Many files:

Remove potential starvation of a full log buffer flush: only flush up to the lsn which was the largest at the time when we requested the full log buffer flush
os0sync.h, os0sync.c:
  Fix a bug in os_event on Unix: even though we signaled the event, some threads could continue waiting if the event became nonsignaled quickly again; this made group commit less efficient than it should be

innobase/btr/btr0btr.c

@@ -822,9 +822,16 @@ btr_page_reorganize_low(
 {
 	page_t*	new_page;
 	ulint	log_mode;
+	ulint	data_size1;
+	ulint	data_size2;
+	ulint	max_ins_size1;
+	ulint	max_ins_size2;
 
 	ut_ad(mtr_memo_contains(mtr, buf_block_align(page),
 			      				MTR_MEMO_PAGE_X_FIX));
+	data_size1 = page_get_data_size(page);
+	max_ins_size1 = page_get_max_insert_size_after_reorganize(page, 1);
+
 	/* Write the log record */
 	mlog_write_initial_log_record(page, MLOG_PAGE_REORGANIZE, mtr);
 
@@ -859,6 +866,19 @@ btr_page_reorganize_low(
 		lock_move_reorganize_page(page, new_page);
 	}
 
+	data_size2 = page_get_data_size(page);
+	max_ins_size2 = page_get_max_insert_size_after_reorganize(page, 1);
+
+	if (data_size1 != data_size2 || max_ins_size1 != max_ins_size2) {
+		buf_page_print(page);
+		buf_page_print(new_page);
+	        fprintf(stderr,
+"InnoDB: Error: page old data size %lu new data size %lu\n"
+"InnoDB: Error: page old max ins size %lu new max ins size %lu\n"
+"InnoDB: Make a detailed bug report and send it to mysql@lists.mysql.com\n",
+			data_size1, data_size2, max_ins_size1, max_ins_size2);
+	}
+
 	buf_frame_free(new_page);
 
 	/* Restore logging mode */
@@ -1945,11 +1965,20 @@ btr_compress(
 
 		btr_page_reorganize(merge_page, mtr);
 
+		max_ins_size = page_get_max_insert_size(merge_page, n_recs);
+
 		ut_ad(page_validate(merge_page, cursor->index));
 		ut_ad(page_get_max_insert_size(merge_page, n_recs)
 							== max_ins_size_reorg);
 	}
 
+	if (data_size > max_ins_size) {
+
+		/* Add fault tolerance, though this should never happen */
+
+		return;
+	}
+
 	btr_search_drop_page_hash_index(page);
 
 	/* Remove the page from the level list */

innobase/include/log0log.h

@@ -173,6 +173,12 @@ log_write_up_to(
 			/* in: TRUE if we want the written log also to be
 			flushed to disk */
 /********************************************************************
+Does a syncronous flush of the log buffer to disk. */
+
+void
+log_buffer_flush_to_disk(void);
+/*==========================*/
+/********************************************************************
 Advances the smallest lsn for which there are unflushed dirty blocks in the
 buffer pool and also may make a new checkpoint. NOTE: this function may only
 be called if the calling thread owns no synchronization objects! */

innobase/include/os0sync.h

@@ -36,8 +36,12 @@ typedef os_event_struct_t*     os_event_t;
 struct os_event_struct {
 	os_fast_mutex_t	os_mutex;	/* this mutex protects the next
 					fields */
-	ibool		is_set;		/* this is TRUE if the next mutex is
+	ibool		is_set;		/* this is TRUE when the event is
-					not reserved */
+					in the signaled state, i.e., a thread
+					does not stop if it tries to wait for
+					this event */
+	ib_longlong	signal_count;	/* this is incremented each time
+					the event becomes signaled */
 	pthread_cond_t	cond_var;	/* condition variable is used in
 					waiting for the event */
 	UT_LIST_NODE_T(os_event_struct_t) os_event_list;

innobase/log/log0log.c

@@ -178,7 +178,8 @@ loop:
 
 		/* Not enough free space, do a syncronous flush of the log
 		buffer */
-		log_write_up_to(ut_dulint_max, LOG_WAIT_ALL_GROUPS, TRUE);
+
+		log_buffer_flush_to_disk();
 
 		count++;
 
@@ -1364,6 +1365,24 @@ do_waits:
 	}			
 }
 
+/********************************************************************
+Does a syncronous flush of the log buffer to disk. */
+
+void
+log_buffer_flush_to_disk(void)
+/*==========================*/
+{
+	dulint	lsn;
+
+	mutex_enter(&(log_sys->mutex));
+
+	lsn = log_sys->lsn;
+
+	mutex_exit(&(log_sys->mutex));
+
+	log_write_up_to(lsn, LOG_WAIT_ALL_GROUPS, TRUE);
+}
+
 /********************************************************************
 Tries to establish a big enough margin of free space in the log buffer, such
 that a new log entry can be catenated without an immediate need for a flush. */
@@ -1374,6 +1393,7 @@ log_flush_margin(void)
 {
 	ibool	do_flush	= FALSE;
 	log_t*	log		= log_sys;
+	dulint	lsn;
 
 	mutex_enter(&(log->mutex));
 
@@ -1384,13 +1404,14 @@ log_flush_margin(void)
 			free space */
 		} else {
 			do_flush = TRUE;
+			lsn = log->lsn;
 		}
 	}
 
 	mutex_exit(&(log->mutex));
 
 	if (do_flush) {
-		log_write_up_to(ut_dulint_max, LOG_NO_WAIT, FALSE);
+		log_write_up_to(lsn, LOG_NO_WAIT, FALSE);
 	}
 }
 

innobase/os/os0sync.c

@@ -143,6 +143,7 @@ os_event_create(
 	ut_a(0 == pthread_cond_init(&(event->cond_var), NULL));
 #endif
 	event->is_set = FALSE;
+	event->signal_count = 0;
 #endif /* __WIN__ */
 
         /* Put to the list of events */
@@ -218,6 +219,7 @@ os_event_set(
 		/* Do nothing */
 	} else {
 		event->is_set = TRUE;
+		event->signal_count += 1;
 		ut_a(0 == pthread_cond_broadcast(&(event->cond_var)));
 	}
 
@@ -310,9 +312,15 @@ os_event_wait(
 	        os_thread_exit(NULL);
 	}
 #else
+	ib_longlong	old_signal_count;
+
 	os_fast_mutex_lock(&(event->os_mutex));
+
+	old_signal_count = event->signal_count;
 loop:
-	if (event->is_set == TRUE) {
+	if (event->is_set == TRUE
+            || event->signal_count != old_signal_count) {
+
 		os_fast_mutex_unlock(&(event->os_mutex));
 
 		if (srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS) {
@@ -326,8 +334,9 @@ loop:
 
 	pthread_cond_wait(&(event->cond_var), &(event->os_mutex));
 
-	/* Solaris manual said that spurious wakeups may occur: we have
+	/* Solaris manual said that spurious wakeups may occur: we have to
-	to check the 'is_set' variable again */
+	check if the event really has been signaled after we came here to
+	wait */
 
 	goto loop;
 #endif

innobase/row/row0mysql.c

@@ -1655,7 +1655,7 @@ row_drop_table_for_mysql_in_background(
 	the InnoDB data dictionary get out-of-sync if the user runs
 	with innodb_flush_log_at_trx_commit = 0 */
 	
-	log_write_up_to(ut_dulint_max, LOG_WAIT_ONE_GROUP, TRUE);
+	log_buffer_flush_to_disk();
 
   	trx_commit_for_mysql(trx);
 

innobase/srv/srv0srv.c

@@ -2836,7 +2836,7 @@ loop:
 		at transaction commit */
 
 		srv_main_thread_op_info = (char*)"flushing log";
-		log_write_up_to(ut_dulint_max, LOG_WAIT_ONE_GROUP, TRUE);
+		log_buffer_flush_to_disk();
 
 		/* If there were less than 5 i/os during the
 		one second sleep, we assume that there is free
@@ -2852,10 +2852,9 @@ loop:
 					(char*)"doing insert buffer merge";
 			ibuf_contract_for_n_pages(TRUE, 5);
 
-			srv_main_thread_op_info =
+			srv_main_thread_op_info = (char*)"flushing log";
-						(char*)"flushing log";
+
-			log_write_up_to(ut_dulint_max, LOG_WAIT_ONE_GROUP,
+			log_buffer_flush_to_disk();
-									TRUE);
 		}
 
 		if (buf_get_modified_ratio_pct() >
@@ -2905,7 +2904,7 @@ loop:
 		buf_flush_batch(BUF_FLUSH_LIST, 100, ut_dulint_max);
 
 		srv_main_thread_op_info = (char*) "flushing log";
-		log_write_up_to(ut_dulint_max, LOG_WAIT_ONE_GROUP, TRUE);
+		log_buffer_flush_to_disk();
 	}
 
 	/* We run a batch of insert buffer merge every 10 seconds,
@@ -2915,7 +2914,7 @@ loop:
 	ibuf_contract_for_n_pages(TRUE, 5);
 
 	srv_main_thread_op_info = (char*)"flushing log";
-	log_write_up_to(ut_dulint_max, LOG_WAIT_ONE_GROUP, TRUE);
+	log_buffer_flush_to_disk();
 
 	/* We run a full purge every 10 seconds, even if the server
 	were active */
@@ -2939,8 +2938,7 @@ loop:
 		if (difftime(current_time, last_flush_time) > 1) {
 			srv_main_thread_op_info = (char*) "flushing log";
 
-		        log_write_up_to(ut_dulint_max, LOG_WAIT_ONE_GROUP,
+		        log_buffer_flush_to_disk();
-									TRUE);
 			last_flush_time = current_time;
 		}
 	}
@@ -3060,6 +3058,10 @@ flush_loop:
 			(char*) "waiting for buffer pool flush to end";
 	buf_flush_wait_batch_end(BUF_FLUSH_LIST);
 
+	srv_main_thread_op_info = (char*) "flushing log";
+
+	log_buffer_flush_to_disk();
+
 	srv_main_thread_op_info = (char*)"making checkpoint";
 
 	log_checkpoint(TRUE, FALSE);

innobase/trx/trx0trx.c

@@ -1524,6 +1524,8 @@ trx_commit_complete_for_mysql(
         dulint  lsn     = trx->commit_lsn;
 
         ut_a(trx);
+	
+	trx->op_info = (char*)"flushing log";
 
         if (srv_flush_log_at_trx_commit == 0) {
                 /* Do nothing */
@@ -1547,6 +1549,8 @@ trx_commit_complete_for_mysql(
                 ut_a(0);
         }
 
+	trx->op_info = (char*)"";
+
         return(0);
 }
 

sql/ha_innodb.cc

@@ -915,7 +915,7 @@ innobase_flush_logs(void)
 
   	DBUG_ENTER("innobase_flush_logs");
 
-	log_write_up_to(ut_dulint_max, LOG_WAIT_ONE_GROUP, TRUE);
+	log_buffer_flush_to_disk();
 
   	DBUG_RETURN(result);
 }
@@ -3538,7 +3538,7 @@ ha_innobase::create(
 	the InnoDB data dictionary get out-of-sync if the user runs
 	with innodb_flush_log_at_trx_commit = 0 */
 
-	log_write_up_to(ut_dulint_max, LOG_WAIT_ONE_GROUP, TRUE);
+	log_buffer_flush_to_disk();
 
 	innobase_table = dict_table_get(norm_name, NULL);
 
@@ -3613,7 +3613,7 @@ ha_innobase::delete_table(
 	the InnoDB data dictionary get out-of-sync if the user runs
 	with innodb_flush_log_at_trx_commit = 0 */
 
-	log_write_up_to(ut_dulint_max, LOG_WAIT_ONE_GROUP, TRUE);
+	log_buffer_flush_to_disk();
 
 	/* Tell the InnoDB server that there might be work for
 	utility threads: */
@@ -3683,7 +3683,7 @@ innobase_drop_database(
 	the InnoDB data dictionary get out-of-sync if the user runs
 	with innodb_flush_log_at_trx_commit = 0 */
 
-	log_write_up_to(ut_dulint_max, LOG_WAIT_ONE_GROUP, TRUE);
+	log_buffer_flush_to_disk();
 
 	/* Tell the InnoDB server that there might be work for
 	utility threads: */
@@ -3755,7 +3755,7 @@ ha_innobase::rename_table(
 	the InnoDB data dictionary get out-of-sync if the user runs
 	with innodb_flush_log_at_trx_commit = 0 */
 
-	log_write_up_to(ut_dulint_max, LOG_WAIT_ONE_GROUP, TRUE);
+	log_buffer_flush_to_disk();
 
 	/* Tell the InnoDB server that there might be work for
 	utility threads: */