BUG#36197: flush tables (or little table cache) can cause crash on slave

When flushing tables, there were a slight chance that the flush was occuring
between processing of two table map events. Since the tables are opened
one by one, it might result in that the tables were not valid and that sub-
sequent locking of tables would cause the slave to crash.

The problem is solved by opening and locking all tables at once using
simple_open_n_lock_tables(). Also, the patch contain a change to open_tables()
so that pre-locking only takes place when the trg_event_map is not zero, which
was not the case before (this caused the lock to be placed in thd->locked_tables
instead of thd->lock since the assumption was that triggers would be called
later and therefore the tables should be pre-locked).


mysql-test/suite/rpl/r/rpl_found_rows.result:
  Result change
mysql-test/suite/rpl/r/rpl_row_inexist_tbl.result:
  Result change
mysql-test/suite/rpl/t/rpl_found_rows.test:
  Adding drop of table that was created in test.
mysql-test/suite/rpl/t/rpl_slave_status.test:
  Adding waits for slave start and stop to ensure that test works.
sql/log_event.cc:
  Replacing table-by-table open and lock with a single call
  to simple_open_n_lock_tables(), which in turn required some
  changes to other code.
sql/log_event_old.cc:
  Replacing table-by-table open and lock with a single call
  to simple_open_n_lock_tables(), which in turn required some
  changes to other code.
sql/sql_base.cc:
  Extending check inside open_tables() so that pre-locking in only done if
  tables->trg_egent_map is non-zero.
mysql-test/include/wait_for_slave_sql_to_start.inc:
  New BitKeeper file ``mysql-test/include/wait_for_slave_sql_to_start.inc''

1 files changed, 32 insertions, 67 deletions

diff --git a/sql/log_event_old.cc b/sql/log_event_old.cc
index 808356a05c7..c6b99b1bd69 100644
--- a/sql/log_event_old.cc
+++ b/sql/log_event_old.cc
@@ -53,81 +53,46 @@ Old_rows_log_event::do_apply_event(Old_rows_log_event *ev, const Relay_log_info
   */
   if (!thd->lock)
   {
-    bool need_reopen= 1; /* To execute the first lap of the loop below */
-
     /*
-      lock_tables() reads the contents of thd->lex, so they must be
-      initialized. Contrary to in
-      Table_map_log_event::do_apply_event() we don't call
-      mysql_init_query() as that may reset the binlog format.
+      Lock_tables() reads the contents of thd->lex, so they must be
+      initialized.
+
+      We also call the mysql_reset_thd_for_next_command(), since this
+      is the logical start of the next "statement". Note that this
+      call might reset the value of current_stmt_binlog_row_based, so
+      we need to do any changes to that value after this function.
     */
     lex_start(thd);
+    mysql_reset_thd_for_next_command(thd);
 
-    while ((error= lock_tables(thd, rli->tables_to_lock,
-                               rli->tables_to_lock_count, &need_reopen)))
+    /*
+      Check if the slave is set to use SBR.  If so, it should switch
+      to using RBR until the end of the "statement", i.e., next
+      STMT_END_F or next error.
+    */
+    if (!thd->current_stmt_binlog_row_based &&
+        mysql_bin_log.is_open() && (thd->options & OPTION_BIN_LOG))
     {
-      if (!need_reopen)
-      {
-        if (thd->is_slave_error || thd->is_fatal_error)
-        {
-          /*
-            Error reporting borrowed from Query_log_event with many excessive
-            simplifications (we don't honour --slave-skip-errors)
-          */
-          uint actual_error= thd->main_da.sql_errno();
-          rli->report(ERROR_LEVEL, actual_error,
-                      "Error '%s' in %s event: when locking tables",
-                      (actual_error ? thd->main_da.message() :
-                       "unexpected success or fatal error"),
-                      ev->get_type_str());
-          thd->is_fatal_error= 1;
-        }
-        else
-        {
-          rli->report(ERROR_LEVEL, error,
-                      "Error in %s event: when locking tables",
-                      ev->get_type_str());
-        }
-        const_cast<Relay_log_info*>(rli)->clear_tables_to_lock();
-        DBUG_RETURN(error);
-      }
-
-      /*
-        So we need to reopen the tables.
-
-        We need to flush the pending RBR event, since it keeps a
-        pointer to an open table.
-
-        ALTERNATIVE SOLUTION (not implemented): Extract a pointer to
-        the pending RBR event and reset the table pointer after the
-        tables has been reopened.
-
-        NOTE: For this new scheme there should be no pending event:
-        need to add code to assert that is the case.
-       */
-      thd->binlog_flush_pending_rows_event(false);
-      TABLE_LIST *tables= rli->tables_to_lock;
-      close_tables_for_reopen(thd, &tables);
+      thd->set_current_stmt_binlog_row_based();
+    }
 
-      uint tables_count= rli->tables_to_lock_count;
-      if ((error= open_tables(thd, &tables, &tables_count, 0)))
+    if (simple_open_n_lock_tables(thd, rli->tables_to_lock))
+    {
+      uint actual_error= thd->main_da.sql_errno();
+      if (thd->is_slave_error || thd->is_fatal_error)
       {
-        if (thd->is_slave_error || thd->is_fatal_error)
-        {
-          /*
-            Error reporting borrowed from Query_log_event with many excessive
-            simplifications (we don't honour --slave-skip-errors)
-          */
-          uint actual_error= thd->main_da.sql_errno();
-          rli->report(ERROR_LEVEL, actual_error,
-                      "Error '%s' on reopening tables",
-                      (actual_error ? thd->main_da.message() :
-                       "unexpected success or fatal error"));
-          thd->is_slave_error= 1;
-        }
-        const_cast<Relay_log_info*>(rli)->clear_tables_to_lock();
-        DBUG_RETURN(error);
+        /*
+          Error reporting borrowed from Query_log_event with many excessive
+          simplifications (we don't honour --slave-skip-errors)
+        */
+        rli->report(ERROR_LEVEL, actual_error,
+                    "Error '%s' on opening tables",
+                    (actual_error ? thd->main_da.message() :
+                     "unexpected success or fatal error"));
+        thd->is_slave_error= 1;
       }
+      const_cast<Relay_log_info*>(rli)->clear_tables_to_lock();
+      DBUG_RETURN(actual_error);
     }
 
     /*