Deploy Alluxio Edge for Trino

Slack Docker Pulls

This document describes how to deploy Alluxio Edge for Trino to work with AWS S3 as UFS. The key step is to place Alluxio jar files to the Trino path, update configuration files, then proceed with the normal deployment of Trino.

Prerequisites

  • Your AWS S3 credentials
  • Preparation of storage for Alluxio Edge: After identifying the storage mounted for Alluxio Edge cache, please note:
    • The size of the local storage to be provisioned for Alluxio Edge alluxio.user.client.cache.size
    • The path where it is mounted alluxio.user.client.cache.dirs
  • Assuming you already have Trino in your environment, and set up similarly as Trino deployment documentation. Note down the installation directory of Trino. We will refer to it as ${TRINO_HOME} throughout this document which you will need to update.
  • A running ETCD cluster; the set of endpoint URLs is a required configuration setting.

Request a trial version of Alluxio Edge. Contact your Alluxio account representative at sales@alluxio.com to request a trial version of Alluxio Edge. Follow their instructions to download the installation tar file into the directory you prepared.

The tar file follows the naming convention alluxio-enterprise-edge-*.tar.gz. For example, if the tarball is named alluxio-enterprise-edge-1.1-6.0.0.tar.gz, the alluxio version is edge-1.1-6.0.0.

Package Alluxio Edge with Trino

Remove any old Alluxio client jar files from the Trino directories

Running a command like this will usually work:

$ find ${TRINO_HOME} -name alluxio*shaded* -exec rm {} \;

Extract Alluxio Edge jars and Place into the Trino Directories

Three Alluxio Edge Java JAR files must be installed on each Trino node.

Extract the Alluxio Edge S3 under store filesystem integration JAR file using this command:

$ tar xf alluxio-enterprise-edge-*.tar.gz alluxio-enterprise-edge-*/lib/alluxio-underfs-s3a-*.jar
$ cp alluxio-enterprise-edge-*/*/alluxio-underfs-s3a-*.jar ${TRINO_HOME}/lib/.

Extract the client and prod jar files from the tarball using this command:

$ tar xf alluxio-enterprise-edge-*.tar.gz alluxio-enterprise-edge-*/client/alluxio-*-client.jar
$ tar xf alluxio-enterprise-edge-*.tar.gz alluxio-enterprise-edge-*/assembly/alluxio-prod-*.jar

Depending on your the Trino version, the jars need to be copied to different destination.

  • For Trino versions 434 or later, copy to ${TRINO_HOME}/plugin/<pluginName>/hdfs
  • For Trino versions older than 434, copy to ${TRINO_HOME}/plugin/<pluginName>

Note there is no impact in copying jars to both destinations

The following example shows the copy commands for version 434+ into the plugin directories for hive, hudi, delta lake, and iceberg.

$ cp alluxio-enterprise-edge-*/*/alluxio-*-client.jar ${TRINO_HOME}/plugin/hive/hdfs
$ cp alluxio-enterprise-edge-*/*/alluxio-prod-*.jar ${TRINO_HOME}/plugin/hive/hdfs
$ cp alluxio-enterprise-edge-*/*/alluxio-*-client.jar ${TRINO_HOME}/plugin/hudi/hdfs
$ cp alluxio-enterprise-edge-*/*/alluxio-prod-*.jar ${TRINO_HOME}/plugin/hudi/hdfs
$ cp alluxio-enterprise-edge-*/*/alluxio-*-client.jar ${TRINO_HOME}/plugin/delta-lake/hdfs
$ cp alluxio-enterprise-edge-*/*/alluxio-prod-*.jar ${TRINO_HOME}/plugin/delta-lake/hdfs
$ cp alluxio-enterprise-edge-*/*/alluxio-*-client.jar ${TRINO_HOME}/plugin/iceberg/hdfs
$ cp alluxio-enterprise-edge-*/*/alluxio-prod-*.jar ${TRINO_HOME}/plugin/iceberg/hdfs

Along with the Alluxio client and prod jars in the plugin directories, an additional third party jar is needed. Download the commons-lang3-3.14.0 jar and place it in the same directories.

$ wget https://repo1.maven.org/maven2/org/apache/commons/commons-lang3/3.14.0/commons-lang3-3.14.0.jar
$ cp commons-lang3-3.14.0.jar ${TRINO_HOME}/plugin/hive/hdfs
$ cp commons-lang3-3.14.0.jar ${TRINO_HOME}/plugin/hudi/hdfs
$ cp commons-lang3-3.14.0.jar ${TRINO_HOME}/plugin/delta-lake/hdfs
$ cp commons-lang3-3.14.0.jar ${TRINO_HOME}/plugin/iceberg/hdfs

Download the Prometheus jar

Prometheus is the supported tool for observing the metrics that Alluxio Edge emits. Download the java agent jar and save it in the ${TRINO_HOME}/lib/ directory.

$ wget https://repo.maven.apache.org/maven2/io/prometheus/jmx/jmx_prometheus_javaagent/0.20.0/jmx_prometheus_javaagent-0.20.0.jar -O ${TRINO_HOME}/lib/jmx_prometheus_javaagent-0.20.0.jar.

Update Configurations

Update Alluxio Configurations in Trino JVM Config

Configure Alluxio configuration properties in Trino’s jvm.config file, which is usually located at ${TRINO_HOME}/etc/. The following settings need to be added:

-Dalluxio.conf.dir=${TRINO_HOME}/etc/alluxio/
--add-opens=java.management/sun.management=ALL-UNNAMED
-Dalluxio.metrics.conf.file=${TRINO_HOME}/etc/alluxio/alluxio-metrics.properties
-javaagent:${TRINO_HOME}/lib/jmx_prometheus_javaagent-0.20.0.jar=9696:${TRINO_HOME}/etc/alluxio/jmx_export_config.yaml
  • alluxio.conf.dir is the directory containing Alluxio specific configuration files
  • The add-opens line allows compatibility with Java 17
  • alluxio.metrics.conf.file is the path to configuration file for metrics
  • javaagent line starts a metrics agent to pull metrics from. In this example, it runs on port 9696 and the jmx_export_config.yaml file defines its configuration.

Note any additional Alluxio properties can be specified following the -D<key>=<value> format.

Create metrics.properties and jmx_export_config.yaml for Prometheus integration with metrics

Create jmx_export_config.yaml in ${TRINO_HOME}/etc/alluxio/ with the following sample content.

---
startDelaySeconds: 0
ssl: false
global:
  scrape_interval:     15s
  evaluation_interval: 15s
rules:
- pattern: ".*"

Also create metrics.properties in ${TRINO_HOME}/etc/alluxio/ with the following sample content.

sink.jmx.class=alluxio.metrics.sink.JmxSink

Create alluxio-site.properties for Alluxio Edge configuration

Create alluxio-site.properties in ${TRINO_HOME}/etc/alluxio/. The following configuration should be set:

alluxio.license=<YOUR LICENSE STRING>
# ex. alluxio.etcd.endpoints=http://trino-edge-etcd1:2379,http://trino-edge-etcd2:2379,http://trino-edge-etcd3:2379
alluxio.etcd.endpoints=<YOUR_ETCD_ENDPOINTS>
alluxio.dora.enabled=false
alluxio.underfs.object.store.breadcrumbs.enabled=false

# Alluxio under file system setup (S3)
# ex. alluxio.dora.client.ufs.root=s3://myBucket
alluxio.dora.client.ufs.root=<YOUR_S3_URL>
# credentials to access S3 bucket, optional if credentials are provided in other ways
s3a.accessKeyId=<MY_KEY_ID>
s3a.secretKey=<MY_SECRET_KEY>

# Enable Alluxio Edge cache on Trino nodes (example with 2 SSD volumes)
alluxio.user.client.cache.enabled=true
alluxio.user.client.cache.size=<PUT_YOUR_SSD1_VOL_SIZE_HERE>GB,<PUT_YOUR_SSD2_VOL_SIZE_HERE>GB
alluxio.user.client.cache.dirs=<PUT_YOUR_SSD1_VOL_MOUNT_POINT_HERE>/alluxio_cache,/<PUT_YOUR_SSD2_VOL_MOUNT_POINT_HERE>/alluxio_cache

# Enable metrics collection
alluxio.user.metrics.collection.enabled=true

Please refer to configuration settings for details.

Create alluxio-core-site.xml for overriding the S3 filesystem

Create alluxio-core-site.xml in ${TRINO_HOME}/etc/alluxio/ to include the fs.s3a.impl property to ensure that Trino uses the S3AFileSystem when a Hive table LOCATION is set to the s3:// or s3a:// scheme.

<configuration>
  <property>
    <name>fs.s3.impl</name>
    <value>alluxio.hadoop.FileSystem</value>
  </property>
  <property>
    <name>fs.s3a.impl</name>
    <value>alluxio.hadoop.FileSystem</value>
  </property>
  <property>
    <name>fs.alluxio.impl</name>
    <value>alluxio.hadoop.FileSystem</value>
  </property>
</configuration>

Update Catalogs

Configure the Trino catalog (such as HIVE and Delta Lake catalog) to include alluxio-core-site.xml file to the resources. You can likely find the catalog files in ${TRINO_HOME}/etc/catalog/. Note that the hive.config.resources property may already be set with existing xml files, ie. /etc/hadoop/conf/core-site.xml,/etc/hadoop/conf/hdfs-site.xml In this case, alluxio-core-site.xml should be appended to the end of the list.


Some optional configurations

    hive.non-managed-table-writes-enabled=true
    hive.s3select-pushdown.enabled=true
    hive.storage-format=PARQUET
    hive.allow-drop-table=true

Redeploy Trino

If Trino was previously running, it must be restarted to pick up the configuration changes applied.