Import wiredtiger: b055c2402564bc521d73c274afa3d3bc989a0d3b from branch mongodb-master

ref: d1261c2d35..b055c24025
for: 6.1.0-rc0

WT-9147       No need to walk the whole update chain in __wt_rec_upd_select except for metadata

2 files changed, 29 insertions, 4 deletions

diff --git a/src/third_party/wiredtiger/import.data b/src/third_party/wiredtiger/import.data
index 92cbfc23bac..cbf3dcbdf58 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-master",
-    "commit": "d1261c2d35dc756bf8b30c0fa0a59a97137ec54d"
+    "commit": "b055c2402564bc521d73c274afa3d3bc989a0d3b"
 }
diff --git a/src/third_party/wiredtiger/src/reconcile/rec_visibility.c b/src/third_party/wiredtiger/src/reconcile/rec_visibility.c
index 4a7153ed196..b2022431e0c 100644
--- a/src/third_party/wiredtiger/src/reconcile/rec_visibility.c
+++ b/src/third_party/wiredtiger/src/reconcile/rec_visibility.c
@@ -418,8 +418,6 @@ __wt_rec_upd_select(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_INSERT *ins, W
         if ((txnid = upd->txnid) == WT_TXN_ABORTED)
             continue;
 
-        upd_memsize += WT_UPDATE_MEMSIZE(upd);
-
         /*
          * Track the first update in the chain that is not aborted and the maximum transaction ID.
          */
@@ -432,6 +430,7 @@ __wt_rec_upd_select(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_INSERT *ins, W
          * Special handling for application threads evicting their own updates.
          */
         if (!is_hs_page && F_ISSET(r, WT_REC_APP_EVICTION_SNAPSHOT) && txnid == session_txnid) {
+            upd_memsize += WT_UPDATE_MEMSIZE(upd);
             has_newer_updates = true;
             continue;
         }
@@ -459,6 +458,17 @@ __wt_rec_upd_select(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_INSERT *ins, W
           !is_hs_page &&
           (F_ISSET(r, WT_REC_VISIBLE_ALL) ? WT_TXNID_LE(r->last_running, txnid) :
                                             !__txn_visible_id(session, txnid))) {
+            /*
+             * Rare case: metadata writes at read uncommitted isolation level, eviction may see a
+             * committed update followed by uncommitted updates. Give up in that case because we
+             * can't discard the uncommitted updates.
+             */
+            if (upd_select->upd != NULL) {
+                WT_ASSERT(session, WT_IS_METADATA(session->dhandle));
+                return (__wt_set_return(session, EBUSY));
+            }
+
+            upd_memsize += WT_UPDATE_MEMSIZE(upd);
             has_newer_updates = true;
             continue;
         }
@@ -468,6 +478,7 @@ __wt_rec_upd_select(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_INSERT *ins, W
           upd->prepare_state == WT_PREPARE_INPROGRESS) {
             WT_ASSERT(session, upd_select->upd == NULL || upd_select->upd->txnid == upd->txnid);
             if (F_ISSET(r, WT_REC_CHECKPOINT)) {
+                upd_memsize += WT_UPDATE_MEMSIZE(upd);
                 has_newer_updates = true;
                 if (upd->start_ts > max_ts)
                     max_ts = upd->start_ts;
@@ -495,7 +506,12 @@ __wt_rec_upd_select(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_INSERT *ins, W
         if (upd_select->upd == NULL)
             upd_select->upd = upd;
 
-        if (!F_ISSET(r, WT_REC_EVICT))
+        /*
+         * We only need to walk the whole update chain if we are evicting metadata as it is written
+         * with read uncommitted isolation and we may see a committed update followed by uncommitted
+         * updates
+         */
+        if (!F_ISSET(r, WT_REC_EVICT) || !WT_IS_METADATA(session->dhandle))
             break;
     }
 
@@ -726,6 +742,15 @@ __wt_rec_upd_select(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_INSERT *ins, W
         supd_restore = F_ISSET(r, WT_REC_EVICT) &&
           (has_newer_updates || F_ISSET(S2C(session), WT_CONN_IN_MEMORY));
 
+        /*
+         * The total update size only contains uncommitted updates. This is wrong for the in memory
+         * case because we cannot discard any update until they are obsolete. Add them to the size.
+         */
+        if (F_ISSET(S2C(session), WT_CONN_IN_MEMORY) && onpage_upd != NULL) {
+            for (upd = tombstone != NULL ? tombstone : onpage_upd; upd != NULL; upd = upd->next)
+                if (upd->txnid != WT_TXN_ABORTED)
+                    upd_memsize += WT_UPDATE_MEMSIZE(upd);
+        }
         WT_RET(__rec_update_save(
           session, r, ins, rip, onpage_upd, tombstone, supd_restore, upd_memsize));