Import wiredtiger: 02329fa1d9ba292f5545d2adc67d4c324d0a2b04 from branch mongodb-5.2

ref: 461b911f9e..02329fa1d9
for: 5.2.0-rc5

WT-8635       Extend VLCS RLE caching test to another set of cases

5 files changed, 20 insertions, 8 deletions

diff --git a/src/third_party/wiredtiger/import.data b/src/third_party/wiredtiger/import.data
index 4c75cc0019a..fc2b568e7d1 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-5.2",
-    "commit": "461b911f9e91e24dad5785dad6bb78eabd3a395f"
+    "commit": "02329fa1d9ba292f5545d2adc67d4c324d0a2b04"
 }
diff --git a/src/third_party/wiredtiger/src/btree/bt_curnext.c b/src/third_party/wiredtiger/src/btree/bt_curnext.c
index bdb3e0149a7..4c0ddf4eeb6 100644
--- a/src/third_party/wiredtiger/src/btree/bt_curnext.c
+++ b/src/third_party/wiredtiger/src/btree/bt_curnext.c
@@ -321,6 +321,10 @@ restart_read:
          * It is only safe to cache the value for other keys in the same RLE cell if it is globally
          * visible. Otherwise, there might be some older timestamp where the value isn't uniform
          * across the cell. Always set cip_saved so it's easy to tell when we change cells.
+         *
+         * Note: it's important that we're checking the on-disk value for global visibility, and not
+         * whatever __wt_txn_read returned, which might be something else. (If it's something else,
+         * we can't cache it; but in that case the on-disk value cannot be globally visible.)
          */
         cbt->cip_saved = cip;
         if (rle > 1 &&
diff --git a/src/third_party/wiredtiger/src/btree/bt_curprev.c b/src/third_party/wiredtiger/src/btree/bt_curprev.c
index dfba334b26d..b93ac62d9bc 100644
--- a/src/third_party/wiredtiger/src/btree/bt_curprev.c
+++ b/src/third_party/wiredtiger/src/btree/bt_curprev.c
@@ -479,6 +479,10 @@ restart_read:
          * It is only safe to cache the value for other keys in the same RLE cell if it is globally
          * visible. Otherwise, there might be some older timestamp where the value isn't uniform
          * across the cell. Always set cip_saved so it's easy to tell when we change cells.
+         *
+         * Note: it's important that we're checking the on-disk value for global visibility, and not
+         * whatever __wt_txn_read returned, which might be something else. (If it's something else,
+         * we can't cache it; but in that case the on-disk value cannot be globally visible.)
          */
         cbt->cip_saved = cip;
         if (rle > 1 &&
diff --git a/src/third_party/wiredtiger/test/suite/test_hs26.py b/src/third_party/wiredtiger/test/suite/test_hs26.py
index 9aab2322c43..acb3bd07790 100644
--- a/src/third_party/wiredtiger/test/suite/test_hs26.py
+++ b/src/third_party/wiredtiger/test/suite/test_hs26.py
@@ -69,6 +69,10 @@ class test_hs26(wttest.WiredTigerTestCase):
     # Other cases of overlapping the key count are still interesting so we still generate
     # the full product of the scenarios.
 
+    visibility_values = [
+        ('not_globally_visible', dict(ts1_globally_visible=False)),
+        ('globally_visible', dict(ts1_globally_visible=True)),
+    ]
     nrows_values = [
         ('more', dict(nrows_1=103, nrows_2=211)),
         ('same', dict(nrows_1=211, nrows_2=211)),
@@ -85,7 +89,8 @@ class test_hs26(wttest.WiredTigerTestCase):
         ('17', dict(value_modulus_2=17)),
     ]
 
-    scenarios = make_scenarios(nrows_values, value_modulus_1_values, value_modulus_2_values)
+    scenarios = make_scenarios(visibility_values,
+        nrows_values, value_modulus_1_values, value_modulus_2_values)
 
     value_1 = 'a' * 500
     value_2 = 'd' * 500
@@ -171,8 +176,12 @@ class test_hs26(wttest.WiredTigerTestCase):
         # Write the first set of values at timestamp_1.
         self.make_updates(ds.uri, ds, self.value_1, self.value_modulus_1, self.nrows_1, self.timestamp_1)
 
+        # Optionally make the first set of values globally visible (and stable).
+        if self.ts1_globally_visible:
+            self.conn.set_timestamp('oldest_timestamp=' + self.timestamp_str(self.timestamp_1) +
+                ',stable_timestamp=' + self.timestamp_str(self.timestamp_1))
+
         # Create a long running read transaction in a separate session.
-        # (Is it necessary for it to be separate? Not sure.)
         session_read = self.conn.open_session()
         session_read.begin_transaction('read_timestamp=' + self.timestamp_str(self.timestamp_1))
 
@@ -186,8 +195,6 @@ class test_hs26(wttest.WiredTigerTestCase):
         self.check(self.session, ds.uri, self.timestamp_1, self.timestamp_1)
         self.check(self.session, ds.uri, self.timestamp_2, self.timestamp_2)
 
-        self.session.breakpoint()
-
         # Now forcibly evict, so that all the pages are RLE-encoded and then read back in.
         # There doesn't seem to be any way to just forcibly evict an entire table, so what
         # I'm going to do is assume that each page can hold at least 41 values, and evict
diff --git a/src/third_party/wiredtiger/test/suite/test_hs27.py b/src/third_party/wiredtiger/test/suite/test_hs27.py
index e685503b8a2..47725808319 100644
--- a/src/third_party/wiredtiger/test/suite/test_hs27.py
+++ b/src/third_party/wiredtiger/test/suite/test_hs27.py
@@ -275,7 +275,6 @@ class test_hs27(wttest.WiredTigerTestCase):
             self.initialize(ds.uri, ds)
 
         # Create a long running read transaction in a separate session.
-        # (Is it necessary for it to be separate? Not sure.)
         session_read = self.conn.open_session()
         session_read.begin_transaction('read_timestamp=' + self.timestamp_str(2))
 
@@ -288,8 +287,6 @@ class test_hs27(wttest.WiredTigerTestCase):
         # Check that the new updates are appropriately visible.
         self.checkall(self.session, ds.uri, ds)
 
-        self.session.breakpoint()
-
         # Now forcibly evict, so that all the pages are RLE-encoded and then read back in.
         # There doesn't seem to be any way to just forcibly evict an entire table, so what
         # I'm going to do is assume that what we care about is evicting the updates (the