Import wiredtiger: 51a5c5ff5b0ceea341f70a386fff5a9f68d2edac from branch mongodb-5.1

ref: e930e8d0cd..51a5c5ff5b
for: 5.1.0-rc1

WT-8226       fix largest_key failed to consider prepared update

6 files changed, 56 insertions, 30 deletions

diff --git a/src/third_party/wiredtiger/import.data b/src/third_party/wiredtiger/import.data
index c338f0f81f0..abf371b6693 100644
--- a/src/third_party/wiredtiger/import.data
+++ b/src/third_party/wiredtiger/import.data
@@ -2,5 +2,5 @@
     "vendor": "wiredtiger",
     "github": "wiredtiger/wiredtiger.git",
     "branch": "mongodb-5.1",
-    "commit": "e930e8d0cd4c3330783e711253eae2923b1e2e3f"
+    "commit": "51a5c5ff5b0ceea341f70a386fff5a9f68d2edac"
 }
diff --git a/src/third_party/wiredtiger/src/cursor/cur_std.c b/src/third_party/wiredtiger/src/cursor/cur_std.c
index de5115845c7..1bf28de3f11 100644
--- a/src/third_party/wiredtiger/src/cursor/cur_std.c
+++ b/src/third_party/wiredtiger/src/cursor/cur_std.c
@@ -1124,23 +1124,24 @@ __wt_cursor_largest_key(WT_CURSOR *cursor)
     WT_DECL_RET;
     WT_SESSION_IMPL *session;
     bool ignore_tombstone;
+    bool visible_all;
 
     ignore_tombstone = F_ISSET(cursor, WT_CURSTD_IGNORE_TOMBSTONE);
+    visible_all = F_ISSET(cursor, WT_CURSTD_VISIBLE_ALL);
     CURSOR_API_CALL(cursor, session, largest_key, NULL);
 
-    if (F_ISSET(session->txn, WT_TXN_SHARED_TS_READ))
-        WT_ERR_MSG(session, EINVAL, "largest key cannot be called with a read timestamp");
-
     WT_ERR(__wt_scr_alloc(session, 0, &key));
 
     /* Reset the cursor to give up the cursor position. */
     WT_ERR(cursor->reset(cursor));
 
-    /* Ignore deletion */
+    /* Ignore deletion. */
     F_SET(cursor, WT_CURSTD_IGNORE_TOMBSTONE);
+    /* Ignore visibility. */
+    F_SET(cursor, WT_CURSTD_VISIBLE_ALL);
 
-    /* Call cursor prev with read uncommitted isolation level. */
-    WT_WITH_TXN_ISOLATION(session, WT_ISO_READ_UNCOMMITTED, ret = cursor->prev(cursor));
+    /* Call cursor prev to get the largest key. */
+    WT_ERR(cursor->prev(cursor));
     WT_ERR(ret);
 
     /* Copy the key as we will reset the cursor after that. */
@@ -1153,6 +1154,8 @@ __wt_cursor_largest_key(WT_CURSOR *cursor)
 err:
     if (!ignore_tombstone)
         F_CLR(cursor, WT_CURSTD_IGNORE_TOMBSTONE);
+    if (!visible_all)
+        F_CLR(cursor, WT_CURSTD_VISIBLE_ALL);
     __wt_scr_free(session, &key);
     if (ret != 0)
         WT_TRET(cursor->reset(cursor));
diff --git a/src/third_party/wiredtiger/src/docs/cursor-ops.dox b/src/third_party/wiredtiger/src/docs/cursor-ops.dox
index 82b763b563f..9618396a77c 100644
--- a/src/third_party/wiredtiger/src/docs/cursor-ops.dox
+++ b/src/third_party/wiredtiger/src/docs/cursor-ops.dox
@@ -108,10 +108,8 @@ record does not previously exist.
 The WT_SESSION::largest_key \c gets the largest key in a table regardless of
 visibility.
 
-It can only be called without a read timestamp, otherwise it returns an invalid
-argument error. Any following prev or next calls will behave as if they were
-invoked on an unpositioned cursor no matter the largest key call is successful
-or not.
+Any following prev or next calls will behave as if they were invoked on an
+unpositioned cursor no matter the largest key call is successful or not.
 
 @section cursor_error Cursor position after error
 
diff --git a/src/third_party/wiredtiger/src/include/txn_inline.h b/src/third_party/wiredtiger/src/include/txn_inline.h
index 908fe1d38dd..a6b13708057 100644
--- a/src/third_party/wiredtiger/src/include/txn_inline.h
+++ b/src/third_party/wiredtiger/src/include/txn_inline.h
@@ -766,9 +766,9 @@ __wt_txn_upd_visible_type(WT_SESSION_IMPL *session, WT_UPDATE *upd)
         if (prepare_state == WT_PREPARE_LOCKED)
             continue;
 
-        if (WT_IS_HS(session->dhandle) && upd->txnid != WT_TXN_ABORTED &&
-          upd->type == WT_UPDATE_STANDARD)
-            /* Entries in the history store are always visible. */
+        /* Entries in the history store are always visible. */
+        if ((WT_IS_HS(session->dhandle) && upd->txnid != WT_TXN_ABORTED &&
+              upd->type == WT_UPDATE_STANDARD))
             return (WT_VISIBLE_TRUE);
 
         upd_visible = __wt_txn_visible(session, upd->txnid, upd->start_ts);
@@ -901,7 +901,10 @@ __wt_txn_read_upd_list_internal(WT_SESSION_IMPL *session, WT_CURSOR_BTREE *cbt,
             continue;
         }
 
-        upd_visible = __wt_txn_upd_visible_type(session, upd);
+        if (F_ISSET(&cbt->iface, WT_CURSTD_VISIBLE_ALL))
+            upd_visible = WT_VISIBLE_TRUE;
+        else
+            upd_visible = __wt_txn_upd_visible_type(session, upd);
 
         if (upd_visible == WT_VISIBLE_TRUE)
             break;
@@ -1038,8 +1041,14 @@ retry:
             cbt->upd_value->tw.prepare = tw.prepare;
         }
 
-        /* If the start time point is visible then we need to return the ondisk value. */
-        if (WT_IS_HS(session->dhandle) || __wt_txn_tw_start_visible(session, &tw)) {
+        /*
+         * We return the onpage value in the following cases:
+         * 1. The record is from the history store.
+         * 2. It has the WT_CURSTD_VISIBLE_ALL flag set.
+         * 3. It is visible to the reader.
+         */
+        if (WT_IS_HS(session->dhandle) || F_ISSET(&cbt->iface, WT_CURSTD_VISIBLE_ALL) ||
+          __wt_txn_tw_start_visible(session, &tw)) {
             if (cbt->upd_value->skip_buf) {
                 cbt->upd_value->buf.data = NULL;
                 cbt->upd_value->buf.size = 0;
diff --git a/src/third_party/wiredtiger/src/include/wiredtiger.in b/src/third_party/wiredtiger/src/include/wiredtiger.in
index 3517c3523d2..ec50d871f90 100644
--- a/src/third_party/wiredtiger/src/include/wiredtiger.in
+++ b/src/third_party/wiredtiger/src/include/wiredtiger.in
@@ -735,6 +735,7 @@ struct __wt_cursor {
 #define	WT_CURSTD_RAW_SEARCH		0x0800000u
 #define	WT_CURSTD_VALUE_EXT		0x1000000u	/* Value points out of tree. */
 #define	WT_CURSTD_VALUE_INT		0x2000000u	/* Value points into tree. */
+#define	WT_CURSTD_VISIBLE_ALL		0x4000000u
 /* AUTOMATIC FLAG VALUE GENERATION STOP 32 */
 #define	WT_CURSTD_KEY_SET	(WT_CURSTD_KEY_EXT | WT_CURSTD_KEY_INT)
 #define	WT_CURSTD_VALUE_SET	(WT_CURSTD_VALUE_EXT | WT_CURSTD_VALUE_INT)
diff --git a/src/third_party/wiredtiger/test/suite/test_cursor17.py b/src/third_party/wiredtiger/test/suite/test_cursor17.py
index 322b82a65f8..becb6fcd5d3 100644
--- a/src/third_party/wiredtiger/test/suite/test_cursor17.py
+++ b/src/third_party/wiredtiger/test/suite/test_cursor17.py
@@ -118,32 +118,47 @@ class test_cursor17(wttest.WiredTigerTestCase):
         session2 = self.setUpSessionOpen(self.conn)
         cursor2 = session2.open_cursor(self.type + self.tablename, None)
         session2.begin_transaction()
-        cursor2[101] = self.ds.value(101)
+        cursor2[200] = self.ds.value(200)
 
         cursor = self.session.open_cursor(self.type + self.tablename, None)
 
         # Verify the largest key.
         self.session.begin_transaction()
         self.assertEqual(cursor.largest_key(), 0)
-        self.assertEqual(cursor.get_key(), 101)
+        self.assertEqual(cursor.get_key(), 200)
         self.session.rollback_transaction()
 
         session2.rollback_transaction()
+    
+    def test_invisible_timestamp(self):
+        self.populate(100)
 
-    def test_read_timestamp(self):
+        cursor = self.session.open_cursor(self.type + self.tablename, None)
+        self.session.begin_transaction()
+        cursor[200] = self.ds.value(200)
+        self.session.commit_transaction("commit_timestamp=" + self.timestamp_str(10))
+
+        # Verify the largest key.
+        self.session.begin_transaction("read_timestamp=" + self.timestamp_str(5))
+        self.assertEqual(cursor.largest_key(), 0)
+        self.assertEqual(cursor.get_key(), 200)
+        self.session.rollback_transaction()
+    
+    def test_prepared_update(self):
         self.populate(100)
 
+        session2 = self.setUpSessionOpen(self.conn)
+        cursor2 = session2.open_cursor(self.type + self.tablename, None)
+        session2.begin_transaction()
+        cursor2[200] = self.ds.value(200)
+        session2.prepare_transaction("prepare_timestamp=" + self.timestamp_str(10))
+
         cursor = self.session.open_cursor(self.type + self.tablename, None)
-        self.session.begin_transaction('read_timestamp=' + self.timestamp_str(5))
-        # Expect the largest key to throw.
-        with self.expectedStderrPattern("largest key cannot be called with a read timestamp"):
-            try:
-                cursor.largest_key()
-            except wiredtiger.WiredTigerError as e:
-                gotException = True
-                self.pr('got expected exception: ' + str(e))
-                self.assertTrue(str(e).find('nvalid argument') >= 0)
-        self.assertTrue(gotException, msg = 'expected exception')
+
+        # Verify the largest key.
+        self.session.begin_transaction("read_timestamp=" + self.timestamp_str(20))
+        self.assertEqual(cursor.largest_key(), 0)
+        self.assertEqual(cursor.get_key(), 200)
         self.session.rollback_transaction()
 
     def test_not_positioned(self):