[Demo+Webinar] New Product Updates to Make Serverless Flink a Developer’s Best Friend | Watch Now
The latest ksqlDB release introduces long-awaited features such as tunable retention and grace period for windowed aggregates, new built-in functions including LATEST_BY_OFFSET, a peek at the new server API under development, and more.
ksqlDB supports window-based aggregation of events. You can use the following query, for instance, to count the number of pageviews per region in each minute-long interval:
CREATE TABLE pageviews_per_region AS SELECT regionid, COUNT(*) from pageviews WINDOW TUMBLING (SIZE 1 MINUTE) GROUP BY regionid EMIT CHANGES;
However, the aggregation results for each (minute-long) window can’t be stored forever as storage space is finite. The amount of time ksqlDB stores windowed results for is the retention time.
With ksqlDB 0.8.0, this retention time is configurable in the aggregation statement itself by specifying a retention clause, such as RETENTION 7 DAYS.
CREATE TABLE pageviews_per_region AS SELECT regionid, COUNT(*) from pageviews WINDOW TUMBLING (SIZE 1 MINUTE, RETENTION 7 DAYS) GROUP BY regionid EMIT CHANGES;
This is very helpful for applications that rely on ksqlDB’s query serving layer. For example, it’s now possible to materialize windowed metrics inside ksqlDB and store the data for weeks or months to directly support a dashboard served by ksqlDB.
When working with windowed aggregates, another parameter of interest is the grace period—the length of time each window continues to accept and process late-arriving events after the end of the window. This is important because event streaming applications often have to deal with events arriving later than the end of their expected window.
With ksqlDB 0.8.0, the grace period for late-arriving events included in windowed computations can be configured by specifying a clause such as GRACE PERIOD 10 MINUTES in the window expression:
CREATE TABLE pageviews_per_region AS SELECT regionid, COUNT(*) from pageviews WINDOW TUMBLING (SIZE 1 MINUTE, RETENTION 7 DAYS, GRACE PERIOD 10 MINUTES) GROUP BY regionid EMIT CHANGES;
For more details, see the documentation.
With the release of ksqlDB 0.8.0, we’re happy to introduce one of the most highly requested built-in functions: LATEST_BY_OFFSET. The LATEST_BY_OFFSET aggregation function is used to track the latest value of a column when aggregating events from a stream into a table. As the function name suggests, latest is determined by offset order rather than timestamp order.
For example, consider a stream of IoT data:
CREATE STREAM sensor_readings (sensorID BIGINT, temp DOUBLE, quality INT) WITH (KAFKA_TOPIC='readings', VALUE_FORMAT='JSON', KEY='sensorID');
…with these example records:
{"sensorId": 1224, "temp": 56.4, "quality": 3}, {"sensorId": 2658, "temp": 56.4, "quality": 3}, {"sensorId": 1224, "temp": 96.4, "quality": 1}, ...
A user interested in capturing the latest temperature and quality value for each sensor can do so with the following query:
CREATE TABLE aggregate_sensor_readings AS SELECT sensorId, LATEST_BY_OFFSET(temp), LATEST_BY_OFFSET(quality) FROM sensor_readings GROUP BY sensorId EMIT CHANGES;
ksqlDB 0.8.0 also includes an early look at the new client and server API proposed in KLIP-15. This internal set of changes lays the groundwork for new features and enhancements to ksqlDB’s developer experience. We’re excited to share more of the KLIP-15 implementation in a future release and in other blog posts coming soon.
This release also made ksqlDB’s integration with Kafka Connect a little easier, as the ksqlDB server and CLI Docker images now ship with confluent-hub, an easy-to-use tool for installing connectors. With the inclusion of confluent-hub in ksqlDB, downloading connectors is as simple as a single command with a Docker image you already have. Check out the tutorial for running Kafka Connect embedded in ksqlDB for an example.
In addition, ksqlDB 0.8.0 introduces two more built-in functions, REGEXP_EXTRACT and ARRAY_LENGTH, as well as various bug fixes and other improvements. See the changelog for the complete list.
If you haven’t already, join us in our #ksqldb Confluent Community Slack channel and get started with ksqlDB today!
At a high level, bad data is data that doesn’t conform to what is expected, and it can cause serious issues and outages for all downstream data users. This blog looks at how bad data may come to be, and how we can deal with it when it comes to event streams.
Versioned key-value state stores, introduced to Kafka Streams in 3.5, enhance stateful processing capabilities by allowing users to store multiple record versions per key, rather than only the single latest version per key as is the case for existing key-value stores today...