qqqttt123/gravitino-playground

A playground to experience Gravitino

Playground introduction

The playground is a complete Apache Gravitino Docker runtime environment with Hive, HDFS, Trino, MySQL, PostgreSQL, Jupyter, and a Gravitino server.

Depending on your network and computer, startup time may take 3-5 minutes. Once the playground environment has started, you can open http://localhost:8090 in a browser to access the Gravitino Web UI.

Prerequisites

Install Git and Docker Compose.

TCP ports used

The playground runs a number of services. The TCP ports used may clash with existing services you run, such as MySQL or Postgres.

Docker container	Ports used
playground-gravitino	8090 9001
playground-hive	3307 19000 19083 60070
playground-mysql	13306
playground-postgresql	15342
playground-trino	18080
playground-jupyter	18888

Start playground

Launch all components of playground

git clone git@github.com:apache/gravitino-playground.git
cd gravitino-playground
./launch-playground.sh

Launch special component or components of playground

git clone git@github.com:apache/gravitino-playground.git
cd gravitino-playground
./launch-playground.sh hive|gravitino|trino|postgresql|mysql|spark|jupyter

Note. Components have dependencies, only launching one or several components cannot experience the full functionality of the playground.

Experiencing Apache Gravitino with Trino SQL

Using Trino CLI in Docker Container

1. Login to the Gravitino playground Trino Docker container using the following command:

docker exec -it playground-trino bash

2. Open the Trino CLI in the container.

trino@container_id:/$ trino

Using Jupyter Notebook

1. Open the Jupyter Notebook in the browser at http://localhost:18888.

2. Open the gravitino-trino-example.ipynb notebook.

3. Start the notebook and run the cells.

Using Spark client

1. Login to the Gravitino playground Spark Docker container using the following command:

docker exec -it playground-spark bash

2. Open the Spark SQL client in the container.

spark@container_id:/$ cd /opt/spark && /bin/bash bin/spark-sql 

Example

Simple Trino queries

You can use simple queries to test in the Trino CLI.

SHOW CATALOGS;

CREATE SCHEMA catalog_hive.company
  WITH (location = 'hdfs://hive:9000/user/hive/warehouse/company.db');

SHOW CREATE SCHEMA catalog_hive.company;

CREATE TABLE catalog_hive.company.employees
(
  name varchar,
  salary decimal(10,2)
)
WITH (
  format = 'TEXTFILE'
);

INSERT INTO catalog_hive.company.employees (name, salary) VALUES ('Sam Evans', 55000);

SELECT * FROM catalog_hive.company.employees;

SHOW SCHEMAS from catalog_hive;

DESCRIBE catalog_hive.company.employees;

SHOW TABLES from catalog_hive.company;

Cross-catalog queries

In a company, there may be different departments using different data stacks. In this example, the HR department uses Apache Hive to store its data and the sales department uses PostgreSQL. You can run some interesting queries by joining the two departments' data together with Gravitino.

To know which employee has the largest sales amount, run this SQL:

SELECT given_name, family_name, job_title, sum(total_amount) AS total_sales
FROM catalog_hive.sales.sales as s,
  catalog_postgres.hr.employees AS e
where s.employee_id = e.employee_id
GROUP BY given_name, family_name, job_title
ORDER BY total_sales DESC
LIMIT 1;

To know the top customers who bought the most by state, run this SQL:

SELECT customer_name, location, SUM(total_amount) AS total_spent
FROM catalog_hive.sales.sales AS s,
  catalog_hive.sales.stores AS l,
  catalog_hive.sales.customers AS c
WHERE s.store_id = l.store_id AND s.customer_id = c.customer_id
GROUP BY location, customer_name
ORDER BY location, SUM(total_amount) DESC;

To know the employee's average performance rating and total sales, run this SQL:

SELECT e.employee_id, given_name, family_name, AVG(rating) AS average_rating, SUM(total_amount) AS total_sales
FROM catalog_postgres.hr.employees AS e,
  catalog_postgres.hr.employee_performance AS p,
  catalog_hive.sales.sales AS s
WHERE e.employee_id = p.employee_id AND p.employee_id = s.employee_id
GROUP BY e.employee_id,  given_name, family_name;

Using Spark and Trino

You might consider generating data with SparkSQL and then querying this data using Trino. Give it a try with Gravitino:

1. Login Spark container and execute the SQLs:

// using Hive catalog to create Hive table
USE catalog_hive;
CREATE DATABASE product;
USE product;

CREATE TABLE IF NOT EXISTS employees (
    id INT,
    name STRING,
    age INT
)
PARTITIONED BY (department STRING)
STORED AS PARQUET;
DESC TABLE EXTENDED employees;

INSERT OVERWRITE TABLE employees PARTITION(department='Engineering') VALUES (1, 'John Doe', 30), (2, 'Jane Smith', 28);
INSERT OVERWRITE TABLE employees PARTITION(department='Marketing') VALUES (3, 'Mike Brown', 32);

2. Login Trino container and execute SQLs:

SELECT * FROM catalog_hive.product.employees WHERE department = 'Engineering';

Using Apache Iceberg REST service

If you want to migrate your business from Hive to Iceberg. Some tables will use Hive, and the other tables will use Iceberg. Gravitino provides an Iceberg REST catalog service, too. You can use Spark to access REST catalog to write the table data. Then, you can use Trino to read the data from the Hive table joining the Iceberg table.

spark-defaults.conf is as follows (It's already configured in the playground):

spark.sql.extensions org.apache.iceberg.spark.extensions.IcebergSparkSessionExtensions
spark.sql.catalog.catalog_rest org.apache.iceberg.spark.SparkCatalog
spark.sql.catalog.catalog_rest.type rest
spark.sql.catalog.catalog_rest.uri http://gravitino:9001/iceberg/
spark.locality.wait.node 0

Please note that catalog_rest in SparkSQL and catalog_iceberg in Gravitino and Trino share the same Iceberg JDBC backend, which implies that they can access the same dataset.

1. Login Spark container and execute the steps.

docker exec -it playground-spark bash

spark@container_id:/$ cd /opt/spark && /bin/bash bin/spark-sql

use catalog_rest;
create database sales;
use sales;
create table customers (customer_id int, customer_name varchar(100), customer_email varchar(100));
describe extended customers;
insert into customers (customer_id, customer_name, customer_email) values (11,'Rory Brown','rory@123.com');
insert into customers (customer_id, customer_name, customer_email) values (12,'Jerry Washington','jerry@dt.com');

2. Login Trino container and execute the steps. You can get all the customers from both the Hive and Iceberg table.

docker exec -it playground-trino bash

trino@container_id:/$ trino

select * from catalog_hive.sales.customers
union
select * from catalog_iceberg.sales.customers;

Using Gravitino with LlamaIndex

Gravitino playground also provides a simple RAG demo with LlamaIndex. This demo will show you the ability of using Gravitino to manage both tabular and non-tabular dataset, connecting to LlamaIndex as a unified data source, then use LlamaIndex and LLM to query both tabular and non-tabular data with one natural language query.

The demo is located in the jupyter folder, you can open the gravitino_llama_index_demo.ipynb demo via Jupyter Notebook by http://localhost:18888.

The scenario of this demo is that basic structured city statistics data is stored in MySQL, and detailed city introductions are stored in PDF files. The user wants to know the answers to the cities both in the structured data and the PDF files.

In this demo, you will use Gravitino to manage the MySQL table using relational catalog, pdf files using fileset catalog, treated Gravitino as a unified data source for LlamaIndex to build indexes on both tabular and non-tabular data. Then you will use LLM to query the data with natural language queries.

Note: to run this demo, you need to set OPENAI_API_KEY in the gravitino_llama_index_demo.ipynb, like below, OPENAI_API_BASE is optional.

import os

os.environ["OPENAI_API_KEY"] = ""
os.environ["OPENAI_API_BASE"] = ""

ASF Incubator disclaimer

Apache Gravitino is an effort undergoing incubation at The Apache Software Foundation (ASF), sponsored by the Apache Incubator. Incubation is required of all newly accepted projects until a further review indicates that the infrastructure, communications, and decision making process have stabilized in a manner consistent with other successful ASF projects. While incubation status is not necessarily a reflection of the completeness or stability of the code, it does indicate that the project has yet to be fully endorsed by the ASF.

_{Apache®, Apache Gravitino™, Apache Hive™, Apache Iceberg™, and Apache Spark™ are either registered trademarks or trademarks of the Apache Software Foundation in the United States and/or other countries.}