最近一直在看 StreamGraph 生成的源码,刚好有点思路,准备动手了发现,
如果不说下 Transformation 后面的 StreamGraph 会差比较多意思,
所以先做点铺垫。
## Transformation
Transformation 类是 Flink 转换算子的基类,实现类有下面这些
AbstractMultipleInputTransformation
CoFeedbackTransformation
FeedbackTransformation
KeyedMultipleInputTransformation
LegacySourceTransformation
MultipleInputTransformation
OneInputTransformation
PartitionTransformation
PhysicalTransformation
SelectTransformation
SideOutputTransformation
SinkTransformation
SourceTransformation
SplitTransformation
TwoInputTransformation
UnionTransformation
类图:
从这些 Transformation 中也可以看出Flink 支持的转换类型: Source、Sink、一个输入、两个输入、多个输入、Union、侧输出、Select、分区 等转换操作
## source Transformation 的起始
env.addSource(new SimpleStringSource)
调用 StreamExecutionEnvironment.scala 的 addSource 方法
def addSource[T: TypeInformation](function: SourceFunction[T]): DataStream[T] = { require(function != null, "Function must not be null.") val cleanFun = scalaClean(function) val typeInfo = implicitly[TypeInformation[T]] asScalaStream(javaEnv.addSource(cleanFun, typeInfo)) }
然后调用 javaEnv.addSource 方法
StreamExecutionEnvironment.java
public <OUT> DataStreamSource<OUT> addSource(SourceFunction<OUT> function, TypeInformation<OUT> typeInfo) { return addSource(function, "Custom Source", typeInfo); } public <OUT> DataStreamSource<OUT> addSource(SourceFunction<OUT> function, String sourceName, TypeInformation<OUT> typeInfo) { TypeInformation<OUT> resolvedTypeInfo = getTypeInfo(function, sourceName, SourceFunction.class, typeInfo); boolean isParallel = function instanceof ParallelSourceFunction; clean(function); // 创建 StreamSource final StreamSource<OUT, ?> sourceOperator = new StreamSource<>(function); // 使用 StreamSource 创建 DataStreamSource 同时创建 Source 的Transformation 了, this 指 env return new DataStreamSource<>(this, resolvedTypeInfo, sourceOperator, isParallel, sourceName); }
DataStreamSource.java 使用输入的 sourceName, operator, outTypeInfo, Parallelism 创建 LegacySourceTransformation
public DataStreamSource( StreamExecutionEnvironment environment, TypeInformation<T> outTypeInfo, StreamSource<T, ?> operator, boolean isParallel, String sourceName) { super(environment, new LegacySourceTransformation<>(sourceName, operator, outTypeInfo, environment.getParallelism())); this.isParallel = isParallel; if (!isParallel) { setParallelism(1); } }
最终调用到 DataStream.java 的 DataStream 方法,将生成的 LegacySourceTransformation 放入到 DataStream 中
public DataStream(StreamExecutionEnvironment environment, Transformation<T> transformation) { this.environment = Preconditions.checkNotNull(environment, "Execution Environment must not be null."); this.transformation = Preconditions.checkNotNull(transformation, "Stream Transformation must not be null."); }
addSource 返回一个 DataStreamSource ,transformation 是 LegacySourceTransformation,并携带 StreamExecutionEnvironment 对象,继续后面算子的调用
## map 算子看 Transformation
stream
.map(str => str)
代码执行到 map 这一行时,会调用到 DataStream.scala 的 map 方法
def map[R: TypeInformation](fun: T => R): DataStream[R] = { if (fun == null) { throw new NullPointerException("Map function must not be null.") } val cleanFun = clean(fun) val mapper = new MapFunction[T, R] { def map(in: T): R = cleanFun(in) } // 又调用 map map(mapper) } def map[R: TypeInformation](mapper: MapFunction[T, R]): DataStream[R] = { if (mapper == null) { throw new NullPointerException("Map function must not be null.") } val outType : TypeInformation[R] = implicitly[TypeInformation[R]] // stream.map 调用到 DataStream.java 中了 asScalaStream(stream.map(mapper, outType).asInstanceOf[JavaStream[R]]) }
注: Flink 主要功能还是在 Java 代码中, Scala Api 就像个外壳,用 Scala 包装了一下,方便 Scala 代码调用,实际上还是会调用到 Java 代码上去
DataStream.java 的 map 方法
这里调用 transform 方法,要构建 Transformation 了,对于这个测试的写法来说, outputType 是 "String", Transformation 名是 "Map"
SimpleOperatorFactory.of(operator)) 获取的工厂类是: SimpleUdfStreamOperatorFactory str => str 就是 Udf
public <R> SingleOutputStreamOperator<R> map(MapFunction<T, R> mapper, TypeInformation<R> outputType) { // 可以看到 transform ,这里的 outputType 是 String 了 return transform("Map", outputType, new StreamMap<>(clean(mapper))); } public <R> SingleOutputStreamOperator<R> transform( String operatorName, TypeInformation<R> outTypeInfo, OneInputStreamOperator<T, R> operator) { return doTransform(operatorName, outTypeInfo, SimpleOperatorFactory.of(operator)); }
DataStream.java 的 doTransform 方法创建 map 算子对于的 OneInputTransformation, 同时创建一个新的 DataStream: SingleOutputStreamOperator
// 真正创建 Transformation protected <R> SingleOutputStreamOperator<R> doTransform( String operatorName, TypeInformation<R> outTypeInfo, StreamOperatorFactory<R> operatorFactory) { // read the output type of the input Transform to coax out errors about MissingTypeInfo // 检验和设置 transformation 输出类型 transformation.getOutputType(); // 创建 一个输入的 Transformation, this.transformation 上一算子的 Transformation 做为 当前算子的 输入 Transformation OneInputTransformation<T, R> resultTransform = new OneInputTransformation<>( this.transformation, operatorName, operatorFactory, outTypeInfo, environment.getParallelism()); @SuppressWarnings({"unchecked", "rawtypes"}) // 创建 一个输出的 StreamOperator 也是 DataStream, 也 携带 environment SingleOutputStreamOperator<R> returnStream = new SingleOutputStreamOperator(environment, resultTransform); // 讲 创建的 Transformation 放到 ExecutionEnvironment 的 transformations 列表中 getExecutionEnvironment().addOperator(resultTransform); // 返回 SingleOutputStreamOperator return returnStream; }
所以,执行完 map 后,返回的也是一个新的 DataStream,这不像有些用户,objectA.methodA().methodB() 每次都返回原来的 objectA
## sink Transformation 的起始
public DataStreamSink<T> addSink(SinkFunction<T> sinkFunction) { // read the output type of the input Transform to coax out errors about MissingTypeInfo // 检验输出和设置输出类型 transformation.getOutputType(); // configure the type if needed // 检查输入方法类型 if (sinkFunction instanceof InputTypeConfigurable) { ((InputTypeConfigurable) sinkFunction).setInputType(getType(), getExecutionConfig()); } // 创建一个 sinkOperator StreamSink<T> sinkOperator = new StreamSink<>(clean(sinkFunction)); // 使用 sinkOperator 创建 DataStreamSink , 同是创建 SinkTransformation DataStreamSink<T> sink = new DataStreamSink<>(this, sinkOperator); // 把 SinkTransformation 添加到 transformations getExecutionEnvironment().addOperator(sink.getTransformation()); // 返回 DataStreamSink return sink; }
DataStreamSink.java 创建 DataStreamSink 的时候,用当前的 DataStream 和 StreamSink 做参数, 当前的 DataStream 做为 StreamSink 的 input Transformation
protected DataStreamSink(DataStream<T> inputStream, StreamSink<T> operator) { this.transformation = new SinkTransformation<T>(inputStream.getTransformation(), "Unnamed", operator, inputStream.getExecutionEnvironment().getParallelism()); }
在创建 Sink 的 DataStream 的时候,将 前一个算子生成的 DataStream 传入 做为了 Sink 的 input Transformation。
## 总结
从 env.addSource.map.addSink 最简单的 Flink 程序,可以看到 Flink 创建 StreamGraph 前的 Transformation 生成过程,其他如: flatMap、filter、union、process 基本类似,其他如 join、window、forward 也相差不大
比如:
val join = process.join(map) .where(str => str) .equalTo(str => str) .window(TumblingProcessingTimeWindows.of(Time.minutes(1))) .apply(new JoinFunction[String, String, String] { override def join(first: String, second: String): String = { first + ";" + second } })
从 apply 追下去,会 看到 在 WindowedStream.java 的 apply 方法中 调用了 input.transform(opName, resultType, operator) 生成了 一个输入的 Transformation
@PublicEvolving public <R> SingleOutputStreamOperator<R> transform( String operatorName, TypeInformation<R> outTypeInfo, OneInputStreamOperator<T, R> operator) { return doTransform(operatorName, outTypeInfo, SimpleOperatorFactory.of(operator)); } protected <R> SingleOutputStreamOperator<R> doTransform( String operatorName, TypeInformation<R> outTypeInfo, StreamOperatorFactory<R> operatorFactory) { // read the output type of the input Transform to coax out errors about MissingTypeInfo transformation.getOutputType(); OneInputTransformation<T, R> resultTransform = new OneInputTransformation<>( this.transformation, operatorName, operatorFactory, outTypeInfo, environment.getParallelism()); @SuppressWarnings({"unchecked", "rawtypes"}) SingleOutputStreamOperator<R> returnStream = new SingleOutputStreamOperator(environment, resultTransform); getExecutionEnvironment().addOperator(resultTransform); return returnStream;
Transformation 就是 用户代码,转换成 Flink 算子的结果,Transformation
欢迎关注Flink菜鸟公众号,会不定期更新Flink(开发技术)相关的推文