Github user vanzin commented on a diff in the pull request: https://github.com/apache/spark/pull/18630#discussion_r128648079 --- Diff: core/src/main/scala/org/apache/spark/deploy/worker/DriverWrapper.scala --- @@ -66,4 +77,68 @@ object DriverWrapper { System.exit(-1) } } + + // R or Python are not supported in cluster mode so download the jars to the driver side + private def setupDependencies(loader: MutableURLClassLoader, userJar: String): Unit = { + val packagesExclusions = sys.props.get("spark.jars.excludes").orNull + val packages = sys.props.get("spark.jars.packages").orNull + val repositories = sys.props.get("spark.jars.repositories").orNull + val hadoopConf = new HadoopConfiguration() + val childClasspath = new ArrayBuffer[String]() + val ivyRepoPath = sys.props.get("spark.jars.ivy").orNull + var jars = sys.props.get("spark.jars").orNull + + val exclusions: Seq[String] = + if (!StringUtils.isBlank(packagesExclusions)) { + packagesExclusions.split(",") + } else { + Nil + } + + // Create the IvySettings, either load from file or build defaults + val ivySettings = sys.props.get("spark.jars.ivySettings").map { ivySettingsFile => + SparkSubmitUtils.loadIvySettings(ivySettingsFile, Option(repositories), + Option(ivyRepoPath)) + }.getOrElse { + SparkSubmitUtils.buildIvySettings(Option(repositories), Option(ivyRepoPath)) + } + + val resolvedMavenCoordinates = SparkSubmitUtils.resolveMavenCoordinates(packages, + ivySettings, exclusions = exclusions) + + if (!StringUtils.isBlank(resolvedMavenCoordinates)) { + jars = SparkSubmit.mergeFileLists(jars, resolvedMavenCoordinates) + } + + val targetDir = Files.createTempDirectory("tmp").toFile + // scalastyle:off runtimeaddshutdownhook + Runtime.getRuntime.addShutdownHook(new Thread() { + override def run(): Unit = { + FileUtils.deleteQuietly(targetDir) + } + }) + // scalastyle:on runtimeaddshutdownhook + + val sparkProperties = new mutable.HashMap[String, String]() + val securityProperties = List("spark.ssl.fs.trustStore", "spark.ssl.trustStore", + "spark.ssl.fs.trustStorePassword", "spark.ssl.trustStorePassword", + "spark.ssl.fs.protocol", "spark.ssl.protocol") + + securityProperties + .map {pName => sys.props.get(pName) + .map{pValue => sparkProperties.put(pName, pValue)}} + + jars = Option(jars).map(SparkSubmit.resolveGlobPaths(_, hadoopConf)).orNull + + // Filter out the user jar + jars = jars.split(",").filterNot(_.contains(userJar.split("/").last)).mkString(",") + jars = Option(jars) + .map(SparkSubmit.downloadFileList(_, targetDir, sparkProperties, hadoopConf)).orNull + + if (jars != null) {childClasspath ++= jars.split(",")} + + for (jar <- childClasspath) { + SparkSubmit.addJarToClasspath(jar, loader) --- End diff -- So now all the stuff in `spark.jars` that actually exists on the node will be automatically added to the driver's classpath, which is a change in behavior from before. It's probably ok, but please call out this kind of thing in the commit message.
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. --- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org