Stream Processing

Creating a collection with an enforced schema over a new or existing Apache Kafka® topic is useful but has limited utility by itself for creating an application. When you declare a collection, you can only work with the events in their current form. But a critical part of creating streaming applications is transforming, filtering, joining, and aggregating events.

In ksqlDB, you manipulate events by deriving new collections from existing ones and describing the changes between them. When a collection is updated with a new event, ksqlDB updates the collections that are derived from it in real-time. This rich form of computing is known formally as stream processing, because it creates programs that operate continually over unbounded streams of events, ad infinitum. These processes stop only when you explicitly terminate them.

The general pattern for stream processing in ksqlDB is to create a new collection by using the SELECT statement on an existing collection. The result of the inner SELECT feeds into the outer declared collection. You don't need to declare a schema when deriving a new collection, because ksqlDB infers the column names and types from the inner SELECT statement. The value of the ROWTIME pseudo column defines the timestamp of the record written to Kafka. The value of system columns can not be set in the SELECT.

Here are a few examples of deriving between the different collection types.

Derive a new stream from an existing stream

Given the following stream:

CREATE STREAM rock_songs (artist VARCHAR, title VARCHAR)
    WITH (kafka_topic='rock_songs', partitions=2, value_format='avro');

You can derive a new stream with all of the song titles transformed to uppercase:

CREATE STREAM title_cased_songs AS
    SELECT artist, UCASE(title) AS capitalized_song
    FROM rock_songs
    EMIT CHANGES;

Each time a new song is inserted into the rock_songs topic, the uppercase version of the title is appended to the title_cased_songs stream.

Deriving a new table from an existing stream

Given the following table and stream:

CREATE TABLE products (product_name VARCHAR PRIMARY KEY, cost DOUBLE)
    WITH (kafka_topic='products', partitions=1, value_format='json');

CREATE STREAM orders (product_name VARCHAR KEY)
    WITH (kafka_topic='orders', partitions=1, value_format='json');

You can create a table that aggregates rows from the orders stream, while also joining the stream on the products table to enrich the orders data:

CREATE TABLE order_metrics AS
    SELECT p.product_name, COUNT(*) AS count, SUM(p.cost) AS revenue
    FROM orders o JOIN products p ON p.product_name = o.product_name
    GROUP BY p.product_name EMIT CHANGES;

This aggregate table keeps track of the total number of orders of each product, along with the total amount of revenue generated by each product.

Deriving a new table from an existing table

Given the following aggregate table:

CREATE TABLE page_view_metrics AS
    SELECT url, location_id, COUNT(*) AS count
    FROM page_views GROUP BY url EMIT CHANGES;

You can derive another table that filters out rows from the page_view_metrics table:

CREATE TABLE page_view_metrics_mountain_view AS
    SELECT url, count FROM page_view_metrics
    WHERE location_id = 42 EMIT CHANGES;

Deriving a new stream from multiple streams

Given the following two streams:

CREATE STREAM impressions (user VARCHAR KEY, impression_id BIGINT, url VARCHAR)
    WITH (kafka_topic='impressions', partitions=1, value_format='json');

CREATE STREAM clicks (user VARCHAR KEY, url VARCHAR)
    WITH (kafka_topic='clicks', partitions=1, value_format='json');

You can create a derived stream that joins the impressions and clicks streams to output rows indicating that a given impression has been clicked within one minute of the initial ad impression:

CREATE STREAM clicked_impressions AS
    SELECT * FROM impressions i JOIN clicks c WITHIN 1 minute ON i.user = c.user
    WHERE i.url = c.url
    EMIT CHANGES;

Any time an impressions row is received, followed within one minute by a clicks row having the same user, a row is emitted into the clicked_impressions stream.

---

Last update: 2020-10-19