[SPARK-12837][SPARK-20666][CORE][FOLLOWUP] getting name should not fail if...

[SPARK-12837][SPARK-20666][CORE][FOLLOWUP] getting name should not fail if accumulator is garbage collected

## What changes were proposed in this pull request?

After https://github.com/apache/spark/pull/17596 , we do not send internal accumulator name to executor side anymore, and always look up the accumulator name in `AccumulatorContext`.

This cause a regression if the accumulator is already garbage collected, this PR fixes this by still sending accumulator name for `SQLMetrics`.

## How was this patch tested?

N/A

Author: Wenchen Fan <wenchen@databricks.com>

Closes #17931 from cloud-fan/bug.

core/src/main/scala/org/apache/spark/util/AccumulatorV2.scala

@@ -84,10 +84,11 @@ abstract class AccumulatorV2[IN, OUT] extends Serializable {
    * Returns the name of this accumulator, can only be called after registration.
    */
   final def name: Option[String] = {
+    assertMetadataNotNull()
+
     if (atDriverSide) {
-      AccumulatorContext.get(id).flatMap(_.metadata.name)
+      metadata.name.orElse(AccumulatorContext.get(id).flatMap(_.metadata.name))
     } else {
-      assertMetadataNotNull()
       metadata.name
     }
   }
@@ -165,13 +166,15 @@ abstract class AccumulatorV2[IN, OUT] extends Serializable {
       }
       val copyAcc = copyAndReset()
       assert(copyAcc.isZero, "copyAndReset must return a zero value copy")
-      val isInternalAcc =
-        (name.isDefined && name.get.startsWith(InternalAccumulator.METRICS_PREFIX)) ||
-          getClass.getSimpleName == "SQLMetric"
+      val isInternalAcc = name.isDefined && name.get.startsWith(InternalAccumulator.METRICS_PREFIX)
       if (isInternalAcc) {
         // Do not serialize the name of internal accumulator and send it to executor.
         copyAcc.metadata = metadata.copy(name = None)
       } else {
+        // For non-internal accumulators, we still need to send the name because users may need to
+        // access the accumulator name at executor side, or they may keep the accumulators sent from
+        // executors and access the name when the registered accumulator is already garbage
+        // collected(e.g. SQLMetrics).
         copyAcc.metadata = metadata
       }
       copyAcc