Import wiredtiger: ff7311530da905f46e26bb1690302fb4b1da1a6f from branch mongodb-5.0

ref: c93ddc3da4..ff7311530d
for: 5.0.11

WT-9500 Fix RTS to use cell time window instead of key/value timestamps of HS update (v5.0 backport) (#8173)

3 files changed, 155 insertions, 33 deletions

diff --git a/src/third_party/wiredtiger/import.data b/src/third_party/wiredtiger/import.data
index 94e9c4fbc85..a78518d2ed9 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-5.0",
-    "commit": "c93ddc3da4f8384a084e9b585960fc284ef5a540"
+    "commit": "ff7311530da905f46e26bb1690302fb4b1da1a6f"
 }
diff --git a/src/third_party/wiredtiger/src/txn/txn_rollback_to_stable.c b/src/third_party/wiredtiger/src/txn/txn_rollback_to_stable.c
index e6209e305cf..1364a563452 100644
--- a/src/third_party/wiredtiger/src/txn/txn_rollback_to_stable.c
+++ b/src/third_party/wiredtiger/src/txn/txn_rollback_to_stable.c
@@ -29,9 +29,7 @@ __rollback_delete_hs(WT_SESSION_IMPL *session, WT_ITEM *key, wt_timestamp_t ts)
     WT_CURSOR *hs_cursor;
     WT_DECL_ITEM(hs_key);
     WT_DECL_RET;
-    wt_timestamp_t hs_start_ts;
-    uint64_t hs_counter;
-    uint32_t hs_btree_id;
+    WT_TIME_WINDOW *hs_tw;
 
     /* Open a history store table cursor. */
     WT_RET(__wt_curhs_open(session, NULL, &hs_cursor));
@@ -53,12 +51,14 @@ __rollback_delete_hs(WT_SESSION_IMPL *session, WT_ITEM *key, wt_timestamp_t ts)
     hs_cursor->set_key(hs_cursor, 4, S2BT(session)->id, key, WT_TS_MAX, UINT64_MAX);
     ret = __wt_curhs_search_near_before(session, hs_cursor);
     for (; ret == 0; ret = hs_cursor->prev(hs_cursor)) {
-        WT_ERR(hs_cursor->get_key(hs_cursor, &hs_btree_id, hs_key, &hs_start_ts, &hs_counter));
-        if (hs_start_ts < ts)
+        /* Retrieve the time window from the history cursor. */
+        __wt_hs_upd_time_window(hs_cursor, &hs_tw);
+        if (hs_tw->start_ts < ts)
             break;
+
         WT_ERR(hs_cursor->remove(hs_cursor));
         WT_STAT_CONN_DATA_INCR(session, txn_rts_hs_removed);
-        if (hs_start_ts == ts)
+        if (hs_tw->start_ts == ts)
             WT_STAT_CONN_DATA_INCR(session, cache_hs_key_truncate_rts);
         else
             WT_STAT_CONN_DATA_INCR(session, cache_hs_key_truncate_rts_unstable);
@@ -401,7 +401,7 @@ __rollback_ondisk_fixup_key(WT_SESSION_IMPL *session, WT_REF *ref, WT_ROW *rip,
          * become obsolete according to the checkpoint.
          */
         if (__rollback_txn_visible_id(session, hs_tw->stop_txn) &&
-          hs_stop_durable_ts <= pinned_ts) {
+          hs_tw->durable_stop_ts <= pinned_ts) {
             __wt_verbose_multi(session, WT_VERB_RECOVERY_RTS(session),
               "history store stop is obsolete with time window: %s and pinned timestamp: %s",
               __wt_time_window_to_string(hs_tw, tw_string),
@@ -423,7 +423,7 @@ __rollback_ondisk_fixup_key(WT_SESSION_IMPL *session, WT_REF *ref, WT_ROW *rip,
          * the written proper timestamp, so comparing against it with history store shouldn't have
          * any problem.
          */
-        if (hs_start_ts <= unpack->tw.start_ts || unpack->tw.prepare) {
+        if (hs_tw->start_ts <= unpack->tw.start_ts || unpack->tw.prepare) {
             if (type == WT_UPDATE_MODIFY)
                 WT_ERR(__wt_modify_apply_item(
                   session, S2BT(session)->value_format, full_value, hs_value->data));
@@ -433,11 +433,9 @@ __rollback_ondisk_fixup_key(WT_SESSION_IMPL *session, WT_REF *ref, WT_ROW *rip,
             }
         } else
             __wt_verbose_multi(session, WT_VERB_RECOVERY_RTS(session),
-              "history store update more recent than on-disk update with start timestamp: %s,"
-              " durable timestamp: %s, stop timestamp: %s and type: %" PRIu8,
-              __wt_timestamp_to_string(hs_start_ts, ts_string[0]),
-              __wt_timestamp_to_string(hs_durable_ts, ts_string[1]),
-              __wt_timestamp_to_string(hs_stop_durable_ts, ts_string[2]), type);
+              "history store update more recent than on-disk update with time window: %s and type: "
+              "%" PRIu8,
+              __wt_time_window_to_string(hs_tw, tw_string), type);
 
         /*
          * Verify the history store timestamps are in order. The start timestamp may be equal to the
@@ -469,32 +467,38 @@ __rollback_ondisk_fixup_key(WT_SESSION_IMPL *session, WT_REF *ref, WT_ROW *rip,
             WT_STAT_CONN_DATA_INCR(session, txn_rts_hs_stop_older_than_newer_start);
 
         /*
+         * Validate the timestamps in the key and the cell are same. This must be validated only
+         * after verifying it's stop time window is not globally visible. The start timestamps of
+         * the time window are cleared when they are globally visible and there will be no stop
+         * timestamp in the history store whenever a prepared update is written to the data store.
+         */
+        WT_ASSERT(session,
+          (hs_tw->start_ts == WT_TS_NONE || hs_tw->start_ts == hs_start_ts) &&
+            (hs_tw->durable_start_ts == WT_TS_NONE || hs_tw->durable_start_ts == hs_durable_ts) &&
+            ((hs_tw->durable_stop_ts == 0 && hs_stop_durable_ts == WT_TS_MAX) ||
+              hs_tw->durable_stop_ts == hs_stop_durable_ts));
+
+        /*
          * Stop processing when we find a stable update according to the given timestamp and
          * transaction id.
          */
         if (__rollback_txn_visible_id(session, hs_tw->start_txn) &&
-          hs_durable_ts <= rollback_timestamp) {
+          hs_tw->durable_start_ts <= rollback_timestamp) {
             __wt_verbose_multi(session, WT_VERB_RECOVERY_RTS(session),
-              "history store update valid with start timestamp: %s, durable timestamp: %s, stop "
-              "timestamp: %s, stable timestamp: %s, txnid: %" PRIu64 " and type: %" PRIu8,
-              __wt_timestamp_to_string(hs_start_ts, ts_string[0]),
-              __wt_timestamp_to_string(hs_durable_ts, ts_string[1]),
-              __wt_timestamp_to_string(hs_stop_durable_ts, ts_string[2]),
-              __wt_timestamp_to_string(rollback_timestamp, ts_string[3]), hs_tw->start_txn, type);
+              "history store update valid with time window: %s, type: %" PRIu8
+              " and stable timestamp: %s",
+              __wt_time_window_to_string(hs_tw, tw_string), type,
+              __wt_timestamp_to_string(rollback_timestamp, ts_string[0]));
             WT_ASSERT(session, unpack->tw.prepare || hs_tw->start_ts <= unpack->tw.start_ts);
             valid_update_found = true;
             break;
         }
 
         __wt_verbose_multi(session, WT_VERB_RECOVERY_RTS(session),
-          "history store update aborted with start timestamp: %s, durable timestamp: %s, stop "
-          "timestamp: %s, stable timestamp: %s, start txnid: %" PRIu64 ", stop txnid: %" PRIu64
-          " and type: %" PRIu8,
-          __wt_timestamp_to_string(hs_start_ts, ts_string[0]),
-          __wt_timestamp_to_string(hs_durable_ts, ts_string[1]),
-          __wt_timestamp_to_string(hs_stop_durable_ts, ts_string[2]),
-          __wt_timestamp_to_string(rollback_timestamp, ts_string[3]), hs_tw->start_txn,
-          hs_tw->stop_txn, type);
+          "history store update aborted with time window: %s,  type: %" PRIu8
+          " and stable timestamp: %s",
+          __wt_time_window_to_string(hs_tw, tw_string), type,
+          __wt_timestamp_to_string(rollback_timestamp, ts_string[3]));
 
         /*
          * Start time point of the current record may be used as stop time point of the previous
@@ -551,7 +555,7 @@ __rollback_ondisk_fixup_key(WT_SESSION_IMPL *session, WT_REF *ref, WT_ROW *rip,
          * timestamp and txnid, we need to restore that as well.
          */
         if (__rollback_txn_visible_id(session, hs_tw->stop_txn) &&
-          hs_stop_durable_ts <= rollback_timestamp) {
+          hs_tw->durable_stop_ts <= rollback_timestamp) {
             /*
              * The restoring tombstone timestamp must be zero or less than previous update start
              * timestamp or the on-disk update is an out of order prepared.
@@ -1285,10 +1289,11 @@ __rollback_to_stable_btree_hs_truncate(WT_SESSION_IMPL *session, uint32_t btree_
     WT_CURSOR *hs_cursor;
     WT_DECL_ITEM(hs_key);
     WT_DECL_RET;
+    WT_TIME_WINDOW *hs_tw;
     wt_timestamp_t hs_start_ts;
     uint64_t hs_counter;
     uint32_t hs_btree_id;
-    char ts_string[WT_TS_INT_STRING_SIZE];
+    char tw_string[WT_TIME_STRING_SIZE];
 
     hs_cursor = NULL;
 
@@ -1307,9 +1312,12 @@ __rollback_to_stable_btree_hs_truncate(WT_SESSION_IMPL *session, uint32_t btree_
         /* We shouldn't cross the btree search space. */
         WT_ASSERT(session, btree_id == hs_btree_id);
 
+        /* Retrieve the time window from the history cursor. */
+        __wt_hs_upd_time_window(hs_cursor, &hs_tw);
+
         __wt_verbose_multi(session, WT_VERB_RECOVERY_RTS(session),
-          "rollback to stable history store cleanup of update with start timestamp: %s",
-          __wt_timestamp_to_string(hs_start_ts, ts_string));
+          "rollback to stable history store cleanup of update with time window: %s",
+          __wt_time_window_to_string(hs_tw, tw_string));
 
         WT_ERR(hs_cursor->remove(hs_cursor));
         WT_STAT_CONN_DATA_INCR(session, txn_rts_hs_removed);
diff --git a/src/third_party/wiredtiger/test/suite/test_rollback_to_stable37.py b/src/third_party/wiredtiger/test/suite/test_rollback_to_stable37.py
new file mode 100644
index 00000000000..c862ffd0e0d
--- /dev/null
+++ b/src/third_party/wiredtiger/test/suite/test_rollback_to_stable37.py
@@ -0,0 +1,114 @@
+#!/usr/bin/env python
+#
+# Public Domain 2014-present MongoDB, Inc.
+# Public Domain 2008-2014 WiredTiger, Inc.
+#
+# This is free and unencumbered software released into the public domain.
+#
+# Anyone is free to copy, modify, publish, use, compile, sell, or
+# distribute this software, either in source code form or as a compiled
+# binary, for any purpose, commercial or non-commercial, and by any
+# means.
+#
+# In jurisdictions that recognize copyright laws, the author or authors
+# of this software dedicate any and all copyright interest in the
+# software to the public domain. We make this dedication for the benefit
+# of the public at large and to the detriment of our heirs and
+# successors. We intend this dedication to be an overt act of
+# relinquishment in perpetuity of all present and future rights to this
+# software under copyright law.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+# EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+# MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+# IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+# OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+# ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+# OTHER DEALINGS IN THE SOFTWARE.
+
+from wiredtiger import stat
+from wtdataset import SimpleDataSet
+from wtscenario import make_scenarios
+from helper import simulate_crash_restart
+from test_rollback_to_stable01 import test_rollback_to_stable_base
+
+# test_rollback_to_stable37.py
+# Test that the rollback to stable to restore proper stable update from history store when a no timestamp
+# update has rewritten the history store data.
+class test_rollback_to_stable37(test_rollback_to_stable_base):
+    conn_config = 'cache_size=1GB,statistics=(all),statistics_log=(json,on_close,wait=1),log=(enabled=false)'
+
+    format_values = [
+        ('column', dict(key_format='r', value_format='S')),
+        ('row_integer', dict(key_format='i', value_format='S')),
+    ]
+
+    scenarios = make_scenarios(format_values)
+
+    def test_rollback_to_stable(self):
+        uri = 'table:test_rollback_to_stable37'
+        nrows = 1000
+
+        if self.value_format == '8t':
+            value_a = 97
+            value_b = 98
+            value_c = 99
+            value_d = 100
+        else:
+            value_a = 'a' * 10
+            value_b = 'b' * 10
+            value_c = 'c' * 10
+            value_d = 'd' * 10
+
+        # Create our table.
+        ds = SimpleDataSet(self, uri, 0, key_format=self.key_format, value_format=self.value_format)
+        ds.populate()
+
+        # Pin oldest and stable to timestamp 1.
+        self.conn.set_timestamp('oldest_timestamp=' + self.timestamp_str(1) +
+            ',stable_timestamp=' + self.timestamp_str(1))
+
+        # Insert 300 updates to the same key.
+        for i in range (20, 320):
+            if self.value_format == '8t':
+                self.large_updates(uri, value_a, ds, nrows, False, i)
+            else:
+                self.large_updates(uri, value_a + str(i), ds, nrows, False, i)
+
+        old_reader_session = self.conn.open_session()
+        old_reader_session.begin_transaction('read_timestamp=' + self.timestamp_str(10))
+
+        self.large_updates(uri, value_b, ds, nrows, False, 2000)
+        self.check(value_b, uri, nrows, 2000)
+
+        self.evict_cursor(uri, nrows, value_b)
+
+        # Insert update without a timestamp.
+        self.large_updates(uri, value_c, ds, nrows, False, 0)
+        self.check(value_c, uri, nrows, 0)
+
+        self.evict_cursor(uri, nrows, value_c)
+
+        self.large_updates(uri, value_d, ds, nrows, False, 3000)
+        self.check(value_d, uri, nrows, 3000)
+
+        old_reader_session.rollback_transaction()
+        self.session.checkpoint()
+
+        self.conn.set_timestamp('stable_timestamp=' + self.timestamp_str(2000))
+        self.session.checkpoint()
+
+        self.conn.rollback_to_stable()
+
+        self.check(value_c, uri, nrows, 1000)
+        self.check(value_c, uri, nrows, 2000)
+        self.check(value_c, uri, nrows, 3000)
+
+        stat_cursor = self.session.open_cursor('statistics:', None, None)
+        keys_removed = stat_cursor[stat.conn.txn_rts_keys_removed][2]
+        stat_cursor.close()
+
+        self.assertEqual(keys_removed, 0)
+
+if __name__ == '__main__':
+    wttest.run()