Interface ProcessorContext<KForward,VForward>

Type Parameters:
KForward - a bound on the types of keys that may be forwarded
VForward - a bound on the types of values that may be forwarded
All Known Implementing Classes:
MockProcessorContext

public interface ProcessorContext<KForward,VForward>
Processor context interface.
  • Method Details

    • applicationId

      String applicationId()
      Return the application id.
      Returns:
      the application id
    • taskId

      TaskId taskId()
      Return the task id.
      Returns:
      the task id
    • recordMetadata

      Optional<RecordMetadata> recordMetadata()
      Return the metadata of the current record if available. Processors may be invoked to process a source record from an input topic, to run a scheduled punctuation (see schedule(Duration, PunctuationType, Punctuator)), or because a parent processor called forward(Record).

      In the case of a punctuation, there is no source record, so this metadata would be undefined. Note that when a punctuator invokes forward(Record), downstream processors will receive the forwarded record as a regular Processor.process(Record) invocation. In other words, it wouldn't be apparent to downstream processors whether or not the record being processed came from an input topic or punctuation and therefore whether or not this metadata is defined. This is why the return type of this method is Optional.

      If there is any possibility of punctuators upstream, any access to this field should consider the case of "recordMetadata().isPresent() == false". Of course, it would be safest to always guard this condition.

    • keySerde

      Serde<?> keySerde()
      Return the default key serde.
      Returns:
      the key serializer
    • valueSerde

      Serde<?> valueSerde()
      Return the default value serde.
      Returns:
      the value serializer
    • stateDir

      File stateDir()
      Return the state directory for the partition.
      Returns:
      the state directory
    • metrics

      StreamsMetrics metrics()
      Return Metrics instance.
      Returns:
      StreamsMetrics
    • getStateStore

      <S extends StateStore> S getStateStore(String name)
      Get the state store given the store name.
      Type Parameters:
      S - The type or interface of the store to return
      Parameters:
      name - The store name
      Returns:
      The state store instance
      Throws:
      ClassCastException - if the return type isn't a type or interface of the actual returned store.
    • schedule

      Cancellable schedule(Duration interval, PunctuationType type, Punctuator callback)
      Schedule a periodic operation for processors. A processor may call this method during initialization or Processor.process(Record) processing} to schedule a periodic callback — called a punctuation — to Punctuator.punctuate(long). The type parameter controls what notion of time is used for punctuation:
      • PunctuationType.STREAM_TIME — uses "stream time", which is advanced by the processing of messages in accordance with the timestamp as extracted by the TimestampExtractor in use. The first punctuation will be triggered by the first record that is processed. NOTE: Only advanced if messages arrive
      • PunctuationType.WALL_CLOCK_TIME — uses system time (the wall-clock time), which is advanced independent of whether new messages arrive. The first punctuation will be triggered after interval has elapsed. NOTE: This is best effort only as its granularity is limited by how long an iteration of the processing loop takes to complete
      Skipping punctuations: Punctuations will not be triggered more than once at any given timestamp. This means that "missed" punctuation will be skipped. It's possible to "miss" a punctuation if:
      Parameters:
      interval - the time interval between punctuations (supported minimum is 1 millisecond)
      type - one of: PunctuationType.STREAM_TIME, PunctuationType.WALL_CLOCK_TIME
      callback - a function consuming timestamps representing the current stream or system time
      Returns:
      a handle allowing cancellation of the punctuation schedule established by this method
      Throws:
      IllegalArgumentException - if the interval is not representable in milliseconds
    • forward

      <K extends KForward, V extends VForward> void forward(Record<K,V> record)
      Forward a record to all child processors.

      Note that the forwarded Record is shared between the parent and child processors. And of course, the parent may forward the same object to multiple children, and the child may forward it to grandchildren, etc. Therefore, you should be mindful of mutability.

      The Record class itself is immutable (all the setter-style methods return an independent copy of the instance). However, the key, value, and headers referenced by the Record may themselves be mutable.

      Some programs may opt to make use of this mutability for high performance, in which case the input record may be mutated and then forwarded by each Processor. However, most applications should instead favor safety.

      Forwarding records safely simply means to make a copy of the record before you mutate it. This is trivial when using the Record.withKey(Object), Record.withValue(Object), and Record.withTimestamp(long) methods, as each of these methods make a copy of the record as a matter of course. But a little extra care must be taken with headers, since the Header class is mutable. The easiest way to safely handle headers is to use the Record constructors to make a copy before modifying headers.

      In other words, this would be considered unsafe: process(Record inputRecord) { inputRecord.headers().add(...); context.forward(inputRecord); } This is unsafe because the parent, and potentially siblings, grandparents, etc., all will see this modification to their shared Headers reference. This is a violation of causality and could lead to undefined behavior.

      A safe usage would look like this: process(Record inputRecord) { // makes a copy of the headers Record toForward = inputRecord.withHeaders(inputRecord.headers()); // Other options to create a safe copy are: // * use any copy-on-write method, which makes a copy of all fields: // toForward = inputRecord.withValue(); // * explicitly copy all fields: // toForward = new Record(inputRecord.key(), inputRecord.value(), inputRecord.timestamp(), inputRecord.headers()); // * create a fresh, empty Headers: // toForward = new Record(inputRecord.key(), inputRecord.value(), inputRecord.timestamp()); // * etc. // now, we are modifying our own independent copy of the headers. toForward.headers().add(...); context.forward(toForward); }

      Parameters:
      record - The record to forward to all children
    • forward

      <K extends KForward, V extends VForward> void forward(Record<K,V> record, String childName)
      Forward a record to the specified child processor. See forward(Record) for considerations.
      Parameters:
      record - The record to forward
      childName - The name of the child processor to receive the record
      See Also:
    • commit

      void commit()
      Request a commit.
    • appConfigs

      Map<String,Object> appConfigs()
      Returns all the application config properties as key/value pairs.

      The config properties are defined in the StreamsConfig object and associated to the ProcessorContext.

      The type of the values is dependent on the type of the property (e.g. the value of DEFAULT_KEY_SERDE_CLASS_CONFIG will be of type Class, even if it was specified as a String to StreamsConfig(Map)).

      Returns:
      all the key/values from the StreamsConfig properties
    • appConfigsWithPrefix

      Map<String,Object> appConfigsWithPrefix(String prefix)
      Return all the application config properties with the given key prefix, as key/value pairs stripping the prefix.

      The config properties are defined in the StreamsConfig object and associated to the ProcessorContext.

      Parameters:
      prefix - the properties prefix
      Returns:
      the key/values matching the given prefix from the StreamsConfig properties.
    • currentSystemTimeMs

      long currentSystemTimeMs()
      Return the current system timestamp (also called wall-clock time) in milliseconds.

      Note: this method returns the internally cached system timestamp from the Kafka Stream runtime. Thus, it may return a different value compared to System.currentTimeMillis().

      Returns:
      the current system timestamp in milliseconds
    • currentStreamTimeMs

      long currentStreamTimeMs()
      Return the current stream-time in milliseconds.

      Stream-time is the maximum observed record timestamp so far (including the currently processed record), i.e., it can be considered a high-watermark. Stream-time is tracked on a per-task basis and is preserved across restarts and during task migration.

      Note: this method is not supported for global processors (cf. Topology.addGlobalStore(org.apache.kafka.streams.state.StoreBuilder<?>, java.lang.String, org.apache.kafka.common.serialization.Deserializer<K>, org.apache.kafka.common.serialization.Deserializer<V>, java.lang.String, java.lang.String, org.apache.kafka.streams.processor.ProcessorSupplier<K, V>) (...) and StreamsBuilder.addGlobalStore(org.apache.kafka.streams.state.StoreBuilder<?>, java.lang.String, org.apache.kafka.streams.kstream.Consumed<K, V>, org.apache.kafka.streams.processor.ProcessorSupplier<K, V>) (...), because there is no concept of stream-time for this case. Calling this method in a global processor will result in an UnsupportedOperationException.

      Returns:
      the current stream-time in milliseconds