Configuration Settings

Slack Docker Pulls GitHub edit source

Alluxio can be configured by setting the values of supported configuration properties . To learn about how users can customize how an application (e.g., a Spark or MapReduce job) interacts with Alluxio, see how to configure Alluxio applications; to learn about how Alluxio admins can customize Alluxio service, see how to configure Alluxio clusters.

Configure Applications

Customizing how an application job interacts with Alluxio service is application specific. Here we provide recommendations for a few common applications.

Alluxio Shell Commands

Alluxio shell users can put JVM system properties -Dproperty=value after fs command and before the subcommand (e.g., copyFromLocal) to specify Alluxio properties from the command line. For example, the following Alluxio shell command sets the write type to CACHE_THROUGH when copying files to Alluxio:

bin/alluxio fs -Dalluxio.user.file.writetype.default=CACHE_THROUGH copyFromLocal /

Spark Jobs

Spark users can use pass JVM system properties to Spark jobs by adding "-Dproperty=value" to spark.executor.extraJavaOptions for Spark executors and spark.driver.extraJavaOptions for Spark drivers. For example, to submit a Spark job with the write CACHE_THROUGH when writing to Alluxio:

spark-submit \
--conf 'spark.driver.extraJavaOptions=-Dalluxio.user.file.writetype.default=CACHE_THROUGH' \
--conf 'spark.executor.extraJavaOptions=-Dalluxio.user.file.writetype.default=CACHE_THROUGH' \

In the Spark Shell, this can be achieved by:

val conf = new SparkConf()
    .set("spark.driver.extraJavaOptions", "-Dalluxio.user.file.writetype.default=CACHE_THROUGH")
    .set("spark.executor.extraJavaOptions", "-Dalluxio.user.file.writetype.default=CACHE_THROUGH")
val sc = new SparkContext(conf)

Hadoop MapReduce Jobs

Hadoop MapReduce users can add "-Dproperty=value" after the hadoop jar or yarn jar command and the properties will be propagated to all the tasks of this job. For example, the following MapReduce job of wordcount sets write type to CACHE_THROUGH when writing to Alluxio:

bin/hadoop jar libexec/share/hadoop/mapreduce/hadoop-mapreduce-examples-2.7.3.jar wordcount \
-Dalluxio.user.file.writetype.default=CACHE_THROUGH \
-libjars /<PATH_TO_ALLUXIO>/client/alluxio-1.6.1-client.jar \

Configure Alluxio Cluster

Alluxio admins can create and customize the property file to configure an Alluxio cluster. If this file does not exist, it can be created from the template file under ${ALLUXIO_HOME}/conf:

cp conf/ conf/

Make sure that this file is distributed to ${ALLUXIO_HOME}/conf on every Alluxio node (masters and workers) before starting the cluster.

Note that, ${ALLUXIO_HOME}/conf/ are only loaded by Alluxio server processes. Applications interacting with Alluxio service through Alluxio client will not load these site property files, unless ${ALLUXIO_HOME}/conf is on the application classpath.

Use Environment variables

Alluxio supports a few frequently used configuration settings via the environment variables, including:

Environment VariableDescription
ALLUXIO_CONF_DIR path to Alluxio configuration directory.
ALLUXIO_LOGS_DIR path to Alluxio logs directory.
ALLUXIO_MASTER_HOSTNAME hostname of Alluxio master, defaults to localhost.
ALLUXIO_UNDERFS_ADDRESS under storage system address, defaults to ${ALLUXIO_HOME}/underFSStorage which is a local file system.
ALLUXIO_RAM_FOLDER the directory where a worker stores in-memory data, defaults to /mnt/ramdisk.
ALLUXIO_JAVA_OPTS Java VM options for both Master, Worker and Alluxio Shell configuration. Note that, by default ALLUXIO_JAVA_OPTS is included in both ALLUXIO_MASTER_JAVA_OPTS, ALLUXIO_WORKER_JAVA_OPTS and ALLUXIO_USER_JAVA_OPTS.
ALLUXIO_MASTER_JAVA_OPTS additional Java VM options for Master configuration.
ALLUXIO_WORKER_JAVA_OPTS additional Java VM options for Worker configuration.
ALLUXIO_USER_JAVA_OPTS additional Java VM options for Alluxio shell configuration.
ALLUXIO_CLASSPATH additional classpath entries for Alluxio processes. This is empty by default.
ALLUXIO_LOGSERVER_HOSTNAME host name of the log server. This is empty by default.
ALLUXIO_LOGSERVER_PORT port number of the log server. This is 45600 by default.
ALLUXIO_LOGSERVER_LOGS_DIR path to the local directory where Alluxio log server stores logs received from Alluxio servers.

For example, if you would like to setup an Alluxio master at localhost that talks to an HDFS cluster with a namenode also running at localhost, and enable Java remote debugging at port 7001, you can do so before starting master process using:

export ALLUXIO_MASTER_HOSTNAME="localhost"
export ALLUXIO_UNDERFS_ADDRESS="hdfs://localhost:9000"
export ALLUXIO_MASTER_JAVA_OPTS="$ALLUXIO_JAVA_OPTS -agentlib:jdwp=transport=dt_socket,server=y,suspend=n,address=7001"

Users can either set these variables through the shell or in conf/ If this file does not exist yet, you can create one by copying the template:

cp conf/ conf/

Order of Configuration Sources

If an Alluxio property is configured in multiple sources, its value is determined by the source earliest in this list:

  1. JVM system properties (i.e., -Dproperty=key)
  2. Environment variables
  3. Property files. When an Alluxio cluster starts, each server process including master and worker searches in a list paths of ${HOME}/.alluxio/, /etc/alluxio/ and ${ALLUXIO_HOME}/conf in order, and will skip the remaining paths once this file is found.

If no above user-specified configuration is found for a property, Alluxio runtime will fallback to its default property value.