branches/zip: row_ext: Fetch the BLOB prefixes already at row_ext_create().

Only add indexed BLOBs to row_ext.

trx_undo_rec_get_partial_row(): Move the BLOB fetching to row_ext_create().

row_build(): Pass only those BLOBs to row_ext_create() that are referenced by
ordering columns of some indexes, similar to trx_undo_rec_get_partial_row().

row_ext_create(): Add the parameter "tuple".  Move the implementation
from row0ext.ic to row0ext.c.

row_ext_lookup_ith(), row_ext_lookup(): Return a const pointer.  Remove
the parameters "field" and "f_len".  Make the row_ext_t* parameter const.

row_ext_t: Remove the field zip_size.

field_ref_zero[]: Declare in btr0types.h instead of btr0cur.h.

row_ext_lookup_low(): Rename to row_ext_cache_fill() and change the
signature.

include/btr0cur.h

@@ -716,11 +716,6 @@ extern ulint	btr_cur_n_sea;
 extern ulint	btr_cur_n_non_sea_old;
 extern ulint	btr_cur_n_sea_old;
 
-/* A BLOB field reference full of zero, for use in assertions and tests.
-Initially, BLOB field references are set to zero, in
-dtuple_convert_big_rec(). */
-extern const byte field_ref_zero[BTR_EXTERN_FIELD_REF_SIZE];
-
 #ifndef UNIV_NONINL
 #include "btr0cur.ic"
 #endif

include/btr0types.h

@@ -23,4 +23,9 @@ The reference is stored at the end of the prefix of the field
 in the index record. */
 #define BTR_EXTERN_FIELD_REF_SIZE	20
 
+/* A BLOB field reference full of zero, for use in assertions and tests.
+Initially, BLOB field references are set to zero, in
+dtuple_convert_big_rec(). */
+extern const byte field_ref_zero[BTR_EXTERN_FIELD_REF_SIZE];
+
 #endif

include/row0ext.h

@@ -11,11 +11,12 @@ Created September 2006 Marko Makela
 
 #include "univ.i"
 #include "row0types.h"
+#include "data0types.h"
 #include "mem0mem.h"
 
 /************************************************************************
 Creates a cache of column prefixes of externally stored columns. */
-UNIV_INLINE
+
 row_ext_t*
 row_ext_create(
 /*===========*/
@@ -25,50 +26,48 @@ row_ext_create(
 				in the InnoDB table object, as reported by
 				dict_col_get_no(); NOT relative to the records
 				in the clustered index */
-	ulint		zip_size,/* compressed page size, or 0 */
+	const dtuple_t*	tuple,	/* in: data tuple containing the field
+				references of the externally stored columns;
+				must be indexed by col_no */
+	ulint		zip_size,/* compressed page size in bytes, or 0 */
 	mem_heap_t*	heap);	/* in: heap where created */
 
 /************************************************************************
 Looks up a column prefix of an externally stored column. */
 UNIV_INLINE
-byte*
+const byte*
 row_ext_lookup_ith(
 /*===============*/
-				/* out: column prefix, or NULL if
-				the column is not stored externally,
-				or pointer to field_ref_zero
-				if the BLOB pointer is unset */
-	row_ext_t*	ext,	/* in/out: column prefix cache */
-	ulint		i,	/* in: ith element of ext */
-	const byte*	field,	/* in: locally stored part of the column */
-	ulint		f_len,	/* in: length of field, in bytes */
-	ulint*		len);	/* out: length of prefix, in bytes,
-				at most REC_MAX_INDEX_COL_LEN */
+					/* out: column prefix, or NULL if
+					the column is not stored externally,
+					or pointer to field_ref_zero
+					if the BLOB pointer is unset */
+	const row_ext_t*	ext,	/* in/out: column prefix cache */
+	ulint			i,	/* in: index of ext->ext[] */
+	ulint*			len);	/* out: length of prefix, in bytes,
+					at most REC_MAX_INDEX_COL_LEN */
 /************************************************************************
 Looks up a column prefix of an externally stored column. */
 UNIV_INLINE
-byte*
+const byte*
 row_ext_lookup(
 /*===========*/
-				/* out: column prefix, or NULL if
-				the column is not stored externally,
-				or pointer to field_ref_zero
-				if the BLOB pointer is unset */
-	row_ext_t*	ext,	/* in/out: column prefix cache */
-	ulint		col,	/* in: column number in the InnoDB
-				table object, as reported by
-				dict_col_get_no(); NOT relative to the
-				records in the clustered index */
-	const byte*	field,	/* in: locally stored part of the column */
-	ulint		f_len,	/* in: length of field, in bytes */
-	ulint*		len);	/* out: length of prefix, in bytes,
-				at most REC_MAX_INDEX_COL_LEN */
+					/* out: column prefix, or NULL if
+					the column is not stored externally,
+					or pointer to field_ref_zero
+					if the BLOB pointer is unset */
+	const row_ext_t*	ext,	/* in: column prefix cache */
+	ulint			col,	/* in: column number in the InnoDB
+					table object, as reported by
+					dict_col_get_no(); NOT relative to the
+					records in the clustered index */
+	ulint*			len);	/* out: length of prefix, in bytes,
+					at most REC_MAX_INDEX_COL_LEN */
 
 /* Prefixes of externally stored columns */
 struct row_ext_struct{
 	ulint		n_ext;	/* number of externally stored columns */
 	const ulint*	ext;	/* col_no's of externally stored columns */
-	ulint		zip_size;/* compressed page size, or 0 */
 	byte*		buf;	/* backing store of the column prefix cache */
 	ulint		len[1];	/* prefix lengths; 0 if not cached */
 };

include/row0ext.ic

@@ -7,118 +7,63 @@ Created September 2006 Marko Makela
 *******************************************************/
 
 #include "rem0types.h"
-
-/************************************************************************
-Looks up and caches a column prefix of an externally stored column. */
-
-byte*
-row_ext_lookup_low(
-/*===============*/
-				/* out: column prefix, or
-				pointer to field_ref_zero
-				if the BLOB pointer is unset */
-	row_ext_t*	ext,	/* in/out: column prefix cache */
-	ulint		i,	/* in: index of ext->ext[] */
-	const byte*	field,	/* in: locally stored part of the column */
-	ulint		f_len,	/* in: length of field, in bytes */
-	ulint*		len);	/* out: length of prefix, in bytes,
-				at most REC_MAX_INDEX_COL_LEN */
-
-/************************************************************************
-Creates a cache of column prefixes of externally stored columns. */
-UNIV_INLINE
-row_ext_t*
-row_ext_create(
-/*===========*/
-				/* out,own: column prefix cache */
-	ulint		n_ext,	/* in: number of externally stored columns */
-	const ulint*	ext,	/* in: col_no's of externally stored columns
-				in the InnoDB table object, as reported by
-				dict_col_get_no(); NOT relative to the records
-				in the clustered index */
-	ulint		zip_size,/* compressed page size, or 0 */
-	mem_heap_t*	heap)	/* in: heap where created */
-{
-	row_ext_t*	ret = mem_heap_alloc(heap, (sizeof *ret)
-					     + (n_ext - 1) * sizeof ret->len);
-
-	ut_ad(ut_is_2pow(zip_size));
-
-	ret->n_ext = n_ext;
-	ret->ext = ext;
-	ret->zip_size = zip_size;
-	ret->buf = mem_heap_alloc(heap, n_ext * REC_MAX_INDEX_COL_LEN);
-#ifdef UNIV_DEBUG
-	memset(ret->buf, 0xaa, n_ext * REC_MAX_INDEX_COL_LEN);
-	UNIV_MEM_ALLOC(ret->buf, n_ext * REC_MAX_INDEX_COL_LEN);
-#endif
-	memset(ret->len, 0, n_ext * sizeof *ret->len);
-
-	return(ret);
-}
+#include "btr0types.h"
 
 /************************************************************************
 Looks up a column prefix of an externally stored column. */
 UNIV_INLINE
-byte*
+const byte*
 row_ext_lookup_ith(
 /*===============*/
-				/* out: column prefix, or NULL if
-				the column is not stored externally,
-				or pointer to field_ref_zero
-				if the BLOB pointer is unset */
-	row_ext_t*	ext,	/* in/out: column prefix cache */
-	ulint		i,	/* in: index of ext->ext[] */
-	const byte*	field,	/* in: locally stored part of the column */
-	ulint		f_len,	/* in: length of field, in bytes */
-	ulint*		len)	/* out: length of prefix, in bytes,
-				at most REC_MAX_INDEX_COL_LEN */
+					/* out: column prefix, or NULL if
+					the column is not stored externally,
+					or pointer to field_ref_zero
+					if the BLOB pointer is unset */
+	const row_ext_t*	ext,	/* in/out: column prefix cache */
+	ulint			i,	/* in: index of ext->ext[] */
+	ulint*			len)	/* out: length of prefix, in bytes,
+					at most REC_MAX_INDEX_COL_LEN */
 {
 	ut_ad(ext);
-	ut_ad(field);
 	ut_ad(len);
 	ut_ad(i < ext->n_ext);
 
-	/* Return from the cache if found */
-	if (ext->len[i]) {
-		*len = ext->len[i];
-		ut_ad(*len > f_len);
+	*len = ext->len[i];
+
+	if (UNIV_UNLIKELY(*len == 0)) {
+		/* The BLOB could not be fetched to the cache. */
+		return(field_ref_zero);
+	} else {
 		return(ext->buf + i * REC_MAX_INDEX_COL_LEN);
 	}
-
-	/* Update the cache */
-	return(row_ext_lookup_low(ext, i, field, f_len, len));
 }
 
 /************************************************************************
 Looks up a column prefix of an externally stored column. */
 UNIV_INLINE
-byte*
+const byte*
 row_ext_lookup(
 /*===========*/
-				/* out: column prefix, or NULL if
-				the column is not stored externally,
-				or pointer to field_ref_zero
-				if the BLOB pointer is unset */
-	row_ext_t*	ext,	/* in/out: column prefix cache */
-	ulint		col,	/* in: column number in the InnoDB
-				table object, as reported by
-				dict_col_get_no(); NOT relative to the
-				records in the clustered index */
-	const byte*	field,	/* in: locally stored part of the column */
-	ulint		f_len,	/* in: length of field, in bytes */
-	ulint*		len)	/* out: length of prefix, in bytes,
-				at most REC_MAX_INDEX_COL_LEN */
+					/* out: column prefix, or NULL if
+					the column is not stored externally,
+					or pointer to field_ref_zero
+					if the BLOB pointer is unset */
+	const row_ext_t*	ext,	/* in: column prefix cache */
+	ulint			col,	/* in: column number in the InnoDB
+					table object, as reported by
+					dict_col_get_no(); NOT relative to the
+					records in the clustered index */
+	ulint*			len)	/* out: length of prefix, in bytes,
+					at most REC_MAX_INDEX_COL_LEN */
 {
 	ulint	i;
 
 	ut_ad(ext);
-	ut_ad(field);
 	ut_ad(len);
 
 	for (i = 0; i < ext->n_ext; i++) {
 		if (col == ext->ext[i]) {
-			return(row_ext_lookup_ith(ext, i, field, f_len, len));
+			return(row_ext_lookup_ith(ext, i, len));
 		}
 	}
 

row/row0ext.c

@@ -15,36 +15,78 @@ Created September 2006 Marko Makela
 #include "btr0cur.h"
 
 /************************************************************************
-Looks up and caches a column prefix of an externally stored column. */
-
-byte*
-row_ext_lookup_low(
+Fills the column prefix cache of an externally stored column. */
+static
+void
+row_ext_cache_fill(
 /*===============*/
-				/* out: column prefix, or
-				pointer to field_ref_zero
-				if the BLOB pointer is unset */
 	row_ext_t*	ext,	/* in/out: column prefix cache */
 	ulint		i,	/* in: index of ext->ext[] */
-	const byte*	field,	/* in: locally stored part of the column */
-	ulint		f_len,	/* in: length of field, in bytes */
-	ulint*		len)	/* out: length of prefix, in bytes,
-				at most REC_MAX_INDEX_COL_LEN */
+	ulint		zip_size,/* compressed page size in bytes, or 0 */
+	const dfield_t*	dfield)	/* in: data field */
 {
-	byte*	buf	= ext->buf + i * REC_MAX_INDEX_COL_LEN;
+	const byte*	field	= dfield_get_data(dfield);
+	ulint		f_len	= dfield_get_len(dfield);
+	byte*		buf	= ext->buf + i * REC_MAX_INDEX_COL_LEN;
 
 	ut_ad(i < ext->n_ext);
+	ut_ad(dfield_is_ext(dfield));
 	ut_a(f_len >= BTR_EXTERN_FIELD_REF_SIZE);
 
 	if (UNIV_UNLIKELY(!memcmp(field_ref_zero,
 				  field + f_len - BTR_EXTERN_FIELD_REF_SIZE,
 				  BTR_EXTERN_FIELD_REF_SIZE))) {
 		/* The BLOB pointer is not set: we cannot fetch it */
-		*len = 0;
-		return((byte*) field_ref_zero);
+		ext->len[i] = 0;
 	}
 
-	*len = ext->len[i] = btr_copy_externally_stored_field_prefix(
+	ext->len[i] = btr_copy_externally_stored_field_prefix(
 		buf,
-		REC_MAX_INDEX_COL_LEN, ext->zip_size, field, f_len);
-	return(buf);
+		REC_MAX_INDEX_COL_LEN, zip_size, field, f_len);
+}
+
+/************************************************************************
+Creates a cache of column prefixes of externally stored columns. */
+UNIV_INLINE
+row_ext_t*
+row_ext_create(
+/*===========*/
+				/* out,own: column prefix cache */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	const ulint*	ext,	/* in: col_no's of externally stored columns
+				in the InnoDB table object, as reported by
+				dict_col_get_no(); NOT relative to the records
+				in the clustered index */
+	const dtuple_t*	tuple,	/* in: data tuple containing the field
+				references of the externally stored columns;
+				must be indexed by col_no */
+	ulint		zip_size,/* compressed page size in bytes, or 0 */
+	mem_heap_t*	heap)	/* in: heap where created */
+{
+	ulint		i;
+	row_ext_t*	ret = mem_heap_alloc(heap, (sizeof *ret)
+					     + (n_ext - 1) * sizeof ret->len);
+
+	ut_ad(ut_is_2pow(zip_size));
+	ut_ad(zip_size <= UNIV_PAGE_SIZE);
+
+	ret->n_ext = n_ext;
+	ret->ext = ext;
+	ret->buf = mem_heap_alloc(heap, n_ext * REC_MAX_INDEX_COL_LEN);
+#ifdef UNIV_DEBUG
+	memset(ret->buf, 0xaa, n_ext * REC_MAX_INDEX_COL_LEN);
+	UNIV_MEM_ALLOC(ret->buf, n_ext * REC_MAX_INDEX_COL_LEN);
+#endif
+
+	/* Fetch the BLOB prefixes */
+	for (i = 0; i < n_ext; i++) {
+		const dfield_t*	dfield;
+		ulint		len;
+		byte*		b;
+
+		dfield = dtuple_get_nth_field(tuple, ext[i]);
+		row_ext_cache_fill(ret, i, zip_size, dfield);
+	}
+
+	return(ret);
 }

row/row0merge.c

@@ -261,7 +261,6 @@ row_merge_buf_add(
 		ulint			col_no;
 		const dfield_t*		row_field;
 		ulint			len;
-		const void*		field_data;
 
 		ifield = dict_index_get_nth_field(index, i);
 		col = ifield->col;
@@ -269,17 +268,15 @@ row_merge_buf_add(
 		row_field = dtuple_get_nth_field(row, col_no);
 		dfield_copy(field, row_field);
 		len = dfield_get_len(field);
-		field_data = dfield_get_data(field);
 
 		if (dfield_is_null(field)) {
 			ut_ad(!(col->prtype & DATA_NOT_NULL));
-			ut_ad(field_data == NULL);
 			continue;
 		} else if (UNIV_LIKELY(!ext)) {
 		} else if (dict_index_is_clust(index)) {
 			/* Flag externally stored fields. */
-			byte*	buf = row_ext_lookup(ext, col_no,
-						     field_data, len, &len);
+			const byte*	buf = row_ext_lookup(ext, col_no,
+							     &len);
 			if (UNIV_LIKELY_NULL(buf)) {
 				ut_a(buf != field_ref_zero);
 				if (i < dict_index_get_n_unique(index)) {
@@ -290,8 +287,8 @@ row_merge_buf_add(
 				}
 			}
 		} else {
-			byte*	buf = row_ext_lookup(ext, col_no,
-						     field_data, len, &len);
+			const byte*	buf = row_ext_lookup(ext, col_no,
+							     &len);
 			if (UNIV_LIKELY_NULL(buf)) {
 				ut_a(buf != field_ref_zero);
 				dfield_set_data(field, buf, len);

row/row0row.c

@@ -121,9 +121,8 @@ row_build_index_entry(
 
 		if (UNIV_LIKELY_NULL(ext) && !dfield_is_null(dfield)) {
 			/* See if the column is stored externally. */
-			byte*	buf = row_ext_lookup(ext, col_no,
-						     dfield_get_data(dfield),
-						     len, &len);
+			const byte*	buf = row_ext_lookup(ext, col_no,
+							     &len);
 			if (UNIV_LIKELY_NULL(buf)) {
 				if (UNIV_UNLIKELY(buf == field_ref_zero)) {
 					return(NULL);
@@ -229,8 +228,10 @@ row_build(
 			= dict_index_get_nth_field(index, i);
 		const dict_col_t*	col
 			= dict_field_get_col(ind_field);
+		ulint			col_no
+			= dict_col_get_no(col);
 		dfield_t*		dfield
-			= dtuple_get_nth_field(row, dict_col_get_no(col));
+			= dtuple_get_nth_field(row, col_no);
 
 		if (ind_field->prefix_len == 0) {
 
@@ -241,16 +242,20 @@ row_build(
 		}
 
 		if (rec_offs_nth_extern(offsets, i)) {
-			ext_cols[j++] = dict_col_get_no(col);
 			dfield_set_ext(dfield);
+			if (col->ord_part) {
+				/* We will have to fetch prefixes of
+				externally stored columns that are
+				referenced by column prefixes. */
+				ext_cols[j++] = col_no;
+			}
 		}
 	}
 
-	ut_ad(j == n_ext_cols);
 	ut_ad(dtuple_check_typed(row));
 
-	if (n_ext_cols) {
-		*ext = row_ext_create(n_ext_cols, ext_cols,
+	if (j) {
+		*ext = row_ext_create(j, ext_cols, row,
 				      dict_table_zip_size(index->table),
 				      heap);
 	} else {

trx/trx0rec.c

@@ -945,30 +945,9 @@ trx_undo_rec_get_partial_row(
 	}
 
 	if (n_ext_cols) {
-		ulint	i;
-
-		*ext = row_ext_create(n_ext_cols, ext_cols,
+		*ext = row_ext_create(n_ext_cols, ext_cols, *row,
 				      dict_table_zip_size(index->table),
 				      heap);
-
-		/* Fetch the BLOB prefixes, because the clustered
-		index record (and the BLOBs) may have been deleted by
-		the time row_ext_lookup() is called later in the purge
-		thread. */
-		for (i = 0; i < n_ext_cols; i++) {
-			const dfield_t*	dfield;
-			ulint		len;
-			byte*		b;
-
-			dfield = dtuple_get_nth_field(*row, ext_cols[i]);
-
-			b = row_ext_lookup_ith(*ext, i,
-					       dfield_get_data(dfield),
-					       dfield_get_len(dfield),
-					       &len);
-			ut_a(b);
-			ut_a(b != field_ref_zero);
-		}
 	} else {
 		*ext = NULL;
 	}