How to log using log4j to local file system inside

Question

I'm building an Apache Spark Streaming application and cannot make it log to a file on the local filesystem when running it on YARN. How can achieve this?

I've set log4.properties file so that it can successfully write to a log file in /tmp directory on the local file system (shown below partially):

log4j.appender.file=org.apache.log4j.FileAppender
log4j.appender.file.File=/tmp/application.log
log4j.appender.file.append=false
log4j.appender.file.layout=org.apache.log4j.PatternLayout
log4j.appender.file.layout.ConversionPattern=%d{yyyy-MM-dd HH:mm:ss} %-5p %c{1}:%L - %m%n

When I run my Spark application locally by using the following command:

spark-submit --class myModule.myClass --master local[2] --deploy-mode client myApp.jar

It runs fine and I can see that log messages are written to /tmp/application.log on my local file system.

But when I run the same application via YARN, e.g.

spark-submit --class myModule.myClass --master yarn-client  --name "myModule" --total-executor-cores 1 --executor-memory 1g myApp.jar

or

spark-submit --class myModule.myClass --master yarn-cluster  --name "myModule" --total-executor-cores 1 --executor-memory 1g myApp.jar

I cannot see any /tmp/application.log on the local file system of the machine that runs YARN.

What am I missing.

Answer 1

Alternatively, you can use PropertyConfigurator of log4j to define your custom log properties.

Ex.

 import com.foo.Bar;

 import org.apache.log4j.Logger;
 import org.apache.log4j.PropertyConfigurator;

 public class MySparkApp {

   static Logger logger = Logger.getLogger(MySparkApp.class.getName());

   public static void main(String[] args) {

     // Location to property file
     PropertyConfigurator.configure(args[0]);

     logger.info("Entering application.");

     logger.info("Exiting application.");
   }
 }

Your properties file shall have the following props,

log4j.appender.file=org.apache.log4j.FileAppender

log4j.appender.file.File=/tmp/application.log

log4j.appender.file.append=false

log4j.appender.file.layout=org.apache.log4j.PatternLayout

log4j.appender.file.layout.ConversionPattern=%d{yyyy-MM-dd HH:mm:ss} %-5p %c{1}:%L - %m%n

EDIT: Updating link to log4j docs. Spark uses log4j 2, not v1.2

Ref : http://logging.apache.org/log4j/2.x/

Answer 2

[Edited to avoid confusion]

It looks like you'll need to append to the JVM arguments used when launching your tasks/jobs.

Try editing conf/spark-defaults.conf as described here

spark.executor.extraJavaOptions=-Dlog4j.configuration=file:/apps/spark-1.2.0/conf/log4j.properties

spark.driver.extraJavaOptions=-Dlog4j.configuration=file:/apps/spark-1.2.0/conf/log4j.properties

Alternatively try editing conf/spark-env.sh as described here to add the same JVM argument, although the entries in conf/spark-defaults.conf should work.

If you are still not getting any joy, you can explicitly pass the location of your log4j.properties file on the command line along with your spark-submit like this if the file is contained within your JAR file and in the root directory of your classpath

spark-submit --class sparky.MyApp --master spark://my.host.com:7077 --conf "spark.executor.extraJavaOptions=-Dlog4j.configuration=log4j-executor.properties" myapp.jar

If the file is not on your classpath use the file: prefix and full path like this

spark-submit ... --conf "spark.executor.extraJavaOptions=-Dlog4j.configuration=file:/apps/spark-1.2.0/conf/log4j-executor.properties" ...

Answer 3

The above options of specifying the log4j.properties using spark.executor.extraJavaOptions, spark.driver.extraJavaOptions would only log it locally and also the log4.properties should be present locally on each node.

As specified in the https://spark.apache.org/docs/1.2.1/running-on-yarn.html documentation, you could alternatively upload log4j.properties along with your applicaiton using --files option. This would do yarn aggregate logging on HDFS and you can access the log using the command

yarn logs -applicationId <application id>

Answer 4

1) To debug how Spark on YARN is interpreting your log4j settings, use log4j.debug flag.

2) Spark will create 2 kind of YARN containers, the driver and the worker. So you want to share a file from where you submit the application with all containers (you cant use a file inside the JAR, since this is not the JAR that really run), so you must use the --files Spark submit directive (this will share file with all workers).

Like this:

spark-submit     
    --class com.X.datahub.djobi.Djobi \
    --files "./log4j.properties" \
    --driver-java-options "-Dlog4j.debug=true -Dlog4j.configuration=log4j.properties" \
    --conf "spark.executor.extraJavaOptions=-Dlog4j.debug=true -Dlog4j.configuration=log4j.properties " \
    ./target/X-1.0.jar "$@"

Where log4j.properties is a project file inside src/main/resources/config folder.

I can see in the console:

log4j: Trying to find [config/log4j.properties] using context 
classloader org.apache.spark.util.MutableURLClassLoader@5bb21b69.
log4j: Using URL [jar:file:/home/hdfs/djobi/latest/lib/djobi-1.0.jar!/config/log4j.properties] for automatic log4j configuration.
log4j: Reading configuration from URL jar:file:/home/hdfs/djobi/latest/lib/djobi-1.0.jar!/config/log4j.properties

So the file is taken in account, you can check on Spark webUI too.

Answer 5

In you log4j.properties file, you should also modify the log4j.rootCategory from INFO,console to INFO,file.

log4j.rootCategory=INFO, console    
log4j.rootCategory=INFO,file