Github user squito commented on a diff in the pull request:

    https://github.com/apache/spark/pull/20424#discussion_r165255299
  
    --- Diff: 
core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala ---
    @@ -191,7 +191,20 @@ private[spark] class PythonWorkerFactory(pythonExec: 
String, envVars: Map[String
             daemon = pb.start()
     
             val in = new DataInputStream(daemon.getInputStream)
    -        daemonPort = in.readInt()
    +        try {
    +          daemonPort = in.readInt()
    +        } catch {
    +          case exc: EOFException =>
    +            throw new IOException(s"No port number in $daemonModule's 
stdout")
    +        }
    +
    +        // test that the returned port number is within a valid range.
    +        // note: this does not cover the case where the port number
    +        // is arbitrary data but is also coincidentally within range
    +        if (daemonPort < 1 || daemonPort > 0xffff) {
    --- End diff --
    
    ah I see, I think you are worried about something other than what bruce and 
I thought.  Your concern is that we might throw an exception for some values 
that are actually perfectly legitimate.  Port 0 being special is a pretty 
standard thing -- its mentioned in the constructor for ServerSocket: 
https://docs.oracle.com/javase/7/docs/api/java/net/ServerSocket.html#ServerSocket%28int%29
    
    which implies that you shouldn't ever open a Socket on port 0, though I 
don't see that officially documented.  At least on my laptop, I get different 
errors if I try to connect to port 0, vs. just connecting to a bogus port:
    
    ```scala
    scala> val s2 = new Socket("localhost", 1234)
    java.net.ConnectException: Connection refused
      at java.net.PlainSocketImpl.socketConnect(Native Method)
      at 
java.net.AbstractPlainSocketImpl.doConnect(AbstractPlainSocketImpl.java:350)
      at 
java.net.AbstractPlainSocketImpl.connectToAddress(AbstractPlainSocketImpl.java:206)
      at 
java.net.AbstractPlainSocketImpl.connect(AbstractPlainSocketImpl.java:188)
      at java.net.SocksSocketImpl.connect(SocksSocketImpl.java:392)
      at java.net.Socket.connect(Socket.java:589)
      at java.net.Socket.connect(Socket.java:538)
      at java.net.Socket.<init>(Socket.java:434)
      at java.net.Socket.<init>(Socket.java:211)
      ... 29 elided
    
    scala> val s3 = new Socket("localhost", 0)
    java.net.NoRouteToHostException: Can't assign requested address
      at java.net.PlainSocketImpl.socketConnect(Native Method)
      at 
java.net.AbstractPlainSocketImpl.doConnect(AbstractPlainSocketImpl.java:350)
      at 
java.net.AbstractPlainSocketImpl.connectToAddress(AbstractPlainSocketImpl.java:206)
      at 
java.net.AbstractPlainSocketImpl.connect(AbstractPlainSocketImpl.java:188)
      at java.net.SocksSocketImpl.connect(SocksSocketImpl.java:392)
      at java.net.Socket.connect(Socket.java:589)
      at java.net.Socket.connect(Socket.java:538)
      at java.net.Socket.<init>(Socket.java:434)
      at java.net.Socket.<init>(Socket.java:211)
      ... 29 elided
    ```
    
    so I think its pretty safe to say that daemon.py (or whatever) shouldn't be 
passing back `0` as the port to bind to.
    
    Still -- it is *clearly* safer to instead have the port written to some 
other file, or (another) socket, so that you we wouldn't have to worry about 
the details of this error handling.


---

---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org

Reply via email to