☰

Getting Started with the Alluxio-Presto Sandbox

By
Tags: , , , , , , , ,

The Alluxio-Presto sandbox is a docker application featuring installations of MySQL, Hadoop, Hive, Presto, and Alluxio. The sandbox lets you easily dive into an interactive environment where you can explore Alluxio, run queries with Presto, and see the performance benefits of using Alluxio in a big data software stack.

In this guide we’ll be using Presto and Alluxio to showcase how Alluxio can improve Presto’s query performance by caching our data locally so that it can be accessed at memory speed!

The goal of this guide is to run the ss-max query from TPC-DS on Presto reading through Alluxio where the data is stored in an Amazon S3 bucket.

The query completes a full table read to find the number of rows, and max of each column in the store_sales table.

Outline

  • Prerequisites
  • Download and launch the container
  • Explore Alluxio
  • Run Queries with Presto on Alluxio

Prerequisites

  • Docker installed on your machine (MacOS or Linux)
  • Minimum 6GB of RAM available on your local machine to run the container. 8GB is recommended
  • Port 8080 and port 19999 should be open and available.
    • If you have an instance of Alluxio running locally, stop it using alluxio-stop.sh
    • If you have a previous sandbox container running, stop it using docker rm -f alluxio-presto-sandbox

Docker on Mac Prerequisites

You’ll need to increase the default Docker VM settings if you’re running docker on MacOS.

  • Go to Docker > Preferences > Advanced > Memory and increase the slider to a minimum of 6GiB. It is recommended to give anywhere between 6.5GiB-8GiB
  • It is recommended to give 4 or more CPUs to the VM

Press Apply & Restart at the bottom of the dialog.

Download and Launch the Container

To get started with the container, pull it from docker hub.

The docker image is ~1.8GB in size. The download time will depend on your network speed. You can use this time to grab a snack 🙂

$ docker pull alluxio/alluxio-presto-sandbox

Once the pull is complete, start up the container with the following command:

$ docker run -d \
  --shm-size 1G \
  -p 19999:19999 \
  -p 8080:8080 \
  --name alluxio-presto-sandbox \
  alluxio/alluxio-presto-sandbox

This command will start up the container and all of the services within it. It includes MySQL, Hadoop, Hive, Presto, and Alluxio.

  • -p 19999:19999 allows us to access the Alluxio web UI from the host machine
  • -p 8080:8080 allows us to access the Presto web UI from the host machine

Open a shell in the container. Use the following command:

$ docker exec -it alluxio-presto-sandbox bash
[root@abcdef12345 ~]#

abcdef12345 will be the 11 leading characters of your docker container id

For the remainder of this guide, assume all terminal commands should be run from within the docker container. The terminal prompt should start with something similar to [root@abcdef12345 ~]#

Service Architecture

The goal of using this container is to read data stored in an S3 bucket through Alluxio such that:

  • A public dataset is hosted on an Amazon S3 bucket such as s3://bucket-name/path
  • The public bucket is mounted into the Alluxio filesystem at alluxio:///path/to/mount
  • The Alluxio mount point is referenced in Hive table definitions via CREATE TABLE table_name ... LOCATION 'alluxio:///path/to/mount/table_name';
  • Presto uses the tables from the hive metastore to run queries with SELECT * FROM table_name;

About the Container Configuration

The Alluxio-Presto Sandbox comes with pre-installed software which can be found in the /opt directory of the container. Alluxio is installed at /opt/alluxio. The configuration parameters for which are in the /opt/alluxio/conf directory.

The configuration files for the rest of the services can be found within their respective directories in /opt/hadoop, /opt/hive, and /opt/presto. Refer to the programs’ specific documentation in order to find out how each is configured.

The services within the container are all managed by a daemon called supervisord. This daemon is responsible for starting, stopping, and restarting each service.

Control the status of each process by using the supervisorctl command.

  • supervisorctl status will show the status of each service
  • supervisorctl stop <service_name> will stop a service
  • supervisorctl start <service_name> will start a service
  • supervisorctl restart <service_name> will restart a service
  • supervisorctl help will list all available commands

Logs for each process can be found in /var/log/supervisor

Explore Alluxio

We’ll use a combination of the Alluxio web UI at http://localhost:19999 and the Alluxio CLI to explore the Alluxio filesystem and cluster status

The container comes with an Amazon S3 bucket pre-mounted in Alluxio at the /scale1 directory. It contains data for TPC-DS benchmarks at the “scale 1” size factor which amounts to about 1GB of data across multiple tables.

Open the Alluxio web UI at http://localhost:19999 to check if the Alluxio master has started successfully. If not, wait a few moments, refresh the page, and it should become available.

You can see the current Alluxio mounts by running alluxio fs mount from within the container.

$ alluxio fs mount


s3://alluxio-public-http-ufs/tpcds/scale1-parquet  on  /scale1  (s3, capacity=-1B, used=-1B, read-only, not shared, properties={aws.secretKey=******, aws.accessKeyId=******})
/opt/alluxio/underFSStorage                        on  /        (local, capacity=58.42GB, used=-1B(0%), not read-only, not shared, properties={})

Next, we’ll investigate more about how to interact with Alluxio and load data into the cache.

Now let’s see what’s in the mounted /scale1 directory

$ alluxio fs ls /scale1
drwx------                                              3       PERSISTED 07-05-2019 19:41:42:054  DIR /scale1/call_center
drwx------                                              3       PERSISTED 07-05-2019 19:41:42:100  DIR /scale1/catalog_page
drwx------                                           2067       PERSISTED 07-05-2019 19:41:52:535  DIR /scale1/catalog_returns
drwx------                                           1832       PERSISTED 07-05-2019 19:41:59:149  DIR /scale1/catalog_sales
drwx------                                              3       PERSISTED 07-05-2019 19:41:59:152  DIR /scale1/customer
drwx------                                              3       PERSISTED 07-05-2019 19:41:59:156  DIR /scale1/customer_address
drwx------                                              3       PERSISTED 07-05-2019 19:41:59:161  DIR /scale1/customer_demographics
drwx------                                              3       PERSISTED 07-05-2019 19:41:59:165  DIR /scale1/date_dim
drwx------                                              3       PERSISTED 07-05-2019 19:41:59:169  DIR /scale1/household_demographics
drwx------                                              3       PERSISTED 07-05-2019 19:41:59:173  DIR /scale1/income_band
drwx------                                            262       PERSISTED 07-05-2019 19:42:00:104  DIR /scale1/inventory
...

The /scale1 directory contains some more directories which are named after common tables used in TPC-DS benchmarking.

Take a look at the cluster usage summary by using the web UI at http://localhost:19999 or by running alluxio fsadmin report

Take note of the Workers Capacity and the Workers Free / Used fields. These describe the state the Alluxio worker’s cache storage.

Load data into memory from one of the tables using alluxio fs load, and afterwards there will be a change in the amount of free and used capacity in the web UI and when running alluxio fsadmin report

$ alluxio fs load /scale1/customer_demographics
/scale1/customer_demographics/part-00000-68449736-ad44-43d2-841f-4d55afd9e0b3-c000.snappy.parquet loaded
/scale1/customer_demographics/part-00000-267c2412-d427-4907-a398-e6de535ff1d4-c000.snappy.parquet loaded
/scale1/customer_demographics/_SUCCESS already in Alluxio fully
/scale1/customer_demographics loaded

This command will download remote data. Depending on the network bandwidth, it may take a few seconds or more to finish, so you can use this time to eat some of that snack you grabbed earlier.

Refresh the Alluxio Web UI. The Workers Free / Used row will have been updated to reflect to newly cached data in the alluxio worker.

You can now see that the Free Capacity has decreased from 1024.00MB from earlier, and the used capacity in the MEM tier has increased to 7.40MB.

Run Queries with Presto on Alluxio

In this next section we’re going to use Presto and Alluxio to show how Alluxio can massively decrease query times by reading cached data.

This guide focuses on using Presto through the command line; however, you can also use the Presto UI at http://localhost:8080 to view the status of your queries.

Run a Presto Query Through Alluxio

Launch the Presto CLI from within the container:

$ presto --catalog hive --debug
presto>

Tip: You can exit at any time by typing exit;

The container comes pre-loaded with tables in Presto. A schema named alluxio has already been defined. The database contains the tables from the TPC-DS benchmark.

presto> show schemas;
       Schema
--------------------
 alluxio
 default
 information_schema
(3 rows)

Use the alluxio schema

presto> use alluxio;
USE
presto:alluxio>

Once you see the prompt presto:alluxio> you will be using the schema. The table definitions can be found in /usr/share/tpcdsData/createAlluxioTpcdsTables.sql.

We’re going to run the ss-max query from the TPC-DS benchmark which can be found at /usr/share/tpcdsData/ss_max.sql within the container.

This query is run against the store_sales table which represents a snapshot of a retail store’s sales information at a specific point in time. The columns in this table are very similar to what a store may collect from its point-of-sale systems.

The query performs the following actions on the store_sales table:

  • gets the total number of rows in the table with count(*)
  • gets the total number of rows in the table with non null transaction date with count(ss_sold_date_sk)
  • gets the number of unique sales with count(distinct ss_sold_date_sk)
  • gets the highest value of each column in the table. This includes fields such as the transaction date, the transaction time, the item, customer, address, etc.

Overall, it provides a short summary of the table’s information

The query might take some time to run, you can use this period to eat some more of that snack

select
  count(*) as total,
  count(ss_sold_date_sk) as not_null_total,
  count(distinct ss_sold_date_sk) as unique_days,
  max(ss_sold_date_sk) as max_ss_sold_date_sk,
  max(ss_sold_time_sk) as max_ss_sold_time_sk,
  max(ss_item_sk) as max_ss_item_sk,
  max(ss_customer_sk) as max_ss_customer_sk,
  max(ss_cdemo_sk) as max_ss_cdemo_sk,
  max(ss_hdemo_sk) as max_ss_hdemo_sk,
  max(ss_addr_sk) as max_ss_addr_sk,
  max(ss_store_sk) as max_ss_store_sk,
  max(ss_promo_sk) as max_ss_promo_sk
from store_sales
;

When the query finishes, you can press q to leave the query results summary. The output should look similar in your terminal.

  total  | not_null_total | unique_days | max_ss_sold_date_sk | max_ss_sold_time_sk | max_ss_item_sk | max_ss_customer_sk | max_ss_cdemo_sk | max
---------+----------------+-------------+---------------------+---------------------+----------------+--------------------+-----------------+----
 2879789 |        2750838 |        1823 |             2452642 |               75599 |          18000 |             100000 |         1920793 |
(1 row)

Query 20190710_051550_00010_7pbtm, FINISHED, 1 node
http://localhost:8080/ui/query.html?20190710_051550_00010_7pbtm
Splits: 58 total, 58 done (100.00%)
CPU Time: 2.8s total, 1.03M rows/s, 6.12MB/s, 17% active
Per Node: 0.2 parallelism,  210K rows/s, 1.25MB/s
Parallelism: 0.2
Peak Memory: 5.71MB
0:14 [2.88M rows, 17.2MB] [210K rows/s, 1.25MB/s]

Note: here that in the last line of the output, 0:14 represents the query time in mm:ss format.

Because this is the first time we’re reading the data, it is pulled from S3 and is then returned through Alluxio. At the same time, the Alluxio worker will be caching the data in memory so that the next time the data is accessed it can be read at memory speed.

If you refresh the Web UI again, you should see that the Alluxio worker usage has increased.

Because the first query had to read all of the data from the store_sales table through Alluxio the data is now cached locally. Running the ss-max query again should be faster.

Let’s run the query again

Good luck trying to finish your snack before the query finishes!

select
  count(*) as total,
  count(ss_sold_date_sk) as not_null_total,
  count(distinct ss_sold_date_sk) as unique_days,
  max(ss_sold_date_sk) as max_ss_sold_date_sk,
  max(ss_sold_time_sk) as max_ss_sold_time_sk,
  max(ss_item_sk) as max_ss_item_sk,
  max(ss_customer_sk) as max_ss_customer_sk,
  max(ss_cdemo_sk) as max_ss_cdemo_sk,
  max(ss_hdemo_sk) as max_ss_hdemo_sk,
  max(ss_addr_sk) as max_ss_addr_sk,
  max(ss_store_sk) as max_ss_store_sk,
  max(ss_promo_sk) as max_ss_promo_sk
from store_sales
;
...

Query 20190710_051959_00011_7pbtm, FINISHED, 1 node
http://localhost:8080/ui/query.html?20190710_051959_00011_7pbtm
Splits: 58 total, 58 done (100.00%)
CPU Time: 2.9s total,  998K rows/s, 5.95MB/s, 52% active
Per Node: 0.6 parallelism,  597K rows/s, 3.56MB/s
Parallelism: 0.6
Peak Memory: 5.71MB
?:?? [2.88M rows, 17.2MB] [597K rows/s, 3.56MB/s]

Tip: exit the Presto shell with exit;

Take note of the execution time of the query. Did you see a change in performance?

If you want replicate the same result, you will need to free the data from Alluxio memory. After exiting the Presto shell, run the alluxio free command to free the data.

$ alluxio fs free /scale1

You can exit the docker terminal by typing exit at the prompt. Then, shut down and delete the container by running docker rm -f alluxio-presto-sandbox.

Next Steps

You can run other queries from the TPC-DS benchmark here