[SPARK-26806][SS] EventTimeStats.merge should handle zeros correctly

## What changes were proposed in this pull request?

Right now, EventTimeStats.merge doesn't handle `zero.merge(zero)` correctly. This will make `avg` become `NaN`. And whatever gets merged with the result of `zero.merge(zero)`, `avg` will still be `NaN`. Then finally, we call `NaN.toLong` and get `0`, and the user will see the following incorrect report:

```
"eventTime" : {
    "avg" : "1970-01-01T00:00:00.000Z",
    "max" : "2019-01-31T12:57:00.000Z",
    "min" : "2019-01-30T18:44:04.000Z",
    "watermark" : "1970-01-01T00:00:00.000Z"
  }
```

This issue was reported by liancheng .

This PR fixes the above issue.

## How was this patch tested?

The new unit tests.

Closes #23718 from zsxwing/merge-zero.

Authored-by: Shixiong Zhu <zsxwing@gmail.com>
Signed-off-by: Shixiong Zhu <zsxwing@gmail.com>
(cherry picked from commit 03a928c)
Signed-off-by: Shixiong Zhu <zsxwing@gmail.com>

sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/EventTimeWatermarkExec.scala

@@ -36,10 +36,19 @@ case class EventTimeStats(var max: Long, var min: Long, var avg: Double, var cou
   }
 
   def merge(that: EventTimeStats): Unit = {
-    this.max = math.max(this.max, that.max)
-    this.min = math.min(this.min, that.min)
-    this.count += that.count
-    this.avg += (that.avg - this.avg) * that.count / this.count
+    if (that.count == 0) {
+      // no-op
+    } else if (this.count == 0) {
+      this.max = that.max
+      this.min = that.min
+      this.count = that.count
+      this.avg = that.avg
+    } else {
+      this.max = math.max(this.max, that.max)
+      this.min = math.min(this.min, that.min)
+      this.count += that.count
+      this.avg += (that.avg - this.avg) * that.count / this.count
+    }
   }
 }
 

sql/core/src/test/scala/org/apache/spark/sql/streaming/EventTimeWatermarkSuite.scala

@@ -38,9 +38,9 @@ class EventTimeWatermarkSuite extends StreamTest with BeforeAndAfter with Matche
     sqlContext.streams.active.foreach(_.stop())
   }
 
-  test("EventTimeStats") {
-    val epsilon = 10E-6
+  private val epsilon = 10E-6
 
+  test("EventTimeStats") {
     val stats = EventTimeStats(max = 100, min = 10, avg = 20.0, count = 5)
     stats.add(80L)
     stats.max should be (100)
@@ -57,7 +57,6 @@ class EventTimeWatermarkSuite extends StreamTest with BeforeAndAfter with Matche
   }
 
   test("EventTimeStats: avg on large values") {
-    val epsilon = 10E-6
     val largeValue = 10000000000L // 10B
     // Make sure `largeValue` will cause overflow if we use a Long sum to calc avg.
     assert(largeValue * largeValue != BigInt(largeValue) * BigInt(largeValue))
@@ -75,6 +74,33 @@ class EventTimeWatermarkSuite extends StreamTest with BeforeAndAfter with Matche
     stats.avg should be ((largeValue + 0.5) +- epsilon)
   }
 
+  test("EventTimeStats: zero merge zero") {
+    val stats = EventTimeStats.zero
+    val stats2 = EventTimeStats.zero
+    stats.merge(stats2)
+    stats should be (EventTimeStats.zero)
+  }
+
+  test("EventTimeStats: non-zero merge zero") {
+    val stats = EventTimeStats(max = 10, min = 1, avg = 5.0, count = 3)
+    val stats2 = EventTimeStats.zero
+    stats.merge(stats2)
+    stats.max should be (10L)
+    stats.min should be (1L)
+    stats.avg should be (5.0 +- epsilon)
+    stats.count should be (3L)
+  }
+
+  test("EventTimeStats: zero merge non-zero") {
+    val stats = EventTimeStats.zero
+    val stats2 = EventTimeStats(max = 10, min = 1, avg = 5.0, count = 3)
+    stats.merge(stats2)
+    stats.max should be (10L)
+    stats.min should be (1L)
+    stats.avg should be (5.0 +- epsilon)
+    stats.count should be (3L)
+  }
+
   test("error on bad column") {
     val inputData = MemoryStream[Int].toDF()
     val e = intercept[AnalysisException] {