Added innodb_locks_unsafe_for_binlog option. This option turns off Innodb

  next-key locking. Using this option the locks InnoDB sets on index 
  records do not affect the ``gap'' before that index record. Thus, this option
  allows phantom problem.


innobase/include/srv0srv.h:
  Added srv_locks_unsafe_for_binlog for innodb_locks_unsafe_for_binlog option.
innobase/row/row0sel.c:
  If innodb_locks_unsafe_for_binlog option is used, we lock only the record, i.e. 
  next-key locking is not used. Therefore, setting lock to the index record
  do not affect the ``gap'' before that index record. Thus, this option
  allows phantom problem, because concurrent insert operations are allowed inside 
  the select range.
innobase/srv/srv0srv.c:
  Added srv_locks_unsafe_for_binlog for innodb_locks_unsafe_for_binlog option.
sql/ha_innodb.cc:
  Added innobase_locks_unsafe_for_binlog and srv_locks_unsafe_for_binlog for 
  innodb_locks_unsafe_for_binlog option.
sql/ha_innodb.h:
  Added innobase_locks_unsafe_for_binlog for innodb_locks_unsafe_for_binlog option.
sql/mysqld.cc:
  Added OPT_INNODB_LOCKS_UNSAFE_FOR_BINLOG, innobase_locks_unsafe_for_binlog for
  innodb_locks_unsafe_for_binlog option.
sql/set_var.cc:
  Added innodb_locks_unsafe_for_binlog and innobase_locks_unsafe_for_binlog for
  innodb_locks_unsafe_for_binlog option.
BitKeeper/etc/logging_ok:
  Logging to logging@openlogging.org accepted

8 files changed, 91 insertions, 8 deletions

diff --git a/BitKeeper/etc/logging_ok b/BitKeeper/etc/logging_ok
index 398a9295579..a9cb6429a35 100644
--- a/BitKeeper/etc/logging_ok
+++ b/BitKeeper/etc/logging_ok
@@ -61,6 +61,7 @@ hf@genie.(none)
 igor@hundin.mysql.fi
 igor@rurik.mysql.com
 ingo@mysql.com
+jan@hundin.mysql.fi
 jani@a80-186-24-72.elisa-laajakaista.fi
 jani@dsl-jkl1657.dial.inet.fi
 jani@dsl-kpogw4gb5.dial.inet.fi
diff --git a/innobase/include/srv0srv.h b/innobase/include/srv0srv.h
index c527d40bc79..b98223cff68 100644
--- a/innobase/include/srv0srv.h
+++ b/innobase/include/srv0srv.h
@@ -42,6 +42,7 @@ extern char*	srv_arch_dir;
 #endif /* UNIV_LOG_ARCHIVE */
 
 extern ibool	srv_file_per_table;
+extern ibool    srv_locks_unsafe_for_binlog;
 
 extern ulint	srv_n_data_files;
 extern char**	srv_data_file_names;
diff --git a/innobase/row/row0sel.c b/innobase/row/row0sel.c
index 8a0da2851a7..bf7f6f1fc3a 100644
--- a/innobase/row/row0sel.c
+++ b/innobase/row/row0sel.c
@@ -631,10 +631,24 @@ row_sel_get_clust_rec(
 
 	if (!node->read_view) {
 		/* Try to place a lock on the index record */
-		
+        
+          /* If innodb_locks_unsafe_for_binlog option is used, 
+             we lock only the record, i.e. next-key locking is
+             not used.
+	  */
+	  if ( srv_locks_unsafe_for_binlog )
+	    {
+		err = lock_clust_rec_read_check_and_lock(0, clust_rec, 
+                        index,node->row_lock_mode, LOCK_REC_NOT_GAP, thr);
+	    }
+	  else
+	    {		
 		err = lock_clust_rec_read_check_and_lock(0, clust_rec, index,
 				node->row_lock_mode, LOCK_ORDINARY, thr);
-		if (err != DB_SUCCESS) {
+
+	    }
+
+	    if (err != DB_SUCCESS) {
 
 			return(err);
 		}
@@ -1184,9 +1198,23 @@ rec_loop:
 		search result set, resulting in the phantom problem. */
 		
 		if (!consistent_read) {
+
+		  /* If innodb_locks_unsafe_for_binlog option is used,
+                     we lock only the record, i.e. next-key locking is
+                     not used.
+		  */
+
+                  if ( srv_locks_unsafe_for_binlog )
+		    {
+			err = sel_set_rec_lock(page_rec_get_next(rec), index,
+				node->row_lock_mode, LOCK_REC_NOT_GAP, thr);
+		    } 
+                    else
+		    {
 			err = sel_set_rec_lock(page_rec_get_next(rec), index,
 				node->row_lock_mode, LOCK_ORDINARY, thr);
-			if (err != DB_SUCCESS) {
+		    }
+		    if (err != DB_SUCCESS) {
 				/* Note that in this case we will store in pcur
 				the PREDECESSOR of the record we are waiting
 				the lock for */
@@ -1211,8 +1239,22 @@ rec_loop:
 	if (!consistent_read) {
 		/* Try to place a lock on the index record */	
 
-		err = sel_set_rec_lock(rec, index, node->row_lock_mode,
+		  /* If innodb_locks_unsafe_for_binlog option is used,
+                     we lock only the record, i.e. next-key locking is
+                     not used.
+		  */
+
+                  if ( srv_locks_unsafe_for_binlog )
+		    {
+		        err = sel_set_rec_lock(rec, index, node->row_lock_mode,
+						LOCK_REC_NOT_GAP, thr);
+		    } 
+		  else
+		    {
+		        err = sel_set_rec_lock(rec, index, node->row_lock_mode,
 						LOCK_ORDINARY, thr);
+		    }
+
 		if (err != DB_SUCCESS) {
 
 			goto lock_wait_or_error;
@@ -3144,10 +3186,24 @@ rec_loop:
 
 			/* Try to place a lock on the index record */	
 
-			err = sel_set_rec_lock(rec, index,
+                        /* If innodb_locks_unsafe_for_binlog option is used, 
+                           we lock only the record, i.e. next-key locking is
+                           not used.
+	                */
+	                if ( srv_locks_unsafe_for_binlog )
+	                {
+			    err = sel_set_rec_lock(rec, index,
+						prebuilt->select_lock_type,
+						LOCK_REC_NOT_GAP, thr);
+			}
+			else
+			{
+			    err = sel_set_rec_lock(rec, index,
 						prebuilt->select_lock_type,
 						LOCK_ORDINARY, thr);
-			if (err != DB_SUCCESS) {
+			}
+			
+                        if (err != DB_SUCCESS) {
 
 				goto lock_wait_or_error;
 			}
@@ -3300,9 +3356,22 @@ rec_loop:
 						prebuilt->select_lock_type,
 						LOCK_REC_NOT_GAP, thr);
 		} else {
-			err = sel_set_rec_lock(rec, index,
+                        /* If innodb_locks_unsafe_for_binlog option is used, 
+                           we lock only the record, i.e. next-key locking is
+                           not used.
+	                */
+	                if ( srv_locks_unsafe_for_binlog )
+	                {
+			    err = sel_set_rec_lock(rec, index,
+						prebuilt->select_lock_type,
+						LOCK_REC_NOT_GAP, thr);
+			}
+			else
+			{
+			    err = sel_set_rec_lock(rec, index,
 						prebuilt->select_lock_type,
 						LOCK_ORDINARY, thr);
+			}
 		}
 		
 		if (err != DB_SUCCESS) {
diff --git a/innobase/srv/srv0srv.c b/innobase/srv/srv0srv.c
index fc46c95a8a6..4c305a76547 100644
--- a/innobase/srv/srv0srv.c
+++ b/innobase/srv/srv0srv.c
@@ -74,6 +74,10 @@ ibool	srv_file_per_table = FALSE;	/* store to its own file each table
 					created by an user; data dictionary
 					tables are in the system tablespace
 					0 */
+ibool   srv_locks_unsafe_for_binlog = FALSE; /* Place locks to records only 
+                                                i.e. do not use next-key locking
+                                                except on duplicate key checking and
+                                                foreign key checking */
 ulint	srv_n_data_files = 0;
 char**	srv_data_file_names = NULL;
 ulint*	srv_data_file_sizes = NULL;	/* size in database pages */ 
diff --git a/sql/ha_innodb.cc b/sql/ha_innodb.cc
index 6eae315e443..a7dce3a6ab8 100644
--- a/sql/ha_innodb.cc
+++ b/sql/ha_innodb.cc
@@ -117,6 +117,7 @@ my_bool innobase_log_archive			= FALSE;/* unused */
 my_bool	innobase_use_native_aio			= FALSE;
 my_bool	innobase_fast_shutdown			= TRUE;
 my_bool	innobase_file_per_table			= FALSE;
+my_bool innobase_locks_unsafe_for_binlog        = FALSE;
 
 static char *internal_innobase_data_file_path	= NULL;
 
@@ -908,6 +909,7 @@ innobase_init(void)
 	srv_fast_shutdown = (ibool) innobase_fast_shutdown;
 
 	srv_file_per_table = (ibool) innobase_file_per_table;
+        srv_locks_unsafe_for_binlog = (ibool) innobase_locks_unsafe_for_binlog;
 
 	srv_max_n_open_files = (ulint) innobase_open_files;
 
diff --git a/sql/ha_innodb.h b/sql/ha_innodb.h
index e09697f7ce6..6815bdd632d 100644
--- a/sql/ha_innodb.h
+++ b/sql/ha_innodb.h
@@ -189,7 +189,7 @@ extern char *innobase_unix_file_flush_method;
 /* The following variables have to be my_bool for SHOW VARIABLES to work */
 extern my_bool innobase_log_archive,
                innobase_use_native_aio, innobase_fast_shutdown,
-	       innobase_file_per_table;
+	       innobase_file_per_table, innobase_locks_unsafe_for_binlog;
 extern "C" {
 extern ulong srv_max_buf_pool_modified_pct;
 }
diff --git a/sql/mysqld.cc b/sql/mysqld.cc
index 4fd13d33bab..869048cee93 100644
--- a/sql/mysqld.cc
+++ b/sql/mysqld.cc
@@ -3880,6 +3880,7 @@ enum options_mysqld
   OPT_INNODB_FLUSH_METHOD,
   OPT_INNODB_FAST_SHUTDOWN,
   OPT_INNODB_FILE_PER_TABLE, OPT_CRASH_BINLOG_INNODB,
+  OPT_INNODB_LOCKS_UNSAFE_FOR_BINLOG,
   OPT_SAFE_SHOW_DB, OPT_INNODB_SAFE_BINLOG,
   OPT_INNODB, OPT_ISAM, OPT_NDBCLUSTER, OPT_SKIP_SAFEMALLOC,
   OPT_TEMP_POOL, OPT_TX_ISOLATION,
@@ -4156,6 +4157,10 @@ Disable with --skip-bdb (will save memory).",
    "Stores each InnoDB table to an .ibd file in the database dir.",
    (gptr*) &innobase_file_per_table,
    (gptr*) &innobase_file_per_table, 0, GET_BOOL, NO_ARG, 0, 0, 0, 0, 0, 0},
+  {"innodb_locks_unsafe_for_binlog", OPT_INNODB_LOCKS_UNSAFE_FOR_BINLOG,
+   "Force Innodb not to use next-key locking. Instead use only row-level locking",
+   (gptr*) &innobase_locks_unsafe_for_binlog,
+   (gptr*) &innobase_locks_unsafe_for_binlog, 0, GET_BOOL, NO_ARG, 0, 0, 0, 0, 0, 0},
 #endif /* End HAVE_INNOBASE_DB */
   {"init-connect", OPT_INIT_CONNECT, "Command(s) that are executed for each new connection",
    (gptr*) &opt_init_connect, (gptr*) &opt_init_connect, 0, GET_STR_ALLOC,
diff --git a/sql/set_var.cc b/sql/set_var.cc
index e1cfb77d297..fb9ff285859 100644
--- a/sql/set_var.cc
+++ b/sql/set_var.cc
@@ -673,6 +673,7 @@ struct show_var_st init_vars[]= {
   {"innodb_fast_shutdown", (char*) &innobase_fast_shutdown, SHOW_MY_BOOL},
   {"innodb_file_io_threads", (char*) &innobase_file_io_threads, SHOW_LONG },
   {"innodb_file_per_table", (char*) &innobase_file_per_table, SHOW_MY_BOOL},
+  {"innodb_locks_unsafe_for_binlog", (char*) &innobase_locks_unsafe_for_binlog, SHOW_MY_BOOL},
   {"innodb_flush_log_at_trx_commit", (char*) &innobase_flush_log_at_trx_commit, SHOW_INT},
   {"innodb_flush_method",    (char*) &innobase_unix_file_flush_method, SHOW_CHAR_PTR},
   {"innodb_force_recovery", (char*) &innobase_force_recovery, SHOW_LONG },