Avoid SnapshotResetXmin() during AtEOXact_Snapshot()

For normal commits and aborts we already reset PgXact->xmin
Avoiding touching highly contented shmem improves concurrent
performance.

Simon Riggs

Discussion: CANP8+jJdXE9b+b9F8CQT-LuxxO0PBCB-SZFfMVAdp+akqo4zfg@mail.gmail.com

src/backend/utils/time/snapmgr.c

@@ -954,7 +954,12 @@ xmin_cmp(const pairingheap_node *a, const pairingheap_node *b, void *arg)
  *
  * If there are no more snapshots, we can reset our PGXACT->xmin to InvalidXid.
  * Note we can do this without locking because we assume that storing an Xid
- * is atomic.
+ * is atomic. We do this because it will allow multi-statement transactions to
+ * reset their xmin and prevent us from holding back removal of dead rows;
+ * this has little purpose when we are dealing with very fast statements in
+ * read committed mode since the xmin will advance quickly anyway. It has no
+ * use at all when we are running single statement transactions since the xmin
+ * is reset as part of end of transaction anyway.
  *
  * Even if there are some remaining snapshots, we may be able to advance our
  * PGXACT->xmin to some degree.  This typically happens when a portal is
@@ -1051,7 +1056,7 @@ AtSubAbort_Snapshot(int level)
  *		Snapshot manager's cleanup function for end of transaction
  */
 void
-AtEOXact_Snapshot(bool isCommit)
+AtEOXact_Snapshot(bool isCommit, bool isPrepare)
 {
 	/*
 	 * In transaction-snapshot mode we must release our privately-managed
@@ -1136,7 +1141,17 @@ AtEOXact_Snapshot(bool isCommit)
 
 	FirstSnapshotSet = false;
 
-	SnapshotResetXmin();
+	/*
+	 * During normal commit and abort processing, we call
+	 * ProcArrayEndTransaction() or ProcArrayClearTransaction() to
+	 * reset the PgXact->xmin. That call happens prior to the call to
+	 * AtEOXact_Snapshot(), so we need not touch xmin here at all,
+	 * accept when we are preparing a transaction.
+	 */
+	if (isPrepare)
+		SnapshotResetXmin();
+
+	Assert(MyPgXact->xmin == 0);
 }