MDEV-12255 innodb_prefix_index_cluster_optimization hits debug build

		assert on UTF-8 columns

Problem:
=======
(1) Multi-byte character cases are not considered during prefix index
cluster optimization check. It leads to fetch of improper results during
read operation.
(2) Strict assert in row_sel_field_store_in_mysql_format_func and it asserts
for prefix index record to mysql conversion.

Solution:
========
(1) Consider the case of multi-byte character during prefix index
cluster optimization check.
(2) Relax the assert in row_sel_field_store_in_mysql_format_func to allow
prefix index record to mysql format conversion.

The patch is taken from
https://github.com/laurynas-biveinis/percona-server/commit/1eee538087ffcf121c37f844b447ba5480faf081

storage/innobase/row/row0sel.cc

@@ -2707,7 +2707,9 @@ row_sel_field_store_in_mysql_format_func(
 		      || !(templ->mysql_col_len % templ->mbmaxlen));
 		ut_ad(len * templ->mbmaxlen >= templ->mysql_col_len
 		      || (field_no == templ->icp_rec_field_no
-			  && field->prefix_len > 0));
+			  && field->prefix_len > 0)
+		      || templ->rec_field_is_prefix);
+
 		ut_ad(!(field->prefix_len % templ->mbmaxlen));
 
 		if (templ->mbminlen == 1 && templ->mbmaxlen != 1) {
@@ -3667,6 +3669,118 @@ row_search_idx_cond_check(
 	return(result);
 }
 
+/** Return the record field length in characters.
+@param[in]	col		table column of the field
+@param[in]	field_no	field number
+@param[in]	rec		physical record
+@param[in]	offsets		field offsets in the physical record
+@return field length in characters. */
+static
+size_t
+rec_field_len_in_chars(
+	const dict_col_t*	col,
+	const ulint		field_no,
+	const rec_t*		rec,
+	const ulint*		offsets)
+{
+	const ulint cset = dtype_get_charset_coll(col->prtype);
+	const CHARSET_INFO* cs = all_charsets[cset];
+	ulint rec_field_len;
+	const char* rec_field = reinterpret_cast<const char *>(
+		rec_get_nth_field(
+			rec, offsets, field_no, &rec_field_len));
+
+	if (UNIV_UNLIKELY(!cs)) {
+		ib_logf(IB_LOG_LEVEL_WARN, "Missing collation " ULINTPF, cset);
+		return SIZE_T_MAX;
+	}
+
+	return(cs->cset->numchars(cs, rec_field, rec_field + rec_field_len));
+}
+
+/** Avoid the clustered index lookup if all the following conditions
+are true:
+1) all columns are in secondary index
+2) all values for columns that are prefix-only indexes are shorter
+than the prefix size. This optimization can avoid many IOs for certain schemas.
+@return true, to avoid clustered index lookup. */
+static
+bool row_search_with_covering_prefix(
+	row_prebuilt_t*	prebuilt,
+	const rec_t*	rec,
+	const ulint*	offsets)
+{
+	const dict_index_t*	index = prebuilt->index;
+	ut_ad(!dict_index_is_clust(index));
+
+	if (!srv_prefix_index_cluster_optimization) {
+		return false;
+	}
+
+	/** Optimization only applicable if there the number of secondary index
+	fields are greater than or equal to number of clustered index fields. */
+	if (prebuilt->n_template > index->n_fields) {
+		return false;
+	}
+
+	for (ulint i = 0; i < prebuilt->n_template; i++) {
+		mysql_row_templ_t* templ = prebuilt->mysql_template + i;
+		ulint j = templ->rec_prefix_field_no;
+
+		/** Condition (1) : is the field in the index. */
+		if (j == ULINT_UNDEFINED) {
+			return false;
+		}
+
+		/** Condition (2): If this is a prefix index then
+		row's value size shorter than prefix length. */
+
+		if (!templ->rec_field_is_prefix) {
+			continue;
+		}
+
+		ulint rec_size = rec_offs_nth_size(offsets, j);
+		const dict_field_t* field = dict_index_get_nth_field(index, j);
+		ulint max_chars = field->prefix_len / templ->mbmaxlen;
+
+		ut_a(field->prefix_len > 0);
+
+		if (rec_size < max_chars) {
+			/* Record in bytes shorter than the index
+			prefix length in char. */
+			continue;
+		}
+
+		if (rec_size * templ->mbminlen >= field->prefix_len) {
+			/* Shortest representation string by the
+			byte length of the record is longer than the
+			maximum possible index prefix. */
+			return false;
+		}
+
+		size_t num_chars = rec_field_len_in_chars(
+			field->col, j, rec, offsets);
+
+		if (num_chars >= max_chars) {
+			/* No of chars to store the record exceeds
+			the index prefix character length. */
+			return false;
+		}
+	}
+
+	/* If prefix index optimization condition satisfied then
+	for all columns above, use rec_prefix_field_no instead of
+	rec_field_no, and skip the clustered lookup below. */
+	for (ulint i = 0; i < prebuilt->n_template; i++) {
+		mysql_row_templ_t* templ = prebuilt->mysql_template + i;
+		templ->rec_field_no = templ->rec_prefix_field_no;
+		ut_a(templ->rec_field_no != ULINT_UNDEFINED);
+	}
+
+	srv_stats.n_sec_rec_cluster_reads_avoided.inc();
+	return true;
+}
+
 /********************************************************************//**
 Searches for rows in the database. This is used in the interface to
 MySQL. This function opens a cursor, and also implements fetch next
@@ -3729,7 +3843,6 @@ row_search_for_mysql(
 	ulint*		offsets				= offsets_;
 	ibool		table_lock_waited		= FALSE;
 	byte*		next_buf			= 0;
-	ibool		use_clustered_index		= FALSE;
 
 	rec_offs_init(offsets_);
 
@@ -4790,69 +4903,10 @@ row_search_for_mysql(
 		break;
 	}
 
-	/* Get the clustered index record if needed, if we did not do the
-	search using the clustered index... */
-
-	use_clustered_index =
-		(index != clust_index && prebuilt->need_to_access_clustered);
-
-	if (use_clustered_index && srv_prefix_index_cluster_optimization
-	    && prebuilt->n_template <= index->n_fields) {
-		/* ...but, perhaps avoid the clustered index lookup if
-		all of the following are true:
-		1) all columns are in the secondary index
-		2) all values for columns that are prefix-only
-		   indexes are shorter than the prefix size
-		This optimization can avoid many IOs for certain schemas.
-		*/
-		ibool row_contains_all_values = TRUE;
-		int i;
-		for (i = 0; i < prebuilt->n_template; i++) {
-			/* Condition (1) from above: is the field in the
-			index (prefix or not)? */
-			mysql_row_templ_t* templ =
-				prebuilt->mysql_template + i;
-			ulint secondary_index_field_no =
-				templ->rec_prefix_field_no;
-			if (secondary_index_field_no == ULINT_UNDEFINED) {
-				row_contains_all_values = FALSE;
-				break;
-			}
-			/* Condition (2) from above: if this is a
-			prefix, is this row's value size shorter
-			than the prefix? */
-			if (templ->rec_field_is_prefix) {
-				ulint record_size = rec_offs_nth_size(
-					offsets,
-					secondary_index_field_no);
-				const dict_field_t *field =
-					dict_index_get_nth_field(
-						index,
-						secondary_index_field_no);
-				ut_a(field->prefix_len > 0);
-				if (record_size >= field->prefix_len) {
-					row_contains_all_values = FALSE;
-					break;
-				}
-			}
-		}
-		/* If (1) and (2) were true for all columns above, use
-		rec_prefix_field_no instead of rec_field_no, and skip
-		the clustered lookup below. */
-		if (row_contains_all_values) {
-			for (i = 0; i < prebuilt->n_template; i++) {
-				mysql_row_templ_t* templ =
-					prebuilt->mysql_template + i;
-				templ->rec_field_no =
-					templ->rec_prefix_field_no;
-				ut_a(templ->rec_field_no != ULINT_UNDEFINED);
-			}
-			use_clustered_index = FALSE;
-			srv_stats.n_sec_rec_cluster_reads_avoided.inc();
+	if (index != clust_index && prebuilt->need_to_access_clustered) {
+		if (row_search_with_covering_prefix(prebuilt, rec, offsets)) {
+			goto use_covering_index;
 		}
-	}
-
-	if (use_clustered_index) {
 requires_clust_rec:
 		ut_ad(index != clust_index);
 		/* We use a 'goto' to the preceding label if a consistent
@@ -4938,6 +4992,7 @@ row_search_for_mysql(
 			}
 		}
 	} else {
+use_covering_index:
 		result_rec = rec;
 	}
 

storage/xtradb/row/row0sel.cc

@@ -3685,6 +3685,117 @@ row_search_idx_cond_check(
 	return(result);
 }
 
+/** Return the record field length in characters.
+@param[in]	col		table column of the field
+@param[in]	field_no	field number
+@param[in]	rec		physical record
+@param[in]	offsets		field offsets in the physical record
+@return field length in characters. */
+static
+size_t
+rec_field_len_in_chars(
+	const dict_col_t*	col,
+	const ulint		field_no,
+	const rec_t*		rec,
+	const ulint*		offsets)
+{
+	const ulint cset = dtype_get_charset_coll(col->prtype);
+	const CHARSET_INFO* cs = all_charsets[cset];
+	ulint rec_field_len;
+	const char* rec_field = reinterpret_cast<const char *>(
+		rec_get_nth_field(
+			rec, offsets, field_no, &rec_field_len));
+
+	if (UNIV_UNLIKELY(!cs)) {
+		ib_logf(IB_LOG_LEVEL_WARN, "Missing collation " ULINTPF, cset);
+		return SIZE_T_MAX;
+	}
+
+	return(cs->cset->numchars(cs, rec_field, rec_field + rec_field_len));
+}
+
+
+/** Avoid the clustered index lookup if all the following conditions
+are true:
+1) all columns are in secondary index
+2) all values for columns that are prefix-only indexes are shorter
+than the prefix size. This optimization can avoid many IOs for certain schemas.
+@return true, to avoid clustered index lookup. */
+static
+bool row_search_with_covering_prefix(
+	row_prebuilt_t*	prebuilt,
+	const rec_t*	rec,
+	const ulint*	offsets)
+{
+	const dict_index_t*	index = prebuilt->index;
+	ut_ad(!dict_index_is_clust(index));
+
+	if (!srv_prefix_index_cluster_optimization) {
+		return false;
+	}
+
+	/** Optimization only applicable if the number of secondary index
+	fields are greater than or equal to number of clustered index fields. */
+	if (prebuilt->n_template > index->n_fields) {
+		return false;
+	}
+
+	for (ulint i = 0; i < prebuilt->n_template; i++) {
+		mysql_row_templ_t* templ = prebuilt->mysql_template + i;
+		ulint j = templ->rec_prefix_field_no;
+
+		/** Condition (1) : is the field in the index. */
+		if (j == ULINT_UNDEFINED) {
+			return false;
+		}
+
+		/** Condition (2): If this is a prefix index then
+		row's value size shorter than prefix length. */
+
+		if (!templ->rec_field_is_prefix) {
+			continue;
+		}
+
+		ulint rec_size = rec_offs_nth_size(offsets, j);
+		const dict_field_t* field = dict_index_get_nth_field(index, j);
+		ulint max_chars = field->prefix_len / templ->mbmaxlen;
+
+		ut_a(field->prefix_len > 0);
+
+		if (rec_size < max_chars) {
+			/* Record in bytes shorter than the index
+			prefix length in char. */
+			continue;
+		}
+
+		if (rec_size * templ->mbminlen >= field->prefix_len) {
+			/* Shortest representation string by the
+			byte length of the record is longer than the
+			maximum possible index prefix. */
+			return false;
+		}
+
+
+		size_t num_chars = rec_field_len_in_chars(
+			field->col, j, rec, offsets);
+
+		if (num_chars >= max_chars) {
+			/* No of chars to store the record exceeds
+			the index prefix character length. */
+			return false;
+		}
+	}
+
+	for (ulint i = 0; i < prebuilt->n_template; i++) {
+		mysql_row_templ_t* templ = prebuilt->mysql_template + i;
+		templ->rec_field_no = templ->rec_prefix_field_no;
+		ut_a(templ->rec_field_no != ULINT_UNDEFINED);
+	}
+
+	srv_stats.n_sec_rec_cluster_reads_avoided.inc();
+	return true;
+}
+
 /********************************************************************//**
 Searches for rows in the database. This is used in the interface to
 MySQL. This function opens a cursor, and also implements fetch next
@@ -3748,7 +3859,6 @@ row_search_for_mysql(
 	ulint*		offsets				= offsets_;
 	ibool		table_lock_waited		= FALSE;
 	byte*		next_buf			= 0;
-	bool		use_clustered_index		= false;
 
 	rec_offs_init(offsets_);
 
@@ -4810,71 +4920,10 @@ row_search_for_mysql(
 		break;
 	}
 
-	/* Get the clustered index record if needed, if we did not do the
-	search using the clustered index... */
-
-	use_clustered_index =
-		(index != clust_index && prebuilt->need_to_access_clustered);
-
-	if (use_clustered_index && srv_prefix_index_cluster_optimization
-	    && prebuilt->n_template <= index->n_fields) {
-		/* ...but, perhaps avoid the clustered index lookup if
-		all of the following are true:
-		1) all columns are in the secondary index
-		2) all values for columns that are prefix-only
-		   indexes are shorter than the prefix size
-		This optimization can avoid many IOs for certain schemas.
-		*/
-		bool row_contains_all_values = true;
-		unsigned int i;
-		for (i = 0; i < prebuilt->n_template; i++) {
-			/* Condition (1) from above: is the field in the
-			index (prefix or not)? */
-			const mysql_row_templ_t* templ =
-				prebuilt->mysql_template + i;
-			ulint secondary_index_field_no =
-				templ->rec_prefix_field_no;
-			if (secondary_index_field_no == ULINT_UNDEFINED) {
-				row_contains_all_values = false;
-				break;
-			}
-			/* Condition (2) from above: if this is a
-			prefix, is this row's value size shorter
-			than the prefix? */
-			if (templ->rec_field_is_prefix) {
-				ulint record_size = rec_offs_nth_size(
-					offsets,
-					secondary_index_field_no);
-				const dict_field_t *field =
-					dict_index_get_nth_field(
-						index,
-						secondary_index_field_no);
-				ut_a(field->prefix_len > 0);
-				if (record_size >= field->prefix_len
-				    / templ->mbmaxlen) {
-					row_contains_all_values = false;
-					break;
-				}
-			}
-		}
-		/* If (1) and (2) were true for all columns above, use
-		rec_prefix_field_no instead of rec_field_no, and skip
-		the clustered lookup below. */
-		if (row_contains_all_values) {
-			for (i = 0; i < prebuilt->n_template; i++) {
-				mysql_row_templ_t* templ =
-					prebuilt->mysql_template + i;
-				templ->rec_field_no =
-					templ->rec_prefix_field_no;
-				ut_a(templ->rec_field_no != ULINT_UNDEFINED);
-			}
-			use_clustered_index = false;
-			srv_stats.n_sec_rec_cluster_reads_avoided.inc();
+	if (index != clust_index && prebuilt->need_to_access_clustered) {
+		if (row_search_with_covering_prefix(prebuilt, rec, offsets)) {
+			goto use_covering_index;
 		}
-	}
-
-	if (use_clustered_index) {
-
 requires_clust_rec:
 		ut_ad(index != clust_index);
 		/* We use a 'goto' to the preceding label if a consistent
@@ -4960,6 +5009,7 @@ row_search_for_mysql(
 			}
 		}
 	} else {
+use_covering_index:
 		result_rec = rec;
 	}