1. Application Development
2. Libraries
3. Event Processing (CEP)

FlinkCEP - Complex event processing for Flink

FlinkCEP is the Complex Event Processing (CEP) library implemented on top of Flink. It allows you to detect event patterns in an endless stream of events, giving you the opportunity to get hold of what’s important in your data.

This page describes the API calls available in Flink CEP. We start by presenting the Pattern API, which allows you to specify the patterns that you want to detect in your stream, before presenting how you can detect and act upon matching event sequences. We then present the assumptions the CEP library makes when dealing with lateness in event time and how you can migrate your job from an older Flink version to Flink-1.3.

• Getting Started
• The Pattern API
  • Individual Patterns
  • Combining Patterns
  • After Match Skip Strategy
• Detecting Patterns
  • Selecting from Patterns
  • Handling Timed Out Partial Patterns
• Handling Lateness in Event Time
• Examples
• Migrating from an older Flink version

Getting Started

If you want to jump right in, set up a Flink program and add the FlinkCEP dependency to the pom.xml of your project.

<dependency>
  <groupId>org.apache.flink</groupId>
  <artifactId>flink-cep_2.11</artifactId>
  <version>1.4.2</version>
</dependency>

<dependency>
  <groupId>org.apache.flink</groupId>
  <artifactId>flink-cep-scala_2.11</artifactId>
  <version>1.4.2</version>
</dependency>

Info FlinkCEP is not part of the binary distribution. See how to link with it for cluster execution here.

Now you can start writing your first CEP program using the Pattern API.

Attention The events in the DataStream to which you want to apply pattern matching must implement proper equals() and hashCode() methods because FlinkCEP uses them for comparing and matching events.

DataStream<Event> input = ...

Pattern<Event, ?> pattern = Pattern.<Event>begin("start").where(
        new SimpleCondition<Event>() {
            @Override
            public boolean filter(Event event) {
                return event.getId() == 42;
            }
        }
    ).next("middle").subtype(SubEvent.class).where(
        new SimpleCondition<Event>() {
            @Override
            public boolean filter(SubEvent subEvent) {
                return subEvent.getVolume() >= 10.0;
            }
        }
    ).followedBy("end").where(
         new SimpleCondition<Event>() {
            @Override
            public boolean filter(Event event) {
                return event.getName().equals("end");
            }
         }
    );

PatternStream<Event> patternStream = CEP.pattern(input, pattern);

DataStream<Alert> result = patternStream.select(
    new PatternSelectFunction<Event, Alert> {
        @Override
        public Alert select(Map<String, List<Event>> pattern) throws Exception {
            return createAlertFrom(pattern);
        }
    }
});

val input: DataStream[Event] = ...

val pattern = Pattern.begin("start").where(_.getId == 42)
  .next("middle").subtype(classOf[SubEvent]).where(_.getVolume >= 10.0)
  .followedBy("end").where(_.getName == "end")

val patternStream = CEP.pattern(input, pattern)

val result: DataStream[Alert] = patternStream.select(createAlert(_))

The Pattern API

The pattern API allows you to define complex pattern sequences that you want to extract from your input stream.

Each complex pattern sequence consists of multiple simple patterns, i.e. patterns looking for individual events with the same properties. From now on, we will call these simple patterns patterns, and the final complex pattern sequence we are searching for in the stream, the pattern sequence. You can see a pattern sequence as a graph of such patterns, where transitions from one pattern to the next occur based on user-specified conditions, e.g. event.getName().equals("start"). A match is a sequence of input events which visits all patterns of the complex pattern graph, through a sequence of valid pattern transitions.

Attention Each pattern must have a unique name, which you use later to identify the matched events.

Attention Pattern names CANNOT contain the character ":".

In the rest of this section we will first describe how to define Individual Patterns, and then how you can combine individual patterns into Complex Patterns.

Individual Patterns

A Pattern can be either a singleton or a looping pattern. Singleton patterns accept a single event, while looping patterns can accept more than one. In pattern matching symbols, the pattern "a b+ c? d" (or "a", followed by one or more "b"’s, optionally followed by a "c", followed by a "d"), a, c?, and d are singleton patterns, while b+ is a looping one. By default, a pattern is a singleton pattern and you can transform it to a looping one by using Quantifiers. Each pattern can have one or more Conditions based on which it accepts events.

Quantifiers

In FlinkCEP, you can specify looping patterns using these methods: pattern.oneOrMore(), for patterns that expect one or more occurrences of a given event (e.g. the b+ mentioned before); and pattern.times(#ofTimes), for patterns that expect a specific number of occurrences of a given type of event, e.g. 4 a’s; and pattern.times(#fromTimes, #toTimes), for patterns that expect a specific minimum number of occurrences and a maximum number of occurrences of a given type of event, e.g. 2-4 as.

You can make looping patterns greedy using the pattern.greedy() method, but you cannot yet make group patterns greedy. You can make all patterns, looping or not, optional using the pattern.optional() method.

For a pattern named start, the following are valid quantifiers:

 // expecting 4 occurrences
 start.times(4);

 // expecting 0 or 4 occurrences
 start.times(4).optional();

 // expecting 2, 3 or 4 occurrences
 start.times(2, 4);

 // expecting 2, 3 or 4 occurrences and repeating as many as possible
 start.times(2, 4).greedy();

 // expecting 0, 2, 3 or 4 occurrences
 start.times(2, 4).optional();

 // expecting 0, 2, 3 or 4 occurrences and repeating as many as possible
 start.times(2, 4).optional().greedy();

 // expecting 1 or more occurrences
 start.oneOrMore();

 // expecting 1 or more occurrences and repeating as many as possible
 start.oneOrMore().greedy();

 // expecting 0 or more occurrences
 start.oneOrMore().optional();

 // expecting 0 or more occurrences and repeating as many as possible
 start.oneOrMore().optional().greedy();

 // expecting 2 or more occurrences
 start.timesOrMore(2);

 // expecting 2 or more occurrences and repeating as many as possible
 start.timesOrMore(2).greedy();

 // expecting 0, 2 or more occurrences and repeating as many as possible
 start.timesOrMore(2).optional().greedy();
 

 // expecting 4 occurrences
 start.times(4)

 // expecting 0 or 4 occurrences
 start.times(4).optional()

 // expecting 2, 3 or 4 occurrences
 start.times(2, 4)

 // expecting 2, 3 or 4 occurrences and repeating as many as possible
 start.times(2, 4).greedy()

 // expecting 0, 2, 3 or 4 occurrences
 start.times(2, 4).optional()

 // expecting 0, 2, 3 or 4 occurrences and repeating as many as possible
 start.times(2, 4).optional().greedy()

 // expecting 1 or more occurrences
 start.oneOrMore()

 // expecting 1 or more occurrences and repeating as many as possible
 start.oneOrMore().greedy()

 // expecting 0 or more occurrences
 start.oneOrMore().optional()

 // expecting 0 or more occurrences and repeating as many as possible
 start.oneOrMore().optional().greedy()

 // expecting 2 or more occurrences
 start.timesOrMore(2)

 // expecting 2 or more occurrences and repeating as many as possible
 start.timesOrMore(2).greedy()

 // expecting 0, 2 or more occurrences
 start.timesOrMore(2).optional()

 // expecting 0, 2 or more occurrences and repeating as many as possible
 start.timesOrMore(2).optional().greedy()
 

Conditions

At every pattern, and to go from one pattern to the next, you can specify additional conditions. You can relate these conditions to:

1. A property of the incoming event, e.g. its value should be larger than 5, or larger than the average value of the previously accepted events.

2. The contiguity of the matching events, e.g. detect pattern a,b,c without non-matching events between any matching ones.

The latter refers to “looping” patterns, i.e. patterns that can accept more than one event, e.g. the b+ in a b+ c, which searches for one or more b’s.

Conditions on Properties

You can specify conditions on the event properties via the pattern.where(), pattern.or() or the pattern.until() method. These can be either IterativeConditions or SimpleConditions.

Iterative Conditions: This is the most general type of condition. This is how you can specify a condition that accepts subsequent events based on properties of the previously accepted events or a statistic over a subset of them.

Below is the code for an iterative condition that accepts the next event for a pattern named “middle” if its name starts with “foo”, and if the sum of the prices of the previously accepted events for that pattern plus the price of the current event do not exceed the value of 5.0. Iterative conditions can be powerful, especially in combination with looping patterns, e.g. oneOrMore().

middle.oneOrMore().where(new IterativeCondition<SubEvent>() {
    @Override
    public boolean filter(SubEvent value, Context<SubEvent> ctx) throws Exception {
        if (!value.getName().startsWith("foo")) {
            return false;
        }

        double sum = value.getPrice();
        for (Event event : ctx.getEventsForPattern("middle")) {
            sum += event.getPrice();
        }
        return Double.compare(sum, 5.0) < 0;
    }
});

middle.oneOrMore().where(
    (value, ctx) => {
        lazy val sum = ctx.getEventsForPattern("middle").asScala.map(_.getPrice).sum
        value.getName.startsWith("foo") && sum + value.getPrice < 5.0
    }
)

Attention The call to context.getEventsForPattern(...) finds all the previously accepted events for a given potential match. The cost of this operation can vary, so when implementing your condition, try to minimize its use.

Simple Conditions: This type of condition extends the aforementioned IterativeCondition class and decides whether to accept an event or not, based only on properties of the event itself.

start.where(new SimpleCondition<Event>() {
    @Override
    public boolean filter(Event value) {
        return value.getName().startsWith("foo");
    }
});

start.where(event => event.getName.startsWith("foo"))

Finally, you can also restrict the type of the accepted event to a subtype of the initial event type (here Event) via the pattern.subtype(subClass) method.

start.subtype(SubEvent.class).where(new SimpleCondition<SubEvent>() {
    @Override
    public boolean filter(SubEvent value) {
        return ... // some condition
    }
});

start.subtype(classOf[SubEvent]).where(subEvent => ... /* some condition */)

Combining Conditions: As shown above, you can combine the subtype condition with additional conditions. This holds for every condition. You can arbitrarily combine conditions by sequentially calling where(). The final result will be the logical AND of the results of the individual conditions. To combine conditions using OR, you can use the or() method, as shown below.

pattern.where(new SimpleCondition<Event>() {
    @Override
    public boolean filter(Event value) {
        return ... // some condition
    }
}).or(new SimpleCondition<Event>() {
    @Override
    public boolean filter(Event value) {
        return ... // or condition
    }
});

pattern.where(event => ... /* some condition */).or(event => ... /* or condition */)

Stop condition: In case of looping patterns (oneOrMore() and oneOrMore().optional()) you can also specify a stop condition, e.g. accept events with value larger than 5 until the sum of values is smaller than 50.

To better understand it, have a look at the following example. Given

• pattern like "(a+ until b)" (one or more "a" until "b")

• a sequence of incoming events "a1" "c" "a2" "b" "a3"

• the library will output results: {a1 a2} {a1} {a2} {a3}.

As you can see {a1 a2 a3} or {a2 a3} are not returned due to the stop condition.

Conditions on Contiguity

FlinkCEP supports the following forms of contiguity between events:

1. Strict Contiguity: Expects all matching events to appear strictly one after the other, without any non-matching events in-between.

2. Relaxed Contiguity: Ignores non-matching events appearing in-between the matching ones.

3. Non-Deterministic Relaxed Contiguity: Further relaxes contiguity, allowing additional matches that ignore some matching events.

To illustrate the above with an example, a pattern sequence "a+ b" (one or more "a"’s followed by a "b") with input "a1", "c", "a2", "b" will have the following results:

1. Strict Contiguity: {a2 b} – the "c" after "a1" causes "a1" to be discarded.

2. Relaxed Contiguity: {a1 b} and {a1 a2 b} – c is ignored.

3. Non-Deterministic Relaxed Contiguity: {a1 b}, {a2 b}, and {a1 a2 b}.

For looping patterns (e.g. oneOrMore() and times()) the default is relaxed contiguity. If you want strict contiguity, you have to explicitly specify it by using the consecutive() call, and if you want non-deterministic relaxed contiguity you can use the allowCombinations() call.

Attention In this section we are talking about contiguity within a single looping pattern, and the consecutive() and allowCombinations() calls need to be understood in that context. Later when looking at Combining Patterns we’ll discuss other calls, such as next() and followedBy(), that are used to specify contiguity conditions between patterns.

pattern.where(new IterativeCondition<Event>() {
    @Override
    public boolean filter(Event value, Context ctx) throws Exception {
        return ... // some condition
    }
});

pattern.where(new IterativeCondition<Event>() {
    @Override
    public boolean filter(Event value, Context ctx) throws Exception {
        return ... // some condition
    }
}).or(new IterativeCondition<Event>() {
    @Override
    public boolean filter(Event value, Context ctx) throws Exception {
        return ... // alternative condition
    }
});

pattern.oneOrMore().until(new IterativeCondition<Event>() {
    @Override
    public boolean filter(Event value, Context ctx) throws Exception {
        return ... // alternative condition
    }
});

pattern.subtype(SubEvent.class);

pattern.oneOrMore();

pattern.timesOrMore(2);

pattern.times(2);

pattern.times(2, 4);

pattern.oneOrMore().optional();

pattern.oneOrMore().greedy();

Pattern.<Event>begin("start").where(new SimpleCondition<Event>() {
  @Override
  public boolean filter(Event value) throws Exception {
    return value.getName().equals("c");
  }
})
.followedBy("middle").where(new SimpleCondition<Event>() {
  @Override
  public boolean filter(Event value) throws Exception {
    return value.getName().equals("a");
  }
}).oneOrMore().consecutive()
.followedBy("end1").where(new SimpleCondition<Event>() {
  @Override
  public boolean filter(Event value) throws Exception {
    return value.getName().equals("b");
  }
});

Pattern.<Event>begin("start").where(new SimpleCondition<Event>() {
  @Override
  public boolean filter(Event value) throws Exception {
    return value.getName().equals("c");
  }
})
.followedBy("middle").where(new SimpleCondition<Event>() {
  @Override
  public boolean filter(Event value) throws Exception {
    return value.getName().equals("a");
  }
}).oneOrMore().allowCombinations()
.followedBy("end1").where(new SimpleCondition<Event>() {
  @Override
  public boolean filter(Event value) throws Exception {
    return value.getName().equals("b");
  }
});

pattern.where(event => ... /* some condition */)

pattern.where(event => ... /* some condition */)
    .or(event => ... /* alternative condition */)

pattern.oneOrMore().until(event => ... /* some condition */)

pattern.subtype(classOf[SubEvent])

pattern.oneOrMore()

pattern.timesOrMore(2)

pattern.times(2)

pattern.times(2, 4)

pattern.oneOrMore().optional()

pattern.oneOrMore().greedy()

Pattern.begin("start").where(_.getName().equals("c"))
  .followedBy("middle").where(_.getName().equals("a"))
                       .oneOrMore().consecutive()
  .followedBy("end1").where(_.getName().equals("b"))

Pattern.begin("start").where(_.getName().equals("c"))
  .followedBy("middle").where(_.getName().equals("a"))
                       .oneOrMore().allowCombinations()
  .followedBy("end1").where(_.getName().equals("b"))

Combining Patterns

Now that you’ve seen what an individual pattern can look like, it is time to see how to combine them into a full pattern sequence.

A pattern sequence has to start with an initial pattern, as shown below:

Pattern<Event, ?> start = Pattern.<Event>begin("start");

val start : Pattern[Event, _] = Pattern.begin("start")

Next, you can append more patterns to your pattern sequence by specifying the desired contiguity conditions between them. In the previous section we described the different contiguity modes supported by Flink, namely strict, relaxed, and non-deterministic relaxed, and how to apply them in looping patterns. To apply them between consecutive patterns, you can use:

1. next(), for strict,
2. followedBy(), for relaxed, and
3. followedByAny(), for non-deterministic relaxed contiguity.

or

1. notNext(), if you do not want an event type to directly follow another
2. notFollowedBy(), if you do not want an event type to be anywhere between two other event types

Attention A pattern sequence cannot end in notFollowedBy().

Attention A NOT pattern cannot be preceded by an optional one.

// strict contiguity
Pattern<Event, ?> strict = start.next("middle").where(...);

// relaxed contiguity
Pattern<Event, ?> relaxed = start.followedBy("middle").where(...);

// non-deterministic relaxed contiguity
Pattern<Event, ?> nonDetermin = start.followedByAny("middle").where(...);

// NOT pattern with strict contiguity
Pattern<Event, ?> strictNot = start.notNext("not").where(...);

// NOT pattern with relaxed contiguity
Pattern<Event, ?> relaxedNot = start.notFollowedBy("not").where(...);

// strict contiguity
val strict: Pattern[Event, _] = start.next("middle").where(...)

// relaxed contiguity
val relaxed: Pattern[Event, _] = start.followedBy("middle").where(...)

// non-deterministic relaxed contiguity
val nonDetermin: Pattern[Event, _] = start.followedByAny("middle").where(...)

// NOT pattern with strict contiguity
val strictNot: Pattern[Event, _] = start.notNext("not").where(...)

// NOT pattern with relaxed contiguity
val relaxedNot: Pattern[Event, _] = start.notFollowedBy("not").where(...)

Relaxed contiguity means that only the first succeeding matching event will be matched, while with non-deterministic relaxed contiguity, multiple matches will be emitted for the same beginning. As an example, a pattern a b, given the event sequence "a", "c", "b1", "b2", will give the following results:

1. Strict Contiguity between a and b: {} (no match), the "c" after "a" causes "a" to be discarded.

2. Relaxed Contiguity between a and b: {a b1}, as relaxed continuity is viewed as “skip non-matching events till the next matching one”.

3. Non-Deterministic Relaxed Contiguity between a and b: {a b1}, {a b2}, as this is the most general form.

It’s also possible to define a temporal constraint for the pattern to be valid. For example, you can define that a pattern should occur within 10 seconds via the pattern.within() method. Temporal patterns are supported for both processing and event time.

Attention A pattern sequence can only have one temporal constraint. If multiple such constraints are defined on different individual patterns, then the smallest is applied.

next.within(Time.seconds(10));

next.within(Time.seconds(10))

It’s also possible to define a pattern sequence as the condition for begin, followedBy, followedByAny and next. The pattern sequence will be considered as the matching condition logically and a GroupPattern will be returned and it is possible to apply oneOrMore(), times(#ofTimes), times(#fromTimes, #toTimes), optional(), consecutive(), allowCombinations() to the GroupPattern.

Pattern<Event, ?> start = Pattern.begin(
    Pattern.<Event>begin("start").where(...).followedBy("start_middle").where(...)
);

// strict contiguity
Pattern<Event, ?> strict = start.next(
    Pattern.<Event>begin("next_start").where(...).followedBy("next_middle").where(...)
).times(3);

// relaxed contiguity
Pattern<Event, ?> relaxed = start.followedBy(
    Pattern.<Event>begin("followedby_start").where(...).followedBy("followedby_middle").where(...)
).oneOrMore();

// non-deterministic relaxed contiguity
Pattern<Event, ?> nonDetermin = start.followedByAny(
    Pattern.<Event>begin("followedbyany_start").where(...).followedBy("followedbyany_middle").where(...)
).optional();

val start: Pattern[Event, _] = Pattern.begin(
    Pattern.begin[Event, _]("start").where(...).followedBy("start_middle").where(...)
)

// strict contiguity
val strict: Pattern[Event, _] = start.next(
    Pattern.begin[Event, _]("next_start").where(...).followedBy("next_middle").where(...)
).times(3)

// relaxed contiguity
val relaxed: Pattern[Event, _] = start.followedBy(
    Pattern.begin[Event, _]("followedby_start").where(...).followedBy("followedby_middle").where(...)
).oneOrMore()

// non-deterministic relaxed contiguity
val nonDetermin: Pattern[Event, _] = start.followedByAny(
    Pattern.begin[Event, _]("followedbyany_start").where(...).followedBy("followedbyany_middle").where(...)
).optional()

Pattern<Event, ?> start = Pattern.<Event>begin("start");

Pattern<Event, ?> start = Pattern.<Event>begin(
    Pattern.<Event>begin("start").where(...).followedBy("middle").where(...)
);

Pattern<Event, ?> next = start.next("middle");

Pattern<Event, ?> next = start.next(
    Pattern.<Event>begin("start").where(...).followedBy("middle").where(...)
);

Pattern<Event, ?> followedBy = start.followedBy("middle");

Pattern<Event, ?> followedBy = start.followedBy(
    Pattern.<Event>begin("start").where(...).followedBy("middle").where(...)
);

Pattern<Event, ?> followedByAny = start.followedByAny("middle");

Pattern<Event, ?> followedByAny = start.followedByAny(
    Pattern.<Event>begin("start").where(...).followedBy("middle").where(...)
);

Pattern<Event, ?> notNext = start.notNext("not");

Pattern<Event, ?> notFollowedBy = start.notFollowedBy("not");

pattern.within(Time.seconds(10));

val start = Pattern.begin[Event]("start")

val next = start.next("middle")

val next = start.next(
    Pattern.begin[Event]("start").where(...).followedBy("middle").where(...)
)

val followedBy = start.followedBy("middle")

val followedBy = start.followedBy(
    Pattern.begin[Event]("start").where(...).followedBy("middle").where(...)
)

val followedByAny = start.followedByAny("middle")

val followedByAny = start.followedByAny(
    Pattern.begin[Event]("start").where(...).followedBy("middle").where(...)
)

val notNext = start.notNext("not")

val notFollowedBy = start.notFollowedBy("not")

pattern.within(Time.seconds(10))

After Match Skip Strategy

For a given pattern, the same event may be assigned to multiple successful matches. To control to how many matches an event will be assigned, you need to specify the skip strategy called AfterMatchSkipStrategy. There are four types of skip strategies, listed as follows:

• NO_SKIP: Every possible match will be emitted.
• SKIP_PAST_LAST_EVENT: Discards every partial match that contains event of the match.
• SKIP_TO_FIRST: Discards every partial match that contains event of the match preceding the first of PatternName.
• SKIP_TO_LAST: Discards every partial match that contains event of the match preceding the last of PatternName.

Notice that when using SKIP_TO_FIRST and SKIP_TO_LAST skip strategy, a valid PatternName should also be specified.

For example, for a given pattern a b{2} and a data stream ab1, ab2, ab3, ab4, ab5, ab6, the differences between these four skip strategies are as follows:

To specify which skip strategy to use, just create an AfterMatchSkipStrategy by calling:

Then apply the skip strategy to a pattern by calling:

AfterMatchSkipStrategy skipStrategy = ...
Pattern.begin("patternName", skipStrategy);

val skipStrategy = ...
Pattern.begin("patternName", skipStrategy)

Detecting Patterns

After specifying the pattern sequence you are looking for, it is time to apply it to your input stream to detect potential matches. To run a stream of events against your pattern sequence, you have to create a PatternStream. Given an input stream input, a pattern pattern and an optional comparator comparator used to sort events with the same timestamp in case of EventTime or that arrived at the same moment, you create the PatternStream by calling:

DataStream<Event> input = ...
Pattern<Event, ?> pattern = ...
EventComparator<Event> comparator = ... // optional

PatternStream<Event> patternStream = CEP.pattern(input, pattern, comparator);

val input : DataStream[Event] = ...
val pattern : Pattern[Event, _] = ...
var comparator : EventComparator[Event] = ... // optional

val patternStream: PatternStream[Event] = CEP.pattern(input, pattern, comparator)

The input stream can be keyed or non-keyed depending on your use-case.

Attention Applying your pattern on a non-keyed stream will result in a job with parallelism equal to 1.

Selecting from Patterns

Once you have obtained a PatternStream you can select from detected event sequences via the select or flatSelect methods.

class MyPatternSelectFunction<IN, OUT> implements PatternSelectFunction<IN, OUT> {
    @Override
    public OUT select(Map<String, List<IN>> pattern) {
        IN startEvent = pattern.get("start").get(0);
        IN endEvent = pattern.get("end").get(0);
        return new OUT(startEvent, endEvent);
    }
}

class MyPatternFlatSelectFunction<IN, OUT> implements PatternFlatSelectFunction<IN, OUT> {
    @Override
    public void flatSelect(Map<String, List<IN>> pattern, Collector<OUT> collector) {
        IN startEvent = pattern.get("start").get(0);
        IN endEvent = pattern.get("end").get(0);

        for (int i = 0; i < startEvent.getValue(); i++ ) {
            collector.collect(new OUT(startEvent, endEvent));
        }
    }
}

def selectFn(pattern : Map[String, Iterable[IN]]): OUT = {
    val startEvent = pattern.get("start").get.next
    val endEvent = pattern.get("end").get.next
    OUT(startEvent, endEvent)
}

def flatSelectFn(pattern : Map[String, Iterable[IN]], collector : Collector[OUT]) = {
    val startEvent = pattern.get("start").get.next
    val endEvent = pattern.get("end").get.next
    for (i <- 0 to startEvent.getValue) {
        collector.collect(OUT(startEvent, endEvent))
    }
}

Handling Timed Out Partial Patterns

Whenever a pattern has a window length attached via the within keyword, it is possible that partial event sequences are discarded because they exceed the window length. To react to these timed out partial matches the select and flatSelect API calls allow you to specify a timeout handler. This timeout handler is called for each timed out partial event sequence. The timeout handler receives all the events that have been matched so far by the pattern, and the timestamp when the timeout was detected.

To treat partial patterns, the select and flatSelect API calls offer an overloaded version which takes as parameters

• PatternTimeoutFunction/PatternFlatTimeoutFunction
• OutputTag for the side output in which the timed out matches will be returned
• and the known PatternSelectFunction/PatternFlatSelectFunction.

PatternStream<Event> patternStream = CEP.pattern(input, pattern);

OutputTag<String> outputTag = new OutputTag<String>("side-output"){};

SingleOutputStreamOperator<ComplexEvent> result = patternStream.select(
    new PatternTimeoutFunction<Event, TimeoutEvent>() {...},
    outputTag,
    new PatternSelectFunction<Event, ComplexEvent>() {...}
);

DataStream<TimeoutEvent> timeoutResult = result.getSideOutput(outputTag);

SingleOutputStreamOperator<ComplexEvent> flatResult = patternStream.flatSelect(
    new PatternFlatTimeoutFunction<Event, TimeoutEvent>() {...},
    outputTag,
    new PatternFlatSelectFunction<Event, ComplexEvent>() {...}
);

DataStream<TimeoutEvent> timeoutFlatResult = flatResult.getSideOutput(outputTag);

val patternStream: PatternStream[Event] = CEP.pattern(input, pattern)

val outputTag = OutputTag[String]("side-output")

val result: SingleOutputStreamOperator[ComplexEvent] = patternStream.select(outputTag){
    (pattern: Map[String, Iterable[Event]], timestamp: Long) => TimeoutEvent()
} {
    pattern: Map[String, Iterable[Event]] => ComplexEvent()
}

val timeoutResult: DataStream<TimeoutEvent> = result.getSideOutput(outputTag)

val patternStream: PatternStream[Event] = CEP.pattern(input, pattern)

val outputTag = OutputTag[String]("side-output")

val result: SingleOutputStreamOperator[ComplexEvent] = patternStream.flatSelect(outputTag){
    (pattern: Map[String, Iterable[Event]], timestamp: Long, out: Collector[TimeoutEvent]) =>
        out.collect(TimeoutEvent())
} {
    (pattern: mutable.Map[String, Iterable[Event]], out: Collector[ComplexEvent]) =>
        out.collect(ComplexEvent())
}

val timeoutResult: DataStream<TimeoutEvent> = result.getSideOutput(outputTag)

Handling Lateness in Event Time

In CEP the order in which elements are processed matters. To guarantee that elements are processed in the correct order when working in event time, an incoming element is initially put in a buffer where elements are sorted in ascending order based on their timestamp, and when a watermark arrives, all the elements in this buffer with timestamps smaller than that of the watermark are processed. This implies that elements between watermarks are processed in event-time order.

Attention The library assumes correctness of the watermark when working in event time.

To guarantee that elements across watermarks are processed in event-time order, Flink’s CEP library assumes correctness of the watermark, and considers as late elements whose timestamp is smaller than that of the last seen watermark. Late elements are not further processed.

Examples

The following example detects the pattern start, middle(name = "error") -> end(name = "critical") on a keyed data stream of Events. The events are keyed by their ids and a valid pattern has to occur within 10 seconds. The whole processing is done with event time.

StreamExecutionEnvironment env = ...
env.setStreamTimeCharacteristic(TimeCharacteristic.EventTime);

DataStream<Event> input = ...

DataStream<Event> partitionedInput = input.keyBy(new KeySelector<Event, Integer>() {
	@Override
	public Integer getKey(Event value) throws Exception {
		return value.getId();
	}
});

Pattern<Event, ?> pattern = Pattern.<Event>begin("start")
	.next("middle").where(new SimpleCondition<Event>() {
		@Override
		public boolean filter(Event value) throws Exception {
			return value.getName().equals("error");
		}
	}).followedBy("end").where(new SimpleCondition<Event>() {
		@Override
		public boolean filter(Event value) throws Exception {
			return value.getName().equals("critical");
		}
	}).within(Time.seconds(10));

PatternStream<Event> patternStream = CEP.pattern(partitionedInput, pattern);

DataStream<Alert> alerts = patternStream.select(new PatternSelectFunction<Event, Alert>() {
	@Override
	public Alert select(Map<String, List<Event>> pattern) throws Exception {
		return createAlert(pattern);
	}
});

val env : StreamExecutionEnvironment = ...
env.setStreamTimeCharacteristic(TimeCharacteristic.EventTime)

val input : DataStream[Event] = ...

val partitionedInput = input.keyBy(event => event.getId)

val pattern = Pattern.begin("start")
  .next("middle").where(_.getName == "error")
  .followedBy("end").where(_.getName == "critical")
  .within(Time.seconds(10))

val patternStream = CEP.pattern(partitionedInput, pattern)

val alerts = patternStream.select(createAlert(_)))

Migrating from an older Flink version

The CEP library in Flink-1.3 ships with a number of new features which have led to some changes in the API. Here we describe the changes that you need to make to your old CEP jobs, in order to be able to run them with Flink-1.3. After making these changes and recompiling your job, you will be able to resume its execution from a savepoint taken with the old version of your job, i.e. without having to re-process your past data.

The changes required are:

1. Change your conditions (the ones in the where(...) clause) to extend the SimpleCondition class instead of implementing the FilterFunction interface.

2. Change your functions provided as arguments to the select(...) and flatSelect(...) methods to expect a list of events associated with each pattern (List in Java, Iterable in Scala). This is because with the addition of the looping patterns, multiple input events can match a single (looping) pattern.

3. The followedBy() in Flink 1.1 and 1.2 implied non-deterministic relaxed contiguity (see here). In Flink 1.3 this has changed and followedBy() implies relaxed contiguity, while followedByAny() should be used if non-deterministic relaxed contiguity is required.

Back to top