[SPARK-26228][MLLIB] OOM issue encountered when computing Gramian matrix

Avoid memory problems in closure cleaning when handling large Gramians (>= 16K rows/cols) by using null as zeroValue

Existing tests.
Note that it's hard to test the case that triggers this issue as it would require a large amount of memory and run a while. I confirmed locally that a 16K x 16K Gramian failed with tons of driver memory before, and didn't fail upfront after this change.

Closes apache#23600 from srowen/SPARK-26228.

Authored-by: Sean Owen <sean.owen@databricks.com>
Signed-off-by: Sean Owen <sean.owen@databricks.com>
(cherry picked from commit 6dcad38)
Signed-off-by: Sean Owen <sean.owen@databricks.com>

Author: srowen

mllib/src/main/scala/org/apache/spark/mllib/linalg/distributed/RowMatrix.scala

@@ -119,11 +119,25 @@ class RowMatrix @Since("1.0.0") (
     val nt = if (n % 2 == 0) ((n / 2) * (n + 1)) else (n * ((n + 1) / 2))
 
     // Compute the upper triangular part of the gram matrix.
-    val GU = rows.treeAggregate(new BDV[Double](nt))(
-      seqOp = (U, v) => {
+    val GU = rows.treeAggregate(null.asInstanceOf[BDV[Double]])(
+      seqOp = (maybeU, v) => {
+        val U =
+          if (maybeU == null) {
+            new BDV[Double](nt)
+          } else {
+            maybeU
+          }
         BLAS.spr(1.0, v, U.data)
         U
-      }, combOp = (U1, U2) => U1 += U2)
+      }, combOp = (U1, U2) =>
+        if (U1 == null) {
+          U2
+        } else if (U2 == null) {
+          U1
+        } else {
+          U1 += U2
+        }
+    )
 
     RowMatrix.triuToFull(n, GU.data)
   }