In our last tutorial, we set up a complete Druid cluster. We created all the Druid dependencies and ingested streaming data. In this tutorial, we will expand upon what we've done in the first two tutorials.
We will be working with the same Wikipedia edits data schema from out previous tutorials.
At this point, you should already have Druid downloaded and be comfortable running a Druid cluster locally. If not, have a look at our second tutorial. If Zookeeper and MySQL are not running, you will have to start them as described in The Druid Cluster.
With real-world data, we recommend having a message bus such as Apache Kafka sit between the data stream and the real-time node. The message bus provides higher availability for production environments. Firehoses are the key abstraction for real-time ingestion.
Druid communicates with Kafka using the KafkaFirehoseFactory. Using this Firehose with the right configuration, we can import data into Druid in real-time without writing any code. To load data to a real-time node via Kafka, we'll first need to initialize Zookeeper and Kafka, and then configure and initialize a Realtime node.
The following quick-start instructions for booting a Zookeeper and then Kafka cluster were adapted from the Apache Kafka quickstart guide.
Download Kafka
For this tutorial we will [download Kafka 0.8.2.1] (https://www.apache.org/dyn/closer.cgi?path=/kafka/0.8.2.1/kafka_2.10-0.8.2.1.tgz)
tar -xzf kafka_2.10-0.8.2.1.tgz cd kafka_2.10-0.8.2.1
Start Kafka
First launch ZooKeeper:
./bin/zookeeper-server-start.sh config/zookeeper.properties
Then start the Kafka server (in a separate console):
./bin/kafka-server-start.sh config/server.properties
Create a topic named wikipedia
./bin/kafka-topics.sh --create --zookeeper localhost:2181 \ --replication-factor 1 --partitions 1 --topic wikipedia
Launch a console producer for that topic (so we can paste in kafka messages in a bit)
./bin/kafka-console-producer.sh --broker-list localhost:9092 --topic wikipedia
The realtime spec for the data source in this tutorial is available under examples/indexing/wikipedia.spec from the Druid download
Launch the realtime node
java -Xmx512m -Duser.timezone=UTC -Dfile.encoding=UTF-8 \ -Ddruid.realtime.specFile=examples/indexing/wikipedia.spec \ -classpath "config/_common:config/realtime:lib/*" \ io.druid.cli.Main server realtime
Copy and paste the following data into the terminal where we launched the Kafka console producer above.
{"timestamp": "2013-08-31T01:02:33Z", "page": "Gypsy Danger", "language" : "en", "user" : "nuclear", "unpatrolled" : "true", "newPage" : "true", "robot": "false", "anonymous": "false", "namespace":"article", "continent":"North America", "country":"United States", "region":"Bay Area", "city":"San Francisco", "added": 57, "deleted": 200, "delta": -143} {"timestamp": "2013-08-31T03:32:45Z", "page": "Striker Eureka", "language" : "en", "user" : "speed", "unpatrolled" : "false", "newPage" : "true", "robot": "true", "anonymous": "false", "namespace":"wikipedia", "continent":"Australia", "country":"Australia", "region":"Cantebury", "city":"Syndey", "added": 459, "deleted": 129, "delta": 330} {"timestamp": "2013-08-31T07:11:21Z", "page": "Cherno Alpha", "language" : "ru", "user" : "masterYi", "unpatrolled" : "false", "newPage" : "true", "robot": "true", "anonymous": "false", "namespace":"article", "continent":"Asia", "country":"Russia", "region":"Oblast", "city":"Moscow", "added": 123, "deleted": 12, "delta": 111} {"timestamp": "2013-08-31T11:58:39Z", "page": "Crimson Typhoon", "language" : "zh", "user" : "triplets", "unpatrolled" : "true", "newPage" : "false", "robot": "true", "anonymous": "false", "namespace":"wikipedia", "continent":"Asia", "country":"China", "region":"Shanxi", "city":"Taiyuan", "added": 905, "deleted": 5, "delta": 900} {"timestamp": "2013-08-31T12:41:27Z", "page": "Coyote Tango", "language" : "ja", "user" : "stringer", "unpatrolled" : "true", "newPage" : "false", "robot": "true", "anonymous": "false", "namespace":"wikipedia", "continent":"Asia", "country":"Japan", "region":"Kanto", "city":"Tokyo", "added": 1, "deleted": 10, "delta": -9}
Note: This config uses a messageTime rejection policy which will accept all events and hand off as long as there is a continuous stream of events. In this particular example, hand-off will not actually occur because we only have a handful of events.
Disclaimer: We recognize the timestamps of these events aren't actually recent.
Watch the events getting ingested and the real-time node announcing a data segment
... 2015-02-17T23:01:50,220 INFO [chief-wikipedia] io.druid.server.coordination.BatchDataSegmentAnnouncer - Announcing segment[wikipedia_2013-08-31T00:00:00.000Z_2013-09-01T00:00:00.000Z_2013-08-31T00:00:00.000Z] at path[/druid/segments/localhost:8084/2015-02-17T23:01:50.219Z0] ...
Issue a query
Issuing a TimeBoundaryQuery to the real-time node should return some results:
curl -XPOST -H'Content-type: application/json' \ "http://localhost:8084/druid/v2/?pretty" \ -d'{"queryType":"timeBoundary","dataSource":"wikipedia"}'
[ { "timestamp" : "2013-08-31T01:02:33.000Z", "result" : { "minTime" : "2013-08-31T01:02:33.000Z", "maxTime" : "2013-08-31T12:41:27.000Z" } } ]
Druid offers an additional method of ingesting streaming data via the indexing service. You may be wondering why a second method is needed. Standalone real-time nodes are sufficient for certain volumes of data and availability tolerances. They pull data from a message queue like Kafka or Rabbit, index data locally, and periodically finalize segments for handoff to historical nodes. They are fairly straightforward to scale, simply taking advantage of the innate scalability of the backing message queue. But they are difficult to make highly available with Kafka, the most popular supported message queue, because its high-level consumer doesn’t provide a way to scale out two replicated consumer groups such that each one gets the same data in the same shard. They also become difficult to manage once you have a lot of them, since every machine needs a unique configuration.
Druid solved the availability problem by switching from a pull-based model to a push-based model; rather than Druid indexers pulling data from Kafka, another process pulls data and pushes the data into Druid. Since with the push based model, we can ensure that the same data makes it into the same shard, we can replicate data. The indexing service encapsulates this functionality, where a task-and-resources model replaces a standalone machine model. In addition to simplifying machine configuration, the model also allows nodes to run in the cloud with an elastic number of machines. If you are interested in this form of real-time ingestion, please check out the client library Tranquility.
Getting data into Druid can definitely be difficult for first time users. Please don't hesitate to ask questions in our IRC channel or on our google groups page.