Import wiredtiger: 9e92b73891510149a24567249223e2ccc34abe91 from branch mongodb-master

ref: 5c6e571501..9e92b73891
for: 5.3.0

WT-8711       Block pinned timestamp from moving if there is an open version cursor

8 files changed, 45 insertions, 35 deletions

diff --git a/src/third_party/wiredtiger/examples/c/ex_cursor.c b/src/third_party/wiredtiger/examples/c/ex_cursor.c
index 5957582ebdf..45381a91956 100644
--- a/src/third_party/wiredtiger/examples/c/ex_cursor.c
+++ b/src/third_party/wiredtiger/examples/c/ex_cursor.c
@@ -229,7 +229,6 @@ main(int argc, char *argv[])
     error_check(cursor->close(cursor));
 
     /* Create a version cursor. */
-    error_check(session->begin_transaction(session, "read_timestamp=1"));
     error_check(
       session->open_cursor(session, "file:map.wt", NULL, "debug=(dump_version=true)", &cursor));
     error_check(version_cursor_dump(cursor));
diff --git a/src/third_party/wiredtiger/import.data b/src/third_party/wiredtiger/import.data
index 6ced12b78e6..92dec15656a 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": "5c6e57150184b77be717446f3184ac32ac42dab9"
+    "commit": "9e92b73891510149a24567249223e2ccc34abe91"
 }
diff --git a/src/third_party/wiredtiger/src/cursor/cur_version.c b/src/third_party/wiredtiger/src/cursor/cur_version.c
index 5fe0cff3754..4088a8dec0d 100644
--- a/src/third_party/wiredtiger/src/cursor/cur_version.c
+++ b/src/third_party/wiredtiger/src/cursor/cur_version.c
@@ -478,8 +478,6 @@ __curversion_search(WT_CURSOR *cursor)
     WT_ROW *rip;
     WT_SESSION_IMPL *session;
     WT_TXN *txn;
-    WT_TXN_SHARED *txn_shared;
-    wt_timestamp_t oldest_ts;
 
     version_cursor = (WT_CURSOR_VERSION *)cursor;
     file_cursor = version_cursor->file_cursor;
@@ -487,22 +485,13 @@ __curversion_search(WT_CURSOR *cursor)
 
     CURSOR_API_CALL(cursor, session, search, CUR2BT(cbt));
     txn = session->txn;
-    txn_shared = WT_SESSION_TXN_SHARED(session);
 
     /*
-     * Check that we have the current transaction's read timestamp pinged as the oldest timestamp to
-     * ensure that the global visibility will not change during the life of this transaction.
+     * We need to run with snapshot isolation to ensure that the globally visibility does not move.
      */
-    WT_ERR_NOTFOUND_OK(
-      __wt_txn_get_pinned_timestamp(session, &oldest_ts, WT_TXN_TS_INCLUDE_OLDEST), true);
-    if (!F_ISSET(txn, WT_TXN_SHARED_TS_READ) ||
-      (ret == 0 && oldest_ts < txn_shared->read_timestamp))
-        WT_ERR(__wt_txn_rollback_required(session,
-          "version cursor can only be called with the read timestamp as the oldest timestamp"));
-    if (ret == WT_NOTFOUND && txn_shared->read_timestamp > 1)
-        WT_ERR(__wt_txn_rollback_required(session,
-          "version cursor can only be called with read timestamp 1 if there is no oldest "
-          "timestamp"));
+    if (txn->isolation != WT_ISO_SNAPSHOT)
+        WT_ERR(__wt_txn_rollback_required(
+          session, "version cursor can only be called with snapshot isolation"));
 
     WT_ERR(__cursor_checkkey(file_cursor));
     if (F_ISSET(file_cursor, WT_CURSTD_KEY_INT))
@@ -571,6 +560,7 @@ __curversion_close(WT_CURSOR *cursor)
     hs_cursor = version_cursor->hs_cursor;
     file_cursor = version_cursor->file_cursor;
     CURSOR_API_CALL(cursor, session, close, NULL);
+
 err:
     version_cursor->next_upd = NULL;
     if (file_cursor != NULL) {
@@ -583,6 +573,7 @@ err:
     }
     __wt_free(session, cursor->value_format);
     __wt_cursor_close(cursor);
+    __wt_atomic_sub32(&S2C(session)->version_cursor_count, 1);
 
     API_END_RET(session, ret);
 }
@@ -620,12 +611,15 @@ __wt_curversion_open(WT_SESSION_IMPL *session, const char *uri, WT_CURSOR *owner
     WT_CURSOR *cursor;
     WT_CURSOR_VERSION *version_cursor;
     WT_DECL_RET;
+    WT_TXN_GLOBAL *txn_global;
+    wt_timestamp_t pinned_ts;
     /* The file cursor is read only. */
     const char *file_cursor_cfg[] = {
       WT_CONFIG_BASE(session, WT_SESSION_open_cursor), "read_only=true", NULL};
     char *version_cursor_value_format;
     size_t format_len;
 
+    txn_global = &S2C(session)->txn_global;
     *cursorp = NULL;
     WT_RET(__wt_calloc_one(session, &version_cursor));
     cursor = (WT_CURSOR *)version_cursor;
@@ -633,6 +627,15 @@ __wt_curversion_open(WT_SESSION_IMPL *session, const char *uri, WT_CURSOR *owner
     cursor->session = (WT_SESSION *)session;
     version_cursor_value_format = NULL;
 
+    /* Freeze pinned timestamp when we open the first version cursor. */
+    __wt_writelock(session, &txn_global->rwlock);
+    if (S2C(session)->version_cursor_count == 0) {
+        __wt_txn_pinned_timestamp(session, &pinned_ts);
+        txn_global->version_cursor_pinned_timestamp = pinned_ts;
+    }
+    (void)__wt_atomic_add32(&S2C(session)->version_cursor_count, 1);
+    __wt_writeunlock(session, &txn_global->rwlock);
+
     /* Open the file cursor to check the key and value format. */
     WT_ERR(__wt_open_cursor(session, uri, NULL, file_cursor_cfg, &version_cursor->file_cursor));
     cursor->key_format = version_cursor->file_cursor->key_format;
diff --git a/src/third_party/wiredtiger/src/include/connection.h b/src/third_party/wiredtiger/src/include/connection.h
index f1677f4f70c..f361eb28311 100644
--- a/src/third_party/wiredtiger/src/include/connection.h
+++ b/src/third_party/wiredtiger/src/include/connection.h
@@ -301,11 +301,12 @@ struct __wt_connection_impl {
 
     /* Locked: handles in each bucket */
     uint64_t *dh_bucket_count;
-    uint64_t dhandle_count;     /* Locked: handles in the queue */
-    u_int open_btree_count;     /* Locked: open writable btree count */
-    uint32_t next_file_id;      /* Locked: file ID counter */
-    uint32_t open_file_count;   /* Atomic: open file handle count */
-    uint32_t open_cursor_count; /* Atomic: open cursor handle count */
+    uint64_t dhandle_count;        /* Locked: handles in the queue */
+    u_int open_btree_count;        /* Locked: open writable btree count */
+    uint32_t next_file_id;         /* Locked: file ID counter */
+    uint32_t open_file_count;      /* Atomic: open file handle count */
+    uint32_t open_cursor_count;    /* Atomic: open cursor handle count */
+    uint32_t version_cursor_count; /* Atomic: open version cursor count */
 
     /*
      * WiredTiger allocates space for 50 simultaneous sessions (threads of control) by default.
diff --git a/src/third_party/wiredtiger/src/include/txn.h b/src/third_party/wiredtiger/src/include/txn.h
index 127ae0c2835..1370f9c59ff 100644
--- a/src/third_party/wiredtiger/src/include/txn.h
+++ b/src/third_party/wiredtiger/src/include/txn.h
@@ -134,6 +134,7 @@ struct __wt_txn_global {
     wt_timestamp_t pinned_timestamp;
     wt_timestamp_t recovery_timestamp;
     wt_timestamp_t stable_timestamp;
+    wt_timestamp_t version_cursor_pinned_timestamp;
     bool has_durable_timestamp;
     bool has_oldest_timestamp;
     bool has_pinned_timestamp;
diff --git a/src/third_party/wiredtiger/src/include/txn_inline.h b/src/third_party/wiredtiger/src/include/txn_inline.h
index 798818d7dab..68f3a523458 100644
--- a/src/third_party/wiredtiger/src/include/txn_inline.h
+++ b/src/third_party/wiredtiger/src/include/txn_inline.h
@@ -515,6 +515,12 @@ __wt_txn_pinned_timestamp(WT_SESSION_IMPL *session, wt_timestamp_t *pinned_tsp)
     if (!txn_global->has_pinned_timestamp)
         return;
 
+    /* If we have a version cursor open, use the pinned timestamp when it is opened. */
+    if (S2C(session)->version_cursor_count > 0) {
+        *pinned_tsp = txn_global->version_cursor_pinned_timestamp;
+        return;
+    }
+
     *pinned_tsp = pinned_ts = txn_global->pinned_timestamp;
 
     /*
diff --git a/src/third_party/wiredtiger/test/suite/test_cursor18.py b/src/third_party/wiredtiger/test/suite/test_cursor18.py
index f4f8c73f23f..2eb7028758f 100644
--- a/src/third_party/wiredtiger/test/suite/test_cursor18.py
+++ b/src/third_party/wiredtiger/test/suite/test_cursor18.py
@@ -77,7 +77,7 @@ class test_cursor18(wttest.WiredTigerTestCase):
         self.session.commit_transaction("commit_timestamp=" + self.timestamp_str(5))
 
         # Open a version cursor
-        self.session.begin_transaction("read_timestamp=" + self.timestamp_str(1))
+        self.session.begin_transaction()
         version_cursor = self.session.open_cursor(self.uri, None, "debug=(dump_version=true)")
         version_cursor.set_key(1)
         self.assertEquals(version_cursor.search(), 0)
@@ -104,7 +104,7 @@ class test_cursor18(wttest.WiredTigerTestCase):
         self.session.rollback_transaction()
 
         # Open a version cursor
-        self.session.begin_transaction("read_timestamp=" + self.timestamp_str(1))
+        self.session.begin_transaction()
         version_cursor = self.session.open_cursor(self.uri, None, "debug=(dump_version=true)")
         version_cursor.set_key(1)
         self.assertEquals(version_cursor.search(), 0)
@@ -138,7 +138,7 @@ class test_cursor18(wttest.WiredTigerTestCase):
         self.session.rollback_transaction()
 
         # Open a version cursor
-        self.session.begin_transaction("read_timestamp=" + self.timestamp_str(1))
+        self.session.begin_transaction()
         version_cursor = self.session.open_cursor(self.uri, None, "debug=(dump_version=true)")
         version_cursor.set_key(1)
         self.assertEquals(version_cursor.search(), 0)
@@ -168,7 +168,7 @@ class test_cursor18(wttest.WiredTigerTestCase):
         self.session.commit_transaction("commit_timestamp=" + self.timestamp_str(5))
 
         # Open a version cursor
-        self.session.begin_transaction("read_timestamp=" + self.timestamp_str(1))
+        self.session.begin_transaction()
         version_cursor = self.session.open_cursor(self.uri, None, "debug=(dump_version=true)")
         version_cursor.set_key(1)
         self.assertEquals(version_cursor.search(), 0)
@@ -197,7 +197,7 @@ class test_cursor18(wttest.WiredTigerTestCase):
         self.session.rollback_transaction()
 
         # Open a version cursor
-        self.session.begin_transaction("read_timestamp=" + self.timestamp_str(1))
+        self.session.begin_transaction()
         version_cursor = self.session.open_cursor(self.uri, None, "debug=(dump_version=true)")
         version_cursor.set_key(1)
         self.assertEquals(version_cursor.search(), 0)
@@ -234,7 +234,7 @@ class test_cursor18(wttest.WiredTigerTestCase):
         self.session.commit_transaction("commit_timestamp=" + self.timestamp_str(10))
 
         # Open a version cursor
-        self.session.begin_transaction("read_timestamp=" + self.timestamp_str(1))
+        self.session.begin_transaction()
         version_cursor = self.session.open_cursor(self.uri, None, "debug=(dump_version=true)")
         version_cursor.set_key(1)
         self.assertEquals(version_cursor.search(), 0)
@@ -269,7 +269,7 @@ class test_cursor18(wttest.WiredTigerTestCase):
         self.session.commit_transaction("commit_timestamp=" + self.timestamp_str(20))
 
         # Ensure that we are able to correctly traverse all versions of this new key.
-        self.session.begin_transaction("read_timestamp=" + self.timestamp_str(1))
+        self.session.begin_transaction()
         version_cursor.set_key(2)
         self.assertEquals(version_cursor.search(), 0)
         self.assertEquals(version_cursor.get_key(), 2)
@@ -304,7 +304,7 @@ class test_cursor18(wttest.WiredTigerTestCase):
         self.session.rollback_transaction()
 
         # Open a version cursor
-        self.session.begin_transaction("read_timestamp=" + self.timestamp_str(1))
+        self.session.begin_transaction()
         version_cursor = self.session.open_cursor(self.uri, None, "debug=(dump_version=true)")
         version_cursor.set_key(1)
         self.assertEquals(version_cursor.search(), 0)
@@ -331,7 +331,7 @@ class test_cursor18(wttest.WiredTigerTestCase):
         self.session.rollback_transaction()
 
         # Open a version cursor
-        self.session.begin_transaction("read_timestamp=" + self.timestamp_str(1))
+        self.session.begin_transaction()
         version_cursor = self.session.open_cursor(self.uri, None, "debug=(dump_version=true)")
         version_cursor.set_key(1)
         self.assertEquals(version_cursor.search(), 0)
@@ -372,7 +372,7 @@ class test_cursor18(wttest.WiredTigerTestCase):
         session2.prepare_transaction("prepare_timestamp=" + self.timestamp_str(2))
 
         # Open a version cursor
-        self.session.begin_transaction("read_timestamp=" + self.timestamp_str(1))
+        self.session.begin_transaction()
         version_cursor = self.session.open_cursor(self.uri, None, "debug=(dump_version=true)")
         version_cursor.set_key(1)
         self.assertEquals(version_cursor.search(), 0)
@@ -390,7 +390,7 @@ class test_cursor18(wttest.WiredTigerTestCase):
         self.session.commit_transaction("commit_timestamp=" + self.timestamp_str(1))
 
         # Open a version cursor
-        self.session.begin_transaction("read_timestamp=" + self.timestamp_str(1))
+        self.session.begin_transaction()
         version_cursor = self.session.open_cursor(self.uri, None, "debug=(dump_version=true)")
         version_cursor.set_key(1)
         self.assertEquals(version_cursor.search(), 0)
@@ -417,7 +417,7 @@ class test_cursor18(wttest.WiredTigerTestCase):
         self.session.commit_transaction("commit_timestamp=" + self.timestamp_str(5))
 
         # Open a version cursor
-        self.session.begin_transaction("read_timestamp=" + self.timestamp_str(1))
+        self.session.begin_transaction()
         version_cursor = self.session.open_cursor(self.uri, None, "debug=(dump_version=true)")
         version_cursor.set_key(1)
         self.assertEquals(version_cursor.search(), 0)
diff --git a/src/third_party/wiredtiger/test/suite/test_cursor19.py b/src/third_party/wiredtiger/test/suite/test_cursor19.py
index dd3c0164451..b3d56564ea7 100644
--- a/src/third_party/wiredtiger/test/suite/test_cursor19.py
+++ b/src/third_party/wiredtiger/test/suite/test_cursor19.py
@@ -127,7 +127,7 @@ class test_cursor19(wttest.WiredTigerTestCase):
         self.session.commit_transaction("commit_timestamp=" + self.timestamp_str(30))
 
         # Open a version cursor
-        self.session.begin_transaction("read_timestamp=" + self.timestamp_str(1))
+        self.session.begin_transaction()
         version_cursor = self.session.open_cursor(self.uri, None, "debug=(dump_version=true)")
         version_cursor.set_key(1)
         self.assertEquals(version_cursor.search(), 0)