Github user latioswang commented on a diff in the pull request:
https://github.com/apache/spark/pull/11628#discussion_r56594485
--- Diff: core/src/main/scala/org/apache/spark/rdd/PipedRDD.scala ---
@@ -118,63 +120,94 @@ private[spark] class PipedRDD[T: ClassTag](
val proc = pb.start()
val env = SparkEnv.get
+ val childThreadException = new AtomicReference[Throwable](null)
// Start a thread to print the process's stderr to ours
- new Thread("stderr reader for " + command) {
- override def run() {
- for (line <- Source.fromInputStream(proc.getErrorStream).getLines)
{
- // scalastyle:off println
- System.err.println(line)
- // scalastyle:on println
+ new Thread(s"stderr reader for $command") {
+ override def run(): Unit = {
+ val err = proc.getErrorStream
+ try {
+ for (line <- Source.fromInputStream(err).getLines) {
+ // scalastyle:off println
+ System.err.println(line)
+ // scalastyle:on println
+ }
+ } catch {
+ case t: Throwable => childThreadException.set(t)
+ } finally {
+ err.close()
}
}
}.start()
// Start a thread to feed the process input from our parent's iterator
- new Thread("stdin writer for " + command) {
- override def run() {
+ new Thread(s"stdin writer for $command") {
+ override def run(): Unit = {
TaskContext.setTaskContext(context)
val out = new PrintWriter(proc.getOutputStream)
-
- // scalastyle:off println
- // input the pipe context firstly
- if (printPipeContext != null) {
- printPipeContext(out.println(_))
- }
- for (elem <- firstParent[T].iterator(split, context)) {
- if (printRDDElement != null) {
- printRDDElement(elem, out.println(_))
- } else {
- out.println(elem)
+ try {
+ // scalastyle:off println
+ // input the pipe context firstly
+ if (printPipeContext != null) {
+ printPipeContext(out.println)
+ }
+ for (elem <- firstParent[T].iterator(split, context)) {
+ if (printRDDElement != null) {
+ printRDDElement(elem, out.println)
+ } else {
+ out.println(elem)
+ }
}
+ // scalastyle:on println
+ } catch {
+ case t: Throwable => childThreadException.set(t)
+ } finally {
+ out.close()
--- End diff --
doh, we should not close the output stream here. as `close()` would signal
the child process that input is complete. The child process might decide to
send a SUCCESS signal to external systems (eg. close immaturely on a partial
output file and upload to some database :/ ).
Instead we should kill the subprocess before closing the stream.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
