Fix tracking of dirty pages. Add statistics for dirty pages.

16 files changed, 135 insertions, 54 deletions

diff --git a/dist/stat_data.py b/dist/stat_data.py
index 2129b72dfa9..21f25b2db32 100644
--- a/dist/stat_data.py
+++ b/dist/stat_data.py
@@ -22,6 +22,8 @@ class Stat:
 connection_stats = [
 	Stat('block_read', 'blocks read from a file'),
 	Stat('block_write', 'blocks written to a file'),
+	Stat('cache_bytes_dirty', 'cache: tracked dirty bytes in the cache'),
+	Stat('cache_bytes_dirty_calc', 'cache: counted dirty bytes in the cache'),
 	Stat('cache_bytes_inuse',
 		'cache: bytes currently held in the cache', perm=1),
 	Stat('cache_bytes_max', 'cache: maximum bytes configured', perm=1),
@@ -32,6 +34,8 @@ connection_stats = [
 	Stat('cache_evict_slow',
 		'cache: eviction server unable to reach eviction goal'),
 	Stat('cache_evict_unmodified', 'cache: unmodified pages evicted'),
+	Stat('cache_pages_dirty', 'cache: tracked dirty pages in the cache'),
+	Stat('cache_pages_dirty_calc', 'cache: counted dirty pages in the cache'),
 	Stat('cache_pages_inuse',
 		'cache: pages currently held in the cache', perm=1),
 	Stat('checkpoint', 'checkpoints'),
diff --git a/examples/c/ex_all.c b/examples/c/ex_all.c
index 710065f32b5..a4512b33f28 100644
--- a/examples/c/ex_all.c
+++ b/examples/c/ex_all.c
@@ -806,7 +806,7 @@ connection_ops(WT_CONNECTION *conn)
 
 	/*! [Setup cache dirty target] */
 	ret = conn->reconfigure(conn, "eviction_dirty_target=75");
-	/*! [Reconfigure a connection] */
+	/*! [Setup cache dirty target] */
 
 	/*! [Get the database home directory] */
 	printf("The database home is %s\n", conn->get_home(conn));
diff --git a/src/btree/bt_handle.c b/src/btree/bt_handle.c
index 64e692d1cbb..99e91566e46 100644
--- a/src/btree/bt_handle.c
+++ b/src/btree/bt_handle.c
@@ -380,7 +380,7 @@ __btree_tree_open_empty(WT_SESSION_IMPL *session, int created)
 	 * or it's actually modified.
 	 */
 	WT_ERR(__wt_page_modify_init(session, leaf));
-	__wt_page_modify_set(leaf);
+	__wt_page_modify_set(session, leaf);
 
 	btree->root_page = root;
 
diff --git a/src/btree/bt_page.c b/src/btree/bt_page.c
index cd74e3cdc87..7ea6318226f 100644
--- a/src/btree/bt_page.c
+++ b/src/btree/bt_page.c
@@ -389,7 +389,7 @@ __inmem_row_int(WT_SESSION_IMPL *session, WT_PAGE *page, size_t *inmem_sizep)
 				WT_ERR(__wt_page_modify_init(session, page));
 				page->modify->first_id = WT_TXN_NONE;
 				if (btree->modified)
-					__wt_page_modify_set(page);
+					__wt_page_modify_set(session, page);
 			}
 
 			++ref;
diff --git a/src/btree/bt_read.c b/src/btree/bt_read.c
index e86f4ccd8d7..0830963eb2c 100644
--- a/src/btree/bt_read.c
+++ b/src/btree/bt_read.c
@@ -33,7 +33,7 @@ __cache_read_row_deleted(
 	WT_RET(__wt_page_modify_init(session, page));
 	page->modify->first_id = ref->txnid;
 	if (btree->modified)
-		__wt_page_modify_set(page);
+		__wt_page_modify_set(session, page);
 
 	/* Allocate the update array. */
 	WT_RET(__wt_calloc_def(session, page->entries, &upd_array));
diff --git a/src/btree/bt_slvg.c b/src/btree/bt_slvg.c
index 4a1e5c4a77e..c332fd8e39e 100644
--- a/src/btree/bt_slvg.c
+++ b/src/btree/bt_slvg.c
@@ -1049,7 +1049,7 @@ __slvg_modify_init(WT_SESSION_IMPL *session, WT_PAGE *page)
 
 	/* The page is dirty. */
 	WT_RET(__wt_page_modify_init(session, page));
-	__wt_page_modify_set(page);
+	__wt_page_modify_set(session, page);
 
 	return (0);
 }
diff --git a/src/btree/bt_sync.c b/src/btree/bt_sync.c
index bb05f966ee4..bf6505da3a7 100644
--- a/src/btree/bt_sync.c
+++ b/src/btree/bt_sync.c
@@ -22,7 +22,7 @@ __wt_bt_cache_force_write(WT_SESSION_IMPL *session)
 
 	/* Dirty the root page to ensure a write. */
 	WT_RET(__wt_page_modify_init(session, page));
-	__wt_page_modify_set(page);
+	__wt_page_modify_set(session, page);
 
 	return (0);
 }
diff --git a/src/btree/bt_walk.c b/src/btree/bt_walk.c
index 7fcedbdd1e2..8c0e033ef80 100644
--- a/src/btree/bt_walk.c
+++ b/src/btree/bt_walk.c
@@ -95,7 +95,7 @@ __tree_walk_delete(
 	 * we write the tree.
 	 */
 	WT_ERR(__wt_page_modify_init(session, page));
-	__wt_page_modify_set(page);
+	__wt_page_modify_set(session, page);
 
 	*skipp = 1;
 
diff --git a/src/btree/rec_write.c b/src/btree/rec_write.c
index 360e184da75..fa789072fda 100644
--- a/src/btree/rec_write.c
+++ b/src/btree/rec_write.c
@@ -476,7 +476,7 @@ __wt_rec_write(WT_SESSION_IMPL *session,
 				break;
 		}
 		WT_RET(__wt_page_modify_init(session, page));
-		__wt_page_modify_set(page);
+		__wt_page_modify_set(session, page);
 
 		return (0);
 	}
@@ -524,7 +524,7 @@ __wt_rec_write(WT_SESSION_IMPL *session,
 	WT_VERBOSE_RET(session, reconcile,
 	    "root page split %p -> %p", page, page->modify->u.split);
 	page = page->modify->u.split;
-	__wt_page_modify_set(page);
+	__wt_page_modify_set(session, page);
 	F_CLR(page->modify, WT_PM_REC_SPLIT_MERGE);
 
 	WT_RET(__wt_rec_write(session, page, NULL, flags));
@@ -1389,7 +1389,7 @@ __wt_rec_bulk_wrapup(WT_CURSOR_BULK *cbulk)
 
 	/* Mark the page's parent dirty. */
 	WT_RET(__wt_page_modify_init(session, page->parent));
-	__wt_page_modify_set(page->parent);
+	__wt_page_modify_set(session, page->parent);
 
 	__wt_rec_destroy(session, &cbulk->reconcile);
 
@@ -3178,6 +3178,7 @@ __rec_write_wrapup(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_PAGE *page)
 	WT_DECL_RET;
 	WT_PAGE_MODIFY *mod;
 	uint32_t i;
+	int was_modified;
 
 	btree = session->btree;
 	mod = page->modify;
@@ -3375,8 +3376,10 @@ err:			__wt_scr_free(&tkey);
 	 * started.
 	 */
 	if (!r->upd_skipped) {
-		__wt_cache_dirty_decr(session, page->memory_footprint);
+		was_modified = __wt_page_is_modified(page);
 		mod->disk_gen = r->orig_write_gen;
+		if (was_modified && !__wt_page_is_modified(page))
+			__wt_cache_dirty_decr(session, page->memory_footprint);
 	}
 
 	return (0);
diff --git a/src/conn/conn_cache.c b/src/conn/conn_cache.c
index 1f8afc3fe97..ed056bb4fe5 100644
--- a/src/conn/conn_cache.c
+++ b/src/conn/conn_cache.c
@@ -7,6 +7,8 @@
 
 #include "wt_internal.h"
 
+static void __cache_stats_traverse(WT_CONNECTION_IMPL *conn);
+
 /*
  * __wt_cache_config --
  *	Configure the underlying cache.
@@ -88,7 +90,7 @@ __wt_cache_create(WT_CONNECTION_IMPL *conn, const char *cfg[])
 	 * We pull some values from the cache statistics (rather than have two
 	 * copies).   Set them.
 	 */
-	__wt_cache_stats_update(conn);
+	__wt_cache_stats_update(conn, 0);
 
 	return (0);
 
@@ -101,7 +103,7 @@ err:	__wt_cache_destroy(conn);
  *	Update the cache statistics for return to the application.
  */
 void
-__wt_cache_stats_update(WT_CONNECTION_IMPL *conn)
+__wt_cache_stats_update(WT_CONNECTION_IMPL *conn, uint32_t flags)
 {
 	WT_CACHE *cache;
 
@@ -112,6 +114,52 @@ __wt_cache_stats_update(WT_CONNECTION_IMPL *conn)
 	    conn->stats, cache_bytes_inuse, __wt_cache_bytes_inuse(cache));
 	WT_STAT_SET(
 	    conn->stats, cache_pages_inuse, __wt_cache_pages_inuse(cache));
+	WT_STAT_SET(
+	    conn->stats, cache_bytes_dirty, __wt_cache_dirty_bytes(cache));
+	WT_STAT_SET(
+	    conn->stats, cache_pages_dirty, cache->pages_dirty);
+
+	if (!LF_ISSET(WT_STATISTICS_FAST))
+		__cache_stats_traverse(conn);
+}
+
+/*
+ * __cache_stats_traverse --
+ *	Generate statistics that require traversing the cache.
+ */
+static void
+__cache_stats_traverse(WT_CONNECTION_IMPL *conn)
+{
+	WT_BTREE *btree;
+	WT_CACHE *cache;
+	WT_DECL_RET;
+	WT_PAGE *page;
+	WT_SESSION_IMPL *session;
+
+	cache = conn->cache;
+	session = conn->default_session;
+	page = NULL;
+	btree = NULL;
+
+	WT_STAT_SET(conn->stats, cache_bytes_dirty_calc, 0);
+	WT_STAT_SET(conn->stats, cache_pages_dirty_calc, 0);
+
+	TAILQ_FOREACH(btree, &conn->btqh, q) {
+		/* Reference the correct WT_BTREE handle. */
+		WT_SET_BTREE_IN_SESSION(session, btree);
+		while ((ret = __wt_tree_walk(
+		    session, &page, WT_TREE_EVICT)) == 0 &&
+		    page != NULL) {
+			if (__wt_page_is_modified(page)) {
+				WT_STAT_INCRV(conn->stats,
+				    cache_bytes_dirty_calc,
+				    page->memory_footprint);
+				WT_STAT_INCRV(conn->stats,
+				    cache_pages_dirty_calc, 1);
+			}
+		}
+		WT_CLEAR_BTREE_IN_SESSION(session);
+	}
 }
 
 /*
diff --git a/src/conn/conn_stat.c b/src/conn/conn_stat.c
index df861665b4b..154689f80de 100644
--- a/src/conn/conn_stat.c
+++ b/src/conn/conn_stat.c
@@ -19,5 +19,5 @@ __wt_conn_stat_init(WT_SESSION_IMPL *session, uint32_t flags)
 	WT_UNUSED(flags);
 	conn = S2C(session);
 
-	__wt_cache_stats_update(conn);
+	__wt_cache_stats_update(conn, flags);
 }
diff --git a/src/include/btree.i b/src/include/btree.i
index c98c3a33355..edc04e4959f 100644
--- a/src/include/btree.i
+++ b/src/include/btree.i
@@ -38,10 +38,13 @@ static inline void
 __wt_cache_page_inmem_decr(
     WT_SESSION_IMPL *session, WT_PAGE *page, size_t size)
 {
-	(void)WT_ATOMIC_SUB(S2C(session)->cache->bytes_inmem, size);
+	WT_CACHE *cache;
+	cache = S2C(session)->cache;
+
+	(void)WT_ATOMIC_SUB(cache->bytes_inmem, size);
 	(void)WT_ATOMIC_SUB(page->memory_footprint, WT_STORE_SIZE(size));
 	if (__wt_page_is_modified(page))
-		(void)WT_ATOMIC_SUB(S2C(session)->cache->bytes_dirty, size);
+		(void)WT_ATOMIC_SUB(cache->bytes_dirty, size);
 }
 
 /*
@@ -53,8 +56,13 @@ static inline void
 __wt_cache_dirty_decr(
     WT_SESSION_IMPL *session, size_t size)
 {
-	(void)WT_ATOMIC_SUB(
-	    S2C(session)->cache->bytes_dirty, size);
+	WT_CACHE *cache;
+	cache = S2C(session)->cache;
+
+	WT_ASSERT(session,
+	    cache->bytes_dirty >= size && cache->pages_dirty > 0);
+	(void)WT_ATOMIC_SUB(cache->bytes_dirty, size);
+	(void)WT_ATOMIC_SUB(cache->pages_dirty, 1);
 }
 
 /*
@@ -91,12 +99,6 @@ __wt_cache_page_evict(WT_SESSION_IMPL *session, WT_PAGE *page)
 	(void)WT_ATOMIC_ADD(cache->pages_evict, 1);
 	(void)WT_ATOMIC_ADD(cache->bytes_evict, page->memory_footprint);
 
-	if (__wt_page_is_modified(page)) {
-		(void)WT_ATOMIC_SUB(cache->pages_dirty, 1);
-		WT_ASSERT(session, cache->bytes_dirty > page->memory_footprint);
-		(void)WT_ATOMIC_SUB(cache->bytes_dirty, page->memory_footprint);
-	}
-
 	page->memory_footprint = 0;
 }
 
@@ -176,11 +178,6 @@ __wt_page_modify_init(WT_SESSION_IMPL *session, WT_PAGE *page)
 	 */
 	if (!WT_ATOMIC_CAS(page->modify, NULL, modify))
 		__wt_free(session, modify);
-	else {
-		(void)WT_ATOMIC_ADD(S2C(session)->cache->pages_dirty, 1);
-		(void)WT_ATOMIC_ADD(
-		    S2C(session)->cache->bytes_dirty, page->memory_footprint);
-	}
 	return (0);
 }
 
@@ -189,8 +186,13 @@ __wt_page_modify_init(WT_SESSION_IMPL *session, WT_PAGE *page)
  *	Mark the page dirty.
  */
 static inline void
-__wt_page_modify_set(WT_PAGE *page)
+__wt_page_modify_set(WT_SESSION_IMPL *session, WT_PAGE *page)
 {
+	if (!__wt_page_is_modified(page)) {
+		(void)WT_ATOMIC_ADD(S2C(session)->cache->pages_dirty, 1);
+		(void)WT_ATOMIC_ADD(
+		    S2C(session)->cache->bytes_dirty, page->memory_footprint);
+	}
 	/*
 	 * Publish: there must be a barrier to ensure all changes to the page
 	 * are flushed before we update the page's write generation, otherwise
@@ -220,7 +222,7 @@ __wt_page_and_tree_modify_set(WT_SESSION_IMPL *session, WT_PAGE *page)
 	 */
 	btree->modified = 1;
 
-	__wt_page_modify_set(page);
+	__wt_page_modify_set(session, page);
 }
 
 /*
diff --git a/src/include/extern.h b/src/include/extern.h
index 09826b23a4d..16f5328b421 100644
--- a/src/include/extern.h
+++ b/src/include/extern.h
@@ -605,7 +605,7 @@ extern const char *__wt_confdfl_wiredtiger_open;
 extern WT_CONFIG_CHECK __wt_confchk_wiredtiger_open[];
 extern int __wt_cache_config(WT_CONNECTION_IMPL *conn, const char *cfg[]);
 extern int __wt_cache_create(WT_CONNECTION_IMPL *conn, const char *cfg[]);
-extern void __wt_cache_stats_update(WT_CONNECTION_IMPL *conn);
+extern void __wt_cache_stats_update(WT_CONNECTION_IMPL *conn, uint32_t flags);
 extern void __wt_cache_destroy(WT_CONNECTION_IMPL *conn);
 extern int __wt_conn_cache_pool_config(WT_SESSION_IMPL *session,
     const char **cfg);
diff --git a/src/include/stat.h b/src/include/stat.h
index 9f8445cc5e8..9fa5e8b3e10 100644
--- a/src/include/stat.h
+++ b/src/include/stat.h
@@ -138,6 +138,8 @@ struct __wt_dsrc_stats {
 struct __wt_connection_stats {
 	WT_STATS block_read;
 	WT_STATS block_write;
+	WT_STATS cache_bytes_dirty;
+	WT_STATS cache_bytes_dirty_calc;
 	WT_STATS cache_bytes_inuse;
 	WT_STATS cache_bytes_max;
 	WT_STATS cache_evict_hazard;
@@ -145,6 +147,8 @@ struct __wt_connection_stats {
 	WT_STATS cache_evict_modified;
 	WT_STATS cache_evict_slow;
 	WT_STATS cache_evict_unmodified;
+	WT_STATS cache_pages_dirty;
+	WT_STATS cache_pages_dirty_calc;
 	WT_STATS cache_pages_inuse;
 	WT_STATS checkpoint;
 	WT_STATS cond_wait;
diff --git a/src/include/wiredtiger.in b/src/include/wiredtiger.in
index dfbab5be762..5d5bdcb1fab 100644
--- a/src/include/wiredtiger.in
+++ b/src/include/wiredtiger.in
@@ -1706,51 +1706,59 @@ extern int wiredtiger_extension_init(WT_SESSION *session,
 #define	WT_STAT_CONN_BLOCK_READ				0
 /*! blocks written to a file */
 #define	WT_STAT_CONN_BLOCK_WRITE			1
+/*! cache: tracked dirty bytes in the cache */
+#define	WT_STAT_CONN_CACHE_BYTES_DIRTY			2
+/*! cache: counted dirty bytes in the cache */
+#define	WT_STAT_CONN_CACHE_BYTES_DIRTY_CALC		3
 /*! cache: bytes currently held in the cache */
-#define	WT_STAT_CONN_CACHE_BYTES_INUSE			2
+#define	WT_STAT_CONN_CACHE_BYTES_INUSE			4
 /*! cache: maximum bytes configured */
-#define	WT_STAT_CONN_CACHE_BYTES_MAX			3
+#define	WT_STAT_CONN_CACHE_BYTES_MAX			5
 /*! cache: pages selected for eviction not evicted because of a hazard
  * reference */
-#define	WT_STAT_CONN_CACHE_EVICT_HAZARD			4
+#define	WT_STAT_CONN_CACHE_EVICT_HAZARD			6
 /*! cache: internal pages evicted */
-#define	WT_STAT_CONN_CACHE_EVICT_INTERNAL		5
+#define	WT_STAT_CONN_CACHE_EVICT_INTERNAL		7
 /*! cache: modified pages evicted */
-#define	WT_STAT_CONN_CACHE_EVICT_MODIFIED		6
+#define	WT_STAT_CONN_CACHE_EVICT_MODIFIED		8
 /*! cache: eviction server unable to reach eviction goal */
-#define	WT_STAT_CONN_CACHE_EVICT_SLOW			7
+#define	WT_STAT_CONN_CACHE_EVICT_SLOW			9
 /*! cache: unmodified pages evicted */
-#define	WT_STAT_CONN_CACHE_EVICT_UNMODIFIED		8
+#define	WT_STAT_CONN_CACHE_EVICT_UNMODIFIED		10
+/*! cache: tracked dirty pages in the cache */
+#define	WT_STAT_CONN_CACHE_PAGES_DIRTY			11
+/*! cache: counted dirty pages in the cache */
+#define	WT_STAT_CONN_CACHE_PAGES_DIRTY_CALC		12
 /*! cache: pages currently held in the cache */
-#define	WT_STAT_CONN_CACHE_PAGES_INUSE			9
+#define	WT_STAT_CONN_CACHE_PAGES_INUSE			13
 /*! checkpoints */
-#define	WT_STAT_CONN_CHECKPOINT				10
+#define	WT_STAT_CONN_CHECKPOINT				14
 /*! condition wait calls */
-#define	WT_STAT_CONN_COND_WAIT				11
+#define	WT_STAT_CONN_COND_WAIT				15
 /*! files currently open */
-#define	WT_STAT_CONN_FILE_OPEN				12
+#define	WT_STAT_CONN_FILE_OPEN				16
 /*! total memory allocations */
-#define	WT_STAT_CONN_MEMALLOC				13
+#define	WT_STAT_CONN_MEMALLOC				17
 /*! total memory frees */
-#define	WT_STAT_CONN_MEMFREE				14
+#define	WT_STAT_CONN_MEMFREE				18
 /*! rwlock readlock calls */
-#define	WT_STAT_CONN_RWLOCK_RDLOCK			15
+#define	WT_STAT_CONN_RWLOCK_RDLOCK			19
 /*! rwlock writelock calls */
-#define	WT_STAT_CONN_RWLOCK_WRLOCK			16
+#define	WT_STAT_CONN_RWLOCK_WRLOCK			20
 /*! total read I/Os */
-#define	WT_STAT_CONN_TOTAL_READ_IO			17
+#define	WT_STAT_CONN_TOTAL_READ_IO			21
 /*! total write I/Os */
-#define	WT_STAT_CONN_TOTAL_WRITE_IO			18
+#define	WT_STAT_CONN_TOTAL_WRITE_IO			22
 /*! ancient transactions */
-#define	WT_STAT_CONN_TXN_ANCIENT			19
+#define	WT_STAT_CONN_TXN_ANCIENT			23
 /*! transactions */
-#define	WT_STAT_CONN_TXN_BEGIN				20
+#define	WT_STAT_CONN_TXN_BEGIN				24
 /*! transactions committed */
-#define	WT_STAT_CONN_TXN_COMMIT				21
+#define	WT_STAT_CONN_TXN_COMMIT				25
 /*! transaction failures due to cache overflow */
-#define	WT_STAT_CONN_TXN_FAIL_CACHE			22
+#define	WT_STAT_CONN_TXN_FAIL_CACHE			26
 /*! transactions rolled-back */
-#define	WT_STAT_CONN_TXN_ROLLBACK			23
+#define	WT_STAT_CONN_TXN_ROLLBACK			27
 
 /*!
  * @}
diff --git a/src/support/stat.c b/src/support/stat.c
index b79f7f3e8e5..11d717824f5 100644
--- a/src/support/stat.c
+++ b/src/support/stat.c
@@ -155,6 +155,10 @@ __wt_stat_alloc_connection_stats(WT_SESSION_IMPL *session, WT_CONNECTION_STATS *
 
 	stats->block_read.desc = "blocks read from a file";
 	stats->block_write.desc = "blocks written to a file";
+	stats->cache_bytes_dirty.desc =
+	    "cache: tracked dirty bytes in the cache";
+	stats->cache_bytes_dirty_calc.desc =
+	    "cache: counted dirty bytes in the cache";
 	stats->cache_bytes_inuse.desc =
 	    "cache: bytes currently held in the cache";
 	stats->cache_bytes_max.desc = "cache: maximum bytes configured";
@@ -165,6 +169,10 @@ __wt_stat_alloc_connection_stats(WT_SESSION_IMPL *session, WT_CONNECTION_STATS *
 	stats->cache_evict_slow.desc =
 	    "cache: eviction server unable to reach eviction goal";
 	stats->cache_evict_unmodified.desc = "cache: unmodified pages evicted";
+	stats->cache_pages_dirty.desc =
+	    "cache: tracked dirty pages in the cache";
+	stats->cache_pages_dirty_calc.desc =
+	    "cache: counted dirty pages in the cache";
 	stats->cache_pages_inuse.desc =
 	    "cache: pages currently held in the cache";
 	stats->checkpoint.desc = "checkpoints";
@@ -195,11 +203,15 @@ __wt_stat_clear_connection_stats(WT_STATS *stats_arg)
 	stats = (WT_CONNECTION_STATS *)stats_arg;
 	stats->block_read.v = 0;
 	stats->block_write.v = 0;
+	stats->cache_bytes_dirty.v = 0;
+	stats->cache_bytes_dirty_calc.v = 0;
 	stats->cache_evict_hazard.v = 0;
 	stats->cache_evict_internal.v = 0;
 	stats->cache_evict_modified.v = 0;
 	stats->cache_evict_slow.v = 0;
 	stats->cache_evict_unmodified.v = 0;
+	stats->cache_pages_dirty.v = 0;
+	stats->cache_pages_dirty_calc.v = 0;
 	stats->checkpoint.v = 0;
 	stats->cond_wait.v = 0;
 	stats->file_open.v = 0;