btr0cur.c Improve range size estimate for big ranges

ha_innobase.cc	Fix Sort aborted bug

BitKeeper/etc/logging_ok

@@ -1 +1,2 @@
 jani@hynda.mysql.fi
+heikki@donna.mysql.fi

innobase/btr/btr0cur.c

@@ -2351,6 +2351,7 @@ btr_estimate_n_rows_in_range(
 	btr_path_t*	slot1;
 	btr_path_t*	slot2;
 	ibool		diverged;
+	ulint           divergence_level;           
 	ulint		n_rows;
 	ulint		i;
 	mtr_t		mtr;
@@ -2393,6 +2394,7 @@ btr_estimate_n_rows_in_range(
 
 	n_rows = 1;
 	diverged = FALSE;
+	divergence_level = 1000000;
 	
 	for (i = 0; ; i++) {
 		ut_ad(i < BTR_PATH_ARRAY_N_SLOTS);
@@ -2403,6 +2405,13 @@ btr_estimate_n_rows_in_range(
 		if (slot1->nth_rec == ULINT_UNDEFINED
 				|| slot2->nth_rec == ULINT_UNDEFINED) {
 
+		        if (i > divergence_level + 1) {
+		                /* In trees whose height is > 1 our algorithm
+		                tends to underestimate: multiply the estimate
+		                by 2: */
+
+		                n_rows = n_rows * 2;
+		        }
 			return(n_rows);
 		}
 
@@ -2417,6 +2426,8 @@ btr_estimate_n_rows_in_range(
 				return(10);
 			}
 
+			divergence_level = i;
+
 			diverged = TRUE;
 		} else if (diverged) {
 			n_rows = (n_rows * (slot1->n_recs + slot2->n_recs))

sql/ha_innobase.cc

@@ -822,11 +822,11 @@ ha_innobase::open(
 
  	if (NULL == (ib_table = dict_table_get(norm_name, NULL))) {
 
-	  fprintf(stderr, "\
+	  fprintf(stderr,
-Cannot find table %s from the internal data dictionary\n\
+"Cannot find table %s from the internal data dictionary\n"
-of InnoDB though the .frm file for the table exists. Maybe you have deleted\n\
+"of InnoDB though the .frm file for the table exists. Maybe you have deleted\n"
-and created again an InnoDB database but forgotten to delete the\n\
+"and created again an InnoDB database but forgotten to delete the\n"
-corresponding .frm files of old InnoDB tables?\n",
+"corresponding .frm files of old InnoDB tables?\n",
 		  norm_name);
 
 	        free_share(share);
@@ -2659,6 +2659,37 @@ ha_innobase::records_in_range(
 	DBUG_RETURN((ha_rows) n_rows);
 }
 
+/*************************************************************************
+Gives an UPPER BOUND to the number of rows in a table. This is used in
+filesort.cc and the upper bound must hold. TODO: Since the number of
+rows in a table may change after this function is called, we still may
+get a 'Sort aborted' error in filesort.cc of MySQL. The ultimate fix is to
+improve the algorithm of filesort.cc. */
+
+ha_rows
+ha_innobase::estimate_number_of_rows(void)
+/*======================================*/
+			/* out: upper bound of rows, currently 32-bit int
+			or uint */
+{
+	row_prebuilt_t* prebuilt	= (row_prebuilt_t*) innobase_prebuilt;
+	dict_table_t*	ib_table;
+
+ 	DBUG_ENTER("info");
+
+ 	ib_table = prebuilt->table;
+
+ 	dict_update_statistics(ib_table);
+
+	data_file_length = ((ulonglong)
+				ib_table->stat_clustered_index_size)
+    					* UNIV_PAGE_SIZE;
+
+	/* The minimum clustered index record size is 20 bytes */
+
+	return((ha_rows) (1000 + data_file_length / 20));
+}
+
 /*************************************************************************
 How many seeks it will take to read through the table. This is to be
 comparable to the number returned by records_in_range so that we can