Hadoop MapReduce Next Generation - Encrypted Shuffle

Introduction

The Encrypted Shuffle capability allows encryption of the MapReduce shuffle using HTTPS and with optional client authentication (also known as bi-directional HTTPS, or HTTPS with client certificates). It comprises:

  • A Hadoop configuration setting for toggling the shuffle between HTTP and HTTPS.
  • A Hadoop configuration settings for specifying the keystore and truststore properties (location, type, passwords) used by the shuffle service and the reducers tasks fetching shuffle data.
  • A way to re-load truststores across the cluster (when a node is added or removed).

Configuration

core-site.xml Properties

To enable encrypted shuffle, set the following properties in core-site.xml of all nodes in the cluster:

Property Default Value Explanation
hadoop.ssl.require.client.cert false Whether client certificates are required
hadoop.ssl.hostname.verifier DEFAULT The hostname verifier to provide for HttpsURLConnections. Valid values are: DEFAULT, STRICT, STRICT_I6, DEFAULT_AND_LOCALHOST and ALLOW_ALL
hadoop.ssl.keystores.factory.class org.apache.hadoop.security.ssl.FileBasedKeyStoresFactory The KeyStoresFactory implementation to use
hadoop.ssl.server.conf ss-server.xml Resource file from which ssl server keystore information will be extracted. This file is looked up in the classpath, typically it should be in Hadoop conf/ directory
hadoop.ssl.client.conf ss-client.xml Resource file from which ssl server keystore information will be extracted. This file is looked up in the classpath, typically it should be in Hadoop conf/ directory
hadoop.ssl.enabled.protocols TLSv1 The supported SSL protocols (JDK6 can use TLSv1, JDK7+ can use TLSv1,TLSv1.1,TLSv1.2)

IMPORTANT: Currently requiring client certificates should be set to false. Refer the Client Certificates section for details.

IMPORTANT: All these properties should be marked as final in the cluster configuration files.

Example:
    ...
    <property>
      <name>hadoop.ssl.require.client.cert</name>
      <value>false</value>
      <final>true</final>
    </property>

    <property>
      <name>hadoop.ssl.hostname.verifier</name>
      <value>DEFAULT</value>
      <final>true</final>
    </property>

    <property>
      <name>hadoop.ssl.keystores.factory.class</name>
      <value>org.apache.hadoop.security.ssl.FileBasedKeyStoresFactory</value>
      <final>true</final>
    </property>

    <property>
      <name>hadoop.ssl.server.conf</name>
      <value>ssl-server.xml</value>
      <final>true</final>
    </property>

    <property>
      <name>hadoop.ssl.client.conf</name>
      <value>ssl-client.xml</value>
      <final>true</final>
    </property>
    ...

mapred-site.xml Properties

To enable encrypted shuffle, set the following property in mapred-site.xml of all nodes in the cluster:

Property Default Value Explanation
mapreduce.shuffle.ssl.enabled false Whether encrypted shuffle is enabled

IMPORTANT: This property should be marked as final in the cluster configuration files.

Example:
    ...
    <property>
      <name>mapreduce.shuffle.ssl.enabled</name>
      <value>true</value>
      <final>true</final>
    </property>
    ...

The Linux container executor should be set to prevent job tasks from reading the server keystore information and gaining access to the shuffle server certificates.

Refer to Hadoop Kerberos configuration for details on how to do this.

Keystore and Truststore Settings

Currently FileBasedKeyStoresFactory is the only KeyStoresFactory implementation. The FileBasedKeyStoresFactory implementation uses the following properties, in the ssl-server.xml and ssl-client.xml files, to configure the keystores and truststores.

ssl-server.xml (Shuffle server) Configuration:

The mapred user should own the ssl-server.xml file and have exclusive read access to it.

Property Default Value Explanation
ssl.server.keystore.type jks Keystore file type
ssl.server.keystore.location NONE Keystore file location. The mapred user should own this file and have exclusive read access to it.
ssl.server.keystore.password NONE Keystore file password
ssl.server.truststore.type jks Truststore file type
ssl.server.truststore.location NONE Truststore file location. The mapred user should own this file and have exclusive read access to it.
ssl.server.truststore.password NONE Truststore file password
ssl.server.truststore.reload.interval 10000 Truststore reload interval, in milliseconds
Example:
<configuration>

  <!-- Server Certificate Store -->
  <property>
    <name>ssl.server.keystore.type</name>
    <value>jks</value>
  </property>
  <property>
    <name>ssl.server.keystore.location</name>
    <value>${user.home}/keystores/server-keystore.jks</value>
  </property>
  <property>
    <name>ssl.server.keystore.password</name>
    <value>serverfoo</value>
  </property>

  <!-- Server Trust Store -->
  <property>
    <name>ssl.server.truststore.type</name>
    <value>jks</value>
  </property>
  <property>
    <name>ssl.server.truststore.location</name>
    <value>${user.home}/keystores/truststore.jks</value>
  </property>
  <property>
    <name>ssl.server.truststore.password</name>
    <value>clientserverbar</value>
  </property>
  <property>
    <name>ssl.server.truststore.reload.interval</name>
    <value>10000</value>
  </property>
</configuration>

ssl-client.xml (Reducer/Fetcher) Configuration:

The mapred user should own the ssl-server.xml file and it should have default permissions.

Property Default Value Explanation
ssl.client.keystore.type jks Keystore file type
ssl.client.keystore.location NONE Keystore file location. The mapred user should own this file and it should have default permissions.
ssl.client.keystore.password NONE Keystore file password
ssl.client.truststore.type jks Truststore file type
ssl.client.truststore.location NONE Truststore file location. The mapred user should own this file and it should have default permissions.
ssl.client.truststore.password NONE Truststore file password
ssl.client.truststore.reload.interval 10000 Truststore reload interval, in milliseconds
Example:
<configuration>

  <!-- Client certificate Store -->
  <property>
    <name>ssl.client.keystore.type</name>
    <value>jks</value>
  </property>
  <property>
    <name>ssl.client.keystore.location</name>
    <value>${user.home}/keystores/client-keystore.jks</value>
  </property>
  <property>
    <name>ssl.client.keystore.password</name>
    <value>clientfoo</value>
  </property>

  <!-- Client Trust Store -->
  <property>
    <name>ssl.client.truststore.type</name>
    <value>jks</value>
  </property>
  <property>
    <name>ssl.client.truststore.location</name>
    <value>${user.home}/keystores/truststore.jks</value>
  </property>
  <property>
    <name>ssl.client.truststore.password</name>
    <value>clientserverbar</value>
  </property>
  <property>
    <name>ssl.client.truststore.reload.interval</name>
    <value>10000</value>
  </property>
</configuration>

Activating Encrypted Shuffle

When you have made the above configuration changes, activate Encrypted Shuffle by re-starting all NodeManagers.

IMPORTANT: Using encrypted shuffle will incur in a significant performance impact. Users should profile this and potentially reserve 1 or more cores for encrypted shuffle.

ClientCertificates Client Certificates

Using Client Certificates does not fully ensure that the client is a reducer task for the job. Currently, Client Certificates (their private key) keystore files must be readable by all users submitting jobs to the cluster. This means that a rogue job could read such those keystore files and use the client certificates in them to establish a secure connection with a Shuffle server. However, unless the rogue job has a proper JobToken, it won't be able to retrieve shuffle data from the Shuffle server. A job, using its own JobToken, can only retrieve shuffle data that belongs to itself.

Reloading Truststores

By default the truststores will reload their configuration every 10 seconds. If a new truststore file is copied over the old one, it will be re-read, and its certificates will replace the old ones. This mechanism is useful for adding or removing nodes from the cluster, or for adding or removing trusted clients. In these cases, the client or NodeManager certificate is added to (or removed from) all the truststore files in the system, and the new configuration will be picked up without you having to restart the NodeManager daemons.

Debugging

NOTE: Enable debugging only for troubleshooting, and then only for jobs running on small amounts of data. It is very verbose and slows down jobs by several orders of magnitude. (You might need to increase mapred.task.timeout to prevent jobs from failing because tasks run so slowly.)

To enable SSL debugging in the reducers, set -Djavax.net.debug=all in the mapreduce.reduce.child.java.opts property; for example:

  <property>
    <name>mapred.reduce.child.java.opts</name>
    <value>-Xmx-200m -Djavax.net.debug=all</value>
  </property>

You can do this on a per-job basis, or by means of a cluster-wide setting in the mapred-site.xml file.

To set this property in NodeManager, set it in the yarn-env.sh file:

  YARN_NODEMANAGER_OPTS="-Djavax.net.debug=all $YARN_NODEMANAGER_OPTS"