Flink SQL makes it simple to develop streaming applications using standard SQL. It is easy to learn Flink if you have ever worked with a database or SQL like system by remaining ANSI-SQL 2011 compliant. This tutorial will help you get started quickly with a Flink SQL development environment.
You only need to have basic knowledge of SQL to follow along. No other programming experience is assumed.
There are multiple ways to install Flink. For experimentation, the most common option is to download the binaries and run them locally. You can follow the steps in local installation to set up an environment for the rest of the tutorial.
Once you’re all set, use the following command to start a local cluster from the installation folder:
Once started, the Flink WebUI on localhost:8081 is available locally, from which you can monitor the different jobs.
The SQL Client is an interactive client to submit SQL queries to Flink and visualize the results.
To start the SQL client, run the
sql-client script from the installation folder.
Once the SQL client, our query editor, is up and running, it’s time to start writing queries. Let’s start with printing ‘Hello World’, using the following simple query:
HELP command lists the full set of supported SQL statements. Let’s run one such command,
SHOW, to see a full list of Flink built-in functions.
These functions provide users with a powerful toolbox of functionality when developing SQL queries.
CURRENT_TIMESTAMP will print the machine’s current system time where it is executed.
As with all SQL engines, Flink queries operate on top of tables. It differs from a traditional database because Flink does not manage data at rest locally; instead, its queries operate continuously over external tables.
Flink data processing pipelines begin with source tables. Source tables produce rows operated over during the query’s execution; they are the tables referenced in the
FROM clause of a query. These could be Kafka topics, databases, filesystems, or any other system that Flink knows how to consume.
Tables can be defined through the SQL client or using environment config file. The SQL client support SQL DDL commands similar to traditional SQL. Standard SQL DDL is used to create, alter, drop tables.
Flink has a support for different connectors and formats that can be used with tables. Following is an example to define a source table backed by a CSV file with
dept_id as columns in a
CREATE table statement.
A continuous query can be defined from this table that reads new rows as they are made available and immediately outputs their results.
For example, we can filter for just those employees who work in department
While not designed initially with streaming semantics in mind, SQL is a powerful tool for building continuous data pipelines. Where Flink SQL differs from traditional database queries is that is continuously consuming rows as the arrives and produces updates to its results.
Aggregations on continuous streams need to store aggregated results continuously during the execution of the query. For example, suppose you need to count the number of employees for each department from an incoming data stream. The query needs to maintain the most up to date count for each department to output timely results as new rows are processed.
Such queries are considered stateful. Flink’s advanced fault-tolerance mechanism will maintain internal state and consistency, so queries always return the correct result, even in the face of hardware failure.
When running this query, the SQL client provides output in real-time but in a read-only fashion. Storing results - to power a report or dashboard - requires writing out to another table. This can be achieved using an
INSERT INTO statement. The table referenced in this clause is known as a sink table. An
INSERT INTO statement will be submitted as a detached query to the Flink cluster.
Once submitted, this will run and store the results into the sink table directly, instead of loading the results into the system memory.
If you get stuck, check out the community support resources. In particular, Apache Flink’s user mailing list consistently ranks as one of the most active of any Apache project and a great way to get help quickly.