sync with extensions rev29

buf/buf0buf.c

@@ -1860,7 +1860,6 @@ buf_page_get_gen(
 	mtr_t*		mtr)	/* in: mini-transaction */
 {
 	buf_block_t*	block;
-	buf_page_t*	bpage= NULL;
 	ibool		accessed;
 	ulint		fix_type;
 	ibool		must_read;
@@ -1883,9 +1882,7 @@ loop:
 	//buf_pool_mutex_enter();
 
 	if (block) {
-		bpage = &block->page;
-
-		block_mutex = buf_page_get_mutex(bpage);
+		block_mutex = buf_page_get_mutex((buf_page_t*)block);
 		mutex_enter(block_mutex);
 
 		/* If the guess is a compressed page descriptor that
@@ -1914,8 +1911,7 @@ loop:
 		mutex_enter(&page_hash_mutex);
 		block = (buf_block_t*) buf_page_hash_get(space, offset);
 		if (block) {
-			bpage = &block->page;
-			block_mutex = buf_page_get_mutex(bpage);
+			block_mutex = buf_page_get_mutex((buf_page_t*)block);
 			mutex_enter(block_mutex);
 		}
 		mutex_exit(&page_hash_mutex);
@@ -1984,12 +1980,11 @@ wait_until_unfixed:
 
 		block = buf_LRU_get_free_block(0);
 		ut_a(block);
-                bpage = &block->page;
-		block_mutex = buf_page_get_mutex(bpage);
+		block_mutex = buf_page_get_mutex((buf_page_t*)block);
 
 		//buf_pool_mutex_enter();
-		mutex_enter(&flush_list_mutex);
 		mutex_enter(&LRU_list_mutex);
+		mutex_enter(&flush_list_mutex);
 		mutex_enter(&page_hash_mutex);
 		mutex_enter(block_mutex);
 
@@ -2006,8 +2001,7 @@ wait_until_unfixed:
 				mutex_exit(block_mutex);
 
 				block = (buf_block_t*) hash_bpage;
-                		bpage = &block->page;
-				block_mutex = buf_page_get_mutex(bpage);
+				block_mutex = buf_page_get_mutex((buf_page_t*)block);
 
 				mutex_enter(block_mutex);
 				mutex_exit(&page_hash_mutex);

buf/buf0flu.c

@@ -1131,9 +1131,15 @@ flush_next:
 
 				old_page_count = page_count;
 
+				if (srv_flush_neighbor_pages) {
 				/* Try to flush also all the neighbors */
 				page_count += buf_flush_try_neighbors(
 					space, offset, flush_type);
+				} else {
+				/* Try to flush the page only */
+				page_count += buf_flush_try_page(
+					space, offset, flush_type);
+				}
 				/* fprintf(stderr,
 				"Flush type %lu, page no %lu, neighb %lu\n",
 				flush_type, offset,

handler/ha_innodb.cc

@@ -2073,6 +2073,8 @@ innobase_init(
 	srv_n_read_io_threads = (ulint) innobase_read_io_threads;
 	srv_n_write_io_threads = (ulint) innobase_write_io_threads;
 
+	srv_read_ahead &= 3;
+
 	srv_force_recovery = (ulint) innobase_force_recovery;
 
 	srv_use_doublewrite_buf = (ibool) innobase_use_doublewrite;
@@ -9586,14 +9588,62 @@ static MYSQL_SYSVAR_ULONG(io_capacity, srv_io_capacity,
   "Number of IO operations per second the server can do. Tunes background IO rate.",
   NULL, NULL, 100, 100, 999999999, 0);
 
-static MYSQL_SYSVAR_ULONG(read_ahead, srv_read_ahead,
+static MYSQL_SYSVAR_LONGLONG(ibuf_max_size, srv_ibuf_max_size,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "The maximum size of the insert buffer. (in bytes)",
+  NULL, NULL, LONGLONG_MAX, 0, LONGLONG_MAX, 0);
+
+static MYSQL_SYSVAR_ULONG(ibuf_active_contract, srv_ibuf_active_contract,
+  PLUGIN_VAR_RQCMDARG,
+  "Enable/Disable active_contract of insert buffer. 0:disable 1:enable",
+  NULL, NULL, 0, 0, 1, 0);
+
+static MYSQL_SYSVAR_ULONG(ibuf_accel_rate, srv_ibuf_accel_rate,
+  PLUGIN_VAR_RQCMDARG,
+  "Tunes amount of insert buffer processing of background, in addition to innodb_io_capacity. (in percentage)",
+  NULL, NULL, 100, 100, 999999999, 0);
+
+static MYSQL_SYSVAR_ULONG(flush_neighbor_pages, srv_flush_neighbor_pages,
+  PLUGIN_VAR_RQCMDARG,
+  "Enable/Disable flushing also neighbor pages. 0:disable 1:enable",
+  NULL, NULL, 1, 0, 1, 0);
+
+static
+void
+innodb_read_ahead_update(
+  THD* thd,
+  struct st_mysql_sys_var*     var,
+  void*        var_ptr,
+  const void*  save)
+{
+  *(long *)var_ptr= (*(long *)save) & 3;
+}
+const char *read_ahead_names[]=
+{
+  "none", /* 0 */
+  "random",
+  "linear",
+  "both", /* 3 */
+  /* For compatibility of the older patch */
+  "0", /* 4 ("none" + 4) */
+  "1",
+  "2",
+  "3", /* 7 ("both" + 4) */
+  NullS
+};
+TYPELIB read_ahead_typelib=
+{
+  array_elements(read_ahead_names) - 1, "read_ahead_typelib",
+  read_ahead_names, NULL
+};
+static MYSQL_SYSVAR_ENUM(read_ahead, srv_read_ahead,
   PLUGIN_VAR_RQCMDARG,
-  "Enable/Diasable read aheads bit0:random bit1:linear",
-  NULL, NULL, 3, 0, 3, 0);
+  "Control read ahead activity. (none, random, linear, [both])",
+  NULL, innodb_read_ahead_update, 3, &read_ahead_typelib);
 
 static MYSQL_SYSVAR_ULONG(adaptive_checkpoint, srv_adaptive_checkpoint,
   PLUGIN_VAR_RQCMDARG,
-  "Enable/Disable flushing along modified age 0:disable 1:enable",
+  "Enable/Disable flushing along modified age. 0:disable 1:enable",
   NULL, NULL, 0, 0, 1, 0);
 
 static MYSQL_SYSVAR_ULONG(read_io_threads, innobase_read_io_threads,
@@ -9656,6 +9706,10 @@ static struct st_mysql_sys_var* innobase_system_variables[]= {
   MYSQL_SYSVAR(show_locks_held),
   MYSQL_SYSVAR(version),
   MYSQL_SYSVAR(io_capacity),
+  MYSQL_SYSVAR(ibuf_max_size),
+  MYSQL_SYSVAR(ibuf_active_contract),
+  MYSQL_SYSVAR(ibuf_accel_rate),
+  MYSQL_SYSVAR(flush_neighbor_pages),
   MYSQL_SYSVAR(read_ahead),
   MYSQL_SYSVAR(adaptive_checkpoint),
   MYSQL_SYSVAR(read_io_threads),

ibuf/ibuf0ibuf.c

@@ -382,8 +382,10 @@ ibuf_init_at_db_start(void)
 	grow in size, as the references on the upper levels of the tree can
 	change */
 
-	ibuf->max_size = buf_pool_get_curr_size() / UNIV_PAGE_SIZE
-		/ IBUF_POOL_SIZE_PER_MAX_SIZE;
+	ibuf->max_size = ut_min( buf_pool_get_curr_size() / UNIV_PAGE_SIZE
+		/ IBUF_POOL_SIZE_PER_MAX_SIZE, (ulint) srv_ibuf_max_size / UNIV_PAGE_SIZE);
+
+	srv_ibuf_max_size = (long long) ibuf->max_size * UNIV_PAGE_SIZE;
 
 	UT_LIST_INIT(ibuf->data_list);
 
@@ -2351,11 +2353,13 @@ ibuf_contract_after_insert(
 
 	mutex_enter(&ibuf_mutex);
 
+	if (!srv_ibuf_active_contract) {
 	if (ibuf->size < ibuf->max_size + IBUF_CONTRACT_ON_INSERT_NON_SYNC) {
 		mutex_exit(&ibuf_mutex);
 
 		return;
 	}
+	}
 
 	sync = FALSE;
 

include/srv0srv.h

@@ -153,6 +153,10 @@ extern ulong	srv_max_purge_lag;
 extern ulong	srv_replication_delay;
 
 extern ulint	srv_io_capacity;
+extern long long	srv_ibuf_max_size;
+extern ulint	srv_ibuf_active_contract;
+extern ulint	srv_ibuf_accel_rate;
+extern ulint	srv_flush_neighbor_pages;
 extern ulint	srv_read_ahead;
 extern ulint	srv_adaptive_checkpoint;
 

srv/srv0srv.c

@@ -153,8 +153,8 @@ UNIV_INTERN ulint	srv_mem_pool_size	= ULINT_MAX;
 UNIV_INTERN ulint	srv_lock_table_size	= ULINT_MAX;
 
 UNIV_INTERN ulint	srv_n_file_io_threads	= ULINT_MAX;
-ulint	srv_n_read_io_threads	= 1;
-ulint	srv_n_write_io_threads	= 1;
+UNIV_INTERN ulint	srv_n_read_io_threads	= 1;
+UNIV_INTERN ulint	srv_n_write_io_threads	= 1;
 
 #ifdef UNIV_LOG_ARCHIVE
 UNIV_INTERN ibool		srv_log_archive_on	= FALSE;
@@ -319,15 +319,22 @@ UNIV_INTERN int	srv_query_thread_priority = 0;
 
 UNIV_INTERN ulong	srv_replication_delay		= 0;
 
-ulint	srv_io_capacity = 100;
+UNIV_INTERN ulint	srv_io_capacity = 100;
 
 /* Returns the number of IO operations that is X percent of the capacity.
 PCT_IO(5) -> returns the number of IO operations that is 5% of the max
 where max is srv_io_capacity. */
 #define PCT_IO(pct) ((ulint) (srv_io_capacity * ((double) pct / 100.0)))
 
-ulint	srv_read_ahead = 3; /* 1: random  2: linear  3: Both */
-ulint	srv_adaptive_checkpoint = 0; /* 0:disable 1:enable */
+UNIV_INTERN long long	srv_ibuf_max_size = 0;
+UNIV_INTERN ulint	srv_ibuf_active_contract = 0; /* 0:disable 1:enable */
+UNIV_INTERN ulint	srv_ibuf_accel_rate = 100;
+#define PCT_IBUF_IO(pct) ((ulint) (srv_io_capacity * srv_ibuf_accel_rate * ((double) pct / 10000.0)))
+
+UNIV_INTERN ulint	srv_flush_neighbor_pages = 1; /* 0:disable 1:enable */
+
+UNIV_INTERN ulint	srv_read_ahead = 3; /* 1: random  2: linear  3: Both */
+UNIV_INTERN ulint	srv_adaptive_checkpoint = 0; /* 0:disable 1:enable */
 /*-------------------------------------------*/
 UNIV_INTERN ulong	srv_n_spin_wait_rounds	= 20;
 UNIV_INTERN ulong	srv_n_free_tickets_to_enter = 500;
@@ -2395,7 +2402,7 @@ loop:
 		if (n_pend_ios < 3 && (n_ios - n_ios_old < PCT_IO(5))) {
 			srv_main_thread_op_info = "doing insert buffer merge";
 			ibuf_contract_for_n_pages(
-				TRUE, PCT_IO((srv_insert_buffer_batch_size / 4)));
+				TRUE, PCT_IBUF_IO((srv_insert_buffer_batch_size / 4)));
 
 			srv_main_thread_op_info = "flushing log";
 
@@ -2431,6 +2438,11 @@ loop:
 
 			} else {
 				if ((log_sys->lsn - oldest_lsn)
+				    > (log_sys->max_checkpoint_age) - ((log_sys->max_checkpoint_age) / 8)) {
+					/* LOG_POOL_PREFLUSH_RATIO_ASYNC is exceeded. */
+					/* We should not flush from here. */
+					mutex_exit(&(log_sys->mutex));
+				} else if ((log_sys->lsn - oldest_lsn)
 				    > (log_sys->max_checkpoint_age) - ((log_sys->max_checkpoint_age) / 4)) {
 
 					/* 2nd defence line (max_checkpoint_age * 3/4) */
@@ -2494,7 +2506,7 @@ loop:
 	even if the server were active */
 
 	srv_main_thread_op_info = "doing insert buffer merge";
-	ibuf_contract_for_n_pages(TRUE, PCT_IO((srv_insert_buffer_batch_size / 4)));
+	ibuf_contract_for_n_pages(TRUE, PCT_IBUF_IO((srv_insert_buffer_batch_size / 4)));
 
 	srv_main_thread_op_info = "flushing log";
 	log_buffer_flush_to_disk();
@@ -2629,7 +2641,7 @@ background_loop:
 		n_bytes_merged = 0;
 	} else {
 		n_bytes_merged = ibuf_contract_for_n_pages(
-			TRUE, PCT_IO((srv_insert_buffer_batch_size * 5)));
+			TRUE, PCT_IBUF_IO((srv_insert_buffer_batch_size * 5)));
 	}
 
 	srv_main_thread_op_info = "reserving kernel mutex";