Merge lp:~gl-az/percona-server/5.1-686534-881001 into lp:percona-server/5.1

[George Ormond Lorch III]

http://jenkins.percona.com/view/PS%205.1/job/percona-server-5.1-param/360/

Fixed issue where LRU dump would hold LRU_list_mutex during the entire
dump process. Now the mutex is only held when navigating the buffer
pool and filling LRU dump pages but not during the LRU dump disk write.

Fixed issue where the last LRU dump page is not written to the LRU dump
file if the page is exactly filled.

Fixed issue where innodb_blocking_lru_restore does not take an optional
bool argument similar to other bool options.

[Laurynas Biveinis (laurynas-biveinis)]

Is "prefill fill" on the line 64 a typo?

review: Approve

[George Ormond Lorch III (gl-az)]

> Is "prefill fill" on the line 64 a typo?
Yes, fixed only that and re-pushed.

=== modified file 'Percona-Server/storage/innodb_plugin/buf/buf0lru.c'
--- Percona-Server/storage/innodb_plugin/buf/buf0lru.c	2012-05-17 17:30:18 +0000
+++ Percona-Server/storage/innodb_plugin/buf/buf0lru.c	2012-08-10 19:00:29 +0000
@@ -2100,6 +2100,11 @@
 Dump the LRU page list to the specific file. */
 #define LRU_DUMP_FILE "ib_lru_dump"
 #define LRU_DUMP_TEMP_FILE "ib_lru_dump.tmp"
+#define LRU_OS_FILE_WRITE() \
+	os_file_write(LRU_DUMP_FILE, dump_file, buffer, \
+		(buffers << UNIV_PAGE_SIZE_SHIFT) & 0xFFFFFFFFUL, \
+		(buffers >> (32 - UNIV_PAGE_SIZE_SHIFT)), \
+		UNIV_PAGE_SIZE)
 
 UNIV_INTERN
 ibool
@@ -2111,17 +2116,19 @@
 	byte*		buffer_base = NULL;
 	byte*		buffer = NULL;
 	buf_page_t*	bpage;
+	buf_page_t*	first_bpage;
 	ulint		buffers;
 	ulint		offset;
-	ibool		ret = FALSE;
+	ulint		pages_written;
 	ulint		i;
+	ulint		total_pages;
 
 	for (i = 0; i < srv_n_data_files; i++) {
 		if (strstr(srv_data_file_names[i], LRU_DUMP_FILE) != NULL) {
 			fprintf(stderr,
 				" InnoDB: The name '%s' seems to be used for"
-				" innodb_data_file_path. Dumping LRU list is not"
-				" done for safeness.\n", LRU_DUMP_FILE);
+				" innodb_data_file_path. Dumping LRU list is"
+				"  not done for safeness.\n", LRU_DUMP_FILE);
 			goto end;
 		}
 	}
@@ -2144,12 +2151,27 @@
 	}
 
 	mutex_enter(&LRU_list_mutex);
-	bpage = UT_LIST_GET_LAST(buf_pool->LRU);
-
-	buffers = offset = 0;
-	while (bpage != NULL) {
+	bpage = first_bpage = UT_LIST_GET_FIRST(buf_pool->LRU);
+	total_pages = UT_LIST_GET_LEN(buf_pool->LRU);
+
+	buffers = offset = pages_written = 0;
+	while (bpage != NULL && (pages_written++ < total_pages)) {
+
+		buf_page_t*	next_bpage = UT_LIST_GET_NEXT(LRU, bpage);
+
+		if (next_bpage == first_bpage) {
+			mutex_exit(&LRU_list_mutex);
+			success = FALSE;
+			fprintf(stderr,
+				"InnoDB: detected cycle in LRU, skipping "
+				"dump\n");
+			goto end;
+		}
+
+		/* prefill with 0xFF so any unused page entries will
+		   stop the reloading process */
 		if (offset == 0) {
-			memset(buffer, 0, UNIV_PAGE_SIZE);
+			memset(buffer, 0xFF, UNIV_PAGE_SIZE);
 		}
 
 		mach_write_to_4(buffer + offset * 4, bpage->space);
@@ -2158,50 +2180,69 @@
 		offset++;
 
 		if (offset == UNIV_PAGE_SIZE/4) {
+			mutex_t		*next_block_mutex = NULL;
+
 			if (srv_shutdown_state != SRV_SHUTDOWN_NONE) {
-				success = 0;
+				mutex_exit(&LRU_list_mutex);
+				success = FALSE;
 				fprintf(stderr,
 					" InnoDB: stopped dumping lru pages"
 					" because of server shutdown.\n");
 				goto end;
 			}
-			success = os_file_write(LRU_DUMP_FILE, dump_file, buffer,
-					(buffers << UNIV_PAGE_SIZE_SHIFT) & 0xFFFFFFFFUL,
-					(buffers >> (32 - UNIV_PAGE_SIZE_SHIFT)),
-					UNIV_PAGE_SIZE);
+
+			/* while writing file, release buffer pool mutex but
+			   keep the next page fixed so we don't worry about
+			   our list iterator becoming invalid */
+			if (next_bpage) {
+				next_block_mutex = buf_page_get_mutex(
+							next_bpage);
+
+				mutex_enter(next_block_mutex);
+				next_bpage->buf_fix_count++;
+				mutex_exit(next_block_mutex);
+			}
+			mutex_exit(&LRU_list_mutex);
+
+			success = LRU_OS_FILE_WRITE();
+
+			/* grab this again here so that next_bpage
+			   can't be purged when we drop the fix_count */
+			mutex_enter(&LRU_list_mutex);
+
+			if (next_bpage) {
+				mutex_enter(next_block_mutex);
+				next_bpage->buf_fix_count--;
+				mutex_exit(next_block_mutex);
+			}
 			if (!success) {
-				mutex_exit(&LRU_list_mutex);
+				mutex_enter(&LRU_list_mutex);
 				fprintf(stderr,
-					" InnoDB: cannot write page %lu of %s\n",
+					" InnoDB: cannot write page"
+					" %lu of %s\n",
 					buffers, LRU_DUMP_FILE);
 				goto end;
 			}
 			buffers++;
 			offset = 0;
+			bpage = next_bpage;
+		} else {
+			bpage = UT_LIST_GET_NEXT(LRU, bpage);
 		}
-
-		bpage = UT_LIST_GET_PREV(LRU, bpage);
-	}
+	} /* while(bpage ...) */
 	mutex_exit(&LRU_list_mutex);
 
+	/* if offset > 0 means that there are some unwritten leftovers */
 	if (offset == 0) {
-		memset(buffer, 0, UNIV_PAGE_SIZE);
-	}
-
-	mach_write_to_4(buffer + offset * 4, 0xFFFFFFFFUL);
-	offset++;
-	mach_write_to_4(buffer + offset * 4, 0xFFFFFFFFUL);
-	offset++;
-
-	success = os_file_write(LRU_DUMP_FILE, dump_file, buffer,
-			(buffers << UNIV_PAGE_SIZE_SHIFT) & 0xFFFFFFFFUL,
-			(buffers >> (32 - UNIV_PAGE_SIZE_SHIFT)),
-			UNIV_PAGE_SIZE);
-	if (!success) {
-		goto end;
-	}
-
-	ret = TRUE;
+		memset(buffer, 0xFF, UNIV_PAGE_SIZE);
+	}
+
+	mach_write_to_4(buffer + offset * 4, 0xFFFFFFFFUL);
+	offset++;
+	mach_write_to_4(buffer + offset * 4, 0xFFFFFFFFUL);
+	offset++;
+
+	success = LRU_OS_FILE_WRITE();
 end:
 	if (dump_file != -1) {
 		if (success) {
@@ -2216,7 +2257,7 @@
 	if (buffer_base)
 		ut_free(buffer_base);
 
-	return(ret);
+	return(success);
 }
 
 typedef struct {

=== modified file 'Percona-Server/storage/innodb_plugin/handler/ha_innodb.cc'
--- Percona-Server/storage/innodb_plugin/handler/ha_innodb.cc	2012-07-02 02:04:45 +0000
+++ Percona-Server/storage/innodb_plugin/handler/ha_innodb.cc	2012-08-10 19:00:29 +0000
@@ -12070,7 +12070,7 @@
   NULL, NULL, 0, 0, UINT_MAX32, 0);
 
 static MYSQL_SYSVAR_BOOL(blocking_lru_restore, innobase_blocking_lru_restore,
-  PLUGIN_VAR_NOCMDARG | PLUGIN_VAR_READONLY,
+  PLUGIN_VAR_OPCMDARG | PLUGIN_VAR_READONLY,
   "Block XtraDB startup process until buffer pool is full restored from a "
   "dump file (if present). Disabled by default.",
   NULL, NULL, FALSE);