Github user JoshRosen commented on a diff in the pull request:
https://github.com/apache/spark/pull/4021#discussion_r25100214
--- Diff: core/src/test/scala/org/apache/spark/AccumulatorSuite.scala ---
@@ -135,5 +137,22 @@ class AccumulatorSuite extends FunSuite with Matchers
with LocalSparkContext {
resetSparkContext()
}
}
+
+ test ("garbage collection") {
+ // Create an accumulator and let it go out of scope to test that it's
properly garbage collected
+ sc = new SparkContext("local", "test")
+ var acc: Accumulable[mutable.Set[Any], Any] = sc.accumulable(new
mutable.HashSet[Any]())
+ val accId = acc.id
+ val ref = WeakReference(acc)
+
+ // Ensure the accumulator is present
+ assert(ref.get.isDefined)
+
+ // Remove the explicit reference to it and allow weak reference to get
garbage collected
+ acc = null
+ System.gc()
+ assert(ref.get.isEmpty)
+ assert(Accumulators.originals.get(accId).isDefined)
--- End diff --
I guess this indicates that we don't garbage-collect the map entry that
points to the weak accumulator reference. This memory leak isn't quite as huge
of a concern as leaking the accumulator itself, since we expect the map entry
to be a small, fixed-size record (a couple hundred bytes, max, probably).
If we do decide to clean this up, though, we can probably do it by adding a
synchronized method to the Accumulators object that removes an accumulator,
then register a cleanup task with ContextCleaner when creating an accumulator.
---
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]
