WT-1982 Fix a window where reconciliation could go back in transaction time.

(cherry picked from commit cd1704d6c4d84c5db8ae6b471c658945ffa226f9)

3 files changed, 22 insertions, 18 deletions

diff --git a/src/include/btmem.h b/src/include/btmem.h
index 77ad31fc3f6..23b17ef2cd3 100644
--- a/src/include/btmem.h
+++ b/src/include/btmem.h
@@ -194,6 +194,11 @@ struct __wt_page_modify {
 	/* The largest update transaction ID (approximate). */
 	uint64_t update_txn;
 
+#ifdef HAVE_DIAGNOSTIC
+	/* Check that transaction time moves forward. */
+	uint64_t last_oldest_id;
+#endif
+
 	/* Dirty bytes added to the cache. */
 	size_t bytes_dirty;
 
diff --git a/src/include/txn.h b/src/include/txn.h
index dbaa11309ab..d2b369a41c4 100644
--- a/src/include/txn.h
+++ b/src/include/txn.h
@@ -25,6 +25,9 @@
 
 #define	WT_SESSION_TXN_STATE(s) (&S2C(s)->txn_global.states[(s)->id])
 
+#define	WT_SESSION_IS_CHECKPOINT(s)					\
+	((s)->id != 0 && (s)->id == S2C(s)->txn_global.checkpoint_id)
+
 struct WT_COMPILER_TYPE_ALIGN(WT_CACHE_LINE_ALIGNMENT) __wt_txn_state {
 	volatile uint64_t id;
 	volatile uint64_t snap_min;
@@ -54,7 +57,7 @@ struct __wt_txn_global {
 	 */
 	volatile uint32_t checkpoint_id;	/* Checkpoint's session ID */
 	volatile uint64_t checkpoint_gen;
-	volatile uint64_t checkpoint_snap_min;
+	volatile uint64_t checkpoint_pinned;
 
 	WT_TXN_STATE *states;		/* Per-session transaction states */
 };
diff --git a/src/include/txn.i b/src/include/txn.i
index f0b0534ff4a..a9b19ca1ff5 100644
--- a/src/include/txn.i
+++ b/src/include/txn.i
@@ -98,8 +98,8 @@ __wt_txn_oldest_id(WT_SESSION_IMPL *session)
 {
 	WT_BTREE *btree;
 	WT_TXN_GLOBAL *txn_global;
-	uint64_t checkpoint_snap_min, oldest_id;
-	uint32_t checkpoint_id;
+	uint64_t checkpoint_pinned, oldest_id;
+	uint32_t checkpoint_gen;
 
 	txn_global = &S2C(session)->txn_global;
 	btree = S2BT_SAFE(session);
@@ -108,9 +108,9 @@ __wt_txn_oldest_id(WT_SESSION_IMPL *session)
 	 * Take a local copy of these IDs in case they are updated while we are
 	 * checking visibility.
 	 */
-	checkpoint_id = txn_global->checkpoint_id;
-	checkpoint_snap_min = txn_global->checkpoint_snap_min;
-	oldest_id = txn_global->oldest_id;
+	WT_ORDERED_READ(oldest_id, txn_global->oldest_id);
+	WT_ORDERED_READ(checkpoint_gen, txn_global->checkpoint_gen);
+	WT_ORDERED_READ(checkpoint_pinned, txn_global->checkpoint_pinned);
 
 	/*
 	 * Checkpoint transactions often fall behind ordinary application
@@ -122,17 +122,13 @@ __wt_txn_oldest_id(WT_SESSION_IMPL *session)
 	 * checkpoint, or this handle is up to date with the active checkpoint
 	 * then it's safe to ignore the checkpoint ID in the visibility check.
 	 */
-	if (checkpoint_snap_min != WT_TXN_NONE &&
-	    checkpoint_id != session->id && (btree == NULL ||
-	    btree->checkpoint_gen != txn_global->checkpoint_gen) &&
-	    TXNID_LT(checkpoint_snap_min, oldest_id))
-		/*
-		 * Use the checkpoint ID for the visibility check if it is the
-		 * oldest ID in the system.
-		 */
-		oldest_id = checkpoint_snap_min;
+	if (checkpoint_pinned == WT_TXN_NONE ||
+	    TXNID_LT(oldest_id, checkpoint_pinned) ||
+	    WT_SESSION_IS_CHECKPOINT(session) ||
+	    (btree != NULL && btree->checkpoint_gen == checkpoint_gen))
+		return (oldest_id);
 
-	return (oldest_id);
+	return (checkpoint_pinned);
 }
 
 /*
@@ -340,7 +336,7 @@ __wt_txn_id_check(WT_SESSION_IMPL *session)
 	if (!F_ISSET(txn, TXN_HAS_ID)) {
 		conn = S2C(session);
 		txn_global = &conn->txn_global;
-		txn_state = &txn_global->states[session->id];
+		txn_state = WT_SESSION_TXN_STATE(session);
 
 		WT_ASSERT(session, txn_state->id == WT_TXN_NONE);
 
@@ -432,7 +428,7 @@ __wt_txn_cursor_op(WT_SESSION_IMPL *session)
 
 	txn = &session->txn;
 	txn_global = &S2C(session)->txn_global;
-	txn_state = &txn_global->states[session->id];
+	txn_state = WT_SESSION_TXN_STATE(session);
 
 	/*
 	 * If there is no transaction running (so we don't have an ID), and no