diff --git a/akka-docs/src/main/paradox/stream/operators/Source-or-Flow/groupBy.md b/akka-docs/src/main/paradox/stream/operators/Source-or-Flow/groupBy.md
index 73c6197ee9..aa822853ab 100644
--- a/akka-docs/src/main/paradox/stream/operators/Source-or-Flow/groupBy.md
+++ b/akka-docs/src/main/paradox/stream/operators/Source-or-Flow/groupBy.md
@@ -14,7 +14,21 @@ Demultiplex the incoming stream into separate output streams.
 
 ## Description
 
-Demultiplex the incoming stream into separate output streams.
+This operation demultiplexes the incoming stream into separate output streams, one for each element key. The
+key is computed for each element using the given function. When a new key is encountered for the first time
+a new substream is opened and subsequently fed with all elements belonging to that key.
+
+Note: If `allowClosedSubstreamRecreation` is set to `true` substream completion and incoming
+elements are subject to race-conditions. If elements arrive for a stream that is in the process
+of closing these elements might get lost.
+
+@@@ warning
+
+If `allowClosedSubstreamRecreation` is set to `false` (default behavior) the stage keeps track of all
+keys of streams that have already been closed. If you expect an infinite number of keys this can cause
+memory issues. Elements belonging to those keys are drained directly and not send to the substream.
+
+@@@
 
 
 @@@div { .callout }
diff --git a/akka-docs/src/main/paradox/stream/stream-substream.md b/akka-docs/src/main/paradox/stream/stream-substream.md
index 6914c16857..76b8df9546 100644
--- a/akka-docs/src/main/paradox/stream/stream-substream.md
+++ b/akka-docs/src/main/paradox/stream/stream-substream.md
@@ -26,6 +26,8 @@ This operation splits the incoming stream into separate output
 streams, one for each element key. The key is computed for each element
 using the given function, which is `f` in the above diagram. When a new key is encountered for the first time
 a new substream is opened and subsequently fed with all elements belonging to that key.
+If `allowClosedSubstreamRecreation` is set to `true` a substream belonging to a specific key
+will be recreated if it was closed before, otherwise elements belonging to that key will be dropped.
 
 If you add a `Sink` or `Flow` right after the `groupBy` stage,
 all transformations are applied to all encountered substreams in the same fashion.
diff --git a/akka-stream-tests/src/test/scala/akka/stream/scaladsl/FlowGroupBySpec.scala b/akka-stream-tests/src/test/scala/akka/stream/scaladsl/FlowGroupBySpec.scala
index bdad7e6606..406790f93f 100644
--- a/akka-stream-tests/src/test/scala/akka/stream/scaladsl/FlowGroupBySpec.scala
+++ b/akka-stream-tests/src/test/scala/akka/stream/scaladsl/FlowGroupBySpec.scala
@@ -459,6 +459,34 @@ class FlowGroupBySpec extends StreamSpec {
       upstream.sendComplete()
     }
 
+    "allow to recreate an already closed substream (#24758)" in assertAllStagesStopped {
+      val (up, down) = Flow[Int]
+        .groupBy(2, identity, true)
+        .take(1) // close the substream after 1 element
+        .mergeSubstreams
+        .runWith(TestSource.probe[Int], TestSink.probe)
+
+      down.request(4)
+
+      // Creates and closes substream "1"
+      up.sendNext(1)
+      down.expectNext(1)
+
+      // Creates and closes substream "2"
+      up.sendNext(2)
+      down.expectNext(2)
+
+      // Recreates and closes substream "1" twice
+      up.sendNext(1)
+      down.expectNext(1)
+      up.sendNext(1)
+      down.expectNext(1)
+
+      // Cleanup, not part of the actual test
+      up.sendComplete()
+      down.expectComplete()
+    }
+
     "cancel if downstream has cancelled & all substreams cancel" in assertAllStagesStopped {
       val upstream = TestPublisher.probe[Int]()
       val downstreamMaster = TestSubscriber.probe[Source[Int, NotUsed]]()
diff --git a/akka-stream/src/main/mima-filters/2.5.12.backwards.excludes b/akka-stream/src/main/mima-filters/2.5.12.backwards.excludes
index c105fe6a6b..2f2a497a45 100644
--- a/akka-stream/src/main/mima-filters/2.5.12.backwards.excludes
+++ b/akka-stream/src/main/mima-filters/2.5.12.backwards.excludes
@@ -1,2 +1,6 @@
 # +str add in-line inspect operator for side effecting #24610
 ProblemFilters.exclude[ReversedMissingMethodProblem]("akka.stream.scaladsl.FlowOps.wireTap")
+
+# #24758 recreate already closed substreams 
+ProblemFilters.exclude[DirectMissingMethodProblem]("akka.stream.impl.fusing.GroupBy.this")
+ProblemFilters.exclude[ReversedMissingMethodProblem]("akka.stream.scaladsl.FlowOps.groupBy")
diff --git a/akka-stream/src/main/scala/akka/stream/impl/fusing/StreamOfStreams.scala b/akka-stream/src/main/scala/akka/stream/impl/fusing/StreamOfStreams.scala
index 4fad974fb0..d2c3104afc 100644
--- a/akka-stream/src/main/scala/akka/stream/impl/fusing/StreamOfStreams.scala
+++ b/akka-stream/src/main/scala/akka/stream/impl/fusing/StreamOfStreams.scala
@@ -4,6 +4,7 @@
 
 package akka.stream.impl.fusing
 
+import java.util.Collections
 import java.util.concurrent.atomic.AtomicReference
 
 import akka.NotUsed
@@ -214,7 +215,7 @@ import scala.collection.JavaConverters._
 /**
  * INTERNAL API
  */
-@InternalApi private[akka] final class GroupBy[T, K](val maxSubstreams: Int, val keyFor: T ⇒ K) extends GraphStage[FlowShape[T, Source[T, NotUsed]]] {
+@InternalApi private[akka] final class GroupBy[T, K](val maxSubstreams: Int, val keyFor: T ⇒ K, val allowClosedSubstreamRecreation: Boolean = false) extends GraphStage[FlowShape[T, Source[T, NotUsed]]] {
   val in: Inlet[T] = Inlet("GroupBy.in")
   val out: Outlet[Source[T, NotUsed]] = Outlet("GroupBy.out")
   override val shape: FlowShape[T, Source[T, NotUsed]] = FlowShape(in, out)
@@ -224,7 +225,7 @@ import scala.collection.JavaConverters._
     parent ⇒
     lazy val decider = inheritedAttributes.mandatoryAttribute[SupervisionStrategy].decider
     private val activeSubstreamsMap = new java.util.HashMap[Any, SubstreamSource]()
-    private val closedSubstreams = new java.util.HashSet[Any]()
+    private val closedSubstreams = if (allowClosedSubstreamRecreation) Collections.unmodifiableSet(Collections.emptySet[Any]) else new java.util.HashSet[Any]()
     private var timeout: FiniteDuration = _
     private var substreamWaitingToBePushed: Option[SubstreamSource] = None
     private var nextElementKey: K = null.asInstanceOf[K]
@@ -334,8 +335,9 @@ import scala.collection.JavaConverters._
     override protected def onTimer(timerKey: Any): Unit = {
       val substreamSource = activeSubstreamsMap.get(timerKey)
       if (substreamSource != null) {
-        substreamSource.timeout(timeout)
-        closedSubstreams.add(timerKey)
+        if (!allowClosedSubstreamRecreation) {
+          closedSubstreams.add(timerKey)
+        }
         activeSubstreamsMap.remove(timerKey)
         if (isClosed(in)) tryCompleteAll()
       }
@@ -349,7 +351,9 @@ import scala.collection.JavaConverters._
       private def completeSubStream(): Unit = {
         complete()
         activeSubstreamsMap.remove(key)
-        closedSubstreams.add(key)
+        if (!allowClosedSubstreamRecreation) {
+          closedSubstreams.add(key)
+        }
       }
 
       private def tryCompleteHandler(): Unit = {
diff --git a/akka-stream/src/main/scala/akka/stream/javadsl/Flow.scala b/akka-stream/src/main/scala/akka/stream/javadsl/Flow.scala
index 68769b5e6a..a4f6047f6a 100755
--- a/akka-stream/src/main/scala/akka/stream/javadsl/Flow.scala
+++ b/akka-stream/src/main/scala/akka/stream/javadsl/Flow.scala
@@ -1711,6 +1711,15 @@ final class Flow[In, Out, Mat](delegate: scaladsl.Flow[In, Out, Mat]) extends Gr
    * a new substream is opened and subsequently fed with all elements belonging to
    * that key.
    *
+   * WARNING: If `allowClosedSubstreamRecreation` is set to `false` (default behavior) the stage
+   * keeps track of all keys of streams that have already been closed. If you expect an infinite
+   * number of keys this can cause memory issues. Elements belonging to those keys are drained
+   * directly and not send to the substream.
+   *
+   * Note: If `allowClosedSubstreamRecreation` is set to `true` substream completion and incoming
+   * elements are subject to race-conditions. If elements arrive for a stream that is in the process
+   * of closing these elements might get lost.
+   *
    * The object returned from this method is not a normal [[Flow]],
    * it is a [[SubFlow]]. This means that after this combinator all transformations
    * are applied to all encountered substreams in the same fashion. Substream mode
@@ -1744,9 +1753,28 @@ final class Flow[In, Out, Mat](delegate: scaladsl.Flow[In, Out, Mat]) extends Gr
    *
    * @param maxSubstreams configures the maximum number of substreams (keys)
    *        that are supported; if more distinct keys are encountered then the stream fails
+   * @param f computes the key for each element
+   * @param allowClosedSubstreamRecreation enables recreation of already closed substreams if elements with their
+   *        corresponding keys arrive after completion
+   */
+  def groupBy[K](maxSubstreams: Int, f: function.Function[Out, K], allowClosedSubstreamRecreation: Boolean): SubFlow[In, Out, Mat] =
+    new SubFlow(delegate.groupBy(maxSubstreams, f.apply, allowClosedSubstreamRecreation))
+
+  /**
+   * This operation demultiplexes the incoming stream into separate output
+   * streams, one for each element key. The key is computed for each element
+   * using the given function. When a new key is encountered for the first time
+   * a new substream is opened and subsequently fed with all elements belonging to
+   * that key.
+   *
+   * WARNING: The stage keeps track of all keys of streams that have already been closed.
+   * If you expect an infinite number of keys this can cause memory issues. Elements belonging
+   * to those keys are drained directly and not send to the substream.
+   *
+   * @see [[#groupBy]]
    */
   def groupBy[K](maxSubstreams: Int, f: function.Function[Out, K]): SubFlow[In, Out, Mat] =
-    new SubFlow(delegate.groupBy(maxSubstreams, f.apply))
+    new SubFlow(delegate.groupBy(maxSubstreams, f.apply, false))
 
   /**
    * This operation applies the given predicate to all incoming elements and
diff --git a/akka-stream/src/main/scala/akka/stream/scaladsl/Flow.scala b/akka-stream/src/main/scala/akka/stream/scaladsl/Flow.scala
index 5fabcd39d4..2d33d42732 100755
--- a/akka-stream/src/main/scala/akka/stream/scaladsl/Flow.scala
+++ b/akka-stream/src/main/scala/akka/stream/scaladsl/Flow.scala
@@ -1768,6 +1768,15 @@ trait FlowOps[+Out, +Mat] {
    * a new substream is opened and subsequently fed with all elements belonging to
    * that key.
    *
+   * WARNING: If `allowClosedSubstreamRecreation` is set to `false` (default behavior) the stage
+   * keeps track of all keys of streams that have already been closed. If you expect an infinite
+   * number of keys this can cause memory issues. Elements belonging to those keys are drained
+   * directly and not send to the substream.
+   *
+   * Note: If `allowClosedSubstreamRecreation` is set to `true` substream completion and incoming
+   * elements are subject to race-conditions. If elements arrive for a stream that is in the process
+   * of closing these elements might get lost.
+   *
    * The object returned from this method is not a normal [[Source]] or [[Flow]],
    * it is a [[SubFlow]]. This means that after this combinator all transformations
    * are applied to all encountered substreams in the same fashion. Substream mode
@@ -1803,20 +1812,38 @@ trait FlowOps[+Out, +Mat] {
    *
    * @param maxSubstreams configures the maximum number of substreams (keys)
    *        that are supported; if more distinct keys are encountered then the stream fails
+   * @param f computes the key for each element
+   * @param allowClosedSubstreamRecreation enables recreation of already closed substreams if elements with their
+   *        corresponding keys arrive after completion
    */
-  def groupBy[K](maxSubstreams: Int, f: Out ⇒ K): SubFlow[Out, Mat, Repr, Closed] = {
+  def groupBy[K](maxSubstreams: Int, f: Out ⇒ K, allowClosedSubstreamRecreation: Boolean): SubFlow[Out, Mat, Repr, Closed] = {
     val merge = new SubFlowImpl.MergeBack[Out, Repr] {
       override def apply[T](flow: Flow[Out, T, NotUsed], breadth: Int): Repr[T] =
-        via(new GroupBy(maxSubstreams, f))
+        via(new GroupBy(maxSubstreams, f, allowClosedSubstreamRecreation))
           .map(_.via(flow))
           .via(new FlattenMerge(breadth))
     }
     val finish: (Sink[Out, NotUsed]) ⇒ Closed = s ⇒
-      via(new GroupBy(maxSubstreams, f))
+      via(new GroupBy(maxSubstreams, f, allowClosedSubstreamRecreation))
         .to(Sink.foreach(_.runWith(s)(GraphInterpreter.currentInterpreter.materializer)))
     new SubFlowImpl(Flow[Out], merge, finish)
   }
 
+  /**
+   * This operation demultiplexes the incoming stream into separate output
+   * streams, one for each element key. The key is computed for each element
+   * using the given function. When a new key is encountered for the first time
+   * a new substream is opened and subsequently fed with all elements belonging to
+   * that key.
+   *
+   * WARNING: The stage keeps track of all keys of streams that have already been closed.
+   * If you expect an infinite number of keys this can cause memory issues. Elements belonging
+   * to those keys are drained directly and not send to the substream.
+   *
+   * @see [[#groupBy]]
+   */
+  def groupBy[K](maxSubstreams: Int, f: Out ⇒ K): SubFlow[Out, Mat, Repr, Closed] = groupBy(maxSubstreams, f, false)
+
   /**
    * This operation applies the given predicate to all incoming elements and
    * emits them to a stream of output streams, always beginning a new one with