A presentation at Berlin Buzzwords in in Berlin, Germany by Hellmar Becker
Let’s Do Data Lineage in Kafka, Flink and Druid by Tracking Aircraft! Hellmar Becker, Senior Sales Engineer ©2022, imply ©2022, Imply 1
About Me Hellmar Becker Sr. Sales Engineer at Imply Lives near Munich hellmar.becker@imply.io https://www.linkedin.com/in/hellmarbecker/ https://blog.hellmar-becker.de/ 2 ©2022, Imply
Agenda ● ● ● ● ● ● ● ● Streaming Data Governance in Streaming Analytics Kafka Headers: A lesser known feature Stream ETL with Flink using Kafka Headers Quick intro to Apache Druid - A Streaming Analytics Database Tracking Aircraft Radar data with Raspberry Pi Let’s put it all together Live Demo Q&A ©2022, Imply
Streaming Data Governance Data governance is a collection of standards, processes, roles, and metrics that ensure that data is usable, accessible, and effective. Data governance with respect to streaming is about ● ● ● Stream Quality Stream Catalog Stream Lineage Today, let’s shine a light on the Lineage aspect! ©2022, Imply
KFD Stack - Kafka Flink Druid Data sources Stream ETL Stream Processor µService Data/Event driven Apps Custom visualizations Streams Database CDC Event Streaming Infrastructure Event Analytics Infrastructure Files Realtime Analytics Messaging BI tools App data Databases Data Lake Root-cause analysis Dashboards & reports
Kafka Headers: A lesser known feature Record headers give you the ability to add some metadata about the Kafka record. They are collections of arbitrary key/value pairs. Kafka views headers as Kafka headers are key-value pairs, where the header key is a java.lang.String type and the header value is a byte array. Flink models headers as a MAP<BYTES, BYTES> virtual (metadata) column. Use DECODE/ENCODE with known character set (usually UTF-8) if you need to process the string values Druid can read Kafka record headers natively: ● Autogenerate table columns from header keys. For ingestion, you specify a prefix to create column names ● For decoding the values, specify the character set ©2022, Imply
Stream ETL with Flink using Kafka Headers Flink views Kafka headers as metadata columns CREATE TABLE adsb-raw ( kafka_timestamp TIMESTAMP_LTZ(3) METADATA FROM ‘timestamp’, kafka_headers MAP<BYTES, BYTES> NOT NULL METADATA FROM ‘headers’, kafka_key STRING, val STRING ) WITH ( ‘connector’ = ‘kafka’, ‘topic’ = ‘adsb-raw’, ‘key.format’ = ‘csv’, ‘key.fields’ = ‘kafka_key’, ‘key.fields-prefix’ = ‘kafka_’, ‘value.format’ = ‘raw’, … );
Metadata columns can be read and written (unless declared virtual)
©2022, Imply
Apache Druid - A Streaming Analytics Database For analytics applications that require: 1 Sub-second queries at any scale 2 High concurrency at the lowest cost 3 Real-time and historical insights Interactive analytics on TB-PBs of data 100s to 1000s QPS via a highly efficient engine True stream ingestion for Kafka and Kinesis Plus, non-stop reliability with automated fault tolerance and continuous backup ©2022, Imply
Apache Druid - A Streaming Analytics Database ©2022, Imply
KFD Stack - Kafka Flink Druid Data sources Stream ETL Stream Processor µService Data/Event driven Apps Custom visualizations Streams Database CDC Event Streaming Infrastructure Event Analytics Infrastructure Files Realtime Analytics Messaging BI tools App data Databases Data Lake Root-cause analysis Dashboards & reports
Tracking Aircraft Radar data with Raspberry Pi #!/bin/bash CC_BOOTSTRAP=”<confluent cloud bootstrap server>” CC_APIKEY=”<api key>” CC_SECRET=”<secret>” CC_SECURE=”-X security.protocol=SASL_SSL -X sasl.mechanism=PLAIN -X sasl.username=${CC_APIKEY} -X sasl.password=${CC_SECRET}” CLIENT_ID=”<client id>” CLIENT_TIMEZONE=$(date +”%Z”) LON=”0.0” LAT=”0.0” TOPIC_NAME=”adsb-raw” nc localhost 30003 \ | awk -F “,” ‘{ print $5 “|” $0 }’ \ | kcat -P \ -t ${TOPIC_NAME} \ -b ${CC_BOOTSTRAP} \ -H “ClientID=${CLIENT_ID}” \ -H “ClientTimezone=${CLIENT_TIMEZONE}” \ -H “ReceiverLon=${LON}” \ -H “ReceiverLat=${LAT}” \ -K “|” \ ${CC_SECURE} ©2022, Imply 11
Let’s put it all together Delivery Processing Storage Query Visualisation Analytics Pipeline Data Production Aircraft transponder data (ADS-B) 12 Kafka as an event streaming platform Preprocessing with Apache Flink:
Live Demo ©2022, Imply 13
Learnings ● ● ● ● ● Lineage is a crucial part of Data Governance We can track lineage by adding metadata to each data record One way to do this is by using Kafka record headers The KFD (Kafka Flink Druid) stack supports tracking data lineage with Kafka headers end-to-end Don’t be scared of “enterprisey” things, you can try this at home with public data sources such as aircraft data! ©2022, Imply 14
Imply Polaris Learn Druid Apache Druid-powered DBaaS https://imply.io/polaris JupyterLabs + learning environment https://github.com/implydata/learn-druid ©2022, Imply
Next week: Kafka x Flink x Druid Meetup in Berlin Monday, June 17th from 5:30pm Remerge | Heidestraße 9, 10557 Berlin, Germany ©2022, Imply
Resources ● ● ● ● Streaming Governance training by Confluent: https://developer.confluent.io/courses/governing-data-streams/overview/ Catalogs in Flink SQL (Decodable blog): https://www.decodable.co/blog/catalogs-in-flink-sql-a-primer Kafka headers: https://www.confluent.io/blog/5-things-every-kafka-developer-should-know/#tip -5-record-headers Github repo: https://github.com/hellmarbecker/plt-airt-2000 ©2022, Imply 17
Questions hellmar.becker@imply.io https://www.linkedin.com/in/hellmarbecker/ https://blog.hellmar-becker.de/ ©2022, imply ©2022, Imply 18
Data lineage means you can track the data bits in your system and know at any time where they come from and how they have been processed. Enterprise systems need to be able to prove lineage for compliance reasons, but in general, lineage is also a significant aspect of data discoverability and governance.
In this talk, Hellmar is going to connect several Raspberry Pi devices that collect ADS-B (aircraft radar) data to a KFD (Kafka-Flink-Druid) stack for analytical processing. He will deliver the data through Kafka, cleanse and enrich it with Flink, and run analytical queries on the results using Druid.
He will demonstrate how to track data lineage through Kafka metadata and show how this information can be maintained throughout the processing pipeline. This relies on using Kafka headers, an underused feature of Kafka that also integrates readily and easily with Druid!
You will learn how to implement data lineage using the open-source KFD stack and readily available data sources. This way, you can try out enterprise-style data lineage processing at home and prepare yourself for a question that will inevitably arise in any enterprise data engineering project!
for free. You
can too.