Fix traversal of half-frozen update chains

alvherre · alvherre · commit 8b6d85f2dc1e · 2017-10-06T17:14:42.000+02:00
When some tuple versions in an update chain are frozen due to them being older than freeze_min_age, the xmax/xmin trail can become broken. This breaks HOT (and probably other things). A subsequent VACUUM can break things in more serious ways, such as leaving orphan heap-only tuples whose root HOT redirect items were removed. This can be seen because index creation (or REINDEX) complain like ERROR: XX000: failed to find parent tuple for heap-only tuple at (0,7) in table "t" Because of relfrozenxid contraints, we cannot avoid the freezing of the early tuples, so we must cope with the results: whenever we see an Xmin of FrozenTransactionId, consider it a match for whatever the previous Xmax value was. This problem seems to have appeared in 9.3 with multixact changes, though strictly speaking it seems unrelated. Since 9.4 we have commit 37484ad "Change the way we mark tuples as frozen", so the fix is simple: just compare the raw Xmin (still stored in the tuple header, since freezing merely set an infomask bit) to the Xmax. But in 9.3 we rewrite the Xmin value to FrozenTransactionId, so the original value is lost and we have nothing to compare the Xmax with. To cope with that case we need to compare the Xmin with FrozenXid, assume it's a match, and hope for the best. Sadly, since you can pg_upgrade a 9.3 instance containing half-frozen pages to newer releases, we need to keep the old check in newer versions too, which seems a bit brittle; I hope we can somehow get rid of that. I didn't optimize the new function for performance. The new coding is probably a bit slower than before, since there is a function call rather than a straight comparison, but I'd rather have it work correctly than be fast but wrong. This is a followup after 20b6552 fixed a few related problems. Apparently, in 9.6 and up there are more ways to get into trouble, but in 9.3 - 9.5 I cannot reproduce a problem anymore with this patch, so there must be a separate bug. Reported-by: Peter Geoghegan Diagnosed-by: Peter Geoghegan, Michael Paquier, Daniel Wood, Yi Wen Wong, Álvaro Discussion: https://postgr.es/m/CAH2-Wznm4rCrhFAiwKPWTpEw2bXDtgROZK7jWWGucXeH3D1fmA@mail.gmail.com
diff --git a/src/backend/access/heap/heapam.c b/src/backend/access/heap/heapam.c
@@ -1737,8 +1737,7 @@ heap_hot_search_buffer(ItemPointer tid, Relation relation, Buffer buffer,
 		 * broken.
 		 */
 		if (TransactionIdIsValid(prev_xmax) &&
-			!TransactionIdEquals(prev_xmax,
-								 HeapTupleHeaderGetXmin(heapTuple->t_data)))
+			!HeapTupleUpdateXmaxMatchesXmin(prev_xmax, heapTuple->t_data))
 			break;
 
 		/*
@@ -1920,7 +1919,7 @@ heap_get_latest_tid(Relation relation,
 		 * tuple.  Check for XMIN match.
 		 */
 		if (TransactionIdIsValid(priorXmax) &&
-		  !TransactionIdEquals(priorXmax, HeapTupleHeaderGetXmin(tp.t_data)))
+			!HeapTupleUpdateXmaxMatchesXmin(priorXmax, tp.t_data))
 		{
 			UnlockReleaseBuffer(buffer);
 			break;
@@ -1952,6 +1951,50 @@ heap_get_latest_tid(Relation relation,
 	}							/* end of loop */
 }
 
+/*
+ * HeapTupleUpdateXmaxMatchesXmin - verify update chain xmax/xmin lineage
+ *
+ * Given the new version of a tuple after some update, verify whether the
+ * given Xmax (corresponding to the previous version) matches the tuple's
+ * Xmin, taking into account that the Xmin might have been frozen after the
+ * update.
+ */
+bool
+HeapTupleUpdateXmaxMatchesXmin(TransactionId xmax, HeapTupleHeader htup)
+{
+	TransactionId	xmin = HeapTupleHeaderGetXmin(htup);
+
+	/*
+	 * If the xmax of the old tuple is identical to the xmin of the new one,
+	 * it's a match.
+	 */
+	if (TransactionIdEquals(xmax, xmin))
+		return true;
+
+	/*
+	 * If the Xmin that was in effect prior to a freeze matches the Xmax,
+	 * it's good too.
+	 */
+	if (HeapTupleHeaderXminFrozen(htup) &&
+		TransactionIdEquals(HeapTupleHeaderGetRawXmin(htup), xmax))
+		return true;
+
+	/*
+	 * When a tuple is frozen, the original Xmin is lost, but we know it's a
+	 * committed transaction.  So unless the Xmax is InvalidXid, we don't know
+	 * for certain that there is a match, but there may be one; and we must
+	 * return true so that a HOT chain that is half-frozen can be walked
+	 * correctly.
+	 *
+	 * We no longer freeze tuples this way, but we must keep this in order to
+	 * interpret pre-pg_upgrade pages correctly.
+	 */
+	if (TransactionIdEquals(xmin, FrozenTransactionId) &&
+		TransactionIdIsValid(xmax))
+		return true;
+
+	return false;
+}
 
 /*
  * UpdateXmaxHintBits - update tuple hint bits after xmax transaction ends
@@ -5240,8 +5283,7 @@ heap_lock_updated_tuple_rec(Relation rel, ItemPointer tid, TransactionId xid,
 		 * end of the chain, we're done, so return success.
 		 */
 		if (TransactionIdIsValid(priorXmax) &&
-			!TransactionIdEquals(HeapTupleHeaderGetXmin(mytup.t_data),
-								 priorXmax))
+			!HeapTupleUpdateXmaxMatchesXmin(priorXmax, mytup.t_data))
 		{
 			UnlockReleaseBuffer(buf);
 			return HeapTupleMayBeUpdated;
diff --git a/src/backend/access/heap/pruneheap.c b/src/backend/access/heap/pruneheap.c
@@ -465,7 +465,7 @@ heap_prune_chain(Relation relation, Buffer buffer, OffsetNumber rootoffnum,
 		 * Check the tuple XMIN against prior XMAX, if any
 		 */
 		if (TransactionIdIsValid(priorXmax) &&
-			!TransactionIdEquals(HeapTupleHeaderGetXmin(htup), priorXmax))
+			!HeapTupleUpdateXmaxMatchesXmin(priorXmax, htup))
 			break;
 
 		/*
@@ -805,7 +805,7 @@ heap_get_root_tuples(Page page, OffsetNumber *root_offsets)
 			htup = (HeapTupleHeader) PageGetItem(page, lp);
 
 			if (TransactionIdIsValid(priorXmax) &&
-				!TransactionIdEquals(priorXmax, HeapTupleHeaderGetXmin(htup)))
+				!HeapTupleUpdateXmaxMatchesXmin(priorXmax, htup))
 				break;
 
 			/* Remember the root line pointer for this item */
diff --git a/src/backend/executor/execMain.c b/src/backend/executor/execMain.c
@@ -2080,8 +2080,7 @@ EvalPlanQualFetch(EState *estate, Relation relation, int lockmode, bool noWait,
 			 * buffer's content lock, since xmin never changes in an existing
 			 * tuple.)
 			 */
-			if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple.t_data),
-									 priorXmax))
+			if (!HeapTupleUpdateXmaxMatchesXmin(priorXmax, tuple.t_data))
 			{
 				ReleaseBuffer(buffer);
 				return NULL;
@@ -2210,8 +2209,7 @@ EvalPlanQualFetch(EState *estate, Relation relation, int lockmode, bool noWait,
 		/*
 		 * As above, if xmin isn't what we're expecting, do nothing.
 		 */
-		if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple.t_data),
-								 priorXmax))
+		if (!HeapTupleUpdateXmaxMatchesXmin(priorXmax, tuple.t_data))
 		{
 			ReleaseBuffer(buffer);
 			return NULL;
diff --git a/src/include/access/heapam.h b/src/include/access/heapam.h
@@ -129,6 +129,9 @@ extern void heap_get_latest_tid(Relation relation, Snapshot snapshot,
 					ItemPointer tid);
 extern void setLastTid(const ItemPointer tid);
 
+extern bool HeapTupleUpdateXmaxMatchesXmin(TransactionId xmax,
+							   HeapTupleHeader htup);
+
 extern BulkInsertState GetBulkInsertState(void);
 extern void FreeBulkInsertState(BulkInsertState);
 
