Analyzing Real-time Wikipedia Changes using Pinot and Kafka
This is a collection of server-sent-event (SSE) replicators that ingest real-time event streams from Wikimedia's event platform APIs into Kafka topics. A Spring Boot application is used to consume Wikipedia's HTTP-based event streams using a reactive SSE client with retries. The contents of each event's payload from a Wikipedia subscription is pushed as a message into a Kafka topic. With the SSE events stored in Kafka, Apache Pinot can be used to ingest events from Kafka using a real-time table and schema. You can then query Pinot using SQL as data from Wikipedia is continuously fed into the cluster.
Run
You can run the example using Docker compose using the supplied docker-compose.yml file. This will ingest events from Wikipedia to a new Kafka cluster.
docker-compose up
This will start Kafka, Zookeeper, Pinot and the wiki-event-stream application. After everything is up and running and you start to see events being output to the log, you can create the Pinot table for the Wikipedia change events.
First, you need to copy the JSON schema and table definition to the Pinot container. Run the following command in another terminal.
docker cp ./pinot pinot-wikipedia-event-stream_pinot_1:.
After the JSON files are copied to the Pinot container, you can create the schema and table.
docker exec pinot-wikipedia-event-stream_pinot_1 bin/pinot-admin.sh AddTable \
-tableConfigFile /pinot/wiki-table-definition.json \
-schemaFile /pinot/wiki-schema-definition.json \
-exec
After creating the schema and table, you can start querying the real-time change feed from Wikipedia.
docker exec pinot-wikipedia-event-stream_pinot_1 bin/pinot-admin.sh PostQuery \
-brokerHost localhost \
-brokerPort 8000 \
-queryType pql \
-query "SELECT title, COUNT(*) from wikiRecentChangeTable GROUP BY title TOP 10"
You can run these queries directly in the Pinot data browser.
http://localhost:9000/#/query
Copy and paste the following query in the query console and run it.
SELECT category, COUNT(*)
FROM wikiRecentChangeTable
GROUP BY category
ORDER BY COUNT(*) DESC
LIMIT 100
You should see something similar to the following result.
category count(*)
Category:All articles with unsourced statements 555
Category:Articles with hCards 421
Category:Articles with short description 1289
Category:Living people 515
Category:Noindexed pages 390
Category:Webarchive template wayback links 392
Category:Wikipedia articles with GND identifiers 346
Category:Wikipedia articles with LCCN identifiers 417
Category:Wikipedia articles with VIAF identifiers 457
Category:Wikipedia articles with WorldCat identifiers 415
Now try searching for virus related articles. Make sure PQL syntax is checked before running this query.
SELECT category, COUNT(*) as rank FROM wikiRecentChangeTable WHERE regexp_like(category, 'virus')
GROUP BY category
ORDER BY rank DESC
LIMIT 10
You should see something similar to the following output.
category count_star
Category:Aphthoviruses 1
Category:B-Class virus articles 2
Category:C-Class virus articles 2
Category:Low-importance virus articles 4
Category:NA-importance virus articles 3
Category:Project-Class virus articles 1
Category:Template-Class virus articles 2
Category:Virus stubs 1
Category:WikiProject Viruses articles 7
Visualizing Changes with D3js
Try this query to get back time series data that shows how articles related to the 2020 coronavirus pandemic are changing minute-by-minute.
SELECT category as name, DISTINCTCOUNT(id) as rank
FROM wikiRecentChangeTable
WHERE text_match(title, "pandemic")
GROUP BY title, dateTimeConvert(changedTime, '1:MILLISECONDS:EPOCH', '1:MINUTES:SIMPLE_DATE_FORMAT:yyyy-MM-dd HH:mm:ss tz(America/Los_Angeles)', '1:MINUTES')
LIMIT 1000
You can download this data as CSV from the Pinot data browser and plug it into a d3js visualization that animates the query results.
https://observablehq.com/@kbastani/coronavirus-wikipedia-article-change-race
To get this to work, you have to make some changes to the CSV.
- Rename the row headers to
name,date,value - Remove all the double quotes
- Check for any commas in the page titles
The CSV file should look like the following.
name,date,value
2020 coronavirus pandemic in Africa,2020-04-13T08:58:00,2
2020 coronavirus pandemic in Algeria,2020-04-13T08:57:00,1
2020 coronavirus pandemic in Assam,2020-04-13T08:58:00,1
2020 coronavirus pandemic in Azerbaijan,2020-04-13T08:31:00,1
2020 coronavirus pandemic in British Columbia,2020-04-13T09:06:00,1
2020 coronavirus pandemic in Bulgaria,2020-04-13T08:47:00,1
2020 coronavirus pandemic in Bulgaria,2020-04-13T08:50:00,1
2020 coronavirus pandemic in Egypt,2020-04-13T08:38:00,1
Running in Kubernetes
Please visit the Apache Pinot documentation to learn how to easily setup a quick start cluster that has Kafka, Zookeeper, Pinot, and Superset.
By following this guide you'll have a functioning Kubernetes cluster that you can deploy the wiki-event-stream replicator to. The replicator will continuously subscribe to events published by the Wikimedia event stream APIs, and will durably retry with backoffs in the case of a failure.
Build
The best way to use this example is to deploy it to a Kubernetes cluster that has Kafka installed. Start by compiling the application with Docker running.
mvn clean install -DskipTests
This will build a docker image. You can modify the DockerHub user that this image will be associated with in the Maven BOM of the parent project, located here: