Merge Bug #56680 from mysql-5.1.

Additional fixes in 5.5:

ibuf_set_del_mark(): Add diagnostics when setting a buffered delete-mark fails.

ibuf_delete(): Correct a misleading comment about non-found records.

rec_print(): Add a const qualifier to the index parameter.

Bug #56680 wrong InnoDB results from a case-insensitive covering index

row_search_for_mysql(): When a secondary index record might not be
visible in the current transaction's read view and we consult the
clustered index and optionally some undo log records, return the
relevant columns of the clustered index record to MySQL instead of the
secondary index record.

ibuf_insert_to_index_page_low(): New function, refactored from
ibuf_insert_to_index_page().

ibuf_insert_to_index_page(): When we are inserting a record in place
of a delete-marked record and some fields of the record differ, update
that record just like row_ins_sec_index_entry_by_modify() would do.

btr_cur_update_alloc_zip(): Make the function public.

mysql_row_templ_t: Add clust_rec_field_no.

row_sel_store_mysql_rec(), row_sel_push_cache_row_for_mysql(): Add the
flag rec_clust, for returning data at clust_rec_field_no instead of
rec_field_no. Resurrect the debug assertion that the record not be
marked for deletion. (Bug #55626)

[UNIV_DEBUG || UNIV_IBUF_DEBUG] ibuf_debug, buf_page_get_gen(),
buf_flush_page_try():
Implement innodb_change_buffering_debug=1 for evicting pages from the
buffer pool, so that change buffering will be attempted more
frequently.

1 files changed, 171 insertions, 61 deletions

diff --git a/storage/innobase/ibuf/ibuf0ibuf.c b/storage/innobase/ibuf/ibuf0ibuf.c
index d560f936cb9..53a11a7f2e0 100644
--- a/storage/innobase/ibuf/ibuf0ibuf.c
+++ b/storage/innobase/ibuf/ibuf0ibuf.c
@@ -49,6 +49,7 @@ Created 7/19/1997 Heikki Tuuri
 #include "btr0cur.h"
 #include "btr0pcur.h"
 #include "btr0btr.h"
+#include "row0upd.h"
 #include "sync0sync.h"
 #include "dict0boot.h"
 #include "fut0lst.h"
@@ -192,6 +193,11 @@ access order rules. */
 /** Operations that can currently be buffered. */
 UNIV_INTERN ibuf_use_t	ibuf_use		= IBUF_USE_ALL;
 
+#if defined UNIV_DEBUG || defined UNIV_IBUF_DEBUG
+/** Flag to control insert buffer debugging. */
+UNIV_INTERN uint	ibuf_debug;
+#endif /* UNIV_DEBUG || UNIV_IBUF_DEBUG */
+
 /** The insert buffer control structure */
 UNIV_INTERN ibuf_t*	ibuf			= NULL;
 
@@ -2761,9 +2767,8 @@ ibuf_get_volume_buffered_count(
 
 	switch (ibuf_op) {
 	case IBUF_OP_INSERT:
-		/* Inserts can be done by
-		btr_cur_set_deleted_flag_for_ibuf().  Because
-		delete-mark and insert operations can be pointing to
+		/* Inserts can be done by updating a delete-marked record.
+		Because delete-mark and insert operations can be pointing to
 		the same records, we must not count duplicates. */
 	case IBUF_OP_DELETE_MARK:
 		/* There must be a record to delete-mark.
@@ -3748,9 +3753,80 @@ During merge, inserts to an index page a secondary index entry extracted
 from the insert buffer. */
 static
 void
+ibuf_insert_to_index_page_low(
+/*==========================*/
+	const dtuple_t*	entry,	/*!< in: buffered entry to insert */
+	buf_block_t*	block,	/*!< in/out: index page where the buffered
+				entry should be placed */
+	dict_index_t*	index,	/*!< in: record descriptor */
+	mtr_t*		mtr,	/*!< in/out: mtr */
+	page_cur_t*	page_cur)/*!< in/out: cursor positioned on the record
+				after which to insert the buffered entry */
+{
+	const page_t*	page;
+	ulint		space;
+	ulint		page_no;
+	ulint		zip_size;
+	const page_t*	bitmap_page;
+	ulint		old_bits;
+
+	if (UNIV_LIKELY
+	    (page_cur_tuple_insert(page_cur, entry, index, 0, mtr) != NULL)) {
+		return;
+	}
+
+	/* If the record did not fit, reorganize */
+
+	btr_page_reorganize(block, index, mtr);
+	page_cur_search(block, index, entry, PAGE_CUR_LE, page_cur);
+
+	/* This time the record must fit */
+
+	if (UNIV_LIKELY
+	    (page_cur_tuple_insert(page_cur, entry, index, 0, mtr) != NULL)) {
+		return;
+	}
+
+	page = buf_block_get_frame(block);
+
+	ut_print_timestamp(stderr);
+
+	fprintf(stderr,
+		"  InnoDB: Error: Insert buffer insert fails;"
+		" page free %lu, dtuple size %lu\n",
+		(ulong) page_get_max_insert_size(page, 1),
+		(ulong) rec_get_converted_size(index, entry, 0));
+	fputs("InnoDB: Cannot insert index record ", stderr);
+	dtuple_print(stderr, entry);
+	fputs("\nInnoDB: The table where this index record belongs\n"
+	      "InnoDB: is now probably corrupt. Please run CHECK TABLE on\n"
+	      "InnoDB: that table.\n", stderr);
+
+	space = page_get_space_id(page);
+	zip_size = buf_block_get_zip_size(block);
+	page_no = page_get_page_no(page);
+
+	bitmap_page = ibuf_bitmap_get_map_page(space, page_no, zip_size, mtr);
+	old_bits = ibuf_bitmap_page_get_bits(bitmap_page, page_no, zip_size,
+					     IBUF_BITMAP_FREE, mtr);
+
+	fprintf(stderr,
+		"InnoDB: space %lu, page %lu, zip_size %lu, bitmap bits %lu\n",
+		(ulong) space, (ulong) page_no,
+		(ulong) zip_size, (ulong) old_bits);
+
+	fputs("InnoDB: Submit a detailed bug report"
+	      " to http://bugs.mysql.com\n", stderr);
+}
+
+/************************************************************************
+During merge, inserts to an index page a secondary index entry extracted
+from the insert buffer. */
+static
+void
 ibuf_insert_to_index_page(
 /*======================*/
-	dtuple_t*	entry,	/*!< in: buffered entry to insert */
+	const dtuple_t*	entry,	/*!< in: buffered entry to insert */
 	buf_block_t*	block,	/*!< in/out: index page where the buffered entry
 				should be placed */
 	dict_index_t*	index,	/*!< in: record descriptor */
@@ -3760,11 +3836,10 @@ ibuf_insert_to_index_page(
 	ulint		low_match;
 	page_t*		page		= buf_block_get_frame(block);
 	rec_t*		rec;
-	page_t*		bitmap_page;
-	ulint		old_bits;
 
 	ut_ad(ibuf_inside());
 	ut_ad(dtuple_check_typed(entry));
+	ut_ad(!buf_block_align(page)->is_hashed);
 
 	if (UNIV_UNLIKELY(dict_table_is_comp(index->table)
 			  != (ibool)!!page_is_comp(page))) {
@@ -3810,71 +3885,87 @@ dump:
 	low_match = page_cur_search(block, index, entry,
 				    PAGE_CUR_LE, &page_cur);
 
-	if (low_match == dtuple_get_n_fields(entry)) {
+	if (UNIV_UNLIKELY(low_match == dtuple_get_n_fields(entry))) {
+		mem_heap_t*	heap;
+		upd_t*		update;
+		ulint*		offsets;
 		page_zip_des_t*	page_zip;
 
 		rec = page_cur_get_rec(&page_cur);
-		page_zip = buf_block_get_page_zip(block);
 
-		btr_cur_set_deleted_flag_for_ibuf(rec, page_zip, FALSE, mtr);
-	} else {
-		rec = page_cur_tuple_insert(&page_cur, entry, index, 0, mtr);
+		/* This is based on
+		row_ins_sec_index_entry_by_modify(BTR_MODIFY_LEAF). */
+		ut_ad(rec_get_deleted_flag(rec, page_is_comp(page)));
+
+		heap = mem_heap_create(1024);
+
+		offsets = rec_get_offsets(rec, index, NULL, ULINT_UNDEFINED,
+					  &heap);
+		update = row_upd_build_sec_rec_difference_binary(
+			index, entry, rec, NULL, heap);
 
-		if (UNIV_LIKELY(rec != NULL)) {
+		page_zip = buf_block_get_page_zip(block);
+
+		if (update->n_fields == 0) {
+			/* The records only differ in the delete-mark.
+			Clear the delete-mark, like we did before
+			Bug #56680 was fixed. */
+			btr_cur_set_deleted_flag_for_ibuf(
+				rec, page_zip, FALSE, mtr);
+updated_in_place:
+			mem_heap_free(heap);
 			return;
 		}
 
-		/* If the record did not fit, reorganize */
+		/* Copy the info bits. Clear the delete-mark. */
+		update->info_bits = rec_get_info_bits(rec, page_is_comp(page));
+		update->info_bits &= ~REC_INFO_DELETED_FLAG;
+
+		/* We cannot invoke btr_cur_optimistic_update() here,
+		because we do not have a btr_cur_t or que_thr_t,
+		as the insert buffer merge occurs at a very low level. */
+		if (!row_upd_changes_field_size_or_external(index, offsets,
+							    update)
+		    && (!page_zip || btr_cur_update_alloc_zip(
+				page_zip, block, index,
+				rec_offs_size(offsets), FALSE, mtr))) {
+			/* This is the easy case. Do something similar
+			to btr_cur_update_in_place(). */
+			row_upd_rec_in_place(rec, index, offsets,
+					     update, page_zip);
+			goto updated_in_place;
+		}
 
-		btr_page_reorganize(block, index, mtr);
-		page_cur_search(block, index, entry, PAGE_CUR_LE, &page_cur);
+		/* A collation may identify values that differ in
+		storage length.
+		Some examples (1 or 2 bytes):
+		utf8_turkish_ci: I = U+0131 LATIN SMALL LETTER DOTLESS I
+		utf8_general_ci: S = U+00DF LATIN SMALL LETTER SHARP S
+		utf8_general_ci: A = U+00E4 LATIN SMALL LETTER A WITH DIAERESIS
 
-		/* This time the record must fit */
-		if (UNIV_UNLIKELY
-		    (!page_cur_tuple_insert(&page_cur, entry, index,
-					    0, mtr))) {
-			ulint	space;
-			ulint	page_no;
-			ulint	zip_size;
+		latin1_german2_ci: SS = U+00DF LATIN SMALL LETTER SHARP S
 
-			ut_print_timestamp(stderr);
+		Examples of a character (3-byte UTF-8 sequence)
+		identified with 2 or 4 characters (1-byte UTF-8 sequences):
 
-			fprintf(stderr,
-				"  InnoDB: Error: Insert buffer insert"
-				" fails; page free %lu,"
-				" dtuple size %lu\n",
-				(ulong) page_get_max_insert_size(
-					page, 1),
-				(ulong) rec_get_converted_size(
-					index, entry, 0));
-			fputs("InnoDB: Cannot insert index record ",
-			      stderr);
-			dtuple_print(stderr, entry);
-			fputs("\nInnoDB: The table where"
-			      " this index record belongs\n"
-			      "InnoDB: is now probably corrupt."
-			      " Please run CHECK TABLE on\n"
-			      "InnoDB: that table.\n", stderr);
+		utf8_unicode_ci: 'II' = U+2171 SMALL ROMAN NUMERAL TWO
+		utf8_unicode_ci: '(10)' = U+247D PARENTHESIZED NUMBER TEN
+		*/
 
-			space = page_get_space_id(page);
-			zip_size = buf_block_get_zip_size(block);
-			page_no = page_get_page_no(page);
+		/* Delete the different-length record, and insert the
+		buffered one. */
 
-			bitmap_page = ibuf_bitmap_get_map_page(
-				space, page_no, zip_size, mtr);
-			old_bits = ibuf_bitmap_page_get_bits(
-				bitmap_page, page_no, zip_size,
-				IBUF_BITMAP_FREE, mtr);
-
-			fprintf(stderr,
-				"InnoDB: space %lu, page %lu,"
-				" zip_size %lu, bitmap bits %lu\n",
-				(ulong) space, (ulong) page_no,
-				(ulong) zip_size, (ulong) old_bits);
+		lock_rec_store_on_page_infimum(block, rec);
+		page_cur_delete_rec(&page_cur, index, offsets, mtr);
+		page_cur_move_to_prev(&page_cur);
+		mem_heap_free(heap);
 
-			fputs("InnoDB: Submit a detailed bug report"
-			      " to http://bugs.mysql.com\n", stderr);
-		}
+		ibuf_insert_to_index_page_low(entry, block, index, mtr,
+					      &page_cur);
+		lock_rec_restore_from_page_infimum(block, rec, block);
+	} else {
+		ibuf_insert_to_index_page_low(entry, block, index, mtr,
+					      &page_cur);
 	}
 }
 
@@ -3906,9 +3997,31 @@ ibuf_set_del_mark(
 		rec = page_cur_get_rec(&page_cur);
 		page_zip = page_cur_get_page_zip(&page_cur);
 
+		if (UNIV_UNLIKELY
+		    (rec_get_deleted_flag(
+			    rec, dict_table_is_comp(index->table)))) {
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: record is already delete-marked\n",
+			      stderr);
+			goto failure;
+		}
+
 		btr_cur_set_deleted_flag_for_ibuf(rec, page_zip, TRUE, mtr);
 	} else {
-		/* This can happen benignly in some situations. */
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: unable to find a record to delete-mark\n",
+		      stderr);
+failure:
+		fputs("InnoDB: tuple ", stderr);
+		dtuple_print(stderr, entry);
+		fputs("\n"
+		      "InnoDB: record ", stderr);
+		rec_print(stderr, page_cur_get_rec(&page_cur), index);
+		putc('\n', stderr);
+		fputs("\n"
+		      "InnoDB: Submit a detailed bug report"
+		      " to http://bugs.mysql.com\n", stderr);
+		ut_ad(0);
 	}
 }
 
@@ -3983,10 +4096,7 @@ ibuf_delete(
 			mem_heap_free(heap);
 		}
 	} else {
-		/* This can happen benignly in some situations: either when
-		we crashed at just the right time, or on database startup
-		when we redo some old log entries (due to worse stored
-		position granularity on disk than in memory). */
+		/* The record must have been purged already. */
 	}
 }