javalibrary / kafkastreams-cep

Complex Event Processing on top of Kafka Streams

Complex Event Processing on top of Kafka Streams Processor API !

Apache Kafka is a high-throughput, distributed, publish-subscribe messaging system.

This library can be used to extend the Kafka Streams API in order to select complex event sequences from streams.

It provides a convenient DSL to build complex event queries.

This library is still evolving and public APIs may change in future.

Quick Start

Add the KafkaStreamsCEP to the dependency of the pom.xml of your project. Note that the last version is currently not available in Maven Central yet.

Maven dependency

For Apache Kafka 0.10.0.1 (deprecated & not stable)

Available in Maven Central

    <dependency>
      <groupId>com.github.fhuss</groupId>
      <artifactId>kafka-streams-cep</artifactId>
      <version>0.1.0</version>
    </dependency>

As of Apache Kafka 1.0.0

    <dependency>
      <groupId>com.github.fhuss</groupId>
      <artifactId>kafka-streams-cep</artifactId>
      <version>1.0.0</version>
    </dependency>

Define a pattern

        ...
        
        Properties streamsConfiguration = new Properties();
        streamsConfiguration.put(StreamsConfig.APPLICATION_ID_CONFIG, "my-first-cep-app");
        streamsConfiguration.put(StreamsConfig.BOOTSTRAP_SERVERS_CONFIG, "localhost:9092");
        streamsConfiguration.put(StreamsConfig.DEFAULT_KEY_SERDE_CLASS_CONFIG, Serdes.StringSerde.class);
        streamsConfiguration.put(StreamsConfig.DEFAULT_VALUE_SERDE_CLASS_CONFIG, Serdes.StringSerde.class);

        Pattern<String, String> pattern = new QueryBuilder<String, String>()
                .select("select-A")
                    .where((event, store) -> event.value().equals("A"))
                    .then()
                .select("select-B")
                    .where(((event, store) -> event.value().equals("B")))
                    .then()
                .select("select-C")
                    .where(((event, store) -> event.value().equals("C")))
                .build();

        StreamsBuilder builder = new StreamsBuilder();

        KStream<String, String> letters = builder.stream("Letters");

        KStream<String, Sequence<String, String>> sequences = new ComplexStreamsBuilder().stream(letters)
                .query("MyLettersQuery", pattern);
        
        sequences.print(Printed.toSysOut());

        KafkaStreams kafkaStreams = new KafkaStreams(builder.build(), streamsConfiguration);
        kafkaStreams.start();
        
        ...

The Pattern API

The Kafka Streams CEP library defines a Pattern API that allow you to define complex event pattern sequence that will be used to select records from input streams.

A complex pattern sequence is made of multiple stages. Each stage consists of a single pattern or a multiple patterns combined together with a logical AND or OR operator.

These simple patterns specify the conditions necessary to select an event from the input streams and to go to the next following stage.

To create a new complex pattern sequence you have to first create a new pattern QueryBuilder instance. A pattern query defines a sequence of events that occur in order but not necessarily in contiguous positions in the input stream.

QueryBuilder<String, Integer> builder = new QueryBuilder<>();

Pattern Builder Methods	Description & Examples
select()	Defines a new stage. select = query.select(); Stage can also be named - by default a stage is named based on its index (0-based) select = query.select("my-first-stage");
optional()	Defines that this pattern is optional. i.e. that at most one event can match this pattern. builder.optional()
zeroOrMore()	Defines that zero or multiple events can match this pattern. builder.zeroOrMore()
oneOrMore()	Defines that at least one event should match this pattern. builder.oneOrMore()
times(#ofTimes)	Defines that a specific number of events should match this pattern. builder.time(3)
where(Matcher)	Defines the condition under which an event should be selected to be added to the pattern sequence. Currently, three types of Matcher can be used to select a relevant event. SimpleMatcher builder.where(new SimpleMatcher() { @Override public boolean matches(Event event) { return event.value().equals("A"); } }); StatefulMatcher builder.where(new StatefulMatcher() { @Override public boolean matches(Event event, States states) { double average = (int)states.get("sum") / (int)states.get("count"); return average >= event.value(); } }); SequenceMatcher builder.where(new SequenceMatcher() { @Override public boolean matches(Event event, Sequence sequence, States states) { double average = StreamSupport.stream(sequence.spliterator(), false) .mapToInt(Event::value) .summaryStatistics() .getAverage(); return average >= event.value(); } }) });
fold(state, aggregator)	The fold() method allows to accumulate some state between each pattern. Then the defined states can be used latter to select relevant events. The aggregator function is invoke only if the input builder.fold("amount", new Aggregator() { @Override public Integer aggregate(String key, Integer value, Integer accumulated) { return accumulated + value; } });
and(Matcher)	Combines two simple patterns in a stage with a logical operator AND. builder.where( (event) -> event.value() > 0).and((event) -> event.value() < 10)
or(Matcher)	Combines two simple patterns in a stage with a logical operator OR. builder.where( (event) -> event.value().equals("A")).or((event) -> event.value().equals("B")))
within(time, TimeUnit)	Defines a time window over the entire complex pattern. builder.within(1, TimeUnit.HOUR) });

Event Selection Strategies

Currently, this library supports the following event selection strategies :

• Strict contiguity : Selected events must be contiguous in the input stream. By default, this is selection strategy used to match complexes sequences.

This strategy can be chosen while defining a new stage :

builder.select("my-stage", Selected.withStrictContiguity())

• Skip till next match :Irrelevant events are skipped until an event matching the next pattern is encountered. If multiple events in the stream can match the next pattern only the first of them is selected.

This strategy can be chosen while defining a new stage :

builder.select("my-stage", Selected.withSkipTilNextMatch())

• Skip till any match : Irrelevant events are skipped until an event matching the next pattern is encountered. All events in the stream that can match a pattern are selected.

This strategy can be chosen while defining a new stage :

builder.select("my-stage", Selected.withSkipTilAnyMatch())

How to match an event from a specific topic ?

Kafka Streams allow you to consume from multiple input topics.

While defining a new stage you can specify the topic name on which the pattern should be applied :

builder.select("my-stage", Selected.fromTopic("my-topic")))

Demonstration

The below example is based on the research paper Efficient Pattern Matching over Event Streams.

Implementation based on https://people.cs.umass.edu/~yanlei/publications/sase-sigmod08.pdf

CEP Query :

     PATTERN SEQ(Stock+ a[ ], Stock b)
       WHERE skip_till_next_match(a[ ], b) {
           [symbol]
       and
           a[1].volume > 1000
       and
           a[i].price > avg(a[..i-1].price)
       and
           b.volume < 80%*a[a.LEN].volume }
       WITHIN 1 hour

Build Query

            new QueryBuilder<String, StockEvent>()
            .select("stage-1")
                .where((event, states) -> event.value().volume > 1000)
                .<Long>fold("avg", (k, v, curr) -> v.price)
                .then()
            .select("stage-2", Selected.withSkipTilNextMatch())
                .zeroOrMore()
                .where((event, states) -> event.value().price > (long) states.get("avg"))
                .<Long>fold("avg", (k, v, curr) -> (curr + v.price) / 2)
                .<Long>fold("volume", (k, v, curr) -> v.volume)
                .then()
            .select("stage-3", Selected.withSkipTilNextMatch())
                .where((event, states) -> event.value().volume < 0.8 * (long) states.getOrElse("volume", 0L))
            .within(1, TimeUnit.HOURS)
            .build();

KStreams API:

        Pattern<Object, StockEvent> pattern = ...

        ComplexStreamsBuilder builder = new ComplexStreamsBuilder();

        CEPStream<String, StockEvent> stream = builder.stream("StockEvents");
        KStream<String, Sequence<String, StockEvent>> stocks = stream.query("Stocks", pattern, Queried.with(Serdes.String(), new StockEventSerde()));
        
        KafkaStreams streams = new KafkaStreams(builder.build(), props);

Processor API:

        Pattern<Object, StockEvent> pattern = ...

        final String queryName = "Stocks";

        Topology topology = new Topology()
                .addSource("source", "StockEvents")
                .addProcessor("cep-processor", () -> new CEPProcessor<>(queryName, pattern), "source")
                .addSink("sink", "Matches", "cep-processor");

        // utility class to register all stores associated with the pattern.
        CEPStoreBuilders<String, StockEvent> builders = new CEPStoreBuilders<>();
        builders.addStateStores(topology, "cep-processor", queryName, pattern, Serdes.String(), new StockEventSerde());
        
        CEPStoreBuilders<K, V> storeBuilders = new CEPStoreBuilders<>(queryName, pattern);
        topology.addStateStore(storeBuilders.getEventBufferStoreBuilder(keySerde, valueSerde), "cep-processor");
        topology.addStateStore(storeBuilders.getNFAStateStoreBuilder(keySerde, valueSerde), "cep-processor");
        topology.addStateStore(storeBuilders.getAggregateStateStores(), "cep-processor");
        
        KafkaStreams streams = new KafkaStreams(topology, props);

States

To track the states of matched sequences, KafkaStreamsCEP needs to create three persistent stores for each query. This state stores will result in the creationf of the following changelog topics :

• <application_id>-<query_name>-streamscep-aggregates-changelog
• <application_id>-<query_name>-streamscep-matched-changelog
• <application_id>-<query_name>-streamscep-states-changelog

Demo

Run the demonstration class CEPStockDemo :

• Create Input/Ouput streams topics

./bin/kafka-topics --create --topic Stocks --partitions 2 --replication-factor 1 --zookeeper localhost:2181
Created topic "Stocks".

/bin/kafka-topics --create --topic Matches --partitions 2 --replication-factor 1 --zookeeper localhost:2181
Created topic "Matches".

• Produce the following json events StockEvents:

./bin/kafka-console-producer --topic Stocks  --property parse.key=true --property key.separator=, --broker-list localhost:9092

• Input

key1,{"name":"e1","price":100,"volume":1010}
key1,{"name":"e2","price":120,"volume":990}
key1,{"name":"e3","price":120,"volume":1005}
key1,{"name":"e4","price":121,"volume":999}
key1,{"name":"e5","price":120,"volume":999}
key1,{"name":"e6","price":125,"volume":750}
key1,{"name":"e7","price":120,"volume":950}
key1,{"name":"e8","price":120,"volume":700}

• Consume from the sink topic "Matches"

./bin/kafka-console-consumer --new-consumer --topic Matches --bootstrap-server localhost:9092

• Output

{"events":[{"name":"stage-1","events":["e1"]},{"name":"stage-2","events":["e2","e3","e4","e5"]},{"name":"stage-3","events":["e6"]}]}
{"events":[{"name":"stage-1","events":["e3"]},{"name":"stage-2","events":["e4"]},{"name":"stage-3","events":["e6"]}]}
{"events":[{"name":"stage-1","events":["e1"]},{"name":"stage-2","events":["e2","e3","e4","e5","e6","e7"]},{"name":"stage-3","events":["e8"]}]}
{"events":[{"name":"stage-1","events":["e3"]},{"name":"stage-2","events":["e4","e6"]},{"name":"stage-3","events":["e8"]}]}

TODO

• Improve test scenarios

Contributions

Any feedback, bug reports and PRs are greatly appreciated! See our guideline

Licence

Licensed to the Apache Software Foundation (ASF) under one or more contributor license agreements. See the NOTICE file distributed with this work for additional information regarding copyright ownership. The ASF licenses this file to you under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License