Configuring Apache Spark Logging with Scala and lo

Question

I am very confused with setting up logging with Apache Spark. Apache spark used Log4j for logging and it generates huge amount of log data. Is there a way to setup log4j for spark logs and use logback for application log. I am quite conversant with logback but it seems spark only support log4j. Below piece of code was working fine till i introduced apache spark. Any help in this regard will be helpful.

import com.typesafe.scalalogging.LazyLogging

import scala.util.{Failure, Success}
import scala.xml.{Elem, XML}



object MainApp extends App with LazyLogging  {



  val currency = new YahooCurrencyLoader() with CurrencyParameters
  val ccy = currency.getXML(currency.ccyUrl) match {

    case Success(v) => XML.save("PreviousRun.xml",v); logger.info("XML has been saved for use")
    case Failure(ex) => logger.error("XML extraction failed. Look at Yahoo extraction class. ${ex.getMessage}" )

  }

  val xmllocation: String = "./PreviousRun.xml"
  val loadxml: Elem = XML.loadFile(xmllocation)
  //print(loadxml)
  //print(currency.findCurrency(loadxml,"GBP"))

  logger.info("USD CAD Cross is " + currency.findCurrency(loadxml,"CAD").head)

Answer 1

I have used the following imports:

import org.slf4j.Logger
import org.slf4j.LoggerFactory

Sample code as shown below.

Object SparkCode {

val logger = LoggerFactory.getLogger(this.getClass.getName)

  def main(args: Array[String]) {

  logger.info("Connection Started . ")

  // Create Spark Context and go on..

  }

}

And you are sorted.

Answer 2

I do not know if you use sbt or maven but that is where it should all start. Myself I use sbt so I will give you an example how we have solved this problem.

1. Apache Spark uses log4j 1.2.xx

That is true and it is really problematic if you do not want to use the same logging implementation.But there is help!

First, exclude the following libs from spark dependencies:

• log4j
• slf4j-log4j12

For sbt (using sbt-assembly) it looks like this:

lazy val spark16 = Seq("spark-core", "spark-sql", "spark-hive")
  .map("org.apache.spark" %% _ % "1.6.1")
  .map(_.excludeAll(
    ExclusionRule(name = "log4j"),
    ExclusionRule(name = "slf4j-log4j12")
  ))

2. Redirect log4j logging to slf4j

A detailed description can be found here: https://www.slf4j.org/legacy.html
And the module that is in our interest is: log4j-over-slf4j

The log4j-over-slf4j module contains replacements of most widely used log4j classes, namely org.apache.log4j.Category, org.apache.log4j.Logger, org.apache.log4j.Priority, org.apache.log4j.Level, org.apache.log4j.MDC, and org.apache.log4j.BasicConfigurator. These replacement classes redirect all work to their corresponding SLF4J classes.

So we can have all the logs redirected back to slf4j from where some other logging implementation could pick it up.

Easy, simply add this dependency to your application

"org.slf4j" % "log4j-over-slf4j" % "1.7.25"

3. Add desired logging implementation

In our case it was (like yours) logback, so we added it as dependency:

"ch.qos.logback" % "logback-classic" % "1.2.3"

Add some logback.xml configuration to your classpath, for example in src/main/resources and enjoy!

spark-submit

If you need help using Logback while deploying your app with spark-submit please follow this answer: https://stackoverflow.com/a/45480145/1549135