Automatic primary key for BDB tables

BitKeeper/etc/logging_ok

-monty@tik.mysql.com
+monty@donna.mysql.com

Docs/manual.texi

@@ -1793,16 +1793,20 @@ Big changes made in @strong{MySQL} Version 3.22.12.
 @item @strong{MyODBC} (uses ODBC SDK 2.5) --- Gamma
 It seems to work well with some programs.
 
-@item Replication -- Alpha
+@item Replication -- Alpha / Beta
 We are still working on replication, so don't expect this to be rock
 solid yet.  On the other hand, some @strong{MySQL} users are already
 using this with good results.
 
-@item BDB Tables -- Alpha
+@item BDB Tables -- Alpha  / Beta
 The Berkeley DB code is very stable, but we are still improving the interface
 between @strong{MySQL} and BDB tables, so it will take some time before this
 is as tested as the other table types.
 
+@item Automatic recovery of MyISAM tables - Alpha.
+This only affects the new code that checks if the table was closed properly
+on open and executes an automatic check/repair of the table if it wasn't.
+
 @end table
 
 MySQL AB provides e-mail support for paying customers, but the @strong{MySQL}
@@ -7979,12 +7983,10 @@ you should also compile your code to be multi-threaded!
 @node Windows and BDB tables., Windows vs Unix, Windows compiling, Windows
 @subsection Windows and BDB Tables
 
-We are working on removing the requirement that one must have a primary
-key in a BDB table. As soon as this is fixed we will throughly test the
-BDB interface by running the @strong{MySQL} benchmark and our internal
-test suite on it.  When the above is done we will start to release binary
-distributions (for Windows and UNIX) of @strong{MySQL} that will include
-support for BDB tables.
+We will shortly do a full test on the new BDB interface on Windows.
+When this is done we will start to release binary distributions (for
+Windows and UNIX) of @strong{MySQL} that will include support for BDB
+tables.
 
 @cindex Windows, versus UNIX
 @cindex operating systems, Windows versus UNIX
@@ -21876,13 +21878,22 @@ Some characteristic of @code{BDB} tables:
 
 @itemize @bullet
 @item
-All @code{BDB} tables must have a primary key.
+@strong{MySQL} requires a @code{PRIMARY KEY} in each BDB table to be
+able to refer to previously read rows; If you don't create on,
+@strong{MySQL} will create an maintain a hidden @code{PRIMARY KEY} for
+you.  The hidden key has a length of 5 bytes and is incremented for each
+insert attempt.
 @item
 If all columns you access in a @code{BDB} tables is part of the same index or
 part of the the primary key then @strong{MySQL} can execute the query
 without having to access the actual row.  In a @code{MyISAM} table the
 above holds only if the columns are part of the same index.
 @item
+The @code{PRIMARY KEY} will be faster than any other key, as the
+@code{PRIMARY KEY} is stored together with the row data.  As the other keys are
+stored as the key data + the @code{PRIMARY KEY}, its important to keep the
+@code{PRIMARY KEY} as short as possible to save disk and get better speed.
+@item
 @code{LOCK TABLES} works on @code{BDB} tables as with other tables.  If
 you don't use @code{LOCK TABLE}, @strong{MYSQL} will issue an internal
 multiple write lock on the table to ensure that the table will be
@@ -37972,6 +37983,11 @@ though, so 3.23 is not released as a stable version yet.
 @appendixsubsec Changes in release 3.23.26
 @itemize @bullet
 @item
+If one don't create a @code{PRIMARY KEY} in a BDB table, a hidden
+@code{PRIMARY KEY} will be created.
+@item
+Added read-only-key optimization to BDB tables.
+@item
 @code{LEFT JOIN} did in some case prefer a full table scan when one
 didn't have a @code{WHERE} clause.
 @item

sql/field.h

@@ -44,7 +44,7 @@ public:
   uint8		null_bit;		// And position to it
   struct st_table *table;		// Pointer for table
   ulong query_id;			// For quick test of used fields
-  key_map key_start,part_of_key;	// Which keys a field is in
+  key_map key_start,part_of_key;	// Key is part of these keys.
   const char *table_name,*field_name;
   utype unireg_check;
   uint32 field_length;			// Length of field

sql/filesort.cc

@@ -331,7 +331,7 @@ static ha_rows find_all_keys(SORTPARAM *param, SQL_SELECT *select,
   if (! indexfile && ! quick_select)
   {
     file->reset();			// QQ; Shouldn't be needed
-    if (table->keyread)			// QQ Can be removed after the reset
+    if (sort_form->key_read)		// QQ Can be removed after the reset
       file->extra(HA_EXTRA_KEYREAD);	// QQ is removed
     next_pos=(byte*) 0;			/* Find records in sequence */
     file->rnd_init();

sql/ha_berkeley.h

@@ -23,15 +23,18 @@
 
 #include <db.h>
 
+#define BDB_HIDDEN_PRIMARY_KEY_LENGTH 5
+
 typedef struct st_berkeley_share {
   ulonglong auto_ident;
   THR_LOCK lock;
   pthread_mutex_t mutex;
   char *table_name;
   uint table_name_length,use_count;
-  my_bool inited;
+  bool primary_key_inited;
 } BDB_SHARE;
 
+
 class ha_berkeley: public handler
 {
   THR_LOCK_DATA lock;
@@ -46,13 +49,15 @@ class ha_berkeley: public handler
   BDB_SHARE *share;
   ulong int_option_flag;
   ulong alloced_rec_buff_length;
-  uint primary_key,last_dup_key;
-  bool fixed_length_row, fixed_length_primary_key, hidden_primary_key;
-
+  uint primary_key,last_dup_key, hidden_primary_key;
+  bool fixed_length_row, fixed_length_primary_key, key_read;
   bool  fix_rec_buff_for_blob(ulong length);
+  byte current_ident[BDB_HIDDEN_PRIMARY_KEY_LENGTH];
+
   ulong max_row_length(const byte *buf);
-  int pack_row(DBT *row,const  byte *record);
+  int pack_row(DBT *row,const  byte *record, bool new_row);
   void unpack_row(char *record, DBT *row);
+  void ha_berkeley::unpack_key(char *record, DBT *key, uint index);
   DBT *pack_key(DBT *key, uint keynr, char *buff, const byte *record);
   DBT *pack_key(DBT *key, uint keynr, char *buff, const byte *key_ptr,
 		uint key_length);
@@ -64,7 +69,7 @@ class ha_berkeley: public handler
   int update_primary_key(DB_TXN *trans, bool primary_key_changed,
 			 const byte * old_row, const byte * new_row,
 			 DBT *prim_key);
-  int read_row(int error, char *buf, uint keynr, DBT *row, bool);
+  int read_row(int error, char *buf, uint keynr, DBT *row, DBT *key, bool);
   DBT *get_pos(DBT *to, byte *pos);
 
  public:
@@ -72,9 +77,8 @@ class ha_berkeley: public handler
     int_option_flag(HA_READ_NEXT | HA_READ_PREV |
 		    HA_REC_NOT_IN_SEQ |
 		    HA_KEYPOS_TO_RNDPOS | HA_READ_ORDER | HA_LASTKEY_ORDER |
-		    HA_LONGLONG_KEYS | HA_NULL_KEY |
-		    HA_BLOB_KEY | 
-		    HA_REQUIRE_PRIMARY_KEY | HA_NOT_EXACT_COUNT | 
+		    HA_LONGLONG_KEYS | HA_NULL_KEY | HA_HAVE_KEY_READ_ONLY |
+		    HA_BLOB_KEY | HA_NOT_EXACT_COUNT | 
 		    HA_PRIMARY_KEY_IN_READ_INDEX | HA_DROP_BEFORE_CREATE),
     last_dup_key((uint) -1)
   {
@@ -84,14 +88,14 @@ class ha_berkeley: public handler
   const char **bas_ext() const;
   ulong option_flag() const { return int_option_flag; }
   uint max_record_length() const { return HA_MAX_REC_LENGTH; }
-  uint max_keys()          const { return MAX_KEY; }
+  uint max_keys()          const { return MAX_KEY-1; }
   uint max_key_parts()     const { return MAX_REF_PARTS; }
   uint max_key_length()    const { return MAX_KEY_LENGTH; }
+  uint extra_rec_buf_length() { return BDB_HIDDEN_PRIMARY_KEY_LENGTH; }
   bool fast_key_read()	   { return 1;}
   bool has_transactions()  { return 1;}
 
   int open(const char *name, int mode, uint test_if_locked);
-  void initialize(void);
   int close(void);
   double scan_time();
   int write_row(byte * buf);
@@ -129,6 +133,16 @@ class ha_berkeley: public handler
   int delete_table(const char *name);
   THR_LOCK_DATA **store_lock(THD *thd, THR_LOCK_DATA **to,
 			     enum thr_lock_type lock_type);
+
+  void update_auto_primary_key();
+  inline void get_auto_primary_key(byte *to)
+  {
+    ulonglong tmp;
+    pthread_mutex_lock(&share->mutex);
+    share->auto_ident++;
+    int5store(to,share->auto_ident);
+    pthread_mutex_unlock(&share->mutex);
+  }
 };
 
 extern bool berkeley_skip;

sql/handler.h

@@ -202,6 +202,7 @@ public:
   virtual double read_time(ha_rows rows) { return rows; }
   virtual bool fast_key_read() { return 0;}
   virtual bool has_transactions(){ return 0;}
+  virtual uint extra_rec_buf_length() { return 0; }
 
   virtual int index_init(uint idx) { active_index=idx; return 0;}
   virtual int index_end() {return 0; }

sql/opt_range.cc

@@ -2347,6 +2347,7 @@ bool QUICK_SELECT::unique_key_range()
 
 QUICK_SELECT *get_quick_select_for_ref(TABLE *table, TABLE_REF *ref)
 {
+  table->file->index_end();			// Remove old cursor
   QUICK_SELECT *quick=new QUICK_SELECT(table, ref->key, 1);
   KEY *key_info = &table->key_info[ref->key];
   KEY_PART *key_part;

sql/sql_base.cc

@@ -1413,11 +1413,17 @@ Field *find_field_in_table(THD *thd,TABLE *table,const char *name,uint length,
     if (field->query_id != thd->query_id)
     {
       field->query_id=thd->query_id;
-      field->table->used_fields++;
+      table->used_fields++;
+      if (field->part_of_key)
+      {
+	if (!(field->part_of_key & table->ref_primary_key))
+	  table->used_keys&=field->part_of_key;
+      }
+      else
+	table->used_keys=0;
     }
     else
       thd->dupp_field=field;
-    field->table->used_keys&=field->part_of_key;
   }
   if (check_grants && !thd->master_access && check_grant_column(thd,table,name,length))
     return WRONG_GRANT;
@@ -1659,19 +1665,19 @@ static bool
 insert_fields(THD *thd,TABLE_LIST *tables, const char *table_name,
 	      List_iterator<Item> *it)
 {
-  TABLE_LIST *table;
   uint found;
   DBUG_ENTER("insert_fields");
 
   found=0;
-  for (table=tables ; table ; table=table->next)
+  for (; tables ; tables=tables->next)
   {
+    TABLE *table=tables->table;
     if (grant_option && !thd->master_access &&
-	check_grant_all_columns(thd,SELECT_ACL,table->table) )
+	check_grant_all_columns(thd,SELECT_ACL,table) )
       DBUG_RETURN(-1);
-    if (!table_name || !strcmp(table_name,table->name))
+    if (!table_name || !strcmp(table_name,tables->name))
     {
-      Field **ptr=table->table->field,*field;
+      Field **ptr=table->field,*field;
       while ((field = *ptr++))
       {
 	Item_field *item= new Item_field(field);
@@ -1682,10 +1688,17 @@ insert_fields(THD *thd,TABLE_LIST *tables, const char *table_name,
 	if (field->query_id == thd->query_id)
 	  thd->dupp_field=field;
 	field->query_id=thd->query_id;
-	field->table->used_keys&=field->part_of_key;
+
+	if (field->part_of_key)
+	{
+	  if (!(field->part_of_key & table->ref_primary_key))
+	    table->used_keys&=field->part_of_key;
+	}
+	else
+	  table->used_keys=0;
       }
       /* All fields are used */
-      table->table->used_fields=table->table->fields;
+      table->used_fields=table->fields;
     }
   }
   if (!found)
@@ -1750,6 +1763,7 @@ int setup_conds(THD *thd,TABLE_LIST *tables,COND **conds)
 	// TODO: This could be optimized to use hashed names if t2 had a hash
 	for (j=0 ; j < t2->fields ; j++)
 	{
+	  key_map tmp_map;
 	  if (!my_strcasecmp(t1->field[i]->field_name,
 			     t2->field[j]->field_name))
 	  {
@@ -1760,8 +1774,20 @@ int setup_conds(THD *thd,TABLE_LIST *tables,COND **conds)
 	    tmp->fix_length_and_dec();	// Update cmp_type
 	    tmp->const_item_cache=0;
 	    cond_and->list.push_back(tmp);
-	    t1->used_keys&= t1->field[i]->part_of_key;
-	    t2->used_keys&= t2->field[j]->part_of_key;
+	    if ((tmp_map=t1->field[i]->part_of_key))
+	    {
+	      if (!(tmp_map & t1->ref_primary_key))
+		t1->used_keys&=tmp_map;
+	    }
+	    else
+	      t1->used_keys=0;
+	    if ((tmp_map=t2->field[j]->part_of_key))
+	    {
+	      if (!(tmp_map & t2->ref_primary_key))
+		t2->used_keys&=tmp_map;
+	    }
+	    else
+	      t2->used_keys=0;
 	    break;
 	  }
 	}

sql/sql_select.cc

@@ -2352,6 +2352,11 @@ make_join_readinfo(JOIN *join,uint options)
       break;
     case JT_EQ_REF:
       table->status=STATUS_NO_RECORD;
+      if (tab->select)
+      {
+	delete tab->select->quick;
+	tab->select->quick=0;
+      }
       delete tab->quick;
       tab->quick=0;
       table->file->index_init(tab->ref.key);
@@ -2365,6 +2370,11 @@ make_join_readinfo(JOIN *join,uint options)
       break;
     case JT_REF:
       table->status=STATUS_NO_RECORD;
+      if (tab->select)
+      {
+	delete tab->select->quick;
+	tab->select->quick=0;
+      }
       delete tab->quick;
       tab->quick=0;
       table->file->index_init(tab->ref.key);

sql/table.cc

@@ -207,49 +207,29 @@ int openfrm(const char *name, const char *alias, uint db_stat, uint prgflag,
   }
 #endif
 
+  /* Allocate handler */
   if (!(outparam->file= get_new_handler(outparam,outparam->db_type)))
     goto err_not_open;
-  error=2;
-  if (db_stat)
-  {
-    int err;
-    if ((err=(outparam->file->
-	      ha_open(index_file,
-		      (db_stat & HA_READ_ONLY ? O_RDONLY : O_RDWR),
-		      (db_stat & HA_OPEN_TEMPORARY ? HA_OPEN_TMP_TABLE :
-		       ((db_stat & HA_WAIT_IF_LOCKED) ||
-			(specialflag & SPECIAL_WAIT_IF_LOCKED)) ?
-		       HA_OPEN_WAIT_IF_LOCKED :
-		       (db_stat & (HA_ABORT_IF_LOCKED | HA_GET_INFO)) ?
-		       HA_OPEN_ABORT_IF_LOCKED :
-		       HA_OPEN_IGNORE_IF_LOCKED) | ha_open_flags))))
-    {
-      /* Set a flag if the table is crashed and it can be auto. repaired */
-      outparam->crashed=(err == HA_ERR_CRASHED &&
-			 outparam->file->auto_repair() &&
-			 !(ha_open_flags & HA_OPEN_FOR_REPAIR));
-      goto err_not_open; /* purecov: inspected */
-    }
-  }
-  outparam->db_low_byte_first=outparam->file->low_byte_first();
 
   error=4;
   outparam->reginfo.lock_type= TL_UNLOCK;
   outparam->current_lock=F_UNLCK;
-  if (db_stat & HA_OPEN_KEYFILE || (prgflag & DELAYED_OPEN)) records=2;
+  if ((db_stat & HA_OPEN_KEYFILE) || (prgflag & DELAYED_OPEN)) records=2;
   else records=1;
   if (prgflag & (READ_ALL+EXTRA_RECORD)) records++;
-  rec_buff_length=ALIGN_SIZE(outparam->reclength+1);
+  /* QQ: TODO, remove the +1 from below */
+  rec_buff_length=ALIGN_SIZE(outparam->reclength+1+
+			     outparam->file->extra_rec_buf_length());
   if (!(outparam->record[0]= (byte*)
 	(record = (char *) alloc_root(&outparam->mem_root,
 				      rec_buff_length * records))))
-    goto err;					/* purecov: inspected */
+    goto err_not_open;				/* purecov: inspected */
   record[outparam->reclength]=0;		// For purify and ->c_ptr()
   outparam->rec_buff_length=rec_buff_length;
   if (my_pread(file,(byte*) record,(uint) outparam->reclength,
 	       (ulong) (uint2korr(head+6)+uint2korr(head+14)),
 	       MYF(MY_NABP)))
-    goto err; /* purecov: inspected */
+    goto err_not_open; /* purecov: inspected */
   for (i=0 ; i < records ; i++, record+=rec_buff_length)
   {
     outparam->record[i]=(byte*) record;
@@ -265,12 +245,12 @@ int openfrm(const char *name, const char *alias, uint db_stat, uint prgflag,
   }
 
   VOID(my_seek(file,pos,MY_SEEK_SET,MYF(0)));
-  if (my_read(file,(byte*) head,288,MYF(MY_NABP))) goto err;
+  if (my_read(file,(byte*) head,288,MYF(MY_NABP))) goto err_not_open;
   if (crypted)
   {
     crypted->decode((char*) head+256,288-256);
     if (sint2korr(head+284) != 0)		// Should be 0
-      goto err;					// Wrong password
+      goto err_not_open;			// Wrong password
   }
 
   outparam->fields= uint2korr(head+258);
@@ -292,13 +272,13 @@ int openfrm(const char *name, const char *alias, uint db_stat, uint prgflag,
 			   (outparam->fields+interval_parts+
 			    keys+3)*sizeof(my_string)+
 			   (n_length+int_length)))))
-    goto err; /* purecov: inspected */
+    goto err_not_open; /* purecov: inspected */
 
   outparam->field=field_ptr;
   read_length=((uint) (outparam->fields*11)+pos+
 	       (uint) (n_length+int_length));
   if (read_string(file,(gptr*) &disk_buff,read_length))
-    goto err; /* purecov: inspected */
+    goto err_not_open; /* purecov: inspected */
   if (crypted)
   {
     crypted->decode((char*) disk_buff,read_length);
@@ -321,6 +301,7 @@ int openfrm(const char *name, const char *alias, uint db_stat, uint prgflag,
   if (keynames)
     fix_type_pointers(&int_array,&outparam->keynames,1,&keynames);
   VOID(my_close(file,MYF(MY_WME)));
+  file= -1;
 
   record=(char*) outparam->record[0]-1;		/* Fieldstart = 1 */
   if (null_field_first)
@@ -426,7 +407,7 @@ int openfrm(const char *name, const char *alias, uint db_stat, uint prgflag,
 						(uint) key_part->length);
 #ifdef EXTRA_DEBUG
 	if (key_part->fieldnr > outparam->fields)
-	  goto err; // sanity check
+	  goto err_not_open; // sanity check
 #endif
 	if (key_part->fieldnr)
 	{					// Should always be true !
@@ -494,6 +475,12 @@ int openfrm(const char *name, const char *alias, uint db_stat, uint prgflag,
 	(outparam->keys_in_use & ((key_map) 1 << primary_key)))
     {
       outparam->primary_key=primary_key;
+      if (outparam->file->option_flag() & HA_PRIMARY_KEY_IN_READ_INDEX)
+	outparam->ref_primary_key= (key_map) 1 << primary_key;
+      /*
+	If we are using an integer as the primary key then allow the user to
+	refer to it as '_rowid'
+      */
       if (outparam->key_info[primary_key].key_parts == 1)
       {
 	Field *field= outparam->key_info[primary_key].key_part[0].field;
@@ -505,6 +492,7 @@ int openfrm(const char *name, const char *alias, uint db_stat, uint prgflag,
       outparam->primary_key = MAX_KEY; // we do not have a primary key
   }
   x_free((gptr) disk_buff);
+  disk_buff=0;
   if (new_field_pack_flag <= 1)
   {			/* Old file format with default null */
     uint null_length=(outparam->null_fields+7)/8;
@@ -523,7 +511,7 @@ int openfrm(const char *name, const char *alias, uint db_stat, uint prgflag,
 	  (Field_blob**) alloc_root(&outparam->mem_root,
 				    (uint) (outparam->blob_fields+1)*
 				    sizeof(Field_blob*))))
-      goto err;
+      goto err_not_open;
     for (ptr=outparam->field ; *ptr ; ptr++)
     {
       if ((*ptr)->flags & BLOB_FLAG)
@@ -535,21 +523,42 @@ int openfrm(const char *name, const char *alias, uint db_stat, uint prgflag,
     outparam->blob_field=
       (Field_blob**) (outparam->field+outparam->fields); // Point at null ptr
 
+  /* The table struct is now initialzed;  Open the table */
+  error=2;
+  if (db_stat)
+  {
+    int err;
+    if ((err=(outparam->file->
+	      ha_open(index_file,
+		      (db_stat & HA_READ_ONLY ? O_RDONLY : O_RDWR),
+		      (db_stat & HA_OPEN_TEMPORARY ? HA_OPEN_TMP_TABLE :
+		       ((db_stat & HA_WAIT_IF_LOCKED) ||
+			(specialflag & SPECIAL_WAIT_IF_LOCKED)) ?
+		       HA_OPEN_WAIT_IF_LOCKED :
+		       (db_stat & (HA_ABORT_IF_LOCKED | HA_GET_INFO)) ?
+		       HA_OPEN_ABORT_IF_LOCKED :
+		       HA_OPEN_IGNORE_IF_LOCKED) | ha_open_flags))))
+    {
+      /* Set a flag if the table is crashed and it can be auto. repaired */
+      outparam->crashed=(err == HA_ERR_CRASHED &&
+			 outparam->file->auto_repair() &&
+			 !(ha_open_flags & HA_OPEN_FOR_REPAIR));
+      goto err_not_open; /* purecov: inspected */
+    }
+  }
+  outparam->db_low_byte_first=outparam->file->low_byte_first();
+
   my_pthread_setspecific_ptr(THR_MALLOC,old_root);
   opened_tables++;
 #ifndef DBUG_OFF
   if (use_hash)
     (void) hash_check(&outparam->name_hash);
 #endif
-  if (db_stat)
-    outparam->file->initialize();
   DBUG_RETURN (0);
 
- err:
-  if (outparam->file && db_stat)
-    (void) outparam->file->close();
  err_not_open:
   x_free((gptr) disk_buff);
+  if (file > 0)
     VOID(my_close(file,MYF(MY_WME)));
 
  err_end:					/* Here when no file */

sql/table.h

@@ -113,7 +113,7 @@ struct st_table {
   byte		*record_pointers;		/* If sorted in memory */
   ha_rows	found_records;			/* How many records in sort */
   ORDER		*group;
-  key_map	quick_keys,used_keys;
+  key_map	quick_keys, used_keys, ref_primary_key;
   ha_rows	quick_rows[MAX_KEY];
   uint		quick_key_parts[MAX_KEY];
   key_part_map  const_key_parts[MAX_KEY];