1 files changed, 161 insertions, 79 deletions

diff --git a/storage/xtradb/srv/srv0srv.c b/storage/xtradb/srv/srv0srv.c
index 6e210071746..6edfbaa7755 100644
--- a/storage/xtradb/srv/srv0srv.c
+++ b/storage/xtradb/srv/srv0srv.c
@@ -58,6 +58,8 @@ Created 10/8/1995 Heikki Tuuri
 *******************************************************/
 
 /* Dummy comment */
+#include "m_string.h" /* for my_sys.h */
+#include "my_sys.h" /* DEBUG_SYNC_C */
 #include "srv0srv.h"
 
 #include "ut0mem.h"
@@ -181,7 +183,7 @@ UNIV_INTERN my_bool	srv_track_changed_pages = TRUE;
 
 UNIV_INTERN ib_uint64_t	srv_max_bitmap_file_size = 100 * 1024 * 1024;
 
-UNIV_INTERN ulonglong	srv_changed_pages_limit = 0;
+UNIV_INTERN ulonglong	srv_max_changed_pages = 0;
 
 /** When TRUE, fake change transcations take S rather than X row locks.
     When FALSE, row locks are not taken at all. */
@@ -312,58 +314,11 @@ UNIV_INTERN ulong srv_purge_batch_size = 20;
 /* the number of rollback segments to use */
 UNIV_INTERN ulong srv_rollback_segments = TRX_SYS_N_RSEGS;
 
-/* variable counts amount of data read in total (in bytes) */
-UNIV_INTERN ulint srv_data_read = 0;
-
 /* Internal setting for "innodb_stats_method". Decides how InnoDB treats
 NULL value when collecting statistics. By default, it is set to
 SRV_STATS_NULLS_EQUAL(0), ie. all NULL value are treated equal */
 ulong srv_innodb_stats_method = SRV_STATS_NULLS_EQUAL;
 
-/* here we count the amount of data written in total (in bytes) */
-UNIV_INTERN ulint srv_data_written = 0;
-
-/* the number of the log write requests done */
-UNIV_INTERN ulint srv_log_write_requests = 0;
-
-/* the number of physical writes to the log performed */
-UNIV_INTERN ulint srv_log_writes = 0;
-
-/* amount of data written to the log files in bytes */
-UNIV_INTERN ulint srv_os_log_written = 0;
-
-/* amount of writes being done to the log files */
-UNIV_INTERN ulint srv_os_log_pending_writes = 0;
-
-/* we increase this counter, when there we don't have enough space in the
-log buffer and have to flush it */
-UNIV_INTERN ulint srv_log_waits = 0;
-
-/* this variable counts the amount of times, when the doublewrite buffer
-was flushed */
-UNIV_INTERN ulint srv_dblwr_writes = 0;
-
-/* here we store the number of pages that have been flushed to the
-doublewrite buffer */
-UNIV_INTERN ulint srv_dblwr_pages_written = 0;
-
-/* in this variable we store the number of write requests issued */
-UNIV_INTERN ulint srv_buf_pool_write_requests = 0;
-
-/* here we store the number of times when we had to wait for a free page
-in the buffer pool. It happens when the buffer pool is full and we need
-to make a flush, in order to be able to read or create a page. */
-UNIV_INTERN ulint srv_buf_pool_wait_free = 0;
-
-/* variable to count the number of pages that were written from buffer
-pool to the disk */
-UNIV_INTERN ulint srv_buf_pool_flushed = 0;
-UNIV_INTERN ulint buf_lru_flush_page_count = 0;
-
-/** Number of buffer pool reads that led to the
-reading of a disk page */
-UNIV_INTERN ulint srv_buf_pool_reads = 0;
-
 /** Time in seconds between automatic buffer pool dumps */
 UNIV_INTERN uint srv_auto_lru_dump = 0;
 
@@ -406,6 +361,9 @@ UNIV_INTERN lint	srv_conc_n_threads	= 0;
 InnoDB */
 UNIV_INTERN ulint	srv_conc_n_waiting_threads = 0;
 
+/* print all user-level transactions deadlocks to mysqld stderr */
+UNIV_INTERN my_bool	srv_print_all_deadlocks = FALSE;
+
 typedef struct srv_conc_slot_struct	srv_conc_slot_t;
 struct srv_conc_slot_struct{
 	os_event_t			event;		/*!< event to wait */
@@ -451,6 +409,10 @@ UNIV_INTERN ulong	srv_sys_stats_root_page = 0;
 #endif
 
 UNIV_INTERN ibool	srv_use_doublewrite_buf	= TRUE;
+UNIV_INTERN ibool	srv_use_atomic_writes = FALSE;
+#ifdef HAVE_POSIX_FALLOCATE
+UNIV_INTERN ibool	srv_use_posix_fallocate = TRUE;
+#endif
 UNIV_INTERN ibool	srv_use_checksums = TRUE;
 UNIV_INTERN ibool	srv_fast_checksum = FALSE;
 
@@ -489,23 +451,83 @@ UNIV_INTERN ibool	srv_print_log_io		= FALSE;
 UNIV_INTERN ibool	srv_print_latch_waits		= FALSE;
 #endif /* UNIV_DEBUG */
 
-UNIV_INTERN ulint		srv_n_rows_inserted		= 0;
-UNIV_INTERN ulint		srv_n_rows_updated		= 0;
-UNIV_INTERN ulint		srv_n_rows_deleted		= 0;
-UNIV_INTERN ulint		srv_n_rows_read			= 0;
-
 static ulint	srv_n_rows_inserted_old		= 0;
 static ulint	srv_n_rows_updated_old		= 0;
 static ulint	srv_n_rows_deleted_old		= 0;
 static ulint	srv_n_rows_read_old		= 0;
 
-UNIV_INTERN ulint		srv_n_lock_deadlock_count	= 0;
-UNIV_INTERN ulint		srv_n_lock_wait_count		= 0;
-UNIV_INTERN ulint		srv_n_lock_wait_current_count	= 0;
-UNIV_INTERN ib_int64_t	srv_n_lock_wait_time		= 0;
-UNIV_INTERN ulint		srv_n_lock_max_wait_time	= 0;
+/* Ensure counters are on separate cache lines */
+
+#define CACHE_LINE_SIZE 64
+#define CACHE_ALIGNED __attribute__ ((aligned (CACHE_LINE_SIZE)))
+
+UNIV_INTERN byte
+counters_pad_start[CACHE_LINE_SIZE] __attribute__((unused)) = {0};
+
+UNIV_INTERN ulint		srv_n_rows_inserted CACHE_ALIGNED	= 0;
+UNIV_INTERN ulint		srv_n_rows_updated CACHE_ALIGNED	= 0;
+UNIV_INTERN ulint		srv_n_rows_deleted CACHE_ALIGNED	= 0;
+UNIV_INTERN ulint		srv_n_rows_read CACHE_ALIGNED		= 0;
+
+UNIV_INTERN ulint		srv_n_lock_deadlock_count CACHE_ALIGNED	= 0;
+UNIV_INTERN ulint		srv_n_lock_wait_count CACHE_ALIGNED	= 0;
+UNIV_INTERN ulint		srv_n_lock_wait_current_count CACHE_ALIGNED = 0;
+UNIV_INTERN ib_int64_t	srv_n_lock_wait_time CACHE_ALIGNED		= 0;
+UNIV_INTERN ulint		srv_n_lock_max_wait_time CACHE_ALIGNED	= 0;
 
-UNIV_INTERN ulint		srv_truncated_status_writes	= 0;
+UNIV_INTERN ulint		srv_truncated_status_writes CACHE_ALIGNED = 0;
+
+/* variable counts amount of data read in total (in bytes) */
+UNIV_INTERN ulint srv_data_read CACHE_ALIGNED			= 0;
+
+/* here we count the amount of data written in total (in bytes) */
+UNIV_INTERN ulint srv_data_written CACHE_ALIGNED		= 0;
+
+/* the number of the log write requests done */
+UNIV_INTERN ulint srv_log_write_requests CACHE_ALIGNED		= 0;
+
+/* the number of physical writes to the log performed */
+UNIV_INTERN ulint srv_log_writes CACHE_ALIGNED			= 0;
+
+/* amount of data written to the log files in bytes */
+UNIV_INTERN ulint srv_os_log_written CACHE_ALIGNED		= 0;
+
+/* amount of writes being done to the log files */
+UNIV_INTERN ulint srv_os_log_pending_writes CACHE_ALIGNED	= 0;
+
+/* we increase this counter, when there we don't have enough space in the
+log buffer and have to flush it */
+UNIV_INTERN ulint srv_log_waits CACHE_ALIGNED			= 0;
+
+/* this variable counts the amount of times, when the doublewrite buffer
+was flushed */
+UNIV_INTERN ulint srv_dblwr_writes CACHE_ALIGNED		= 0;
+
+/* here we store the number of pages that have been flushed to the
+doublewrite buffer */
+UNIV_INTERN ulint srv_dblwr_pages_written CACHE_ALIGNED		= 0;
+
+/* in this variable we store the number of write requests issued */
+UNIV_INTERN ulint srv_buf_pool_write_requests CACHE_ALIGNED	= 0;
+
+/* here we store the number of times when we had to wait for a free page
+in the buffer pool. It happens when the buffer pool is full and we need
+to make a flush, in order to be able to read or create a page. */
+UNIV_INTERN ulint srv_buf_pool_wait_free CACHE_ALIGNED		= 0;
+
+/** Number of buffer pool reads that led to the
+reading of a disk page */
+UNIV_INTERN ulint srv_buf_pool_reads CACHE_ALIGNED		= 0;
+
+/* variable to count the number of pages that were written from buffer
+pool to the disk */
+UNIV_INTERN ulint srv_buf_pool_flushed CACHE_ALIGNED		= 0;
+
+/* variable to count the number of LRU flushed pages */
+UNIV_INTERN ulint buf_lru_flush_page_count CACHE_ALIGNED	= 0;
+
+UNIV_INTERN byte
+counters_pad_end[CACHE_LINE_SIZE] __attribute__((unused)) = {0};
 
 /*
   Set the following to 0 if you want InnoDB to write messages on
@@ -1438,7 +1460,7 @@ retry:
 	ut_ad(!sync_thread_levels_nonempty_trx(trx->has_search_latch));
 #endif /* UNIV_SYNC_DEBUG */
 
-	if (innobase_get_slow_log() && trx->take_stats) {
+	if (UNIV_UNLIKELY(trx->take_stats)) {
 		ut_usectime(&sec, &ms);
 		start_time = (ib_uint64_t)sec * 1000000 + ms;
 	} else {
@@ -1453,7 +1475,7 @@ retry:
 
 	trx->op_info = "";
 
-	if (innobase_get_slow_log() && trx->take_stats && start_time) {
+	if (UNIV_UNLIKELY(start_time != 0)) {
 		ut_usectime(&sec, &ms);
 		finish_time = (ib_uint64_t)sec * 1000000 + ms;
 		trx->innodb_que_wait_timer += (ulint)(finish_time - start_time);
@@ -1764,6 +1786,10 @@ srv_suspend_mysql_thread(
 
 	trx = thr_get_trx(thr);
 
+	if (trx->mysql_thd != 0) {
+		DEBUG_SYNC_C("srv_suspend_mysql_thread_enter");
+	}
+
 	os_event_set(srv_lock_timeout_thread_event);
 
 	mutex_enter(&kernel_mutex);
@@ -2193,6 +2219,8 @@ srv_printf_innodb_monitor(
 		(long) srv_conc_n_threads,
 		(ulong) srv_conc_n_waiting_threads);
 
+	mutex_enter(&kernel_mutex);
+
 	fprintf(file, "%lu read views open inside InnoDB\n",
 		UT_LIST_GET_LEN(trx_sys->view_list));
 
@@ -2206,6 +2234,8 @@ srv_printf_innodb_monitor(
 		}
 	}
 
+	mutex_exit(&kernel_mutex);
+
 	n_reserved = fil_space_get_n_reserved_extents(0);
 	if (n_reserved > 0) {
 		fprintf(file,
@@ -2290,16 +2320,18 @@ void
 srv_export_innodb_status(void)
 /*==========================*/
 {
-	buf_pool_stat_t	stat;
-	ulint		LRU_len;
-	ulint		free_len;
-	ulint		flush_list_len;
-	ulint		mem_adaptive_hash, mem_dictionary;
-	read_view_t*	oldest_view;
-	ulint		i;
+	buf_pool_stat_t		stat;
+	buf_pools_list_size_t	buf_pools_list_size;
+	ulint			LRU_len;
+	ulint			free_len;
+	ulint			flush_list_len;
+	ulint			mem_adaptive_hash, mem_dictionary;
+	read_view_t*		oldest_view;
+	ulint			i;
 
 	buf_get_total_stat(&stat);
 	buf_get_total_list_len(&LRU_len, &free_len, &flush_list_len);
+	buf_get_total_list_size_in_bytes(&buf_pools_list_size);
 
 	if (btr_search_sys && btr_search_sys->hash_index[0]->heap) {
 		mem_adaptive_hash = mem_heap_get_size(btr_search_sys->hash_index[0]->heap);
@@ -2364,7 +2396,12 @@ srv_export_innodb_status(void)
 	export_vars.innodb_buffer_pool_read_ahead_evicted
 		= stat.n_ra_pages_evicted;
 	export_vars.innodb_buffer_pool_pages_data = LRU_len;
+	export_vars.innodb_buffer_pool_bytes_data =
+		buf_pools_list_size.LRU_bytes
+		+ buf_pools_list_size.unzip_LRU_bytes;
 	export_vars.innodb_buffer_pool_pages_dirty = flush_list_len;
+	export_vars.innodb_buffer_pool_bytes_dirty =
+		buf_pools_list_size.flush_list_bytes;
 	export_vars.innodb_buffer_pool_pages_free = free_len;
 	export_vars.innodb_deadlocks = srv_n_lock_deadlock_count;
 #ifdef UNIV_DEBUG
@@ -2497,6 +2534,23 @@ srv_export_innodb_status(void)
 	export_vars.innodb_rows_deleted = srv_n_rows_deleted;
 	export_vars.innodb_truncated_status_writes = srv_truncated_status_writes;
 
+#ifdef UNIV_DEBUG
+	if (trx_sys->max_trx_id < purge_sys->done_trx_no) {
+		export_vars.innodb_purge_trx_id_age = 0;
+	} else {
+		export_vars.innodb_purge_trx_id_age =
+		  trx_sys->max_trx_id - purge_sys->done_trx_no;
+	}
+
+	if (!purge_sys->view
+	    || trx_sys->max_trx_id < purge_sys->view->up_limit_id) {
+		export_vars.innodb_purge_view_trx_id_age = 0;
+	} else {
+		export_vars.innodb_purge_view_trx_id_age =
+		  trx_sys->max_trx_id - purge_sys->view->up_limit_id;
+	}
+#endif /* UNIV_DEBUG */
+
 	mutex_exit(&srv_innodb_monitor_mutex);
 }
 
@@ -3080,11 +3134,19 @@ srv_redo_log_follow_thread(
 		os_event_reset(srv_checkpoint_completed_event);
 
 		if (srv_shutdown_state < SRV_SHUTDOWN_LAST_PHASE) {
-			log_online_follow_redo_log();
+			if (!log_online_follow_redo_log()) {
+				/* TODO: sync with I_S log tracking status? */
+				fprintf(stderr,
+					"InnoDB: Error: log tracking bitmap "
+					"write failed, stopping log tracking "
+					"thread!\n");
+				break;
+			}
 		}
 
 	} while (srv_shutdown_state < SRV_SHUTDOWN_LAST_PHASE);
 
+	srv_track_changed_pages = FALSE;
 	log_online_read_shutdown();
 	os_event_set(srv_redo_log_thread_finished_event);
 
@@ -3329,6 +3391,26 @@ loop:
 	for (i = 0; i < 10; i++) {
 		ulint	cur_time = ut_time_ms();
 
+#ifdef UNIV_DEBUG
+		if (btr_cur_limit_optimistic_insert_debug
+		    && srv_n_purge_threads == 0) {
+			/* If btr_cur_limit_optimistic_insert_debug is enabled
+			and no purge_threads, purge opportunity is increased
+			by x100 (1purge/100msec), to speed up debug scripts
+			which should wait for purged. */
+			next_itr_time -= 900;
+
+			srv_main_thread_op_info = "master purging";
+
+			srv_master_do_purge();
+
+			if (srv_fast_shutdown && srv_shutdown_state > 0) {
+
+				goto background_loop;
+			}
+		}
+#endif /* UNIV_DEBUG */
+
 		n_pages_flushed = 0; /* initialize */
 
 		/* ALTER TABLE in MySQL requires on Unix that the table handler
@@ -3487,8 +3569,7 @@ loop:
 
 						buf_pool = buf_pool_from_array(j);
 
-						/* The scanning flush_list is optimistic here */
-
+						buf_flush_list_mutex_enter(buf_pool);
 						level = 0;
 						n_blocks = 0;
 						bpage = UT_LIST_GET_FIRST(buf_pool->flush_list);
@@ -3502,6 +3583,7 @@ loop:
 							bpage = UT_LIST_GET_NEXT(flush_list, bpage);
 							n_blocks++;
 						}
+						buf_flush_list_mutex_exit(buf_pool);
 
 						if (level) {
 							bpl += ((ib_uint64_t) n_blocks * n_blocks
@@ -3567,30 +3649,25 @@ retry_flush_batch:
 
 				/* prev_flush_info[j] should be the previous loop's */
 				for (j = 0; j < srv_buf_pool_instances; j++) {
-					lint	blocks_num, new_blocks_num, flushed_blocks_num;
-					ibool	found;
+					lint	blocks_num, new_blocks_num = 0;
+					lint	flushed_blocks_num;
 
 					buf_pool = buf_pool_from_array(j);
+					buf_flush_list_mutex_enter(buf_pool);
 
 					blocks_num = UT_LIST_GET_LEN(buf_pool->flush_list);
 					bpage = UT_LIST_GET_FIRST(buf_pool->flush_list);
-					new_blocks_num = 0;
 
-					found = FALSE;
 					while (bpage != NULL) {
 						if (prev_flush_info[j].space == bpage->space
 						    && prev_flush_info[j].offset == bpage->offset
 						    && prev_flush_info[j].oldest_modification
 								== bpage->oldest_modification) {
-							found = TRUE;
 							break;
 						}
 						bpage = UT_LIST_GET_NEXT(flush_list, bpage);
 						new_blocks_num++;
 					}
-					if (!found) {
-						new_blocks_num = blocks_num;
-					}
 
 					flushed_blocks_num = new_blocks_num + prev_flush_info[j].count
 								- blocks_num;
@@ -3605,7 +3682,9 @@ retry_flush_batch:
 						prev_flush_info[j].space = bpage->space;
 						prev_flush_info[j].offset = bpage->offset;
 						prev_flush_info[j].oldest_modification = bpage->oldest_modification;
+						buf_flush_list_mutex_exit(buf_pool);
 					} else {
+						buf_flush_list_mutex_exit(buf_pool);
 						prev_flush_info[j].space = 0;
 						prev_flush_info[j].offset = 0;
 						prev_flush_info[j].oldest_modification = 0;
@@ -3631,6 +3710,7 @@ retry_flush_batch:
 				/* store previous first pages of the flush_list */
 				for (j = 0; j < srv_buf_pool_instances; j++) {
 					buf_pool = buf_pool_from_array(j);
+					buf_flush_list_mutex_enter(buf_pool);
 
 					bpage = UT_LIST_GET_FIRST(buf_pool->flush_list);
 
@@ -3639,7 +3719,9 @@ retry_flush_batch:
 						prev_flush_info[j].space = bpage->space;
 						prev_flush_info[j].offset = bpage->offset;
 						prev_flush_info[j].oldest_modification = bpage->oldest_modification;
+						buf_flush_list_mutex_exit(buf_pool);
 					} else {
+						buf_flush_list_mutex_exit(buf_pool);
 						prev_flush_info[j].space = 0;
 						prev_flush_info[j].offset = 0;
 						prev_flush_info[j].oldest_modification = 0;