Added option "AND DISABLE CHECKPOINT" to "FLUSH TABLES WITH READ LOCK"

This makes it possible to do safe multi volume snapshots as long as one snapshots the volume with the transaction logs last.


include/mysql_com.h:
  Added REFRESH_CHECKPOINT
mysql-test/r/flush.result:
  Added test of new FLUSH TABLES syntax + calls to checkpoint_status handler calls
mysql-test/t/flush.test:
  Added test of new FLUSH TABLES syntax + calls to checkpoint_status handler calls
sql/handler.cc:
  Added code to call checkpoint_state for all handlertons that supports it
sql/handler.h:
  Added new checkpoint_state() handlerton call to temporarly disable checkpoints.
sql/lex.h:
  Added CHECKPOINT keyword
sql/sql_yacc.yy:
  Added support for FLUSH TABLES WITH READ LOCK AND DISABLE CHECKPOINT
storage/maria/ha_maria.cc:
  Added handlerton call to disable checkpoints.
storage/maria/ma_checkpoint.c:
  Don't do checkpoint if checkpoints are disabled.
storage/maria/ma_static.c:
  Added maria_checkpoint_disabled
storage/maria/maria_def.h:
  Added maria_checkpoint_disabled
storage/xtradb/handler/ha_innodb.cc:
  Added handlerton call to disable checkpoints.
storage/xtradb/include/log0log.h:
  Added option to log_checkpoint() to allow one to ignore not critical checkpoints during the time checkpoints are disabled.
storage/xtradb/log/log0log.c:
  Added code to allow one to disable checkpoints during a FLUSH TABLES ... DISABLE CHECKPOINT
  This was done by adding a new argument to log_checkpoint() which tells us when the checkpoint is called by srv_master_thread (which are safe to ignore)
storage/xtradb/srv/srv0srv.c:
  Tell log_checkpoint() that checkpoints from srv_master_thread() are safe to ignore (will just delay recovery time a bit).

include/mysql_com.h

@@ -136,6 +136,7 @@ enum enum_server_command
 #define REFRESH_QUERY_CACHE_FREE 0x20000L /* pack query cache */
 #define REFRESH_DES_KEY_FILE	0x40000L
 #define REFRESH_USER_RESOURCES	0x80000L
+#define REFRESH_CHECKPOINT	0x100000L /* Don't do checkpoints */
 
 #define CLIENT_LONG_PASSWORD	1	/* new more secure passwords */
 #define CLIENT_FOUND_ROWS	2	/* Found instead of affected rows */

mysql-test/r/flush.result

@@ -6,7 +6,7 @@ insert into t2 values(3);
 select * from t1;
 n
 3
-flush tables with read lock;
+flush tables with read lock and disable checkpoint;
 drop table t2;
 ERROR HY000: Can't execute the query because you have a conflicting read lock
 drop table t2;

mysql-test/t/flush.test

@@ -28,7 +28,7 @@ enable_query_log;
 connection con1;
 select * from t1;
 connection con2;
-flush tables with read lock;
+flush tables with read lock and disable checkpoint;
 --error 1223
 drop table t2;
 connection con1;

sql/handler.cc

@@ -598,6 +598,23 @@ void ha_drop_database(char* path)
 }
 
 
+static my_bool checkpoint_state_handlerton(THD *unused1, plugin_ref plugin,
+                                           void *disable)
+{
+  handlerton *hton= plugin_data(plugin, handlerton *);
+  if (hton->state == SHOW_OPTION_YES && hton->checkpoint_state)
+    hton->checkpoint_state(hton, (int) *(bool*) disable);
+  return FALSE;
+}
+
+
+void ha_checkpoint_state(bool disable)
+{
+  plugin_foreach(NULL, checkpoint_state_handlerton, MYSQL_STORAGE_ENGINE_PLUGIN, &disable);
+}
+
+
+
 static my_bool closecon_handlerton(THD *thd, plugin_ref plugin,
                                    void *unused)
 {

sql/handler.h

@@ -865,6 +865,19 @@ struct handlerton
    int  (*recover)(handlerton *hton, XID *xid_list, uint len);
    int  (*commit_by_xid)(handlerton *hton, XID *xid);
    int  (*rollback_by_xid)(handlerton *hton, XID *xid);
+  /*
+    "Disable or enable checkpointing internal to the storage engine. This is
+    used for FLUSH TABLES WITH READ LOCK AND DISABLE CHECKPOINT to ensure that
+    the engine will never start any recovery from a time between
+    FLUSH TABLES ... ; UNLOCK TABLES.
+
+    While checkpointing is disabled, the engine should pause any background
+    write activity (such as tablespace checkpointing) that require consistency
+    between different files (such as transaction log and tablespace files) for
+    crash recovery to succeed. The idea is to use this to make safe
+    multi-volume LVM snapshot backups.
+  */
+   int  (*checkpoint_state)(handlerton *hton, bool disabled);
    void *(*create_cursor_read_view)(handlerton *hton, THD *thd);
    void (*set_cursor_read_view)(handlerton *hton, THD *thd, void *read_view);
    void (*close_cursor_read_view)(handlerton *hton, THD *thd, void *read_view);
@@ -2629,6 +2642,7 @@ int ha_panic(enum ha_panic_function flag);
 void ha_close_connection(THD* thd);
 bool ha_flush_logs(handlerton *db_type);
 void ha_drop_database(char* path);
+void ha_checkpoint_state(bool disable);
 int ha_create_table(THD *thd, const char *path,
                     const char *db, const char *table_name,
                     HA_CREATE_INFO *create_info,

sql/lex.h

@@ -104,6 +104,7 @@ static SYMBOL symbols[] = {
   { "CHARACTER",	SYM(CHAR_SYM)},
   { "CHARSET",		SYM(CHARSET)},
   { "CHECK",		SYM(CHECK_SYM)},
+  { "CHECKPOINT",	SYM(CHECKPOINT_SYM)},
   { "CHECKSUM",		SYM(CHECKSUM_SYM)},
   { "CIPHER",		SYM(CIPHER_SYM)},
   { "CLIENT",		SYM(CLIENT_SYM)},

sql/lock.cc

@@ -1449,7 +1449,7 @@ static void print_lock_error(int error, const char *table)
 
 ****************************************************************************/
 
-volatile uint global_read_lock=0;
+volatile uint global_read_lock=0, global_disable_checkpoint= 0;
 volatile uint global_read_lock_blocks_commit=0;
 static volatile uint protect_against_global_read_lock=0;
 static volatile uint waiting_for_read_lock=0;
@@ -1508,6 +1508,14 @@ void unlock_global_read_lock(THD *thd)
   tmp= --global_read_lock;
   if (thd->global_read_lock == MADE_GLOBAL_READ_LOCK_BLOCK_COMMIT)
     --global_read_lock_blocks_commit;
+  if (thd->global_disable_checkpoint)
+  {
+    thd->global_disable_checkpoint= 0;
+    if (!--global_disable_checkpoint)
+    {
+      ha_checkpoint_state(0);                   // Enable checkpoints
+    }
+  }
   pthread_mutex_unlock(&LOCK_global_read_lock);
   /* Send the signal outside the mutex to avoid a context switch */
   if (!tmp)
@@ -1629,6 +1637,24 @@ bool make_global_read_lock_block_commit(THD *thd)
 }
 
 
+/**
+   Disable checkpoints for all handlers
+   This is released in unlock_global_read_lock()
+*/
+
+void disable_checkpoints(THD *thd)
+{
+  pthread_mutex_lock(&LOCK_global_read_lock);
+  if (!thd->global_disable_checkpoint)
+  {
+    thd->global_disable_checkpoint= 1;
+    if (!global_disable_checkpoint++)
+      ha_checkpoint_state(1);                   // Disable checkpoints
+  }
+  pthread_mutex_unlock(&LOCK_global_read_lock);
+}
+
+
 /**
   Broadcast COND_refresh and COND_global_read_lock.
 

sql/mysql_priv.h

@@ -2094,7 +2094,7 @@ extern bool opt_ignore_builtin_innodb;
 extern my_bool opt_character_set_client_handshake;
 extern bool volatile abort_loop, shutdown_in_progress;
 extern bool in_bootstrap;
-extern uint volatile thread_count, thread_running, global_read_lock;
+extern uint volatile thread_count, thread_running, global_read_lock, global_disable_checkpoint;
 extern ulong thread_created;
 extern uint thread_handling;
 extern uint connection_count, extra_connection_count;
@@ -2257,6 +2257,7 @@ bool wait_if_global_read_lock(THD *thd, bool abort_on_refresh,
                               bool is_not_commit);
 void start_waiting_global_read_lock(THD *thd);
 bool make_global_read_lock_block_commit(THD *thd);
+void disable_checkpoints(THD *thd);
 bool set_protect_against_global_read_lock(void);
 void unset_protect_against_global_read_lock(void);
 void broadcast_refresh(void);

sql/sql_class.cc

@@ -674,6 +674,7 @@ THD::THD()
    stmt_depends_on_first_successful_insert_id_in_prev_stmt(FALSE),
    examined_row_count(0),
    global_read_lock(0),
+   global_disable_checkpoint(0),
    is_fatal_error(0),
    transaction_rollback_request(0),
    is_fatal_sub_stmt_error(0),

sql/sql_class.h

@@ -1920,7 +1920,7 @@ public:
   ulong      query_plan_fsort_passes; 
   pthread_t  real_id;                           /* For debugging */
   my_thread_id  thread_id;
-  uint	     tmp_table, global_read_lock;
+  uint	     tmp_table, global_read_lock, global_disable_checkpoint;
   uint	     server_status,open_options;
   enum enum_thread_type system_thread;
   uint       select_number;             //number of select (used for EXPLAIN)

sql/sql_parse.cc

@@ -7058,6 +7058,8 @@ bool reload_acl_and_cache(THD *thd, ulong options, TABLE_LIST *tables,
         unlock_global_read_lock(thd);
         return 1;
       }
+      if (options & REFRESH_CHECKPOINT)
+        disable_checkpoints(thd);
     }
     else
     {

sql/sql_yacc.yy

@@ -758,6 +758,7 @@ bool my_yyoverflow(short **a, YYSTYPE **b, ulong *yystacksize);
 %token  CHANGED
 %token  CHARSET
 %token  CHAR_SYM                      /* SQL-2003-R */
+%token  CHECKPOINT_SYM
 %token  CHECKSUM_SYM
 %token  CHECK_SYM                     /* SQL-2003-R */
 %token  CIPHER_SYM
@@ -1336,6 +1337,7 @@ bool my_yyoverflow(short **a, YYSTYPE **b, ulong *yystacksize);
         opt_natural_language_mode opt_query_expansion
         opt_ev_status opt_ev_on_completion ev_on_completion opt_ev_comment
         ev_alter_on_schedule_completion opt_ev_rename_to opt_ev_sql_stmt
+        optional_flush_tables_arguments
 
 %type <ulong_num>
         ulong_num real_ulong_num merge_insert_types
@@ -10784,8 +10786,8 @@ flush_option:
           table_or_tables
           { Lex->type|= REFRESH_TABLES; }
           opt_table_list {}
-        | TABLES WITH READ_SYM LOCK_SYM
-          { Lex->type|= REFRESH_TABLES | REFRESH_READ_LOCK; }
+        | TABLES WITH READ_SYM LOCK_SYM optional_flush_tables_arguments
+          { Lex->type|= REFRESH_TABLES | REFRESH_READ_LOCK | $5; }
         | QUERY_SYM CACHE_SYM
           { Lex->type|= REFRESH_QUERY_CACHE_FREE; }
         | HOSTS_SYM
@@ -10821,6 +10823,10 @@ opt_table_list:
         | table_list {}
         ;
 
+optional_flush_tables_arguments:
+          /* empty */        {$$= 0;}
+        | AND_SYM DISABLE_SYM CHECKPOINT_SYM {$$= REFRESH_CHECKPOINT; } 
+
 reset:
           RESET_SYM
           {
@@ -11859,6 +11865,7 @@ keyword:
         | CACHE_SYM             {}
         | CHARSET               {}
         | CHECKSUM_SYM          {}
+        | CHECKPOINT_SYM        {}
         | CLOSE_SYM             {}
         | COMMENT_SYM           {}
         | COMMIT_SYM            {}

storage/maria/ha_maria.cc

@@ -3167,6 +3167,14 @@ bool maria_flush_logs(handlerton *hton)
 }
 
 
+int maria_checkpoint_state(handlerton *hton, bool disabled)
+{
+  maria_checkpoint_disabled= (my_bool) disabled;
+  return 0;
+}
+
+
+
 #define SHOW_MSG_LEN (FN_REFLEN + 20)
 /**
   @brief show status handler
@@ -3355,6 +3363,7 @@ static int ha_maria_init(void *p)
   maria_hton->panic= maria_hton_panic;
   maria_hton->commit= maria_commit;
   maria_hton->rollback= maria_rollback;
+  maria_hton->checkpoint_state= maria_checkpoint_state;
 #ifdef MARIA_CANNOT_ROLLBACK
   maria_hton->commit= 0;
 #endif

storage/maria/ma_checkpoint.c

@@ -576,6 +576,12 @@ pthread_handler_t ma_checkpoint_background(void *arg)
     switch (sleeps % interval)
     {
     case 0:
+      /* If checkpoints are disabled, wait 1 second and try again */
+      if (maria_checkpoint_disabled)
+      {
+        sleep_time= 1;
+        break;
+      }
       /*
         With background flushing evenly distributed over the time
         between two checkpoints, we should have only little flushing to do
@@ -590,6 +596,7 @@ pthread_handler_t ma_checkpoint_background(void *arg)
         want to checkpoint every minute, hence the positive
         checkpoint_min_activity.
       */
+
       if (((translog_get_horizon() - log_horizon_at_last_checkpoint) +
            (maria_pagecache->global_cache_write -
             pagecache_flushes_at_last_checkpoint) *

storage/maria/ma_static.c

@@ -39,6 +39,7 @@ my_bool maria_inited= FALSE;
 my_bool maria_in_ha_maria= FALSE; /* If used from ha_maria or not */
 my_bool maria_recovery_changed_data= 0, maria_recovery_verbose= 0;
 my_bool maria_assert_if_crashed_table= 0;
+my_bool maria_checkpoint_disabled= 0;
 
 pthread_mutex_t THR_LOCK_maria;
 #if defined(THREAD) && !defined(DONT_USE_RW_LOCKS)

storage/maria/maria_def.h

@@ -820,7 +820,7 @@ extern uint maria_quick_table_bits;
 extern char *maria_data_root;
 extern uchar maria_zero_string[];
 extern my_bool maria_inited, maria_in_ha_maria, maria_recovery_changed_data;
-extern my_bool maria_recovery_verbose;
+extern my_bool maria_recovery_verbose, maria_checkpoint_disabled;
 extern my_bool maria_assert_if_crashed_table;
 extern HASH maria_stored_state;
 extern int (*maria_create_trn_hook)(MARIA_HA *);

storage/xtradb/handler/ha_innodb.cc

@@ -496,6 +496,17 @@ bool innobase_show_status(handlerton *hton, THD* thd,
                           stat_print_fn* stat_print,
                           enum ha_stat_type stat_type);
 
+/* Enable / disable checkpoints */
+static int innobase_checkpoint_state(handlerton *hton, bool disable)
+{
+  if (disable)
+    (void) log_disable_checkpoint();
+  else
+     log_enable_checkpoint();
+  return 0;
+}
+
+
 /*****************************************************************//**
 Commits a transaction in an InnoDB database. */
 static
@@ -2065,6 +2076,7 @@ innobase_init(
         innobase_hton->recover=innobase_xa_recover;
         innobase_hton->commit_by_xid=innobase_commit_by_xid;
         innobase_hton->rollback_by_xid=innobase_rollback_by_xid;
+        innobase_hton->checkpoint_state= innobase_checkpoint_state;
         innobase_hton->create_cursor_read_view=innobase_create_cursor_view;
         innobase_hton->set_cursor_read_view=innobase_set_cursor_view;
         innobase_hton->close_cursor_read_view=innobase_close_cursor_view;

storage/xtradb/include/log0log.h

@@ -249,12 +249,15 @@ log_checkpoint(
 /*===========*/
 	ibool	sync,		/*!< in: TRUE if synchronous operation is
 				desired */
-	ibool	write_always);	/*!< in: the function normally checks if the
+	ibool	write_always,	/*!< in: the function normally checks if the
 				the new checkpoint would have a greater
 				lsn than the previous one: if not, then no
 				physical write is done; by setting this
 				parameter TRUE, a physical write will always be
 				made to log files */
+        ibool   safe_to_ignore);/*!< in: TRUE if checkpoint can be ignored in
+                                  the case checkpoint's are disabled */
+
 /****************************************************************//**
 Makes a checkpoint at a given lsn or later. */
 UNIV_INTERN
@@ -271,6 +274,18 @@ log_make_checkpoint_at(
 					by setting this parameter TRUE, a
 					physical write will always be made to
 					log files */
+/****************************************************************//**
+Disable checkpoints. This is used when doing a volume snapshot
+to ensure that we don't get checkpoint between snapshoting two
+different volumes */
+UNIV_INTERN
+ibool log_disable_checkpoint();
+
+/****************************************************************//**
+Enable checkpoints that was disabled with log_disable_checkpoint() */
+UNIV_INTERN
+void log_enable_checkpoint();
+
 /****************************************************************//**
 Makes a checkpoint at the latest lsn and writes it to first page of each
 data file in the database, so that we know that the file spaces contain

storage/xtradb/log/log0log.c

@@ -97,6 +97,8 @@ archive */
 UNIV_INTERN byte	log_archive_io;
 #endif /* UNIV_LOG_ARCHIVE */
 
+UNIV_INTERN ulint       log_disable_checkpoint_active= 0;
+
 /* A margin for free space in the log buffer before a log entry is catenated */
 #define LOG_BUF_WRITE_MARGIN	(4 * OS_FILE_LOG_BLOCK_SIZE)
 
@@ -174,7 +176,7 @@ log_fsp_current_free_limit_set_and_checkpoint(
 	success = FALSE;
 
 	while (!success) {
-		success = log_checkpoint(TRUE, TRUE);
+          success = log_checkpoint(TRUE, TRUE, FALSE);
 	}
 }
 
@@ -1988,12 +1990,14 @@ log_checkpoint(
 /*===========*/
 	ibool	sync,		/*!< in: TRUE if synchronous operation is
 				desired */
-	ibool	write_always)	/*!< in: the function normally checks if the
+	ibool	write_always,	/*!< in: the function normally checks if the
 				the new checkpoint would have a greater
 				lsn than the previous one: if not, then no
 				physical write is done; by setting this
 				parameter TRUE, a physical write will always be
 				made to log files */
+        ibool   safe_to_ignore) /*!< in: TRUE if checkpoint can be ignored in
+                                  the case checkpoint's are disabled */
 {
 	ib_uint64_t	oldest_lsn;
 
@@ -2024,14 +2028,27 @@ log_checkpoint(
 
 	mutex_enter(&(log_sys->mutex));
 
+        /* Return if this is not a forced checkpoint and either there is no
+           need for a checkpoint or if checkpoints are disabled */
 	if (!write_always
-	    && log_sys->last_checkpoint_lsn >= oldest_lsn) {
+	    && (log_sys->last_checkpoint_lsn >= oldest_lsn ||
+                (safe_to_ignore && log_disable_checkpoint_active)))
+        {
 
 		mutex_exit(&(log_sys->mutex));
 
 		return(TRUE);
 	}
 
+        if (log_disable_checkpoint_active)
+        {
+          	/* Wait until we are allowed to do a checkpoint */
+		mutex_exit(&(log_sys->mutex));
+		rw_lock_s_lock(&(log_sys->checkpoint_lock));
+		rw_lock_s_unlock(&(log_sys->checkpoint_lock));
+                mutex_enter(&(log_sys->mutex));
+        }
+
 	ut_ad(log_sys->flushed_to_disk_lsn >= oldest_lsn);
 
 	if (log_sys->n_pending_checkpoint_writes > 0) {
@@ -2092,7 +2109,73 @@ log_make_checkpoint_at(
 
 	while (!log_preflush_pool_modified_pages(lsn, TRUE));
 
-	while (!log_checkpoint(TRUE, write_always));
+	while (!log_checkpoint(TRUE, write_always, FALSE));
+}
+
+/****************************************************************//**
+Disable checkpoints. This is used when doing a volumne snapshot
+to ensure that we don't get checkpoint between snapshoting two
+different volumes */
+
+UNIV_INTERN
+ibool log_disable_checkpoint()
+{
+  mutex_enter(&(log_sys->mutex));
+
+  /*
+    Wait if a checkpoint write is running.
+    This is the same code that is used in log_checkpoint() to ensure
+    that two checkpoints are not happening at the same time.
+  */
+  while (log_sys->n_pending_checkpoint_writes > 0)
+  {
+    mutex_exit(&(log_sys->mutex));
+    rw_lock_s_lock(&(log_sys->checkpoint_lock));
+    rw_lock_s_unlock(&(log_sys->checkpoint_lock));
+    mutex_enter(&(log_sys->mutex));
+  }
+  /*
+    The following should never be true; It's is here just in case of
+    wrong usage of this function. (Better safe than sorry).
+  */
+
+  if (log_disable_checkpoint_active)
+  {
+    mutex_exit(&(log_sys->mutex));
+    return 1;                                   /* Already disabled */
+  }
+  /*
+    Take the checkpoint lock to ensure we will not get any checkpoints
+    running
+  */
+  rw_lock_x_lock_gen(&(log_sys->checkpoint_lock), LOG_CHECKPOINT);
+  log_disable_checkpoint_active= 1;
+  mutex_exit(&(log_sys->mutex));
+  return 0;
+}
+
+
+/****************************************************************//**
+Enable checkpoints that was disabled with log_disable_checkpoint()
+This lock is called by MariaDB and only when we have done call earlier
+to log_disable_checkpoint().
+
+Note: We can't take a log->mutex lock here running log_checkpoint()
+which is waiting (log_sys->checkpoint_lock may already have it.
+This is however safe to do without a mutex as log_disable_checkpoint
+is protected by log_sys->checkpoint_lock.
+*/
+
+UNIV_INTERN
+void log_enable_checkpoint()
+{
+  ut_ad(log_disable_checkpoint_active);
+  /* Test variable, mostly to protect against wrong usage */
+  if (log_disable_checkpoint_active)
+  {
+    log_disable_checkpoint_active= 0;
+    rw_lock_x_unlock_gen(&(log_sys->checkpoint_lock), LOG_CHECKPOINT);
+  }
 }
 
 /****************************************************************//**
@@ -2189,7 +2272,7 @@ loop:
 	}
 
 	if (do_checkpoint) {
-		log_checkpoint(checkpoint_sync, FALSE);
+                log_checkpoint(checkpoint_sync, FALSE, FALSE);
 
 		if (checkpoint_sync) {
 
@@ -3099,6 +3182,10 @@ logs_empty_and_mark_files_at_shutdown(void)
 		ut_print_timestamp(stderr);
 		fprintf(stderr, "  InnoDB: Starting shutdown...\n");
 	}
+
+        /* Enable checkpoints if someone had turned them off */
+        log_enable_checkpoint();
+
 	/* Wait until the master thread and all other operations are idle: our
 	algorithm only works if the server is idle at shutdown */
 

storage/xtradb/srv/srv0srv.c

@@ -3232,7 +3232,7 @@ retry_flush_batch:
 
 	/* Make a new checkpoint about once in 10 seconds */
 
-	log_checkpoint(TRUE, FALSE);
+	log_checkpoint(TRUE, FALSE, TRUE);
 
 	srv_main_thread_op_info = "reserving kernel mutex";
 
@@ -3353,7 +3353,7 @@ flush_loop:
 
 	srv_main_thread_op_info = "making checkpoint";
 
-	log_checkpoint(TRUE, FALSE);
+	log_checkpoint(TRUE, FALSE, TRUE);
 
 	if (buf_get_modified_ratio_pct() > srv_max_buf_pool_modified_pct) {