if you need to compile against a different version.
  2. Copy config/shiro.ini.template to shiro.ini and edit as appropriate. NOTE: only required when authentication = on
  3. Copy config/local.conf.template to <environment>.conf and edit as appropriate.
  4. bin/server_deploy.sh <environment> -- this packages the job server along with config files and pushes it to the remotes you have configured in <environment>.sh
  5. On the remote server, start it in the deployed directory with server_start.sh and stop it with server_stop.sh

The server_start.sh script uses spark-submit under the hood and may be passed any of the standard extra arguments from spark-submit.

NOTE: by default the assembly jar from job-server-extras, which includes support for SQLContext and HiveContext, is used. If you face issues with all the extra dependencies, consider modifying the install scripts to invoke sbt job-server/assembly instead, which doesn't include the extra dependencies.

Context per JVM

Each context can be a separate process launched using spark-submit, via the included manager_start.sh script, if context-per-jvm is set to true. You may want to set deploy.manager-start-cmd to the correct path to your start script and customize the script. This can be especially desirable when you want to run many contexts at once, or for certain types of contexts such as StreamingContexts which really need their own processes.

Also, the extra processes talk to the master HTTP process via random ports using the Akka Cluster gossip protocol. If for some reason the separate processes causes issues, set spark.jobserver.context-per-jvm to false, which will cause the job server to use a single JVM for all contexts.

Among the known issues:

  • Launched contexts do not shut down by themselves. You need to manually kill each separate process, or do -X DELETE /contexts/<context-name>

Log files are separated out for each context (assuming context-per-jvm is true) in their own subdirs under the LOG_DIR configured in settings.sh in the deployed directory.

Note: to test out the deploy to a local staging dir, or package the job server for Mesos, use bin/server_package.sh <environment>.

Configuring Spark Jobserver meta data Database backend

By default, H2 database is used for storing Spark Jobserver related meta data. But this can be overridden. For example, to use PostgreSQL as backend add the following configuration to local.conf. Ensure that you have spark_jobserver database created with necessary rights granted to user.

sqldao {
  # Slick database driver, full classpath
  slick-driver = slick.driver.PostgresDriver

  # JDBC driver, full classpath
  jdbc-driver = org.postgresql.Driver

  # Directory where default H2 driver stores its data. Only needed for H2.
  rootdir = "/var/spark-jobserver/sqldao/data"

  jdbc {
    url = "jdbc:postgresql://db_host/spark_jobserver"
    user = "secret"
    password = "secret"
  }

  dbcp {
    maxactive = 20
    maxidle = 10
    initialsize = 10
  }
}

If you are using context-per-jvm = true, be sure to add AUTO_MIXED_MODE to your H2 JDBC URL; this allows multiple processes to share the same H2 database using a lock file.

Also add the following line at the root level.

flyway.locations="db/postgresql/migration"

It is also important that any dependent jars are to be added to Job Server class path.

In a yarn-client mode if using H2 the below is advised.

jdbc {
        url = "jdbc:h2:tcp://localhost/db_host/spark_jobserver"
        user = "secret"
        password = "secret"
      }

Chef

There is also a Chef cookbook which can be used to deploy Spark Jobserver.

Architecture

The job server is intended to be run as one or more independent processes, separate from the Spark cluster (though it very well may be collocated with say the Master).

At first glance, it seems many of these functions (eg job management) could be integrated into the Spark standalone master. While this is true, we believe there are many significant reasons to keep it separate:

  • We want the job server to work for Mesos and YARN as well
  • Spark and Mesos masters are organized around "applications" or contexts, but the job server supports running many discrete "jobs" inside a single context
  • We want it to support Shark functionality in the future
  • Loose coupling allows for flexible HA arrangements (multiple job servers targeting same standalone master, or possibly multiple Spark clusters per job server)

Flow diagrams are checked in in the doc/ subdirectory. .diagram files are for websequencediagrams.com... check them out, they really will help you understand the flow of messages between actors.

API

Binaries

GET /binaries               - lists all current binaries
POST /binaries/<appName>    - upload a new binary file
DELETE /binaries/<appName>  - delete defined binary

When POSTing new binaries, the content-type header must be set to one of the types supported by the subclasses of the BinaryType trait. e.g. "application/java-archive" or application/python-archive"

Jars (deprecated)

GET /jars                   - lists all the jars and the last upload timestamp
POST /jars/<appName>        - uploads a new jar under <appName>

These routes are kept for legacy purposes but are deprecated in favour of the /binaries routes

Contexts

GET /contexts               - lists all current contexts
POST /contexts/<name>       - creates a new context
DELETE /contexts/<name>     - stops a context and all jobs running in it
PUT /contexts?reset=reboot  - kills all contexts and re-loads only the contexts from config

Spark context configuration params can follow POST /contexts/<name> as query params. See section below for more details.

Jobs

Jobs submitted to the job server must implement a SparkJob trait. It has a main runJob method which is passed a SparkContext and a typesafe Config object. Results returned by the method are made available through the REST API.

GET /jobs                - Lists the last N jobs
POST /jobs               - Starts a new job, use ?sync=true to wait for results
GET /jobs/<jobId>        - Gets the result or status of a specific job
DELETE /jobs/<jobId>     - Kills the specified job
GET /jobs/<jobId>/config - Gets the job configuration

For details on the Typesafe config format used for input (JSON also works), see the Typesafe Config docs.

Data

It is sometime necessary to programmatically upload files to the server. Use these paths to manage such files:

GET /data                - Lists previously uploaded files that were not yet deleted
POST /data/<prefix>      - Uploads a new file, the full path of the file on the server is returned, the
                           prefix is the prefix of the actual filename used on the server (a timestamp is
                           added to ensure uniqueness)
DELETE /data/<filename>  - Deletes the specified file (only if under control of the JobServer)

These files are uploaded to the server and are stored in a local temporary directory where the JobServer runs. The POST command returns the full pathname and filename of the uploaded file so that later jobs can work with this just the same as with any other server-local file. A job could therefore add this file to HDFS or distribute it to worker nodes via the SparkContext.addFile command. For files that are larger than a few hundred MB, it is recommended to manually upload these files to the server or to directly add them to your HDFS.

Data API Example

$ curl -d "Test data file api" http://localhost:8090/data/test_data_file_upload.txt
{
  "result": {
    "filename": "/tmp/spark-jobserver/upload/test_data_file_upload.txt-2016-07-04T09_09_57.928+05_30.dat"
  }
}

$ curl http://localhost:8090/data
["/tmp/spark-jobserver/upload/test_data_file_upload.txt-2016-07-04T09_09_57.928+05_30.dat"]

$ curl -X DELETE http://localhost:8090/data/%2Ftmp%2Fspark-jobserver%2Fupload%2Ftest_data_file_upload.txt-2016-07-04T09_09_57.928%2B05_30.dat
OK

$ curl http://localhost:8090/data
[]

Note: Both POST and DELETE requests takes URI encoded file names.

Context configuration

A number of context-specific settings can be controlled when creating a context (POST /contexts) or running an ad-hoc job (which creates a context on the spot). For example, add urls of dependent jars for a context.

POST '/contexts/my-new-context?dependent-jar-uris=file:///some/path/of/my-foo-lib.jar'

NOTE: Only the latest dependent-jar-uris (btw it’s jar-uris, not jars-uri) takes effect. You can specify multiple URIs by comma-separating them. So like this:

&dependent-jar-uris=file:///path/a.jar,file:///path/b.jar

When creating a context via POST /contexts, the query params are used to override the default configuration in spark.context-settings. For example,

POST /contexts/my-new-context?num-cpu-cores=10

would override the default spark.context-settings.num-cpu-cores setting.

When starting a job, and the context= query param is not specified, then an ad-hoc context is created. Any settings specified in spark.context-settings will override the defaults in the job server config when it is started up.

Any spark configuration param can be overridden either in POST /contexts query params, or through spark .context-settings job configuration. In addition, num-cpu-cores maps to spark.cores.max, and mem-per- node maps to spark.executor.memory. Therefore the following are all equivalent:

POST /contexts/my-new-context?num-cpu-cores=10

POST /contexts/my-new-context?spark.cores.max=10

or in the job config when using POST /jobs,

spark.context-settings {
    spark.cores.max = 10
}

User impersonation for an already Kerberos authenticated user is supported via spark.proxy.user query param:

POST /contexts/my-new-context?spark.proxy.user=

However, whenever the flag shiro.use-as-proxy-user is set to on (and authentication is on) then this parameter is ignored and the name of the authenticated user is always used as the value of the spark.proxy.user parameter when creating contexts.

To pass settings directly to the sparkConf that do not use the "spark." prefix "as-is", use the "passthrough" section.

spark.context-settings {
    spark.cores.max = 10
    passthrough {
      some.custom.hadoop.config = "192.168.1.1"
    }
}

To add to the underlying Hadoop configuration in a Spark context, add the hadoop section to the context settings

spark.context-settings {
    hadoop {
        mapreduce.framework.name = "Foo"
    }
}

For the exact context configuration parameters, see JobManagerActor docs as well as application.conf.

Also see the yarn doc for more tips.

Other configuration settings

For all of the Spark Job Server configuration settings, see job-server/src/main/resources/application.conf.

Job Result Serialization

The result returned by the SparkJob runJob method is serialized by the job server into JSON for routes that return the result (GET /jobs with sync=true, GET /jobs/). Currently the following types can be serialized properly:

  • String, Int, Long, Double, Float, Boolean
  • Scala Map's with string key values (non-string keys may be converted to strings)
  • Scala Seq's
  • Array's
  • Anything that implements Product (Option, case classes) -- they will be serialized as lists
  • Subclasses of java.util.List
  • Subclasses of java.util.Map with string key values (non-string keys may be converted to strings)
  • Maps, Seqs, Java Maps and Java Lists may contain nested values of any of the above
  • If a job result is of scala's Stream[Byte] type it will be serialised directly as a chunk encoded stream. This is useful if your job result payload is large and may cause a timeout serialising as objects. Beware, this will not currently work as desired with context-per-jvm=true configuration, since it would require serialising Stream[_] blob between processes. For now use Stream[_] job results in context-per-jvm=false configuration, pending potential future enhancements to support this in context-per-jvm=true mode.

If we encounter a data type that is not supported, then the entire result will be serialized to a string.

HTTP Override

Spark Job Server offers HTTP override functionality. Often reverse proxies and firewall implement access limitations to, for example, DELETE and PUT requests. HTTP override allows overcoming these limitations by wrapping, for example, a DELETE request into a POST request.

Requesting the destruction of a context can be accomplished through HTTP override using the following syntax:

$ curl -X POST "localhost:8090/contexts/test_context?_method=DELETE"

Here, a DELETE request is passed to Spark Job Server "through" a POST request.

Clients

Spark Jobserver project has a python binding package. This can be used to quickly develop python applications that can interact with Spark Jobserver programmatically.

Contribution and Development

Contributions via Github Pull Request are welcome. Please start by taking a look at the contribution guidelines and check the TODO for some contribution ideas.

  • If you need to build with a specific scala version use ++x.xx.x followed by the regular command, for instance: sbt ++2.11.6 job-server/compile
  • From the "master" project, please run "test" to ensure nothing is broken.
    • You may need to set SPARK_LOCAL_IP to localhost to ensure Akka port can bind successfully
    • Note for Windows users: very few tests fail on Windows. Thus, run testOnly -- -l WindowsIgnore from SBT shell to ignore them.
  • Logging for tests goes to "job-server-test.log"
  • Run scoverage:test to check the code coverage and improve it.
    • Windows users: run ; coverage ; testOnly -- -l WindowsIgnore ; coverageReport from SBT shell.
  • Please run scalastyle to ensure your code changes don't break the style guide.
  • Do "reStart" from SBT for quick restarts of the job server process
  • Please update the g8 template if you change the SparkJob API

Profiling software generously provided by

YourKit supports open source projects with its full-featured Java Profiler.

Publishing packages

In the root project, do release cross.

To announce the release on ls.implicit.ly, use Herald after adding release notes in the notes/ dir. Also regenerate the catalog with lsWriteVersion SBT task and lsync, in project job-server.

Contact

For user/dev questions, we are using google group for discussions: https://groups.google.com/forum/#!forum/spark-jobserver

Please report bugs/problems to: https://github.com/spark-jobserver/spark-jobserver/issues

License

Apache 2.0, see LICENSE.md

TODO

  • More debugging for classpath issues

  • Add Swagger support. See the spray-swagger project.

  • Implement an interactive SQL window. See: spark-admin

  • Stream the current job progress via a Listener

  • Add routes to return stage info for a job. Persist it via DAO so that we can always retrieve stage / performance info even for historical jobs. This would be pretty kickass.