Fix for bug #12828477 - "MDL SUBSYSTEM CREATES BIG OVERHEAD

FOR CERTAIN QUERIES TO INFORMATION_SCHEMA".

The problem was that metadata locking subsystem introduced
too much overhead for queries to I_S which were processed by
opening only .FRM or .TRG files and had to scanned a lot of
tables (e.g. SELECT COUNT(*) FROM I_S.TRIGGERS was affected). 
The same effect was not observed for similar queries which 
performed full-blown table open in order to fill I_S table.

The problem stemmed from the fact that in case when I_S 
implementation opened only .FRM or .TRG file for each table 
processed it didn't release metadata lock it has acquired on 
the table after finishing its processing. As result, list
of acquired metadata locks were growing until the end of 
statement. Since acquisition of each new lock required 
search in the list of already acquired locks performance
degraded.

The same effect is not observed when I_S implementation
performs full-blown table open for each table being
processed, as in the latter cases metadata lock on the
table is released right after table processing.

This fix addressed the problem by ensuring that I_S 
implementation releases metadata lock after processing
the table in both cases of full-blown table open and in 
case when only .FRM or .TRG file is read.

mysql-test/r/information_schema.result:
  Added coverage for bug #12828477 - "MDL SUBSYSTEM CREATES BIG
  OVERHEAD FOR CERTAIN QUERIES TO INFORMATION_SCHEMA".
mysql-test/t/information_schema.test:
  Added coverage for bug #12828477 - "MDL SUBSYSTEM CREATES BIG
  OVERHEAD FOR CERTAIN QUERIES TO INFORMATION_SCHEMA".
sql/sql_show.cc:
  Changed fill_schema_table_from_frm() to release metadata lock
  it has acquired after processing the .FRM or .TRG file for
  table. 
  Without this step metadata locks acquired for each table 
  processed will be accumulated. In situation when a lot of 
  tables are processed by I_S query this will result in
  transaction with too many metadata locks. As result
  performance of acquisition of new lock will degrade.

1 files changed, 28 insertions, 5 deletions

diff --git a/sql/sql_show.cc b/sql/sql_show.cc
index fb5ed62ecdf..887115b38ad 100644
--- a/sql/sql_show.cc
+++ b/sql/sql_show.cc
@@ -3157,6 +3157,10 @@ end:
   */
   thd->temporary_tables= NULL;
   close_thread_tables(thd);
+  /*
+    Release metadata lock we might have acquired.
+    See comment in fill_schema_table_from_frm() for details.
+  */
   thd->mdl_context.rollback_to_savepoint(open_tables_state_backup->mdl_system_tables_svp);
 
   thd->lex= old_lex;
@@ -3339,6 +3343,9 @@ try_acquire_high_prio_shared_mdl_lock(THD *thd, TABLE_LIST *table,
   @param[in]      db_name                  database name
   @param[in]      table_name               table name
   @param[in]      schema_table_idx         I_S table index
+  @param[in]      open_tables_state_backup Open_tables_state object which is used
+                                           to save/restore original state of metadata
+                                           locks.
   @param[in]      can_deadlock             Indicates that deadlocks are possible
                                            due to metadata locks, so to avoid
                                            them we should not wait in case if
@@ -3356,6 +3363,7 @@ static int fill_schema_table_from_frm(THD *thd, TABLE_LIST *tables,
                                       LEX_STRING *db_name,
                                       LEX_STRING *table_name,
                                       enum enum_schema_tables schema_table_idx,
+                                      Open_tables_backup *open_tables_state_backup,
                                       bool can_deadlock)
 {
   TABLE *table= tables->table;
@@ -3501,13 +3509,27 @@ end_share:
 
 end_unlock:
   mysql_mutex_unlock(&LOCK_open);
-  /*
-    Don't release the MDL lock, it can be part of a transaction.
-    If it is not, it will be released by the call to
-    MDL_context::rollback_to_savepoint() in the caller.
-  */
 
 end:
+  /*
+    Release metadata lock we might have acquired.
+
+    Without this step metadata locks acquired for each table processed
+    will be accumulated. In situation when a lot of tables are processed
+    by I_S query this will result in transaction with too many metadata
+    locks. As result performance of acquisition of new lock will suffer.
+
+    Of course, the fact that we don't hold metadata lock on tables which
+    were processed till the end of I_S query makes execution less isolated
+    from concurrent DDL. Consequently one might get 'dirty' results from
+    such a query. But we have never promised serializability of I_S queries
+    anyway.
+
+    We don't have any tables open since we took backup, so rolling back to
+    savepoint is safe.
+  */
+  DBUG_ASSERT(thd->open_tables == NULL);
+  thd->mdl_context.rollback_to_savepoint(open_tables_state_backup->mdl_system_tables_svp);
   thd->clear_error();
   return res;
 }
@@ -3758,6 +3780,7 @@ int get_all_tables(THD *thd, TABLE_LIST *tables, COND *cond)
               int res= fill_schema_table_from_frm(thd, tables, schema_table,
                                                   db_name, table_name,
                                                   schema_table_idx,
+                                                  &open_tables_state_backup,
                                                   can_deadlock);
 
               thd->pop_internal_handler();