Hi all,
What is the proper way to call a Python function in WindowFunction.apply()? I want to apply a Python function to values in a fixed-side sliding window. I'm trying it because - I'm currently working on time-series prediction using deep learning, which is why I need a sliding window to get the latest N items from the unbound data stream. - I already have a DNN written using Keras on top of Theano (Keras and Theano are Python libraries) in order to exploit Nvidia's CUDA library . - There is no Python DataStream API, so I tried to use Scala DataStream API. - PySpark's structured streaming does not allow me to define UDAF (see a question I posted on stackoverflow about it: http://stackoverflow.com/questions/42747236/how-to-define-udaf-over-event-time-windows-in-pyspark-2-1-0) - Spark DStream API does not look promising to this case due to the lack of support in count window. For these reasons, I thoughtlessly wrote a toy example to see the feasibility of applying Python methods to values in the sliding window. -------- import jep.Jep import org.apache.flink.streaming.api.scala._ import org.apache.flink.streaming.api.scala.StreamExecutionEnvironment import org.apache.flink.util.Collector import org.apache.flink.streaming.api.scala.function.AllWindowFunction import org.apache.flink.streaming.api.windowing.windows.GlobalWindow class WindowFunction extends AllWindowFunction[String, String, GlobalWindow] { val jep = new Jep() jep.runScript("prediction.py") override def apply(window: GlobalWindow, iter: Iterable[String], out: Collector[String]): Unit = { // ... } } object main { def main(args: Array[String]): Unit = { val env = StreamExecutionEnvironment.getExecutionEnvironment env.socketTextStream("localhost", 9999) .countWindowAll(5, 1) .apply(new WindowFunction()) .print() env.execute() } } -------- Now I'm facing with serializable error with the following error messages: -------- Exception in thread "main" org.apache.flink.api.common.InvalidProgramException: Task not serializable at org.apache.flink.api.scala.ClosureCleaner$.ensureSerializable(ClosureCleaner.scala:172) at org.apache.flink.api.scala.ClosureCleaner$.clean(ClosureCleaner.scala:164) at org.apache.flink.streaming.api.scala.StreamExecutionEnvironment.scalaClean(StreamExecutionEnvironment.scala:666) at org.apache.flink.streaming.api.scala.AllWindowedStream.clean(AllWindowedStream.scala:568) at org.apache.flink.streaming.api.scala.AllWindowedStream.apply(AllWindowedStream.scala:315) at main$.main(main.scala:23) at main.main(main.scala) at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method) at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62) at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43) at java.lang.reflect.Method.invoke(Method.java:497) at com.intellij.rt.execution.application.AppMain.main(AppMain.java:147) Caused by: java.io.NotSerializableException: jep.Jep at java.io.ObjectOutputStream.writeObject0(ObjectOutputStream.java:1184) at java.io.ObjectOutputStream.defaultWriteFields(ObjectOutputStream.java:1548) at java.io.ObjectOutputStream.writeSerialData(ObjectOutputStream.java:1509) at java.io.ObjectOutputStream.writeOrdinaryObject(ObjectOutputStream.java:1432) at java.io.ObjectOutputStream.writeObject0(ObjectOutputStream.java:1178) at java.io.ObjectOutputStream.writeObject(ObjectOutputStream.java:348) at org.apache.flink.util.InstantiationUtil.serializeObject(InstantiationUtil.java:317) at org.apache.flink.api.scala.ClosureCleaner$.ensureSerializable(ClosureCleaner.scala:170) ... 11 more -------- Apparently, the source of problem is the third party library called Jep which helps call Python scripts. Do I have to make the third party library serializable? Or there's a way to figure out this sort of thing in a totally different way in Flink? Any help (even other frameworks than Flink) will be appreciated :-) Thanks you. - Dongwon |
Hello,
I would suggest implementing the RichWindowFunction instead, and instantiate Jep within open(), or maybe do some lazy instantiation within apply. Regards, Chesnay On 14.03.2017 15:47, 김동원 wrote: Hi all, |
Hey,
Naturally this would imply that you're script is available on all nodes, so you will have to distribute it manually. On 14.03.2017 17:23, Chesnay Schepler wrote:
|
Alright, it works perfectly. I checked that my Python methods are properly executed inside RichWindowFunction. Thanks a lot! p.s. for those who wonder why I use Jep, refer to https://sushant-hiray.me/posts/python-in-scala-stack/ to grasp the idea of using Python inside Java through Jep instead of Jython and JyNI. ------------ class WindowFunction extends RichAllWindowFunction[String, String, GlobalWindow] { var jep: Option[Jep] = None override def open(parameters: Configuration): Unit = { jep = Some(new Jep()) jep map (_.runScript("prediction.py")) } override def apply(window: GlobalWindow, iter: Iterable[String], out: Collector[String]): Unit = { ... }
|
Free forum by Nabble | Edit this page |