This guide describes how to implement custom data streamer and describes basic concepts and implementation details.
It is useful to take a look at:
SocketStreamer
- reference implementation of custom data streamer as TCP socket server.WordsSocketStreamerServer
and WordsSocketStreamerClient
- example of SocketStreamer
usage.The purpose of streamers is data ingesting from various sources (e.g. TCP socket or Kafka) and streaming them into Ignite cache for further processing. Ignite provides IgniteDataStreamer
API that allows stream data into cache and convenient StreamAdapter
that wraps data streamer instance and provides basis for custom streamer implementation.
Data stream consists of messages that can be represented as Java objects of some type that should be converted to cache entries (key-value pairs or tuples). StreamTupleExtractor
is responsible for such conversion.
The following parts of custom data streamer are missing and should be implemented by developer:
String
);From connection initiating stand point data streamer can be implemented as a server or a client with respect of data source and requirements. For example, HTTP data streamer implemented as a client can request the data from external web services. On the other hand, HTTP data streamer implemented as a server can process a large number of request from external HTTP clients.
In order to implement custom data streamer a developer should optionally extend StreamAdapter
class, add functionality related with particular data source and streamer life cycle logic if needed. As mentioned above
wraps StreamAdapter
IgniteDataStreamer
instance and also needs StreamTupleExtractor
instance, that could be provided via constructor or corresponding setters.
The central StreamAdapater
's method is addMessage(T msg)
that just converts message to cache entry using StreamTupleExtractor
and streams entry into cache. This method doesn't block current thread due to the nature of IgniteDataStreamer
.
StreamTupleExtractor
interface expose the extract(T msg)
method that returns a cache entry as Map.Entry<K, V>
instance. For example, if message represents a string with two values that separated by '=' symbol where left part is a word and right one - integer number, then extractor could be implemented as:
public class WordCountTupleExtractor implements StreamTupleExtractor<String, String, Integer> { @Override public Map.Entry<String, Integer> extract(String msg) { String[] tokens = msg.split("="); return new IgniteBiTuple<>(tokens[0], Integer.valueOf(tokens[1])); } }
Ignite provides SocketStreamer
- reference custom data streamer implementation. SocketStreamer
is NIO-based TCP socket server that accepts client connections on configured port, receives data represented as bytes array, converts it to user defined type using SocketMessageConverter
and streams it into cache.
The default SocketMessageConverter
implementation uses standard Java serialization/deserialization mechanism assuming that data stream contains serialized Java objects.
SocketStreamer
supports two communication protocols:
SocketStreamer.setDelimiter(byte[] delim)
method).Streamer life cycle depends on chosen implementation model (server or client) and requirements. Usually streamer can be in one of the following states: created, initialized, shutdown and may be one of transition states. StreamAdapter
doesn't provide any life cycle managements methods. Correct life cycle management implementation is completely responsibility of data streamer developer.
See SocketStreamer
implementation with life cycle management start
and stop
methods that initialize and shutdown TCP socket server respectively.