Patch that refactors global read lock implementation and fixes

bug #57006 "Deadlock between HANDLER and FLUSH TABLES WITH READ
LOCK" and bug #54673 "It takes too long to get readlock for
'FLUSH TABLES WITH READ LOCK'".

The first bug manifested itself as a deadlock which occurred
when a connection, which had some table open through HANDLER
statement, tried to update some data through DML statement
while another connection tried to execute FLUSH TABLES WITH
READ LOCK concurrently.

What happened was that FTWRL in the second connection managed
to perform first step of GRL acquisition and thus blocked all
upcoming DML. After that it started to wait for table open
through HANDLER statement to be flushed. When the first connection
tried to execute DML it has started to wait for GRL/the second
connection creating deadlock.

The second bug manifested itself as starvation of FLUSH TABLES
WITH READ LOCK statements in cases when there was a constant
stream of concurrent DML statements (in two or more
connections).

This has happened because requests for protection against GRL
which were acquired by DML statements were ignoring presence of
pending GRL and thus the latter was starved.

This patch solves both these problems by re-implementing GRL
using metadata locks.

Similar to the old implementation acquisition of GRL in new
implementation is two-step. During the first step we block
all concurrent DML and DDL statements by acquiring global S
metadata lock (each DML and DDL statement acquires global IX
lock for its duration). During the second step we block commits
by acquiring global S lock in COMMIT namespace (commit code
acquires global IX lock in this namespace).

Note that unlike in old implementation acquisition of
protection against GRL in DML and DDL is semi-automatic.
We assume that any statement which should be blocked by GRL
will either open and acquires write-lock on tables or acquires
metadata locks on objects it is going to modify. For any such
statement global IX metadata lock is automatically acquired
for its duration.

The first problem is solved because waits for GRL become
visible to deadlock detector in metadata locking subsystem
and thus deadlocks like one in the first bug become impossible.

The second problem is solved because global S locks which
are used for GRL implementation are given preference over
IX locks which are acquired by concurrent DML (and we can
switch to fair scheduling in future if needed).

Important change:
FTWRL/GRL no longer blocks DML and DDL on temporary tables.
Before this patch behavior was not consistent in this respect:
in some cases DML/DDL statements on temporary tables were
blocked while in others they were not. Since the main use cases
for FTWRL are various forms of backups and temporary tables are
not preserved during backups we have opted for consistently
allowing DML/DDL on temporary tables during FTWRL/GRL.

Important change:
This patch changes thread state names which are used when
DML/DDL of FTWRL is waiting for global read lock. It is now
either "Waiting for global read lock" or "Waiting for commit
lock" depending on the stage on which FTWRL is.

Incompatible change:
To solve deadlock in events code which was exposed by this
patch we have to replace LOCK_event_metadata mutex with
metadata locks on events. As result we have to prohibit
DDL on events under LOCK TABLES.

This patch also adds extensive test coverage for interaction
of DML/DDL and FTWRL.

Performance of new and old global read lock implementations
in sysbench tests were compared. There were no significant
difference between new and old implementations.

1 files changed, 117 insertions, 68 deletions

diff --git a/sql/mdl.h b/sql/mdl.h
index 7938d833eac..ff30746e739 100644
--- a/sql/mdl.h
+++ b/sql/mdl.h
@@ -150,6 +150,15 @@ enum enum_mdl_type {
   MDL_TYPE_END};
 
 
+/** Duration of metadata lock. */
+
+enum enum_mdl_duration { MDL_STATEMENT= 0,
+                         MDL_TRANSACTION,
+                         MDL_EXPLICIT,
+                         /* This should be the last ! */
+                         MDL_DURATION_END };
+
+
 /** Maximal length of key for metadata locking subsystem. */
 #define MAX_MDLKEY_LENGTH (1 + NAME_LEN + 1 + NAME_LEN + 1)
 
@@ -167,13 +176,16 @@ class MDL_key
 {
 public:
   /**
-    Object namespaces
+    Object namespaces.
+    Sic: when adding a new member to this enum make sure to
+    update m_namespace_to_wait_state_name array in mdl.cc!
 
     Different types of objects exist in different namespaces
      - TABLE is for tables and views.
      - FUNCTION is for stored functions.
      - PROCEDURE is for stored procedures.
      - TRIGGER is for triggers.
+     - EVENT is for event scheduler events
     Note that although there isn't metadata locking on triggers,
     it's necessary to have a separate namespace for them since
     MDL_key is also used outside of the MDL subsystem.
@@ -184,6 +196,8 @@ public:
                             FUNCTION,
                             PROCEDURE,
                             TRIGGER,
+                            EVENT,
+                            COMMIT,
                             /* This should be the last ! */
                             NAMESPACE_END };
 
@@ -305,6 +319,8 @@ class MDL_request
 public:
   /** Type of metadata lock. */
   enum          enum_mdl_type type;
+  /** Duration for requested lock. */
+  enum enum_mdl_duration duration;
 
   /**
     Pointers for participating in the list of lock requests for this context.
@@ -327,17 +343,16 @@ public:
 
   void init(MDL_key::enum_mdl_namespace namespace_arg,
             const char *db_arg, const char *name_arg,
-            enum_mdl_type mdl_type_arg);
-  void init(const MDL_key *key_arg, enum_mdl_type mdl_type_arg);
+            enum_mdl_type mdl_type_arg,
+            enum_mdl_duration mdl_duration_arg);
+  void init(const MDL_key *key_arg, enum_mdl_type mdl_type_arg,
+            enum_mdl_duration mdl_duration_arg);
   /** Set type of lock request. Can be only applied to pending locks. */
   inline void set_type(enum_mdl_type type_arg)
   {
     DBUG_ASSERT(ticket == NULL);
     type= type_arg;
   }
-  static MDL_request *create(MDL_key::enum_mdl_namespace mdl_namespace,
-                             const char *db, const char *name,
-                             enum_mdl_type mdl_type, MEM_ROOT *root);
 
   /*
     This is to work around the ugliness of TABLE_LIST
@@ -363,6 +378,7 @@ public:
 
   MDL_request(const MDL_request *rhs)
     :type(rhs->type),
+    duration(rhs->duration),
     ticket(NULL),
     key(&rhs->key)
   {}
@@ -484,17 +500,35 @@ public:
 private:
   friend class MDL_context;
 
-  MDL_ticket(MDL_context *ctx_arg, enum_mdl_type type_arg)
+  MDL_ticket(MDL_context *ctx_arg, enum_mdl_type type_arg
+#ifndef DBUG_OFF
+             , enum_mdl_duration duration_arg
+#endif
+            )
    : m_type(type_arg),
+#ifndef DBUG_OFF
+     m_duration(duration_arg),
+#endif
      m_ctx(ctx_arg),
      m_lock(NULL)
   {}
 
-  static MDL_ticket *create(MDL_context *ctx_arg, enum_mdl_type type_arg);
+  static MDL_ticket *create(MDL_context *ctx_arg, enum_mdl_type type_arg
+#ifndef DBUG_OFF
+                            , enum_mdl_duration duration_arg
+#endif
+                            );
   static void destroy(MDL_ticket *ticket);
 private:
   /** Type of metadata lock. Externally accessible. */
   enum enum_mdl_type m_type;
+#ifndef DBUG_OFF
+  /**
+    Duration of lock represented by this ticket.
+    Context private. Debug-only.
+  */
+  enum_mdl_duration m_duration;
+#endif
   /**
     Context of the owner of the metadata lock ticket. Externally accessible.
   */
@@ -512,6 +546,39 @@ private:
 
 
 /**
+  Savepoint for MDL context.
+
+  Doesn't include metadata locks with explicit duration as
+  they are not released during rollback to savepoint.
+*/
+
+class MDL_savepoint
+{
+public:
+  MDL_savepoint() {};
+
+private:
+  MDL_savepoint(MDL_ticket *stmt_ticket, MDL_ticket *trans_ticket)
+    : m_stmt_ticket(stmt_ticket), m_trans_ticket(trans_ticket)
+  {}
+
+  friend class MDL_context;
+
+private:
+  /**
+    Pointer to last lock with statement duration which was taken
+    before creation of savepoint.
+  */
+  MDL_ticket *m_stmt_ticket;
+  /**
+    Pointer to last lock with transaction duration which was taken
+    before creation of savepoint.
+  */
+  MDL_ticket *m_trans_ticket;
+};
+
+
+/**
   A reliable way to wait on an MDL lock.
 */
 
@@ -559,9 +626,7 @@ public:
   typedef I_P_List<MDL_ticket,
                    I_P_List_adapter<MDL_ticket,
                                     &MDL_ticket::next_in_context,
-                                    &MDL_ticket::prev_in_context>,
-                   I_P_List_null_counter,
-                   I_P_List_fast_push_back<MDL_ticket> >
+                                    &MDL_ticket::prev_in_context> >
           Ticket_list;
 
   typedef Ticket_list::Iterator Ticket_iterator;
@@ -584,37 +649,28 @@ public:
                      const char *db, const char *name,
                      enum_mdl_type mdl_type);
 
-  bool has_lock(MDL_ticket *mdl_savepoint, MDL_ticket *mdl_ticket);
+  bool has_lock(const MDL_savepoint &mdl_savepoint, MDL_ticket *mdl_ticket);
 
   inline bool has_locks() const
   {
-    return !m_tickets.is_empty();
-  }
-
-  MDL_ticket *mdl_savepoint()
-  {
-    /*
-      NULL savepoint represents the start of the transaction.
-      Checking for m_trans_sentinel also makes sure we never
-      return a pointer to HANDLER ticket as a savepoint.
-    */
-    return m_tickets.front() == m_trans_sentinel ? NULL : m_tickets.front();
+    return !(m_tickets[MDL_STATEMENT].is_empty() &&
+             m_tickets[MDL_TRANSACTION].is_empty() &&
+             m_tickets[MDL_EXPLICIT].is_empty());
   }
 
-  void set_trans_sentinel()
+  MDL_savepoint mdl_savepoint()
   {
-    m_trans_sentinel= m_tickets.front();
+    return MDL_savepoint(m_tickets[MDL_STATEMENT].front(),
+                         m_tickets[MDL_TRANSACTION].front());
   }
-  MDL_ticket *trans_sentinel() const { return m_trans_sentinel; }
 
-  void reset_trans_sentinel(MDL_ticket *sentinel_arg)
-  {
-    m_trans_sentinel= sentinel_arg;
-  }
-  void move_ticket_after_trans_sentinel(MDL_ticket *mdl_ticket);
+  void set_explicit_duration_for_all_locks();
+  void set_transaction_duration_for_all_locks();
+  void set_lock_duration(MDL_ticket *mdl_ticket, enum_mdl_duration duration);
 
+  void release_statement_locks();
   void release_transactional_locks();
-  void rollback_to_savepoint(MDL_ticket *mdl_savepoint);
+  void rollback_to_savepoint(const MDL_savepoint &mdl_savepoint);
 
   inline THD *get_thd() const { return m_thd; }
 
@@ -655,46 +711,43 @@ public:
   MDL_wait m_wait;
 private:
   /**
-    All MDL tickets acquired by this connection.
-
-    The order of tickets in m_tickets list.
-    ---------------------------------------
-    The entire set of locks acquired by a connection
-    can be separated in two subsets: transactional and
-    non-transactional locks.
-
-    Transactional locks are locks with automatic scope. They
-    are accumulated in the course of a transaction, and
-    released only on COMMIT, ROLLBACK or ROLLBACK TO SAVEPOINT.
-    They must not be (and never are) released manually,
+    Lists of all MDL tickets acquired by this connection.
+
+    Lists of MDL tickets:
+    ---------------------
+    The entire set of locks acquired by a connection can be separated
+    in three subsets according to their: locks released at the end of
+    statement, at the end of transaction and locks are released
+    explicitly.
+
+    Statement and transactional locks are locks with automatic scope.
+    They are accumulated in the course of a transaction, and released
+    either at the end of uppermost statement (for statement locks) or
+    on COMMIT, ROLLBACK or ROLLBACK TO SAVEPOINT (for transactional
+    locks). They must not be (and never are) released manually,
     i.e. with release_lock() call.
 
-    Non-transactional locks are taken for locks that span
+    Locks with explicit duration are taken for locks that span
     multiple transactions or savepoints.
     These are: HANDLER SQL locks (HANDLER SQL is
     transaction-agnostic), LOCK TABLES locks (you can COMMIT/etc
     under LOCK TABLES, and the locked tables stay locked), and
-    SET GLOBAL READ_ONLY=1 global shared lock.
+    locks implementing "global read lock".
 
-    Transactional locks are always prepended to the beginning
-    of the list. In other words, they are stored in reverse
-    temporal order. Thus, when we rollback to a savepoint,
-    we start popping and releasing tickets from the front
-    until we reach the last ticket acquired after the
-    savepoint.
+    Statement/transactional locks are always prepended to the
+    beginning of the appropriate list. In other words, they are
+    stored in reverse temporal order. Thus, when we rollback to
+    a savepoint, we start popping and releasing tickets from the
+    front until we reach the last ticket acquired after the savepoint.
 
-    Non-transactional locks are always stored after
-    transactional ones, and among each other can be
-    split into three sets:
+    Locks with explicit duration stored are not stored in any
+    particular order, and among each other can be split into
+    three sets:
 
     [LOCK TABLES locks] [HANDLER locks] [GLOBAL READ LOCK locks]
 
     The following is known about these sets:
 
-    * we can never have both HANDLER and LOCK TABLES locks
-      together -- HANDLER statements are prohibited under LOCK
-      TABLES, entering LOCK TABLES implicitly closes all open
-      HANDLERs.
     * GLOBAL READ LOCK locks are always stored after LOCK TABLES
       locks and after HANDLER locks. This is because one can't say
       SET GLOBAL read_only=1 or FLUSH TABLES WITH READ LOCK
@@ -709,14 +762,9 @@ private:
       However, one can open a few HANDLERs after entering the
       read only mode.
     * LOCK TABLES locks include intention exclusive locks on
-      involved schemas.
-  */
-  Ticket_list m_tickets;
-  /**
-    Separates transactional and non-transactional locks
-    in m_tickets list, @sa m_tickets.
+      involved schemas and global intention exclusive lock.
   */
-  MDL_ticket *m_trans_sentinel;
+  Ticket_list m_tickets[MDL_DURATION_END];
   THD *m_thd;
   /**
     TRUE -  if for this context we will break protocol and try to
@@ -747,8 +795,9 @@ private:
   MDL_wait_for_subgraph *m_waiting_for;
 private:
   MDL_ticket *find_ticket(MDL_request *mdl_req,
-                          bool *is_transactional);
-  void release_locks_stored_before(MDL_ticket *sentinel);
+                          enum_mdl_duration *duration);
+  void release_locks_stored_before(enum_mdl_duration duration, MDL_ticket *sentinel);
+  void release_lock(enum_mdl_duration duration, MDL_ticket *ticket);
   bool try_acquire_lock_impl(MDL_request *mdl_request,
                              MDL_ticket **out_ticket);