Backport into build-201002030816-5.0.87sp1

> ------------------------------------------------------------
> revno: 2818.1.19
> revision-id: kostja@sun.com-20091103165854-7di545xruez8w207
> parent: li-bing.song@sun.com-20091103090041-zj7nedx6ok5jgges
> committer: Konstantin Osipov <kostja@sun.com>
> branch nick: 5.0-41756
> timestamp: Tue 2009-11-03 19:58:54 +0300
> message:
>   A fix and a test case for
>   Bug#41756 "Strange error messages about locks from InnoDB".
>   
>   In JT_EQ_REF (join_read_key()) access method,
>   don't try to unlock rows in the handler, unless certain that
>   a) they were locked
>   b) they are not used.
>   
>   Unlocking of rows is done by the logic of the nested join loop,
>   and is unaware of the possible caching that the access method may
>   have. This could lead to double unlocking, when a row
>   was unlocked first after reading into the cache, and then
>   when taken from cache, as well as to unlocking of rows which
>   were actually used (but taken from cache).
>   
>   Delegate part of the unlocking logic to the access method,
>   and in JT_EQ_REF count how many times a record was actually
>   used in the join. Unlock it only if it's usage count is 0.
>   
>   Implemented review comments.

5 files changed, 79 insertions, 7 deletions

diff --git a/sql/item_subselect.cc b/sql/item_subselect.cc
index 805669b3cfa..80fbc2c74d3 100644
--- a/sql/item_subselect.cc
+++ b/sql/item_subselect.cc
@@ -1869,6 +1869,7 @@ int subselect_single_select_engine::exec()
               tab->read_record.record= tab->table->record[0];
               tab->read_record.thd= join->thd;
               tab->read_record.ref_length= tab->table->file->ref_length;
+              tab->read_record.unlock_row= rr_unlock_row;
               *(last_changed_tab++)= tab;
               break;
             }
diff --git a/sql/records.cc b/sql/records.cc
index d5c3a421cd9..e0219fac06e 100644
--- a/sql/records.cc
+++ b/sql/records.cc
@@ -60,6 +60,7 @@ void init_read_record_idx(READ_RECORD *info, THD *thd, TABLE *table,
   info->file=  table->file;
   info->record= table->record[0];
   info->print_error= print_error;
+  info->unlock_row= rr_unlock_row;
 
   table->status=0;			/* And it's always found */
   if (!table->file->inited)
@@ -134,6 +135,7 @@ void init_read_record(READ_RECORD *info,THD *thd, TABLE *table,
   }
   info->select=select;
   info->print_error=print_error;
+  info->unlock_row= rr_unlock_row;
   info->ignore_not_found_rows= 0;
   table->status=0;			/* And it's always found */
 
diff --git a/sql/sql_select.cc b/sql/sql_select.cc
index ca7f20f5b21..c3cd1f41a41 100644
--- a/sql/sql_select.cc
+++ b/sql/sql_select.cc
@@ -140,6 +140,7 @@ static int join_read_const_table(JOIN_TAB *tab, POSITION *pos);
 static int join_read_system(JOIN_TAB *tab);
 static int join_read_const(JOIN_TAB *tab);
 static int join_read_key(JOIN_TAB *tab);
+static void join_read_key_unlock_row(st_join_table *tab);
 static int join_read_always_key(JOIN_TAB *tab);
 static int join_read_last_key(JOIN_TAB *tab);
 static int join_no_more_records(READ_RECORD *info);
@@ -5376,7 +5377,9 @@ static bool create_ref_for_key(JOIN *join, JOIN_TAB *j, KEYUSE *org_keyuse,
   }
   j->ref.key_buff2=j->ref.key_buff+ALIGN_SIZE(length);
   j->ref.key_err=1;
+  j->ref.has_record= FALSE;
   j->ref.null_rejecting= 0;
+  j->ref.use_count= 0;
   keyuse=org_keyuse;
 
   store_key **ref_key= j->ref.key_copy;
@@ -6176,6 +6179,20 @@ make_join_select(JOIN *join,SQL_SELECT *select,COND *cond)
   DBUG_RETURN(0);
 }
 
+
+/**
+  The default implementation of unlock-row method of READ_RECORD,
+  used in all access methods.
+*/
+
+void rr_unlock_row(st_join_table *tab)
+{
+  READ_RECORD *info= &tab->read_record;
+  info->file->unlock_row();
+}
+
+
+
 static void
 make_join_readinfo(JOIN *join, ulonglong options)
 {
@@ -6191,6 +6208,7 @@ make_join_readinfo(JOIN *join, ulonglong options)
     TABLE *table=tab->table;
     tab->read_record.table= table;
     tab->read_record.file=table->file;
+    tab->read_record.unlock_row= rr_unlock_row;
     tab->next_select=sub_select;		/* normal select */
 
     /*
@@ -6234,6 +6252,7 @@ make_join_readinfo(JOIN *join, ulonglong options)
       delete tab->quick;
       tab->quick=0;
       tab->read_first_record= join_read_key;
+      tab->read_record.unlock_row= join_read_key_unlock_row;
       tab->read_record.read_record= join_no_more_records;
       if (table->used_keys.is_set(tab->ref.key) &&
 	  !table->no_keyread)
@@ -10933,7 +10952,7 @@ evaluate_join_record(JOIN *join, JOIN_TAB *join_tab,
         return NESTED_LOOP_NO_MORE_ROWS;
     }
     else
-      join_tab->read_record.file->unlock_row();
+      join_tab->read_record.unlock_row(join_tab);
   }
   else
   {
@@ -10943,7 +10962,7 @@ evaluate_join_record(JOIN *join, JOIN_TAB *join_tab,
     */
     join->examined_rows++;
     join->thd->row_count++;
-    join_tab->read_record.file->unlock_row();
+    join_tab->read_record.unlock_row(join_tab);
   }
   return NESTED_LOOP_OK;
 }
@@ -11296,17 +11315,55 @@ join_read_key(JOIN_TAB *tab)
       table->status=STATUS_NOT_FOUND;
       return -1;
     }
+    /*
+      Moving away from the current record. Unlock the row
+      in the handler if it did not match the partial WHERE.
+    */
+    if (tab->ref.has_record && tab->ref.use_count == 0)
+    {
+      tab->read_record.file->unlock_row();
+      tab->ref.has_record= FALSE;
+    }
+
     error=table->file->index_read(table->record[0],
 				  tab->ref.key_buff,
 				  tab->ref.key_length,HA_READ_KEY_EXACT);
     if (error && error != HA_ERR_KEY_NOT_FOUND)
       return report_error(table, error);
+
+    if (! error)
+    {
+      tab->ref.has_record= TRUE;
+      tab->ref.use_count= 1;
+    }
+  }
+  else if (table->status == 0)
+  {
+    DBUG_ASSERT(tab->ref.has_record);
+    tab->ref.use_count++;
   }
   table->null_row=0;
   return table->status ? -1 : 0;
 }
 
 
+/**
+  Since join_read_key may buffer a record, do not unlock
+  it if it was not used in this invocation of join_read_key().
+  Only count locks, thus remembering if the record was left unused,
+  and unlock already when pruning the current value of
+  TABLE_REF buffer.
+  @sa join_read_key()
+*/
+
+static void
+join_read_key_unlock_row(st_join_table *tab)
+{
+  DBUG_ASSERT(tab->ref.use_count);
+  if (tab->ref.use_count)
+    tab->ref.use_count--;
+}
+
 /*
   ref access method implementation: "read_first" function
 
diff --git a/sql/sql_select.h b/sql/sql_select.h
index a217d199a30..346d98aae58 100644
--- a/sql/sql_select.h
+++ b/sql/sql_select.h
@@ -53,6 +53,8 @@ class store_key;
 typedef struct st_table_ref
 {
   bool		key_err;
+  /** True if something was read into buffer in join_read_key.  */
+  bool          has_record;
   uint          key_parts;                // num of ...
   uint          key_length;               // length of key_buff
   int           key;                      // key no
@@ -79,7 +81,11 @@ typedef struct st_table_ref
   key_part_map  null_rejecting;
   table_map	depend_map;		  // Table depends on these tables.
   byte          *null_ref_key;		  // null byte position in the key_buf.
-  					  // used for REF_OR_NULL optimization.
+  /*
+    The number of times the record associated with this key was used
+    in the join.
+  */
+  ha_rows       use_count;
 } TABLE_REF;
 
 /*
diff --git a/sql/structs.h b/sql/structs.h
index be55527eaa0..3a7b2f49b2a 100644
--- a/sql/structs.h
+++ b/sql/structs.h
@@ -117,16 +117,22 @@ typedef struct st_reginfo {		/* Extra info about reg */
 } REGINFO;
 
 
-struct st_read_record;				/* For referense later */
 class SQL_SELECT;
 class THD;
 class handler;
+struct st_join_table;
+
+void rr_unlock_row(st_join_table *tab);
 
-typedef struct st_read_record {			/* Parameter to read_record */
+struct READ_RECORD {			/* Parameter to read_record */
+  typedef int (*Read_func)(READ_RECORD*);
+  typedef void (*Unlock_row_func)(st_join_table *);
   struct st_table *table;			/* Head-form */
   handler *file;
   struct st_table **forms;			/* head and ref forms */
-  int (*read_record)(struct st_read_record *);
+
+  Read_func read_record;
+  Unlock_row_func unlock_row;
   THD *thd;
   SQL_SELECT *select;
   uint cache_records;
@@ -138,7 +144,7 @@ typedef struct st_read_record {			/* Parameter to read_record */
   byte	*cache,*cache_pos,*cache_end,*read_positions;
   IO_CACHE *io_cache;
   bool print_error, ignore_not_found_rows;
-} READ_RECORD;
+};
 
 
 /*