Percona-Server-5.5.35-rel33.0.tar.gz

btr/btr0cur.c

@@ -61,6 +61,7 @@ Created 10/16/1994 Heikki Tuuri
 #include "row0upd.h"
 #include "trx0rec.h"
 #include "trx0roll.h" /* trx_is_recv() */
+#include "trx0undo.h"
 #include "que0que.h"
 #include "row0row.h"
 #include "srv0srv.h"
@@ -1757,7 +1758,7 @@ btr_cur_upd_lock_and_undo(
 
 /***********************************************************//**
 Writes a redo log record of updating a record in-place. */
-UNIV_INLINE
+UNIV_INTERN
 void
 btr_cur_update_in_place_log(
 /*========================*/
@@ -1785,18 +1786,30 @@ btr_cur_update_in_place_log(
 		return;
 	}
 
-	/* The code below assumes index is a clustered index: change index to
-	the clustered index if we are updating a secondary index record (or we
-	could as well skip writing the sys col values to the log in this case
-	because they are not needed for a secondary index record update) */
-
-	index = dict_table_get_first_index(index->table);
+	/* For secondary indexes, we could skip writing the dummy system fields
+	to the redo log but we have to change redo log parsing of
+	MLOG_REC_UPDATE_IN_PLACE/MLOG_COMP_REC_UPDATE_IN_PLACE or we have to add
+	new redo log record. For now, just write dummy sys fields to the redo
+	log if we are updating a secondary index record.
+	*/
 
 	mach_write_to_1(log_ptr, flags);
 	log_ptr++;
 
-	log_ptr = row_upd_write_sys_vals_to_log(index, trx, roll_ptr, log_ptr,
-						mtr);
+	if (dict_index_is_clust(index)) {
+		log_ptr = row_upd_write_sys_vals_to_log(
+				index, trx, roll_ptr, log_ptr, mtr);
+	} else {
+		/* Dummy system fields for a secondary index */
+		/* TRX_ID Position */
+		log_ptr += mach_write_compressed(log_ptr, 0);
+		/* ROLL_PTR */
+		trx_write_roll_ptr(log_ptr, 0);
+		log_ptr += DATA_ROLL_PTR_LEN;
+		/* TRX_ID */
+		log_ptr += mach_ull_write_compressed(log_ptr, 0);
+	}
+
 	mach_write_to_2(log_ptr, page_offset(rec));
 	log_ptr += 2;
 

btr/btr0pcur.c

 /*****************************************************************************
 
-Copyright (c) 1996, 2012, Oracle and/or its affiliates. All Rights Reserved.
+Copyright (c) 1996, 2013, Oracle and/or its affiliates. All Rights Reserved.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -111,7 +111,7 @@ btr_pcur_store_position(
 	page_t*		page;
 	ulint		offs;
 
-	ut_a(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
 	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
 
 	block = btr_pcur_get_block(cursor);
@@ -128,7 +128,6 @@ btr_pcur_store_position(
 
 	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_S_FIX)
 	      || mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
-	ut_a(cursor->latch_mode != BTR_NO_LATCHES);
 
 	if (UNIV_UNLIKELY(page_get_n_recs(page) == 0)) {
 		/* It must be an empty index tree; NOTE that in this case
@@ -239,21 +238,12 @@ btr_pcur_restore_position_func(
 
 	ut_ad(mtr);
 	ut_ad(mtr->state == MTR_ACTIVE);
+	ut_ad(cursor->old_stored == BTR_PCUR_OLD_STORED);
+	ut_ad(cursor->pos_state == BTR_PCUR_WAS_POSITIONED
+	      || cursor->pos_state == BTR_PCUR_IS_POSITIONED);
 
 	index = btr_cur_get_index(btr_pcur_get_btr_cur(cursor));
 
-	if (UNIV_UNLIKELY(cursor->old_stored != BTR_PCUR_OLD_STORED)
-	    || UNIV_UNLIKELY(cursor->pos_state != BTR_PCUR_WAS_POSITIONED
-			     && cursor->pos_state != BTR_PCUR_IS_POSITIONED)) {
-		ut_print_buf(stderr, cursor, sizeof(btr_pcur_t));
-		putc('\n', stderr);
-		if (cursor->trx_if_known) {
-			trx_print(stderr, cursor->trx_if_known, 0);
-		}
-
-		ut_error;
-	}
-
 	if (UNIV_UNLIKELY
 	    (cursor->rel_pos == BTR_PCUR_AFTER_LAST_IN_TREE
 	     || cursor->rel_pos == BTR_PCUR_BEFORE_FIRST_IN_TREE)) {
@@ -277,14 +267,14 @@ btr_pcur_restore_position_func(
 
 	if (UNIV_LIKELY(latch_mode == BTR_SEARCH_LEAF)
 	    || UNIV_LIKELY(latch_mode == BTR_MODIFY_LEAF)) {
-		/* Try optimistic restoration */
+		/* Try optimistic restoration. */
 
-		if (UNIV_LIKELY(buf_page_optimistic_get(
-					latch_mode,
-					cursor->block_when_stored,
-					cursor->modify_clock,
-					file, line, mtr))) {
+		if (buf_page_optimistic_get(latch_mode,
+					    cursor->block_when_stored,
+					    cursor->modify_clock,
+					    file, line, mtr)) {
 			cursor->pos_state = BTR_PCUR_IS_POSITIONED;
+			cursor->latch_mode = latch_mode;
 
 			buf_block_dbg_add_level(
 				btr_pcur_get_block(cursor),
@@ -296,9 +286,6 @@ btr_pcur_restore_position_func(
 				const rec_t*	rec;
 				const ulint*	offsets1;
 				const ulint*	offsets2;
-#endif /* UNIV_DEBUG */
-				cursor->latch_mode = latch_mode;
-#ifdef UNIV_DEBUG
 				rec = btr_pcur_get_rec(cursor);
 
 				heap = mem_heap_create(256);
@@ -316,7 +303,13 @@ btr_pcur_restore_position_func(
 #endif /* UNIV_DEBUG */
 				return(TRUE);
 			}
-
+			/* This is the same record as stored,
+			may need to be adjusted for BTR_PCUR_BEFORE/AFTER,
+			depending on search mode and direction. */
+			if (btr_pcur_is_on_user_rec(cursor)) {
+				cursor->pos_state
+					= BTR_PCUR_IS_POSITIONED_OPTIMISTIC;
+			}
 			return(FALSE);
 		}
 	}
@@ -418,7 +411,7 @@ btr_pcur_move_to_next_page(
 	buf_block_t*	next_block;
 	page_t*		next_page;
 
-	ut_a(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
 	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
 	ut_ad(btr_pcur_is_after_last_on_page(cursor));
 
@@ -484,7 +477,6 @@ btr_pcur_move_backward_from_page(
 	ulint		latch_mode;
 	ulint		latch_mode2;
 
-	ut_a(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
 	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
 	ut_ad(btr_pcur_is_before_first_on_page(cursor));
 	ut_ad(!btr_pcur_is_before_first_in_tree(cursor, mtr));

buf/buf0buf.c

@@ -4754,12 +4754,16 @@ buf_get_latched_pages_number_instance(
 		case BUF_BLOCK_FILE_PAGE:
 			/* uncompressed page */
 			break;
+		case BUF_BLOCK_REMOVE_HASH:
+			/* We hold flush list but not LRU list mutex here.
+			Thus encountering BUF_BLOCK_REMOVE_HASH pages is
+			possible.  */
+			break;
 		case BUF_BLOCK_ZIP_FREE:
 		case BUF_BLOCK_ZIP_PAGE:
 		case BUF_BLOCK_NOT_USED:
 		case BUF_BLOCK_READY_FOR_USE:
 		case BUF_BLOCK_MEMORY:
-		case BUF_BLOCK_REMOVE_HASH:
 			ut_error;
 			break;
 		}

buf/buf0flu.c

@@ -1605,6 +1605,16 @@ buf_flush_page_and_try_neighbors(
 		ut_ad(block_mutex);
 	}
 
+	if (UNIV_UNLIKELY(buf_page_get_state(bpage)
+			  == BUF_BLOCK_REMOVE_HASH)) {
+
+		/* In case we don't hold the LRU list mutex, we may see a page
+		that is about to be relocated on the flush list.  Do not
+		attempt to flush it.  */
+		ut_ad(flush_type == BUF_FLUSH_LIST);
+		return (flushed);
+	}
+
 	ut_a(buf_page_in_file(bpage));
 
 	if (buf_flush_ready_for_flush(bpage, flush_type)) {

dict/dict0dict.c

@@ -36,6 +36,11 @@ UNIV_INTERN dict_index_t*	dict_ind_redundant;
 /** dummy index for ROW_FORMAT=COMPACT supremum and infimum records */
 UNIV_INTERN dict_index_t*	dict_ind_compact;
 
+#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 */
+
 #ifndef UNIV_HOTBACKUP
 #include "buf0buf.h"
 #include "data0type.h"
@@ -4855,6 +4860,8 @@ dict_update_statistics(
 	dict_index_t*	index;
 	ulint		sum_of_index_sizes	= 0;
 
+	DBUG_EXECUTE_IF("skip_innodb_statistics", return;);
+
 	if (table->ibd_file_missing) {
 		ut_print_timestamp(stderr);
 		fprintf(stderr,
@@ -4916,6 +4923,12 @@ dict_update_statistics(
 			continue;
 		}
 
+#if defined UNIV_DEBUG || defined UNIV_IBUF_DEBUG
+		if (ibuf_debug && !dict_index_is_clust(index)) {
+			goto fake_statistics;
+		}
+#endif /* UNIV_DEBUG || UNIV_IBUF_DEBUG */
+
 		if (UNIV_LIKELY
 		    (srv_force_recovery < SRV_FORCE_NO_IBUF_MERGE
 		     || (srv_force_recovery < SRV_FORCE_NO_LOG_REDO

fil/fil0fil.c

@@ -4369,7 +4369,7 @@ fil_load_single_table_tablespace(
 
 		if (check_msg) {
 			fprintf(stderr,
-				"InnoDB: Error: %s in file %s",
+				"InnoDB: Error: %s in file %s\n",
 				check_msg, filepath);
 			goto func_exit;
 		}
@@ -4967,6 +4967,7 @@ fil_extend_space_to_desired_size(
 			space->size += (size_after_extend - start_page_no);
 			os_has_said_disk_full = FALSE;
 		}
+		fil_node_complete_io(node, fil_system, OS_FILE_READ);
 		goto complete_io;
 	}
 #endif

handler/ha_innodb.cc

@@ -55,6 +55,7 @@ this program; if not, write to the Free Software Foundation, Inc.,
 #include <mysql/innodb_priv.h>
 #include <mysql/psi/psi.h>
 #include <my_sys.h>
+#include <my_check_opt.h>
 
 #ifdef MYSQL_SERVER
 #include <rpl_mi.h>
@@ -8810,6 +8811,10 @@ ha_innobase::records_in_range(
 	/* There exists possibility of not being able to find requested
 	index due to inconsistency between MySQL and InoDB dictionary info.
 	Necessary message should have been printed in innobase_get_index() */
+	if (prebuilt->table->ibd_file_missing) {
+		n_rows = HA_POS_ERROR;
+		goto func_exit;
+	}
 	if (UNIV_UNLIKELY(!index)) {
 		n_rows = HA_POS_ERROR;
 		goto func_exit;
@@ -9533,8 +9538,7 @@ int
 ha_innobase::check(
 /*===============*/
 	THD*		thd,		/*!< in: user thread handle */
-	HA_CHECK_OPT*	check_opt)	/*!< in: check options, currently
-					ignored */
+	HA_CHECK_OPT*	check_opt)	/*!< in: check options */
 {
 	dict_index_t*	index;
 	ulint		n_rows;
@@ -9591,11 +9595,6 @@ ha_innobase::check(
 	do additional check */
 	prebuilt->table->corrupted = FALSE;
 
-	/* Enlarge the fatal lock wait timeout during CHECK TABLE. */
-	mutex_enter(&kernel_mutex);
-	srv_fatal_semaphore_wait_threshold += SRV_SEMAPHORE_WAIT_EXTENSION;
-	mutex_exit(&kernel_mutex);
-
 	for (index = dict_table_get_first_index(prebuilt->table);
 	     index != NULL;
 	     index = dict_table_get_next_index(index)) {
@@ -9608,20 +9607,41 @@ ha_innobase::check(
 
 		/* If this is an index being created, break */
 		if (*index->name == TEMP_INDEX_PREFIX) {
-			break;
-		}  else if (!btr_validate_index(index, prebuilt->trx)) {
-			is_ok = FALSE;
+			continue;
+		}
+		if (!(check_opt->flags & T_QUICK)) {
+			/* Enlarge the fatal lock wait timeout during
+			CHECK TABLE. */
+			mutex_enter(&kernel_mutex);
+			srv_fatal_semaphore_wait_threshold +=
+				SRV_SEMAPHORE_WAIT_EXTENSION;
+			mutex_exit(&kernel_mutex);
+
+			ibool	valid = TRUE;
+			valid = btr_validate_index(index, prebuilt->trx);
+
+			/* Restore the fatal lock wait timeout after
+			CHECK TABLE. */
+			mutex_enter(&kernel_mutex);
+			srv_fatal_semaphore_wait_threshold -=
+				SRV_SEMAPHORE_WAIT_EXTENSION;
+			mutex_exit(&kernel_mutex);
+
+			if (!valid) {
+				is_ok = FALSE;
 
-			innobase_format_name(
-				index_name, sizeof index_name,
-				prebuilt->index->name, TRUE);
+				innobase_format_name(
+					index_name, sizeof index_name,
+					index->name, TRUE);
+				push_warning_printf(thd,
+					MYSQL_ERROR::WARN_LEVEL_WARN,
+					ER_NOT_KEYFILE,
+					"InnoDB: The B-tree of"
+					" index %s is corrupted.",
+					index_name);
 
-			push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
-					    ER_NOT_KEYFILE,
-					    "InnoDB: The B-tree of"
-					    " index %s is corrupted.",
-					    index_name);
-			continue;
+				continue;
+			}
 		}
 
 		/* Instead of invoking change_active_index(), set up
@@ -9725,21 +9745,17 @@ ha_innobase::check(
 	/* Restore the original isolation level */
 	prebuilt->trx->isolation_level = old_isolation_level;
 
-	/* We validate also the whole adaptive hash index for all tables
-	at every CHECK TABLE */
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	/* We validate the whole adaptive hash index for all tables
+	at every CHECK TABLE only when QUICK flag is not present. */
 
-	if (!btr_search_validate()) {
+	if (!(check_opt->flags & T_QUICK) && !btr_search_validate()) {
 		push_warning(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
 			     ER_NOT_KEYFILE,
 			     "InnoDB: The adaptive hash index is corrupted.");
 		is_ok = FALSE;
 	}
-
-	/* Restore the fatal lock wait timeout after CHECK TABLE. */
-	mutex_enter(&kernel_mutex);
-	srv_fatal_semaphore_wait_threshold -= SRV_SEMAPHORE_WAIT_EXTENSION;
-	mutex_exit(&kernel_mutex);
-
+#endif /* defined UNIV_AHI_DEBUG || defined UNIV_DEBUG */
 	prebuilt->trx->op_info = "";
 	if (thd_killed(user_thd)) {
 		my_error(ER_QUERY_INTERRUPTED, MYF(0));
@@ -10623,6 +10639,7 @@ innodb_show_status(
 	const long		MAX_STATUS_SIZE = 1048576;
 	ulint			trx_list_start = ULINT_UNDEFINED;
 	ulint			trx_list_end = ULINT_UNDEFINED;
+	bool			ret_val;
 
 	DBUG_ENTER("innodb_show_status");
 	DBUG_ASSERT(hton == innodb_hton_ptr);
@@ -10687,12 +10704,13 @@ innodb_show_status(
 
 	mutex_exit(&srv_monitor_file_mutex);
 
-	stat_print(thd, innobase_hton_name, (uint) strlen(innobase_hton_name),
-		   STRING_WITH_LEN(""), str, flen);
+	ret_val= stat_print(thd, innobase_hton_name,
+				(uint) strlen(innobase_hton_name),
+				STRING_WITH_LEN(""), str, flen);
 
 	my_free(str);
 
-	DBUG_RETURN(FALSE);
+	DBUG_RETURN(ret_val);
 }
 
 /************************************************************************//**

handler/i_s.cc

@@ -48,14 +48,18 @@ extern "C" {
 #include "btr0types.h"
 #include "buf0buddy.h"
 #include "buf0buf.h"
+#include "buf0lru.h"
 #include "ibuf0ibuf.h"
 #include "dict0mem.h"
 #include "dict0types.h"
 #include "dict0boot.h"
+#include "dict0load.h"
 #include "ha_prototypes.h"
 #include "srv0start.h"
+#include "srv0srv.h"
 #include "trx0i_s.h"
 #include "trx0rseg.h"
+#include "trx0trx.h"
 #include "trx0undo.h"
 #include "log0online.h"
 #include "btr0btr.h"

ibuf/ibuf0ibuf.c

@@ -194,11 +194,6 @@ 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;
 
@@ -2693,6 +2688,12 @@ ibuf_contract_ext(
 		return(0);
 	}
 
+#if defined UNIV_DEBUG || defined UNIV_IBUF_DEBUG
+	if (ibuf_debug) {
+		return(0);
+	}
+#endif /* UNIV_DEBUG || UNIV_IBUF_DEBUG */
+
 	ibuf_mtr_start(&mtr);
 
 	/* Open a cursor to a randomly chosen leaf of the tree, at a random
@@ -4066,6 +4067,24 @@ updated_in_place:
 			to btr_cur_update_in_place(). */
 			row_upd_rec_in_place(rec, index, offsets,
 					     update, page_zip);
+
+			/* Log the update in place operation. During recovery
+			MLOG_COMP_REC_UPDATE_IN_PLACE/MLOG_REC_UPDATE_IN_PLACE
+			expects trx_id, roll_ptr for secondary indexes. So we
+			just write dummy trx_id(0), roll_ptr(0) */
+			btr_cur_update_in_place_log(BTR_KEEP_SYS_FLAG, rec,
+						    index, update,
+						    NULL, 0, mtr);
+			DBUG_EXECUTE_IF(
+				"crash_after_log_ibuf_upd_inplace",
+				log_buffer_flush_to_disk();
+				fprintf(stderr,
+					"InnoDB: Wrote log record for ibuf "
+					"update in place operation\n");
+				DBUG_SUICIDE();
+			);
+
+
 			goto updated_in_place;
 		}
 

include/btr0cur.h

@@ -637,6 +637,21 @@ btr_cur_set_deleted_flag_for_ibuf(
 					uncompressed */
 	ibool		val,		/*!< in: value to set */
 	mtr_t*		mtr);		/*!< in/out: mini-transaction */
+
+/***********************************************************//**
+Writes a redo log record of updating a record in-place. */
+UNIV_INTERN
+void
+btr_cur_update_in_place_log(
+/*========================*/
+	ulint		flags,		/*!< in: flags */
+	rec_t*		rec,		/*!< in: record */
+	dict_index_t*	index,		/*!< in: index where cursor positioned */
+	const upd_t*	update,		/*!< in: update vector */
+	trx_t*		trx,		/*!< in: transaction */
+	roll_ptr_t	roll_ptr,	/*!< in: roll ptr */
+	mtr_t*		mtr);		/*!< in: mtr */
+
 /*######################################################################*/
 
 /** In the pessimistic delete, if the page data size drops below this

include/btr0pcur.h

 /*****************************************************************************
 
-Copyright (c) 1996, 2011, Oracle and/or its affiliates. All Rights Reserved.
+Copyright (c) 1996, 2013, Oracle and/or its affiliates. All Rights Reserved.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -447,6 +447,27 @@ btr_pcur_move_to_prev_on_page(
 /*==========================*/
 	btr_pcur_t*	cursor);/*!< in/out: persistent cursor */
 
+/** Position state of persistent B-tree cursor. */
+enum pcur_pos_t {
+	/** The persistent cursor is not positioned. */
+	BTR_PCUR_NOT_POSITIONED = 0,
+	/** The persistent cursor was previously positioned.
+	TODO: currently, the state can be BTR_PCUR_IS_POSITIONED,
+	though it really should be BTR_PCUR_WAS_POSITIONED,
+	because we have no obligation to commit the cursor with
+	mtr; similarly latch_mode may be out of date. This can
+	lead to problems if btr_pcur is not used the right way;
+	all current code should be ok. */
+	BTR_PCUR_WAS_POSITIONED,
+	/** The persistent cursor is positioned by optimistic get to the same
+	record as it was positioned at. Not used for rel_pos == BTR_PCUR_ON.
+	It may need adjustment depending on previous/current search direction
+	and rel_pos. */
+	BTR_PCUR_IS_POSITIONED_OPTIMISTIC,
+	/** The persistent cursor is positioned by index search.
+	Or optimistic get for rel_pos == BTR_PCUR_ON. */
+	BTR_PCUR_IS_POSITIONED
+};
 
 /* The persistent B-tree cursor structure. This is used mainly for SQL
 selects, updates, and deletes. */
@@ -480,10 +501,8 @@ struct btr_pcur_struct{
 	ib_uint64_t	modify_clock;	/*!< the modify clock value of the
 					buffer block when the cursor position
 					was stored */
-	ulint		pos_state;	/*!< see TODO note below!
-					BTR_PCUR_IS_POSITIONED,
-					BTR_PCUR_WAS_POSITIONED,
-					BTR_PCUR_NOT_POSITIONED */
+	enum pcur_pos_t	pos_state;	/*!< btr_pcur_store_position() and
+					btr_pcur_restore_position() state. */
 	ulint		search_mode;	/*!< PAGE_CUR_G, ... */
 	trx_t*		trx_if_known;	/*!< the transaction, if we know it;
 					otherwise this field is not defined;
@@ -499,21 +518,6 @@ struct btr_pcur_struct{
 					is not NULL */
 };
 
-#define BTR_PCUR_IS_POSITIONED	1997660512	/* TODO: currently, the state
-						can be BTR_PCUR_IS_POSITIONED,
-						though it really should be
-						BTR_PCUR_WAS_POSITIONED,
-						because we have no obligation
-						to commit the cursor with
-						mtr; similarly latch_mode may
-						be out of date. This can
-						lead to problems if btr_pcur
-						is not used the right way;
-						all current code should be
-						ok. */
-#define BTR_PCUR_WAS_POSITIONED	1187549791
-#define BTR_PCUR_NOT_POSITIONED 1328997689
-
 #define BTR_PCUR_OLD_STORED	908467085
 #define BTR_PCUR_OLD_NOT_STORED	122766467
 

include/btr0pcur.ic

 /*****************************************************************************
 
-Copyright (c) 1996, 2011, Oracle and/or its affiliates. All Rights Reserved.
+Copyright (c) 1996, 2013, Oracle and/or its affiliates. All Rights Reserved.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -379,7 +379,7 @@ btr_pcur_commit_specify_mtr(
 	btr_pcur_t*	pcur,	/*!< in: persistent cursor */
 	mtr_t*		mtr)	/*!< in: mtr to commit */
 {
-	ut_a(pcur->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(pcur->pos_state == BTR_PCUR_IS_POSITIONED);
 
 	pcur->latch_mode = BTR_NO_LATCHES;
 

include/btr0sea.h

@@ -196,8 +196,6 @@ UNIV_INTERN
 ibool
 btr_search_validate(void);
 /*======================*/
-#else
-# define btr_search_validate()	TRUE
 #endif /* defined UNIV_AHI_DEBUG || defined UNIV_DEBUG */
 
 /********************************************************************//**

include/dict0types.h

@@ -63,4 +63,9 @@ typedef enum dict_err_ignore		dict_err_ignore_t;
 #define TEMP_TABLE_PREFIX                "#sql"
 #define TEMP_TABLE_PATH_PREFIX           "/" TEMP_TABLE_PREFIX
 
+#if defined UNIV_DEBUG || defined UNIV_IBUF_DEBUG
+/** Flag to control insert buffer debugging. */
+extern uint		ibuf_debug;
+#endif /* UNIV_DEBUG || UNIV_IBUF_DEBUG */
+
 #endif

include/ibuf0ibuf.h

@@ -63,11 +63,6 @@ typedef enum {
 /** Operations that can currently be buffered. */
 extern ibuf_use_t	ibuf_use;
 
-#if defined UNIV_DEBUG || defined UNIV_IBUF_DEBUG
-/** Flag to control insert buffer debugging. */
-extern uint		ibuf_debug;
-#endif /* UNIV_DEBUG || UNIV_IBUF_DEBUG */
-
 /** The insert buffer control structure */
 extern ibuf_t*		ibuf;
 

include/log0log.h

@@ -822,6 +822,8 @@ struct log_struct{
 					later; this is advanced when a flush
 					operation is completed to all the log
 					groups */
+	volatile ibool	is_extending;	/*!< this is set to true during extend
+					the log buffer size */
 	ib_uint64_t	written_to_some_lsn;
 					/*!< first log sequence number not yet
 					written to any log group; for this to

include/univ.i

@@ -64,7 +64,7 @@ component, i.e. we show M.N.P as M.N */
 	(INNODB_VERSION_MAJOR << 8 | INNODB_VERSION_MINOR)
 
 #ifndef PERCONA_INNODB_VERSION
-#define PERCONA_INNODB_VERSION 29.3
+#define PERCONA_INNODB_VERSION 33.0
 #endif
 
 #define INNODB_VERSION_STR	MYSQL_SERVER_VERSION "-" IB_TO_STR(PERCONA_INNODB_VERSION)

log/log0log.c

@@ -262,6 +262,85 @@ log_check_tracking_margin(
 	return tracked_lsn_age + lsn_advance > log_sys->max_checkpoint_age;
 }
 
+/** Extends the log buffer.
+@param[in] len	requested minimum size in bytes */
+static
+void
+log_buffer_extend(
+	ulint	len)
+{
+	ulint	move_start;
+	ulint	move_end;
+	byte	tmp_buf[OS_FILE_LOG_BLOCK_SIZE];
+
+	mutex_enter(&(log_sys->mutex));
+
+	while (log_sys->is_extending) {
+		/* Another thread is trying to extend already.
+		Needs to wait for. */
+		mutex_exit(&(log_sys->mutex));
+
+		log_buffer_flush_to_disk();
+
+		mutex_enter(&(log_sys->mutex));
+
+		if (srv_log_buffer_size > len / UNIV_PAGE_SIZE) {
+			/* Already extended enough by the others */
+			mutex_exit(&(log_sys->mutex));
+			return;
+		}
+	}
+
+	log_sys->is_extending = TRUE;
+
+	while (log_sys->n_pending_writes != 0
+	       || ut_calc_align_down(log_sys->buf_free,
+				     OS_FILE_LOG_BLOCK_SIZE)
+		  != ut_calc_align_down(log_sys->buf_next_to_write,
+					OS_FILE_LOG_BLOCK_SIZE)) {
+		/* Buffer might have >1 blocks to write still. */
+		mutex_exit(&(log_sys->mutex));
+
+		log_buffer_flush_to_disk();
+
+		mutex_enter(&(log_sys->mutex));
+	}
+
+	move_start = ut_calc_align_down(
+		log_sys->buf_free,
+		OS_FILE_LOG_BLOCK_SIZE);
+	move_end = log_sys->buf_free;
+
+	/* store the last log block in buffer */
+	ut_memcpy(tmp_buf, log_sys->buf + move_start,
+		  move_end - move_start);
+
+	log_sys->buf_free -= move_start;
+	log_sys->buf_next_to_write -= move_start;
+
+	/* reallocate log buffer */
+	srv_log_buffer_size = len / UNIV_PAGE_SIZE + 1;
+	mem_free(log_sys->buf_ptr);
+	log_sys->buf_ptr = mem_alloc(LOG_BUFFER_SIZE + OS_FILE_LOG_BLOCK_SIZE);
+	log_sys->buf = ut_align(log_sys->buf_ptr, OS_FILE_LOG_BLOCK_SIZE);
+	log_sys->buf_size = LOG_BUFFER_SIZE;
+	memset(log_sys->buf, '\0', LOG_BUFFER_SIZE);
+	log_sys->max_buf_free = log_sys->buf_size / LOG_BUF_FLUSH_RATIO
+		- LOG_BUF_FLUSH_MARGIN;
+
+	/* restore the last log block */
+	ut_memcpy(log_sys->buf, tmp_buf, move_end - move_start);
+
+	ut_ad(log_sys->is_extending);
+	log_sys->is_extending = FALSE;
+
+	mutex_exit(&(log_sys->mutex));
+
+	fprintf(stderr,
+		"InnoDB: innodb_log_buffer_size was extended to %lu.\n",
+		LOG_BUFFER_SIZE);
+}
+
 /************************************************************//**
 Opens the log for log_write_low. The log must be closed with log_close.
 @return	start lsn of the log record */
@@ -279,10 +358,38 @@ log_open(
 #endif /* UNIV_LOG_ARCHIVE */
 	ulint	count			= 0;
 
-	ut_a(len < log->buf_size / 2);
+	if (len >= log->buf_size / 2) {
+		DBUG_EXECUTE_IF("ib_log_buffer_is_short_crash",
+				DBUG_SUICIDE(););
+
+		/* log_buffer is too small. try to extend instead of crash. */
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			" InnoDB: Warning: "
+			"The transaction log size is too large"
+			" for innodb_log_buffer_size (%lu >= %lu / 2). "
+			"Trying to extend it.\n",
+			len, LOG_BUFFER_SIZE);
+
+		log_buffer_extend((len + 1) * 2);
+	}
 loop:
 	ut_ad(!recv_no_log_write);
 
+	if (log->is_extending) {
+
+		mutex_exit(&(log->mutex));
+
+		/* Log buffer size is extending. Writing up to the next block
+		should wait for the extending finished. */
+
+		os_thread_sleep(100000);
+
+		ut_ad(++count < 50);
+
+		goto loop;
+	}
+
 	/* Calculate an upper limit for the space the string may take in the
 	log buffer */
 
@@ -899,6 +1006,7 @@ log_init(void)
 	log_sys->buf = ut_align(log_sys->buf_ptr, OS_FILE_LOG_BLOCK_SIZE);
 
 	log_sys->buf_size = LOG_BUFFER_SIZE;
+	log_sys->is_extending = FALSE;
 
 	memset(log_sys->buf, '\0', LOG_BUFFER_SIZE);
 

log/log0online.c

@@ -1790,7 +1790,8 @@ log_online_purge_changed_page_bitmaps(
 		mutex_enter(&log_bmp_sys->mutex);
 	}
 
-	if (!log_online_setup_bitmap_file_range(&bitmap_files, 0, lsn)) {
+	if (!log_online_setup_bitmap_file_range(&bitmap_files, 0,
+						IB_ULONGLONG_MAX)) {
 		if (srv_track_changed_pages) {
 			mutex_exit(&log_bmp_sys->mutex);
 		}
@@ -1805,8 +1806,20 @@ log_online_purge_changed_page_bitmaps(
 	}
 
 	for (i = 0; i < bitmap_files.count; i++) {
-		if (bitmap_files.files[i].seq_num == 0
-		    || bitmap_files.files[i].start_lsn >= lsn) {
+
+		/* We consider the end LSN of the current bitmap, derived from
+		the start LSN of the subsequent bitmap file, to determine
+		whether to remove the current bitmap.  Note that bitmap_files
+		does not contain an entry for the bitmap past the given LSN so
+		we must check the boundary conditions as well.  For example,
+		consider 1_0.xdb and 2_10.xdb and querying LSN 5.  bitmap_files
+		will only contain 1_0.xdb and we must not delete it since it
+		represents LSNs 0-9. */
+		if ((i + 1 == bitmap_files.count
+		     || bitmap_files.files[i + 1].seq_num == 0
+		     || bitmap_files.files[i + 1].start_lsn > lsn)
+		    && (lsn != IB_ULONGLONG_MAX)) {
+
 			break;
 		}
 		if (!os_file_delete_if_exists(bitmap_files.files[i].name)) {

row/row0sel.c

@@ -3099,48 +3099,78 @@ sel_restore_position_for_mysql(
 	mtr_t*		mtr)		/*!< in: mtr; CAUTION: may commit
 					mtr temporarily! */
 {
-	ibool	success;
-	ulint	relative_position;
-
-	relative_position = pcur->rel_pos;
+	ibool		success;
 
 	success = btr_pcur_restore_position(latch_mode, pcur, mtr);
 
 	*same_user_rec = success;
 
-	if (relative_position == BTR_PCUR_ON) {
-		if (success) {
-			return(FALSE);
-		}
-
-		if (moves_up) {
-			btr_pcur_move_to_next(pcur, mtr);
-		}
-
-		return(TRUE);
+	ut_ad(!success || pcur->rel_pos == BTR_PCUR_ON);
+#ifdef UNIV_DEBUG
+	if (pcur->pos_state == BTR_PCUR_IS_POSITIONED_OPTIMISTIC) {
+		ut_ad(pcur->rel_pos == BTR_PCUR_BEFORE
+		      || pcur->rel_pos == BTR_PCUR_AFTER);
+	} else {
+		ut_ad(pcur->pos_state == BTR_PCUR_IS_POSITIONED);
+		ut_ad((pcur->rel_pos == BTR_PCUR_ON)
+		      == btr_pcur_is_on_user_rec(pcur));
 	}
+#endif
 
-	if (relative_position == BTR_PCUR_AFTER
-	    || relative_position == BTR_PCUR_AFTER_LAST_IN_TREE) {
+	/* The position may need be adjusted for rel_pos and moves_up. */
 
-		if (moves_up) {
+	switch (pcur->rel_pos) {
+	case BTR_PCUR_ON:
+		if (!success && moves_up) {
+next:
+			btr_pcur_move_to_next(pcur, mtr);
 			return(TRUE);
 		}
-
-		if (btr_pcur_is_on_user_rec(pcur)) {
+		return(!success);
+	case BTR_PCUR_AFTER_LAST_IN_TREE:
+	case BTR_PCUR_BEFORE_FIRST_IN_TREE:
+		return(TRUE);
+	case BTR_PCUR_AFTER:
+		/* positioned to record after pcur->old_rec. */
+		pcur->pos_state = BTR_PCUR_IS_POSITIONED;
+prev:
+		if (btr_pcur_is_on_user_rec(pcur) && !moves_up) {
 			btr_pcur_move_to_prev(pcur, mtr);
 		}
-
 		return(TRUE);
+	case BTR_PCUR_BEFORE:
+		/* For non optimistic restoration:
+		The position is now set to the record before pcur->old_rec.
+
+		For optimistic restoration:
+		The position also needs to take the previous search_mode into
+		consideration. */
+
+		switch (pcur->pos_state) {
+		case BTR_PCUR_IS_POSITIONED_OPTIMISTIC:
+			pcur->pos_state = BTR_PCUR_IS_POSITIONED;
+			if (pcur->search_mode == PAGE_CUR_GE) {
+				/* Positioned during Greater or Equal search
+				with BTR_PCUR_BEFORE. Optimistic restore to
+				the same record. If scanning for lower then
+				we must move to previous record.
+				This can happen with:
+				HANDLER READ idx a = (const);
+				HANDLER READ idx PREV; */
+				goto prev;
+			}
+			return(TRUE);
+		case BTR_PCUR_IS_POSITIONED:
+			if (moves_up && btr_pcur_is_on_user_rec(pcur)) {
+				goto next;
+			}
+			return(TRUE);
+		case BTR_PCUR_WAS_POSITIONED:
+		case BTR_PCUR_NOT_POSITIONED:
+			break;
+		}
 	}
-
-	ut_ad(relative_position == BTR_PCUR_BEFORE
-	      || relative_position == BTR_PCUR_BEFORE_FIRST_IN_TREE);
-
-	if (moves_up && btr_pcur_is_on_user_rec(pcur)) {
-		btr_pcur_move_to_next(pcur, mtr);
-	}
-
+	ut_ad(0);
 	return(TRUE);
 }
 
@@ -4141,6 +4171,14 @@ wrong_offs:
 
 			btr_pcur_store_position(pcur, &mtr);
 
+			/* The found record was not a match, but may be used
+			as NEXT record (index_next). Set the relative position
+			to BTR_PCUR_BEFORE, to reflect that the position of
+			the persistent cursor is before the found/stored row
+			(pcur->old_rec). */
+			ut_ad(pcur->rel_pos == BTR_PCUR_ON);
+			pcur->rel_pos = BTR_PCUR_BEFORE;
+
 			err = DB_RECORD_NOT_FOUND;
 			/* ut_print_name(stderr, index->name);
 			fputs(" record not found 3\n", stderr); */
@@ -4180,6 +4218,14 @@ wrong_offs:
 
 			btr_pcur_store_position(pcur, &mtr);
 
+			/* The found record was not a match, but may be used
+			as NEXT record (index_next). Set the relative position
+			to BTR_PCUR_BEFORE, to reflect that the position of
+			the persistent cursor is before the found/stored row
+			(pcur->old_rec). */
+			ut_ad(pcur->rel_pos == BTR_PCUR_ON);
+			pcur->rel_pos = BTR_PCUR_BEFORE;
+
 			err = DB_RECORD_NOT_FOUND;
 			/* ut_print_name(stderr, index->name);
 			fputs(" record not found 4\n", stderr); */
@@ -4757,6 +4803,7 @@ normal_return:
 	if (prebuilt->n_fetch_cached > 0) {
 		row_sel_pop_cached_row_for_mysql(buf, prebuilt);
 
+		DEBUG_SYNC_C("row_search_cached_row");
 		err = DB_SUCCESS;
 	}
 

trx/trx0trx.c

@@ -1306,7 +1306,15 @@ trx_cleanup_at_db_startup(
 	}
 
 	trx->state = TRX_NOT_STARTED;
+
+	/* This code is executed in a single threaded context, but we acquire
+	kernel_mutex to satisfy a debug assertion in
+	trx_release_descriptor(). */
+
+	mutex_enter(&kernel_mutex);
 	trx_release_descriptor(trx);
+	mutex_exit(&kernel_mutex);
+
 	trx->rseg = NULL;
 	trx->undo_no = 0;
 	trx->last_sql_stat_start.least_undo_no = 0;