Spark Custom Logging

Question

I have multiple spark projects in my IDE. By default spark is picking log4j.properties file in spark/conf folder.

As I have multiple spark projects, I want have multiple log4j.properties files(per project one). Probably as part of the project code(resources folder)

Is there a way we can pickup specified log4j.properries instead of default log4j.properties.

Note: I tried this

--driver-java-options "-Dlog4j.configuration=file:///usr/local/Cellar/apache-spark/2.4.1/libexec/conf/driver_log4j.properties"

and it worked without any issues, however I'm looking for something like below.

however I want to load log4j.properties file which is in resource folder while creating the spark logger.

class SparkLogger():
    def __init__(self, app_name, sparksession = None):
        self._spark = sparksession
        self.log4jLogger = None

        if self._spark is not None:
            sparkContext =self._spark.sparkContext
            self.log4jLogger = sparkContext._jvm.org.apache.log4j
            self.log4jLogger = self.log4jLogger.LogManager.getLogger(app_name)

    def info(self, info):
        if self.log4jLogger:
            self.log4jLogger.info(str(info))

    def error(self, info):
        if self.log4jLogger:
            self.log4jLogger.error(str(info))

    def warn(self, info):
        if self.log4jLogger:
            self.log4jLogger.warn(str(info))

    def debug(self, info):
        if self.log4jLogger:
            self.log4jLogger.debug(str(info))

Answer 1

I have attempted to build my custom Logging just like what u described in your question but failed at last. I have to say it was totally a waste.

Finally I chose java.util.logging instead of log4j. Actually it is an original Logging util within JDK. The purpose I use it is that I wanna log information only for myself into a specified file.

So the class is like below.

package org.apache.spark.internal

import java.io.File
import java.text.SimpleDateFormat
import java.util.Date
import java.util.logging._

import scala.collection.mutable

protected [spark] object YLogger extends Serializable with Logging {

  private var ylogs_ = new mutable.HashMap[String, Logger]()

  private def initializeYLogging(className: String): Unit = {
    // Here we set log file onto user's home.
    val rootPath = System.getProperty("user.home")
    val logPath = rootPath + File.separator + className + ".log"
    logInfo(s"Create ylogger for class [${className}] with log file named [${logPath}]")
    val log_ = Logger.getLogger(className)
    val fileHandler = new FileHandler(logPath, false)
    val formatter = new Formatter {
      override def format(record: LogRecord): String = {
        val time = new SimpleDateFormat("yyyy-MM-dd HH:mm:ss").format(new Date)
        new StringBuilder()
          .append("[")
          .append(className)
          .append("]")
          .append("[")
          .append(time)
          .append("]")
          .append(":: ")
          .append(record.getMessage)
          .append("\r\n")
          .toString
      }
    }
    fileHandler.setFormatter(formatter)
    log_.addHandler(fileHandler)
    ylogs_.put(className, log_)
  }

  private def ylog(logName: String): Logger = {
    if (!ylogs_.contains(logName)) {
      initializeYLogging(logName)
    }
    ylogs_.get(logName).get
  }

  def ylogInfo(logName: String)(info: String): Unit = {
    if (ylog(logName).isLoggable(Level.INFO)) ylog(logName).info(info)
  }

  def ylogWarning(logName: String)(warning: String): Unit = {
    if (ylog(logName).isLoggable(Level.WARNING)) ylog(logName).warning(warning)
  }
}

And you can use it like below.

YLogger.ylogInfo("logFileName") ("This is a log.")

It's quite simple to use, I hope my answer could help u.

Answer 2

You have to define application_name log logger properties in log4j file. When you call get logger method using applicaiton_name, you will able to access customized application basis logs generation.