[packages/mysql.git] / innodb_dict_size_limit.patch

# name       : innodb_dict_size_limit.patch
# introduced : 11 or before
# maintainer : Yasufumi
#
#!!! notice !!!
# Any small change to this file in the main branch
# should be done or reviewed by the maintainer!
--- a/storage/innobase/btr/btr0sea.c
+++ b/storage/innobase/btr/btr0sea.c
@@ -1187,6 +1187,178 @@
 	mem_free(folds);
 }
 
+/************************************************************************
+Drops a page hash index based on index */
+UNIV_INTERN
+void
+btr_search_drop_page_hash_index_on_index(
+/*=====================================*/
+	dict_index_t*	index)		/* in: record descriptor */
+{
+
+	hash_table_t*	table;
+	buf_block_t*	block;
+	ulint		n_fields;
+	ulint		n_bytes;
+	const page_t*		page;
+	const rec_t*		rec;
+	ulint		fold;
+	ulint		prev_fold;
+	index_id_t	index_id;
+	ulint		n_cached;
+	ulint		n_recs;
+	ulint*		folds;
+	ulint		i, j;
+	mem_heap_t*	heap	= NULL;
+	ulint*		offsets;
+	ibool		released_search_latch;
+
+	rw_lock_s_lock(&btr_search_latch);
+
+	table = btr_search_sys->hash_index;
+
+	for (j = 0; j < srv_buf_pool_instances; j++) {
+		buf_pool_t*	buf_pool;
+
+		buf_pool = buf_pool_from_array(j);
+
+		do {
+			buf_chunk_t*	chunks	= buf_pool->chunks;
+			buf_chunk_t*	chunk	= chunks + buf_pool->n_chunks;
+
+			released_search_latch = FALSE;
+
+			while (--chunk >= chunks) {
+				block	= chunk->blocks;
+				i	= chunk->size;
+
+retry:
+				for (; i--; block++) {
+					if (buf_block_get_state(block)
+					    != BUF_BLOCK_FILE_PAGE
+					    || block->index != index
+					    || !block->index) {
+						continue;
+					}
+
+					page = block->frame;
+
+					/* from btr_search_drop_page_hash_index() */
+					n_fields = block->curr_n_fields;
+					n_bytes = block->curr_n_bytes;
+
+
+					/* keeping latch order */
+					rw_lock_s_unlock(&btr_search_latch);
+					released_search_latch = TRUE;
+					rw_lock_x_lock(&block->lock);
+
+
+					ut_a(n_fields + n_bytes > 0);
+
+					n_recs = page_get_n_recs(page);
+
+					/* Calculate and cache fold values into an array for fast deletion
+					from the hash index */
+
+					folds = mem_alloc(n_recs * sizeof(ulint));
+
+					n_cached = 0;
+
+					rec = page_get_infimum_rec(page);
+					rec = page_rec_get_next_low(rec, page_is_comp(page));
+
+					index_id = btr_page_get_index_id(page);
+	
+					ut_a(index_id == index->id);
+
+					prev_fold = 0;
+
+					offsets = NULL;
+
+					while (!page_rec_is_supremum(rec)) {
+						offsets = rec_get_offsets(rec, index, offsets,
+									n_fields + (n_bytes > 0), &heap);
+						ut_a(rec_offs_n_fields(offsets) == n_fields + (n_bytes > 0));
+						fold = rec_fold(rec, offsets, n_fields, n_bytes, index_id);
+
+						if (fold == prev_fold && prev_fold != 0) {
+
+							goto next_rec;
+						}
+
+						/* Remove all hash nodes pointing to this page from the
+						hash chain */
+
+						folds[n_cached] = fold;
+						n_cached++;
+next_rec:
+						rec = page_rec_get_next_low(rec, page_rec_is_comp(rec));
+						prev_fold = fold;
+					}
+
+					if (UNIV_LIKELY_NULL(heap)) {
+						mem_heap_empty(heap);
+					}
+
+					rw_lock_x_lock(&btr_search_latch);
+
+					if (UNIV_UNLIKELY(!block->index)) {
+						goto cleanup;
+					}
+
+					ut_a(block->index == index);
+
+					if (UNIV_UNLIKELY(block->curr_n_fields != n_fields)
+					    || UNIV_UNLIKELY(block->curr_n_bytes != n_bytes)) {
+						rw_lock_x_unlock(&btr_search_latch);
+						rw_lock_x_unlock(&block->lock);
+
+						mem_free(folds);
+
+						rw_lock_s_lock(&btr_search_latch);
+						goto retry;
+					}
+
+					for (i = 0; i < n_cached; i++) {
+
+						ha_remove_all_nodes_to_page(table, folds[i], page);
+					}
+
+					ut_a(index->search_info->ref_count > 0);
+					index->search_info->ref_count--;
+
+					block->index = NULL;
+
+cleanup:
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+					if (UNIV_UNLIKELY(block->n_pointers)) {
+						/* Corruption */
+						ut_print_timestamp(stderr);
+						fprintf(stderr,
+"InnoDB: The adaptive hash index is corrupted. After dropping\n"
+"InnoDB: the hash index to a page of %s, %lu hash nodes still remain.\n",
+							index->name, (ulong) block->n_pointers);
+					}
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+					rw_lock_x_unlock(&btr_search_latch);
+					rw_lock_x_unlock(&block->lock);
+
+					mem_free(folds);
+
+					rw_lock_s_lock(&btr_search_latch);
+				}
+			}
+		} while (released_search_latch);
+	}
+
+	rw_lock_s_unlock(&btr_search_latch);
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+}
+
 /********************************************************************//**
 Drops a possible page hash index when a page is evicted from the buffer pool
 or freed in a file segment. */
--- a/storage/innobase/buf/buf0buf.c
+++ b/storage/innobase/buf/buf0buf.c
@@ -294,14 +294,14 @@
 # endif /* !PFS_SKIP_BUFFER_MUTEX_RWLOCK */
 #endif /* UNIV_PFS_MUTEX || UNIV_PFS_RWLOCK */
 
-/** A chunk of buffers.  The buffer pool is allocated in chunks. */
-struct buf_chunk_struct{
-	ulint		mem_size;	/*!< allocated size of the chunk */
-	ulint		size;		/*!< size of frames[] and blocks[] */
-	void*		mem;		/*!< pointer to the memory area which
-					was allocated for the frames */
-	buf_block_t*	blocks;		/*!< array of buffer control blocks */
-};
+/** A chunk of buffers.  The buffer pool is allocated in chunks. (moved to buf0buf.h)*/
+//struct buf_chunk_struct{
+//	ulint		mem_size;	/*!< allocated size of the chunk */
+//	ulint		size;		/*!< size of frames[] and blocks[] */
+//	void*		mem;		/*!< pointer to the memory area which
+//					was allocated for the frames */
+//	buf_block_t*	blocks;		/*!< array of buffer control blocks */
+//};
 #endif /* !UNIV_HOTBACKUP */
 
 /********************************************************************//**
--- a/storage/innobase/dict/dict0boot.c
+++ b/storage/innobase/dict/dict0boot.c
@@ -284,6 +284,7 @@
 	system tables */
 	/*-------------------------*/
 	table = dict_mem_table_create("SYS_TABLES", DICT_HDR_SPACE, 8, 0);
+	table->n_mysql_handles_opened = 1; /* for pin */
 
 	dict_mem_table_add_col(table, heap, "NAME", DATA_BINARY, 0, 0);
 	dict_mem_table_add_col(table, heap, "ID", DATA_BINARY, 0, 0);
@@ -336,6 +337,7 @@
 
 	/*-------------------------*/
 	table = dict_mem_table_create("SYS_COLUMNS", DICT_HDR_SPACE, 7, 0);
+	table->n_mysql_handles_opened = 1; /* for pin */
 
 	dict_mem_table_add_col(table, heap, "TABLE_ID", DATA_BINARY, 0, 0);
 	dict_mem_table_add_col(table, heap, "POS", DATA_INT, 0, 4);
@@ -368,6 +370,7 @@
 
 	/*-------------------------*/
 	table = dict_mem_table_create("SYS_INDEXES", DICT_HDR_SPACE, 7, 0);
+	table->n_mysql_handles_opened = 1; /* for pin */
 
 	dict_mem_table_add_col(table, heap, "TABLE_ID", DATA_BINARY, 0, 0);
 	dict_mem_table_add_col(table, heap, "ID", DATA_BINARY, 0, 0);
@@ -413,6 +416,7 @@
 
 	/*-------------------------*/
 	table = dict_mem_table_create("SYS_FIELDS", DICT_HDR_SPACE, 3, 0);
+	table->n_mysql_handles_opened = 1; /* for pin */
 
 	dict_mem_table_add_col(table, heap, "INDEX_ID", DATA_BINARY, 0, 0);
 	dict_mem_table_add_col(table, heap, "POS", DATA_INT, 0, 4);
--- a/storage/innobase/dict/dict0crea.c
+++ b/storage/innobase/dict/dict0crea.c
@@ -1209,6 +1209,9 @@
 		/* Foreign constraint system tables have already been
 		created, and they are ok */
 
+		table1->n_mysql_handles_opened = 1; /* for pin */
+		table2->n_mysql_handles_opened = 1; /* for pin */
+
 		mutex_exit(&(dict_sys->mutex));
 
 		return(DB_SUCCESS);
@@ -1290,6 +1293,11 @@
 
 	trx_commit_for_mysql(trx);
 
+	table1 = dict_table_get_low("SYS_FOREIGN");
+	table2 = dict_table_get_low("SYS_FOREIGN_COLS");
+	table1->n_mysql_handles_opened = 1; /* for pin */
+	table2->n_mysql_handles_opened = 1; /* for pin */
+
 	row_mysql_unlock_data_dictionary(trx);
 
 	trx_free_for_mysql(trx);
--- a/storage/innobase/dict/dict0dict.c
+++ b/storage/innobase/dict/dict0dict.c
@@ -626,6 +626,8 @@
 
 	table = dict_table_get_on_id_low(table_id);
 
+	dict_table_LRU_trim(table);
+
 	mutex_exit(&(dict_sys->mutex));
 
 	return(table);
@@ -744,6 +746,8 @@
 		table->n_mysql_handles_opened++;
 	}
 
+	dict_table_LRU_trim(table);
+
 	mutex_exit(&(dict_sys->mutex));
 
 	if (table != NULL) {
@@ -1264,6 +1268,64 @@
 	dict_mem_table_free(table);
 }
 
+/**************************************************************************
+Frees tables from the end of table_LRU if the dictionary cache occupies
+too much space. */
+UNIV_INTERN
+void
+dict_table_LRU_trim(
+/*================*/
+	dict_table_t*	self)
+{
+	dict_table_t*	table;
+	dict_table_t*	prev_table;
+	dict_foreign_t*	foreign;
+	ulint		n_removed;
+	ulint		n_have_parent;
+	ulint		cached_foreign_tables;
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+#endif /* UNIV_SYNC_DEBUG */
+
+retry:
+	n_removed = n_have_parent = 0;
+	table = UT_LIST_GET_LAST(dict_sys->table_LRU);
+
+	while ( srv_dict_size_limit && table
+		&& ((dict_sys->table_hash->n_cells
+		     + dict_sys->table_id_hash->n_cells) * sizeof(hash_cell_t)
+		    + dict_sys->size) > srv_dict_size_limit ) {
+		prev_table = UT_LIST_GET_PREV(table_LRU, table);
+
+		if (table == self || table->n_mysql_handles_opened)
+			goto next_loop;
+
+		cached_foreign_tables = 0;
+		foreign = UT_LIST_GET_FIRST(table->foreign_list);
+		while (foreign != NULL) {
+			if (foreign->referenced_table)
+				cached_foreign_tables++;
+			foreign = UT_LIST_GET_NEXT(foreign_list, foreign);
+		}
+
+		if (cached_foreign_tables == 0) {
+			dict_table_remove_from_cache(table);
+			n_removed++;
+		} else {
+			n_have_parent++;
+		}
+next_loop:
+		table = prev_table;
+	}
+
+	if ( srv_dict_size_limit && n_have_parent && n_removed
+		&& ((dict_sys->table_hash->n_cells
+		     + dict_sys->table_id_hash->n_cells) * sizeof(hash_cell_t)
+		    + dict_sys->size) > srv_dict_size_limit )
+		goto retry;
+}
+
 /****************************************************************//**
 If the given column name is reserved for InnoDB system columns, return
 TRUE.
@@ -1768,6 +1830,11 @@
 	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
 	ut_ad(mutex_own(&(dict_sys->mutex)));
 
+	/* remove all entry of the index from adaptive hash index,
+	because removing from adaptive hash index needs dict_index */
+	if (btr_search_enabled && srv_dict_size_limit)
+		btr_search_drop_page_hash_index_on_index(index);
+
 	/* We always create search info whether or not adaptive
 	hash index is enabled or not. */
 	info = index->search_info;
--- a/storage/innobase/handler/ha_innodb.cc
+++ b/storage/innobase/handler/ha_innodb.cc
@@ -677,6 +677,8 @@
   (char*) &export_vars.innodb_dblwr_pages_written,	  SHOW_LONG},
   {"dblwr_writes",
   (char*) &export_vars.innodb_dblwr_writes,		  SHOW_LONG},
+  {"dict_tables",
+  (char*) &export_vars.innodb_dict_tables,		  SHOW_LONG},
   {"have_atomic_builtins",
   (char*) &export_vars.innodb_have_atomic_builtins,	  SHOW_BOOL},
   {"log_waits",
@@ -11813,6 +11815,11 @@
   NULL, NULL, 0, 0, 1, 0);
 #endif
 
+static MYSQL_SYSVAR_ULONG(dict_size_limit, srv_dict_size_limit,
+  PLUGIN_VAR_RQCMDARG,
+  "Limit the allocated memory for dictionary cache. (0: unlimited)",
+  NULL, NULL, 0, 0, LONG_MAX, 0);
+
 static struct st_mysql_sys_var* innobase_system_variables[]= {
   MYSQL_SYSVAR(additional_mem_pool_size),
   MYSQL_SYSVAR(autoextend_increment),
@@ -11882,6 +11889,7 @@
   MYSQL_SYSVAR(flush_neighbor_pages),
   MYSQL_SYSVAR(read_ahead),
   MYSQL_SYSVAR(adaptive_flushing_method),
+  MYSQL_SYSVAR(dict_size_limit),
   MYSQL_SYSVAR(use_sys_malloc),
   MYSQL_SYSVAR(use_native_aio),
   MYSQL_SYSVAR(change_buffering),
--- a/storage/innobase/ibuf/ibuf0ibuf.c
+++ b/storage/innobase/ibuf/ibuf0ibuf.c
@@ -575,6 +575,7 @@
 
 	/* Use old-style record format for the insert buffer. */
 	table = dict_mem_table_create(IBUF_TABLE_NAME, IBUF_SPACE_ID, 1, 0);
+	table->n_mysql_handles_opened = 1; /* for pin */
 
 	dict_mem_table_add_col(table, heap, "DUMMY_COLUMN", DATA_BINARY, 0, 0);
 
--- a/storage/innobase/include/btr0sea.h
+++ b/storage/innobase/include/btr0sea.h
@@ -140,6 +140,13 @@
 				s- or x-latched, or an index page
 				for which we know that
 				block->buf_fix_count == 0 */
+/************************************************************************
+Drops a page hash index based on index */
+UNIV_INTERN
+void
+btr_search_drop_page_hash_index_on_index(
+/*=====================================*/
+	dict_index_t*	index);		/* in: record descriptor */
 /********************************************************************//**
 Drops a possible page hash index when a page is evicted from the buffer pool
 or freed in a file segment. */
--- a/storage/innobase/include/buf0buf.h
+++ b/storage/innobase/include/buf0buf.h
@@ -1595,6 +1595,15 @@
 #define BUF_POOL_ZIP_FOLD_BPAGE(b) BUF_POOL_ZIP_FOLD((buf_block_t*) (b))
 /* @} */
 
+/** A chunk of buffers.  The buffer pool is allocated in chunks. */
+struct buf_chunk_struct{
+	ulint		mem_size;	/*!< allocated size of the chunk */
+	ulint		size;		/*!< size of frames[] and blocks[] */
+	void*		mem;		/*!< pointer to the memory area which
+					was allocated for the frames */
+	buf_block_t*	blocks;		/*!< array of buffer control blocks */
+};
+
 /** @brief The buffer pool statistics structure. */
 struct buf_pool_stat_struct{
 	ulint	n_page_gets;	/*!< number of page gets performed;
--- a/storage/innobase/include/dict0dict.h
+++ b/storage/innobase/include/dict0dict.h
@@ -1200,6 +1200,12 @@
 /*====================================*/
 	dict_table_t*	table,	/*!< in: table */
 	const char*	name);	/*!< in: name of the index to find */
+
+UNIV_INTERN
+void
+dict_table_LRU_trim(
+/*================*/
+	dict_table_t*	self);
 /* Buffers for storing detailed information about the latest foreign key
 and unique key errors */
 extern FILE*	dict_foreign_err_file;
--- a/storage/innobase/include/dict0dict.ic
+++ b/storage/innobase/include/dict0dict.ic
@@ -825,6 +825,13 @@
 	HASH_SEARCH(name_hash, dict_sys->table_hash, table_fold,
 		    dict_table_t*, table, ut_ad(table->cached),
 		    !strcmp(table->name, table_name));
+
+	/* make young in table_LRU */
+	if (table) {
+		UT_LIST_REMOVE(table_LRU, dict_sys->table_LRU, table);
+		UT_LIST_ADD_FIRST(table_LRU, dict_sys->table_LRU, table);
+	}
+
 	return(table);
 }
 
@@ -918,6 +925,12 @@
 		table = dict_load_table_on_id(table_id);
 	}
 
+	/* make young in table_LRU */
+	if (table) {
+		UT_LIST_REMOVE(table_LRU, dict_sys->table_LRU, table);
+		UT_LIST_ADD_FIRST(table_LRU, dict_sys->table_LRU, table);
+	}
+
 	ut_ad(!table || table->cached);
 
 	/* TODO: should get the type information from MySQL */
--- a/storage/innobase/include/srv0srv.h
+++ b/storage/innobase/include/srv0srv.h
@@ -234,6 +234,7 @@
 extern ulint	srv_read_ahead;
 extern ulint	srv_adaptive_flushing_method;
 
+extern ulint	srv_dict_size_limit;
 /*-------------------------------------------*/
 
 extern ulint	srv_n_rows_inserted;
@@ -717,6 +718,7 @@
 	ulint innodb_data_writes;		/*!< I/O write requests */
 	ulint innodb_data_written;		/*!< Data bytes written */
 	ulint innodb_data_reads;		/*!< I/O read requests */
+	ulint innodb_dict_tables;
 	ulint innodb_buffer_pool_pages_total;	/*!< Buffer pool size */
 	ulint innodb_buffer_pool_pages_data;	/*!< Data pages */
 	ulint innodb_buffer_pool_pages_dirty;	/*!< Dirty data pages */
--- a/storage/innobase/srv/srv0srv.c
+++ b/storage/innobase/srv/srv0srv.c
@@ -417,6 +417,8 @@
 UNIV_INTERN ulint	srv_enable_unsafe_group_commit = 0; /* 0:disable 1:enable */
 UNIV_INTERN ulint	srv_read_ahead = 3; /* 1: random  2: linear  3: Both */
 UNIV_INTERN ulint	srv_adaptive_flushing_method = 0; /* 0: native  1: estimate  2: keep_average */
+
+UNIV_INTERN ulint	srv_dict_size_limit = 0;
 /*-------------------------------------------*/
 UNIV_INTERN ulong	srv_n_spin_wait_rounds	= 30;
 UNIV_INTERN ulong	srv_n_free_tickets_to_enter = 500;
@@ -2225,6 +2227,7 @@
 	export_vars.innodb_data_reads = os_n_file_reads;
 	export_vars.innodb_data_writes = os_n_file_writes;
 	export_vars.innodb_data_written = srv_data_written;
+	export_vars.innodb_dict_tables= (dict_sys ? UT_LIST_GET_LEN(dict_sys->table_LRU) : 0);
 	export_vars.innodb_buffer_pool_read_requests = stat.n_page_gets;
 	export_vars.innodb_buffer_pool_write_requests
 		= srv_buf_pool_write_requests;
--- /dev/null
+++ b/mysql-test/suite/sys_vars/r/innodb_dict_size_limit_basic.result
@@ -0,0 +1,3 @@
+SELECT @@global.innodb_dict_size_limit;
+@@global.innodb_dict_size_limit
+0
--- /dev/null
+++ b/mysql-test/suite/sys_vars/t/innodb_dict_size_limit_basic.test
@@ -0,0 +1 @@
+SELECT @@global.innodb_dict_size_limit;
Commit	Line	Data
b4e1fa2c AM	1	# name : innodb_dict_size_limit.patch
	2	# introduced : 11 or before
	3	# maintainer : Yasufumi
	4	#
	5	#!!! notice !!!
	6	# Any small change to this file in the main branch
	7	# should be done or reviewed by the maintainer!
db82db79 AM	8	--- a/storage/innobase/btr/btr0sea.c
db82db79 AM	9	+++ b/storage/innobase/btr/btr0sea.c
13ceb006	10	@@ -1187,6 +1187,178 @@
b4e1fa2c AM	11	mem_free(folds);
	12	}
	13
	14	+/************************************************************************
	15	+Drops a page hash index based on index */
	16	+UNIV_INTERN
	17	+void
	18	+btr_search_drop_page_hash_index_on_index(
	19	+/=====================================/
	20	+ dict_index_t* index) /* in: record descriptor */
	21	+{
d8778560	22	+
b4e1fa2c AM	23	+ hash_table_t* table;
	24	+ buf_block_t* block;
	25	+ ulint n_fields;
	26	+ ulint n_bytes;
	27	+ const page_t* page;
	28	+ const rec_t* rec;
	29	+ ulint fold;
	30	+ ulint prev_fold;
	31	+ index_id_t index_id;
	32	+ ulint n_cached;
	33	+ ulint n_recs;
	34	+ ulint* folds;
	35	+ ulint i, j;
	36	+ mem_heap_t* heap = NULL;
	37	+ ulint* offsets;
d8778560	38	+ ibool released_search_latch;
b4e1fa2c	39	+
d8778560	40	+ rw_lock_s_lock(&btr_search_latch);
b4e1fa2c AM	41	+
	42	+ table = btr_search_sys->hash_index;
	43	+
	44	+ for (j = 0; j < srv_buf_pool_instances; j++) {
	45	+ buf_pool_t* buf_pool;
	46	+
	47	+ buf_pool = buf_pool_from_array(j);
	48	+
d8778560 AM	49	+ do {
	50	+ buf_chunk_t* chunks = buf_pool->chunks;
	51	+ buf_chunk_t* chunk = chunks + buf_pool->n_chunks;
	52	+
	53	+ released_search_latch = FALSE;
	54	+
	55	+ while (--chunk >= chunks) {
	56	+ block = chunk->blocks;
	57	+ i = chunk->size;
	58	+
	59	+retry:
	60	+ for (; i--; block++) {
	61	+ if (buf_block_get_state(block)
	62	+ != BUF_BLOCK_FILE_PAGE
	63	+ \|\| block->index != index
13ceb006	64	+ \|\| !block->index) {
d8778560 AM	65	+ continue;
	66	+ }
	67	+
	68	+ page = block->frame;
	69	+
	70	+ /* from btr_search_drop_page_hash_index() */
	71	+ n_fields = block->curr_n_fields;
	72	+ n_bytes = block->curr_n_bytes;
	73	+
b4e1fa2c	74	+
d8778560 AM	75	+ /* keeping latch order */
	76	+ rw_lock_s_unlock(&btr_search_latch);
	77	+ released_search_latch = TRUE;
	78	+ rw_lock_x_lock(&block->lock);
b4e1fa2c	79	+
b4e1fa2c	80	+
d8778560	81	+ ut_a(n_fields + n_bytes > 0);
b4e1fa2c	82	+
d8778560	83	+ n_recs = page_get_n_recs(page);
b4e1fa2c	84	+
d8778560 AM	85	+ /* Calculate and cache fold values into an array for fast deletion
d8778560 AM	86	+ from the hash index */
b4e1fa2c	87	+
d8778560	88	+ folds = mem_alloc(n_recs * sizeof(ulint));
b4e1fa2c	89	+
d8778560	90	+ n_cached = 0;
b4e1fa2c	91	+
d8778560 AM	92	+ rec = page_get_infimum_rec(page);
d8778560 AM	93	+ rec = page_rec_get_next_low(rec, page_is_comp(page));
b4e1fa2c	94	+
d8778560	95	+ index_id = btr_page_get_index_id(page);
b4e1fa2c	96	+
d8778560	97	+ ut_a(index_id == index->id);
b4e1fa2c	98	+
d8778560	99	+ prev_fold = 0;
b4e1fa2c	100	+
d8778560	101	+ offsets = NULL;
b4e1fa2c	102	+
d8778560 AM	103	+ while (!page_rec_is_supremum(rec)) {
	104	+ offsets = rec_get_offsets(rec, index, offsets,
	105	+ n_fields + (n_bytes > 0), &heap);
	106	+ ut_a(rec_offs_n_fields(offsets) == n_fields + (n_bytes > 0));
	107	+ fold = rec_fold(rec, offsets, n_fields, n_bytes, index_id);
b4e1fa2c	108	+
d8778560	109	+ if (fold == prev_fold && prev_fold != 0) {
b4e1fa2c	110	+
d8778560 AM	111	+ goto next_rec;
d8778560 AM	112	+ }
b4e1fa2c	113	+
d8778560 AM	114	+ /* Remove all hash nodes pointing to this page from the
d8778560 AM	115	+ hash chain */
b4e1fa2c	116	+
d8778560 AM	117	+ folds[n_cached] = fold;
d8778560 AM	118	+ n_cached++;
b4e1fa2c	119	+next_rec:
d8778560 AM	120	+ rec = page_rec_get_next_low(rec, page_rec_is_comp(rec));
	121	+ prev_fold = fold;
	122	+ }
b4e1fa2c	123	+
d8778560 AM	124	+ if (UNIV_LIKELY_NULL(heap)) {
	125	+ mem_heap_empty(heap);
	126	+ }
b4e1fa2c	127	+
d8778560	128	+ rw_lock_x_lock(&btr_search_latch);
b4e1fa2c	129	+
13ceb006	130	+ if (UNIV_UNLIKELY(!block->index)) {
d8778560 AM	131	+ goto cleanup;
d8778560 AM	132	+ }
b4e1fa2c	133	+
d8778560 AM	134	+ ut_a(block->index == index);
	135	+
	136	+ if (UNIV_UNLIKELY(block->curr_n_fields != n_fields)
	137	+ \|\| UNIV_UNLIKELY(block->curr_n_bytes != n_bytes)) {
	138	+ rw_lock_x_unlock(&btr_search_latch);
	139	+ rw_lock_x_unlock(&block->lock);
	140	+
	141	+ mem_free(folds);
	142	+
	143	+ rw_lock_s_lock(&btr_search_latch);
	144	+ goto retry;
	145	+ }
	146	+
	147	+ for (i = 0; i < n_cached; i++) {
	148	+
	149	+ ha_remove_all_nodes_to_page(table, folds[i], page);
	150	+ }
	151	+
	152	+ ut_a(index->search_info->ref_count > 0);
	153	+ index->search_info->ref_count--;
	154	+
d8778560 AM	155	+ block->index = NULL;
	156	+
	157	+cleanup:
b4e1fa2c	158	+#if defined UNIV_AHI_DEBUG \|\| defined UNIV_DEBUG
d8778560 AM	159	+ if (UNIV_UNLIKELY(block->n_pointers)) {
	160	+ /* Corruption */
	161	+ ut_print_timestamp(stderr);
	162	+ fprintf(stderr,
	163	+"InnoDB: The adaptive hash index is corrupted. After dropping\n"
	164	+"InnoDB: the hash index to a page of %s, %lu hash nodes still remain.\n",
	165	+ index->name, (ulong) block->n_pointers);
	166	+ }
b4e1fa2c	167	+#endif /* UNIV_AHI_DEBUG \|\| UNIV_DEBUG */
d8778560 AM	168	+ rw_lock_x_unlock(&btr_search_latch);
d8778560 AM	169	+ rw_lock_x_unlock(&block->lock);
b4e1fa2c	170	+
d8778560	171	+ mem_free(folds);
b4e1fa2c	172	+
d8778560 AM	173	+ rw_lock_s_lock(&btr_search_latch);
	174	+ }
	175	+ }
	176	+ } while (released_search_latch);
b4e1fa2c AM	177	+ }
b4e1fa2c AM	178	+
d8778560	179	+ rw_lock_s_unlock(&btr_search_latch);
b4e1fa2c AM	180	+
	181	+ if (UNIV_LIKELY_NULL(heap)) {
	182	+ mem_heap_free(heap);
	183	+ }
	184	+}
	185	+
	186	/******************************************************************//
13ceb006 AM	187	Drops a possible page hash index when a page is evicted from the buffer pool
13ceb006 AM	188	or freed in a file segment. */
db82db79 AM	189	--- a/storage/innobase/buf/buf0buf.c
db82db79 AM	190	+++ b/storage/innobase/buf/buf0buf.c
d8778560 AM	191	@@ -294,14 +294,14 @@
	192	# endif /* !PFS_SKIP_BUFFER_MUTEX_RWLOCK */
	193	#endif /* UNIV_PFS_MUTEX \|\| UNIV_PFS_RWLOCK */
	194
	195	-/** A chunk of buffers. The buffer pool is allocated in chunks. */
	196	-struct buf_chunk_struct{
	197	- ulint mem_size; /!< allocated size of the chunk /
	198	- ulint size; /!< size of frames[] and blocks[] /
	199	- void* mem; /*!< pointer to the memory area which
	200	- was allocated for the frames */
	201	- buf_block_t* blocks; /!< array of buffer control blocks /
	202	-};
	203	+/** A chunk of buffers. The buffer pool is allocated in chunks. (moved to buf0buf.h)*/
	204	+//struct buf_chunk_struct{
	205	+// ulint mem_size; /!< allocated size of the chunk /
	206	+// ulint size; /!< size of frames[] and blocks[] /
	207	+// void* mem; /*!< pointer to the memory area which
	208	+// was allocated for the frames */
	209	+// buf_block_t* blocks; /!< array of buffer control blocks /
	210	+//};
	211	#endif /* !UNIV_HOTBACKUP */
	212
	213	/******************************************************************//
db82db79 AM	214	--- a/storage/innobase/dict/dict0boot.c
db82db79 AM	215	+++ b/storage/innobase/dict/dict0boot.c
b4e1fa2c AM	216	@@ -284,6 +284,7 @@
	217	system tables */
	218	/-------------------------/
	219	table = dict_mem_table_create("SYS_TABLES", DICT_HDR_SPACE, 8, 0);
	220	+ table->n_mysql_handles_opened = 1; /* for pin */
	221
	222	dict_mem_table_add_col(table, heap, "NAME", DATA_BINARY, 0, 0);
	223	dict_mem_table_add_col(table, heap, "ID", DATA_BINARY, 0, 0);
	224	@@ -336,6 +337,7 @@
	225
	226	/-------------------------/
	227	table = dict_mem_table_create("SYS_COLUMNS", DICT_HDR_SPACE, 7, 0);
	228	+ table->n_mysql_handles_opened = 1; /* for pin */
	229
	230	dict_mem_table_add_col(table, heap, "TABLE_ID", DATA_BINARY, 0, 0);
	231	dict_mem_table_add_col(table, heap, "POS", DATA_INT, 0, 4);
	232	@@ -368,6 +370,7 @@
	233
	234	/-------------------------/
	235	table = dict_mem_table_create("SYS_INDEXES", DICT_HDR_SPACE, 7, 0);
	236	+ table->n_mysql_handles_opened = 1; /* for pin */
	237
	238	dict_mem_table_add_col(table, heap, "TABLE_ID", DATA_BINARY, 0, 0);
	239	dict_mem_table_add_col(table, heap, "ID", DATA_BINARY, 0, 0);
	240	@@ -413,6 +416,7 @@
	241
	242	/-------------------------/
	243	table = dict_mem_table_create("SYS_FIELDS", DICT_HDR_SPACE, 3, 0);
	244	+ table->n_mysql_handles_opened = 1; /* for pin */
	245
	246	dict_mem_table_add_col(table, heap, "INDEX_ID", DATA_BINARY, 0, 0);
	247	dict_mem_table_add_col(table, heap, "POS", DATA_INT, 0, 4);
db82db79 AM	248	--- a/storage/innobase/dict/dict0crea.c
	249	+++ b/storage/innobase/dict/dict0crea.c
	250	@@ -1209,6 +1209,9 @@
b4e1fa2c AM	251	/* Foreign constraint system tables have already been
	252	created, and they are ok */
	253
	254	+ table1->n_mysql_handles_opened = 1; /* for pin */
	255	+ table2->n_mysql_handles_opened = 1; /* for pin */
	256	+
	257	mutex_exit(&(dict_sys->mutex));
	258
	259	return(DB_SUCCESS);
db82db79	260	@@ -1290,6 +1293,11 @@
b4e1fa2c AM	261
	262	trx_commit_for_mysql(trx);
	263
	264	+ table1 = dict_table_get_low("SYS_FOREIGN");
	265	+ table2 = dict_table_get_low("SYS_FOREIGN_COLS");
	266	+ table1->n_mysql_handles_opened = 1; /* for pin */
	267	+ table2->n_mysql_handles_opened = 1; /* for pin */
	268	+
	269	row_mysql_unlock_data_dictionary(trx);
	270
	271	trx_free_for_mysql(trx);
db82db79 AM	272	--- a/storage/innobase/dict/dict0dict.c
db82db79 AM	273	+++ b/storage/innobase/dict/dict0dict.c
adf0fb13	274	@@ -626,6 +626,8 @@
b4e1fa2c AM	275
	276	table = dict_table_get_on_id_low(table_id);
	277
	278	+ dict_table_LRU_trim(table);
	279	+
	280	mutex_exit(&(dict_sys->mutex));
	281
	282	return(table);
adf0fb13	283	@@ -744,6 +746,8 @@
b4e1fa2c AM	284	table->n_mysql_handles_opened++;
	285	}
	286
	287	+ dict_table_LRU_trim(table);
	288	+
	289	mutex_exit(&(dict_sys->mutex));
	290
	291	if (table != NULL) {
29ffd636	292	@@ -1264,6 +1268,64 @@
b4e1fa2c AM	293	dict_mem_table_free(table);
	294	}
	295
	296	+/**************************************************************************
	297	+Frees tables from the end of table_LRU if the dictionary cache occupies
	298	+too much space. */
	299	+UNIV_INTERN
	300	+void
	301	+dict_table_LRU_trim(
	302	+/================/
	303	+ dict_table_t* self)
	304	+{
	305	+ dict_table_t* table;
	306	+ dict_table_t* prev_table;
	307	+ dict_foreign_t* foreign;
	308	+ ulint n_removed;
	309	+ ulint n_have_parent;
	310	+ ulint cached_foreign_tables;
	311	+
	312	+#ifdef UNIV_SYNC_DEBUG
	313	+ ut_ad(mutex_own(&(dict_sys->mutex)));
	314	+#endif /* UNIV_SYNC_DEBUG */
	315	+
	316	+retry:
	317	+ n_removed = n_have_parent = 0;
	318	+ table = UT_LIST_GET_LAST(dict_sys->table_LRU);
	319	+
	320	+ while ( srv_dict_size_limit && table
	321	+ && ((dict_sys->table_hash->n_cells
	322	+ + dict_sys->table_id_hash->n_cells) * sizeof(hash_cell_t)
	323	+ + dict_sys->size) > srv_dict_size_limit ) {
	324	+ prev_table = UT_LIST_GET_PREV(table_LRU, table);
	325	+
	326	+ if (table == self \|\| table->n_mysql_handles_opened)
	327	+ goto next_loop;
	328	+
	329	+ cached_foreign_tables = 0;
	330	+ foreign = UT_LIST_GET_FIRST(table->foreign_list);
	331	+ while (foreign != NULL) {
	332	+ if (foreign->referenced_table)
	333	+ cached_foreign_tables++;
	334	+ foreign = UT_LIST_GET_NEXT(foreign_list, foreign);
	335	+ }
	336	+
	337	+ if (cached_foreign_tables == 0) {
	338	+ dict_table_remove_from_cache(table);
	339	+ n_removed++;
	340	+ } else {
	341	+ n_have_parent++;
	342	+ }
	343	+next_loop:
	344	+ table = prev_table;
	345	+ }
	346	+
	347	+ if ( srv_dict_size_limit && n_have_parent && n_removed
	348	+ && ((dict_sys->table_hash->n_cells
	349	+ + dict_sys->table_id_hash->n_cells) * sizeof(hash_cell_t)
	350	+ + dict_sys->size) > srv_dict_size_limit )
	351	+ goto retry;
	352	+}
	353	+
	354	/**************************************************************//
	355	If the given column name is reserved for InnoDB system columns, return
	356	TRUE.
29ffd636	357	@@ -1768,6 +1830,11 @@
b4e1fa2c AM	358	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
	359	ut_ad(mutex_own(&(dict_sys->mutex)));
	360
	361	+ /* remove all entry of the index from adaptive hash index,
	362	+ because removing from adaptive hash index needs dict_index */
	363	+ if (btr_search_enabled && srv_dict_size_limit)
	364	+ btr_search_drop_page_hash_index_on_index(index);
	365	+
	366	/* We always create search info whether or not adaptive
	367	hash index is enabled or not. */
	368	info = index->search_info;
db82db79 AM	369	--- a/storage/innobase/handler/ha_innodb.cc
db82db79 AM	370	+++ b/storage/innobase/handler/ha_innodb.cc
734d6226	371	@@ -677,6 +677,8 @@
b4e1fa2c AM	372	(char*) &export_vars.innodb_dblwr_pages_written, SHOW_LONG},
	373	{"dblwr_writes",
	374	(char*) &export_vars.innodb_dblwr_writes, SHOW_LONG},
	375	+ {"dict_tables",
	376	+ (char*) &export_vars.innodb_dict_tables, SHOW_LONG},
	377	{"have_atomic_builtins",
	378	(char*) &export_vars.innodb_have_atomic_builtins, SHOW_BOOL},
	379	{"log_waits",
29ffd636 AM	380	@@ -11813,6 +11815,11 @@
	381	NULL, NULL, 0, 0, 1, 0);
	382	#endif
b4e1fa2c AM	383
	384	+static MYSQL_SYSVAR_ULONG(dict_size_limit, srv_dict_size_limit,
	385	+ PLUGIN_VAR_RQCMDARG,
	386	+ "Limit the allocated memory for dictionary cache. (0: unlimited)",
	387	+ NULL, NULL, 0, 0, LONG_MAX, 0);
	388	+
	389	static struct st_mysql_sys_var* innobase_system_variables[]= {
	390	MYSQL_SYSVAR(additional_mem_pool_size),
	391	MYSQL_SYSVAR(autoextend_increment),
29ffd636	392	@@ -11882,6 +11889,7 @@
11822e22	393	MYSQL_SYSVAR(flush_neighbor_pages),
d8778560	394	MYSQL_SYSVAR(read_ahead),
b4e1fa2c	395	MYSQL_SYSVAR(adaptive_flushing_method),
b4e1fa2c AM	396	+ MYSQL_SYSVAR(dict_size_limit),
	397	MYSQL_SYSVAR(use_sys_malloc),
	398	MYSQL_SYSVAR(use_native_aio),
	399	MYSQL_SYSVAR(change_buffering),
db82db79 AM	400	--- a/storage/innobase/ibuf/ibuf0ibuf.c
db82db79 AM	401	+++ b/storage/innobase/ibuf/ibuf0ibuf.c
1bfc1981	402	@@ -575,6 +575,7 @@
b4e1fa2c AM	403
	404	/* Use old-style record format for the insert buffer. */
	405	table = dict_mem_table_create(IBUF_TABLE_NAME, IBUF_SPACE_ID, 1, 0);
	406	+ table->n_mysql_handles_opened = 1; /* for pin */
	407
	408	dict_mem_table_add_col(table, heap, "DUMMY_COLUMN", DATA_BINARY, 0, 0);
	409
db82db79 AM	410	--- a/storage/innobase/include/btr0sea.h
db82db79 AM	411	+++ b/storage/innobase/include/btr0sea.h
b4e1fa2c AM	412	@@ -140,6 +140,13 @@
	413	s- or x-latched, or an index page
	414	for which we know that
	415	block->buf_fix_count == 0 */
	416	+/************************************************************************
	417	+Drops a page hash index based on index */
	418	+UNIV_INTERN
	419	+void
	420	+btr_search_drop_page_hash_index_on_index(
	421	+/=====================================/
	422	+ dict_index_t* index); /* in: record descriptor */
	423	/******************************************************************//
13ceb006 AM	424	Drops a possible page hash index when a page is evicted from the buffer pool
13ceb006 AM	425	or freed in a file segment. */
db82db79 AM	426	--- a/storage/innobase/include/buf0buf.h
db82db79 AM	427	+++ b/storage/innobase/include/buf0buf.h
29ffd636	428	@@ -1595,6 +1595,15 @@
d8778560 AM	429	#define BUF_POOL_ZIP_FOLD_BPAGE(b) BUF_POOL_ZIP_FOLD((buf_block_t*) (b))
	430	/* @} */
	431
	432	+/** A chunk of buffers. The buffer pool is allocated in chunks. */
	433	+struct buf_chunk_struct{
	434	+ ulint mem_size; /!< allocated size of the chunk /
	435	+ ulint size; /!< size of frames[] and blocks[] /
	436	+ void* mem; /*!< pointer to the memory area which
	437	+ was allocated for the frames */
	438	+ buf_block_t* blocks; /!< array of buffer control blocks /
	439	+};
	440	+
	441	/** @brief The buffer pool statistics structure. */
	442	struct buf_pool_stat_struct{
	443	ulint n_page_gets; /*!< number of page gets performed;
db82db79 AM	444	--- a/storage/innobase/include/dict0dict.h
db82db79 AM	445	+++ b/storage/innobase/include/dict0dict.h
734d6226	446	@@ -1200,6 +1200,12 @@
b4e1fa2c AM	447	/====================================/
	448	dict_table_t* table, /!< in: table /
	449	const char* name); /!< in: name of the index to find /
	450	+
	451	+UNIV_INTERN
	452	+void
	453	+dict_table_LRU_trim(
	454	+/================/
	455	+ dict_table_t* self);
	456	/* Buffers for storing detailed information about the latest foreign key
	457	and unique key errors */
	458	extern FILE* dict_foreign_err_file;
db82db79 AM	459	--- a/storage/innobase/include/dict0dict.ic
db82db79 AM	460	+++ b/storage/innobase/include/dict0dict.ic
734d6226	461	@@ -825,6 +825,13 @@
b4e1fa2c AM	462	HASH_SEARCH(name_hash, dict_sys->table_hash, table_fold,
	463	dict_table_t*, table, ut_ad(table->cached),
	464	!strcmp(table->name, table_name));
	465	+
	466	+ /* make young in table_LRU */
	467	+ if (table) {
	468	+ UT_LIST_REMOVE(table_LRU, dict_sys->table_LRU, table);
	469	+ UT_LIST_ADD_FIRST(table_LRU, dict_sys->table_LRU, table);
	470	+ }
	471	+
	472	return(table);
	473	}
	474
734d6226	475	@@ -918,6 +925,12 @@
b4e1fa2c AM	476	table = dict_load_table_on_id(table_id);
	477	}
	478
	479	+ /* make young in table_LRU */
	480	+ if (table) {
	481	+ UT_LIST_REMOVE(table_LRU, dict_sys->table_LRU, table);
	482	+ UT_LIST_ADD_FIRST(table_LRU, dict_sys->table_LRU, table);
	483	+ }
	484	+
	485	ut_ad(!table \|\| table->cached);
	486
	487	/* TODO: should get the type information from MySQL */
db82db79 AM	488	--- a/storage/innobase/include/srv0srv.h
db82db79 AM	489	+++ b/storage/innobase/include/srv0srv.h
734d6226	490	@@ -234,6 +234,7 @@
11822e22	491	extern ulint srv_read_ahead;
b4e1fa2c AM	492	extern ulint srv_adaptive_flushing_method;
b4e1fa2c AM	493
b4e1fa2c AM	494	+extern ulint srv_dict_size_limit;
	495	/-------------------------------------------/
	496
	497	extern ulint srv_n_rows_inserted;
29ffd636	498	@@ -717,6 +718,7 @@
b4e1fa2c AM	499	ulint innodb_data_writes; /!< I/O write requests /
	500	ulint innodb_data_written; /!< Data bytes written /
	501	ulint innodb_data_reads; /!< I/O read requests /
	502	+ ulint innodb_dict_tables;
	503	ulint innodb_buffer_pool_pages_total; /!< Buffer pool size /
	504	ulint innodb_buffer_pool_pages_data; /!< Data pages /
	505	ulint innodb_buffer_pool_pages_dirty; /!< Dirty data pages /
db82db79 AM	506	--- a/storage/innobase/srv/srv0srv.c
db82db79 AM	507	+++ b/storage/innobase/srv/srv0srv.c
734d6226	508	@@ -417,6 +417,8 @@
11822e22 AM	509	UNIV_INTERN ulint srv_enable_unsafe_group_commit = 0; /* 0:disable 1:enable */
11822e22 AM	510	UNIV_INTERN ulint srv_read_ahead = 3; /* 1: random 2: linear 3: Both */
b4e1fa2c	511	UNIV_INTERN ulint srv_adaptive_flushing_method = 0; /* 0: native 1: estimate 2: keep_average */
11822e22	512	+
b4e1fa2c AM	513	+UNIV_INTERN ulint srv_dict_size_limit = 0;
	514	/-------------------------------------------/
	515	UNIV_INTERN ulong srv_n_spin_wait_rounds = 30;
	516	UNIV_INTERN ulong srv_n_free_tickets_to_enter = 500;
29ffd636	517	@@ -2225,6 +2227,7 @@
b4e1fa2c AM	518	export_vars.innodb_data_reads = os_n_file_reads;
	519	export_vars.innodb_data_writes = os_n_file_writes;
	520	export_vars.innodb_data_written = srv_data_written;
	521	+ export_vars.innodb_dict_tables= (dict_sys ? UT_LIST_GET_LEN(dict_sys->table_LRU) : 0);
	522	export_vars.innodb_buffer_pool_read_requests = stat.n_page_gets;
	523	export_vars.innodb_buffer_pool_write_requests
	524	= srv_buf_pool_write_requests;
13ceb006 AM	525	--- /dev/null
	526	+++ b/mysql-test/suite/sys_vars/r/innodb_dict_size_limit_basic.result
	527	@@ -0,0 +1,3 @@
	528	+SELECT @@global.innodb_dict_size_limit;
	529	+@@global.innodb_dict_size_limit
	530	+0
	531	--- /dev/null
	532	+++ b/mysql-test/suite/sys_vars/t/innodb_dict_size_limit_basic.test
	533	@@ -0,0 +1 @@
	534	+SELECT @@global.innodb_dict_size_limit;