Import wiredtiger-wiredtiger-mongodb-3.2-rc4-56-g7a4f325.tar.gz from wiredtiger branch mongodb-3.2

ref: 8326df6..7a4f325

e731ef8 WT-2251 Free addresses when discarding deleted page references.
0e93d60 SERVER-21691 Avoid insert stalls
264ec21 WT-2249 Keep eviction stuck until cache usage is under 100%.
dca1411 WT-2250 Minor fix: use SET instead of INCRV for stat.

8 files changed, 80 insertions, 70 deletions

diff --git a/src/third_party/wiredtiger/src/btree/bt_discard.c b/src/third_party/wiredtiger/src/btree/bt_discard.c
index 7cd97831044..54d9761c487 100644
--- a/src/third_party/wiredtiger/src/btree/bt_discard.c
+++ b/src/third_party/wiredtiger/src/btree/bt_discard.c
@@ -269,11 +269,7 @@ __wt_free_ref(
 	 * Free any address allocation; if there's no linked WT_REF page, it
 	 * must be allocated.
 	 */
-	if (ref->addr != NULL &&
-	    (ref->home == NULL || __wt_off_page(ref->home, ref->addr))) {
-		__wt_free(session, ((WT_ADDR *)ref->addr)->addr);
-		__wt_free(session, ref->addr);
-	}
+	__wt_ref_free_addr(session, ref);
 
 	/* Free any page-deleted information. */
 	if (ref->page_del != NULL) {
diff --git a/src/third_party/wiredtiger/src/btree/bt_slvg.c b/src/third_party/wiredtiger/src/btree/bt_slvg.c
index 80e467b5707..e4a860bb421 100644
--- a/src/third_party/wiredtiger/src/btree/bt_slvg.c
+++ b/src/third_party/wiredtiger/src/btree/bt_slvg.c
@@ -1290,9 +1290,7 @@ __slvg_col_build_leaf(WT_SESSION_IMPL *session, WT_TRACK *trk, WT_REF *ref)
 	 * would have been lost.)  Clear the reference addr so eviction doesn't
 	 * free the underlying blocks.
 	 */
-	__wt_free(session, ((WT_ADDR *)ref->addr)->addr);
-	__wt_free(session, ref->addr);
-	ref->addr = NULL;
+	__wt_ref_free_addr(session, ref);
 
 	/* Write the new version of the leaf page to disk. */
 	WT_ERR(__slvg_modify_init(session, page));
@@ -2013,9 +2011,7 @@ __slvg_row_build_leaf(
 	 * would have been lost.)  Clear the reference addr so eviction doesn't
 	 * free the underlying blocks.
 	 */
-	__wt_free(session, ((WT_ADDR *)ref->addr)->addr);
-	__wt_free(session, ref->addr);
-	ref->addr = NULL;
+	__wt_ref_free_addr(session, ref);
 
 	/* Write the new version of the leaf page to disk. */
 	WT_ERR(__slvg_modify_init(session, page));
diff --git a/src/third_party/wiredtiger/src/btree/bt_split.c b/src/third_party/wiredtiger/src/btree/bt_split.c
index a0dfbf32cad..e2031553aed 100644
--- a/src/third_party/wiredtiger/src/btree/bt_split.c
+++ b/src/third_party/wiredtiger/src/btree/bt_split.c
@@ -895,6 +895,7 @@ __split_parent(WT_SESSION_IMPL *session, WT_REF *ref, WT_REF **ref_new,
 		 */
 		WT_ASSERT(session, next_ref->page_del == NULL);
 
+		__wt_ref_free_addr(session, next_ref);
 		WT_TRET(__split_safe_free(
 		    session, split_gen, exclusive, next_ref, sizeof(WT_REF)));
 		parent_decr += sizeof(WT_REF);
@@ -1182,8 +1183,8 @@ err:	/*
  *	Lock an internal page.
  */
 static int
-__split_internal_lock(
-    WT_SESSION_IMPL *session, WT_REF *ref, WT_PAGE **parentp, bool *hazardp)
+__split_internal_lock(WT_SESSION_IMPL *session, WT_REF *ref, bool trylock,
+    WT_PAGE **parentp, bool *hazardp)
 {
 	WT_DECL_RET;
 	WT_PAGE *parent;
@@ -1201,7 +1202,7 @@ __split_internal_lock(
 	 * loop until the exclusive lock is resolved). If we want to split
 	 * the parent, give up to avoid that deadlock.
 	 */
-	if (S2BT(session)->checkpointing != WT_CKPT_OFF)
+	if (!trylock && S2BT(session)->checkpointing != WT_CKPT_OFF)
 		return (EBUSY);
 
 	/*
@@ -1226,7 +1227,10 @@ __split_internal_lock(
 		if (F_ISSET_ATOMIC(parent, WT_PAGE_SPLIT_BLOCK))
 			return (EBUSY);
 
-		WT_RET(__wt_fair_lock(session, &parent->page_lock));
+		if (trylock)
+			WT_RET(__wt_fair_trylock(session, &parent->page_lock));
+		else
+			WT_RET(__wt_fair_lock(session, &parent->page_lock));
 		if (parent == ref->home)
 			break;
 		WT_RET(__wt_fair_unlock(session, &parent->page_lock));
@@ -1370,7 +1374,7 @@ __split_parent_climb(WT_SESSION_IMPL *session, WT_PAGE *page, bool page_hazard)
 		 * locks, lock-coupling up the tree.
 		 */
 		WT_ERR(__split_internal_lock(
-		    session, ref, &parent, &parent_hazard));
+		    session, ref, true, &parent, &parent_hazard));
 		ret = __split_internal(session, parent, page);
 		WT_TRET(__split_internal_unlock(session, page, page_hazard));
 
@@ -1668,6 +1672,12 @@ __split_insert(WT_SESSION_IMPL *session, WT_REF *ref)
 	child->addr = ref->addr;
 
 	/*
+	 * The address has moved to the replacement WT_REF.  Make sure it isn't
+	 * freed when the original ref is discarded.
+	 */
+	ref->addr = NULL;
+
+	/*
 	 * Copy the first key from the original page into first ref in the new
 	 * parent.  Pages created in memory always have a "smallest" insert
 	 * list, so look there first.  If we don't find one, get the first key
@@ -1856,6 +1866,11 @@ __split_insert(WT_SESSION_IMPL *session, WT_REF *ref)
 	return (0);
 
 err:	if (split_ref[0] != NULL) {
+		/*
+		 * The address was moved to the replacement WT_REF, restore it.
+		 */
+		ref->addr = split_ref[0]->addr;
+
 		__wt_free(session, split_ref[0]->key.ikey);
 		__wt_free(session, split_ref[0]);
 	}
@@ -1883,7 +1898,7 @@ __wt_split_insert(WT_SESSION_IMPL *session, WT_REF *ref)
 	WT_RET(__wt_verbose(
 	    session, WT_VERB_SPLIT, "%p: split-insert", ref->page));
 
-	WT_RET(__split_internal_lock(session, ref, &parent, &hazard));
+	WT_RET(__split_internal_lock(session, ref, true, &parent, &hazard));
 	if ((ret = __split_insert(session, ref)) != 0) {
 		WT_TRET(__split_internal_unlock(session, parent, hazard));
 		return (ret);
@@ -1975,7 +1990,7 @@ __wt_split_multi(WT_SESSION_IMPL *session, WT_REF *ref, int closing)
 	WT_RET(__wt_verbose(
 	    session, WT_VERB_SPLIT, "%p: split-multi", ref->page));
 
-	WT_RET(__split_internal_lock(session, ref, &parent, &hazard));
+	WT_RET(__split_internal_lock(session, ref, false, &parent, &hazard));
 	if ((ret = __split_multi(session, ref, closing)) != 0 || closing) {
 		WT_TRET(__split_internal_unlock(session, parent, hazard));
 		return (ret);
@@ -2004,7 +2019,7 @@ __wt_split_reverse(WT_SESSION_IMPL *session, WT_REF *ref)
 	WT_RET(__wt_verbose(
 	    session, WT_VERB_SPLIT, "%p: reverse-split", ref->page));
 
-	WT_RET(__split_internal_lock(session, ref, &parent, &hazard));
+	WT_RET(__split_internal_lock(session, ref, false, &parent, &hazard));
 	ret = __split_parent(session, ref, NULL, 0, 0, false, true);
 	WT_TRET(__split_internal_unlock(session, parent, hazard));
 	return (ret);
diff --git a/src/third_party/wiredtiger/src/evict/evict_lru.c b/src/third_party/wiredtiger/src/evict/evict_lru.c
index 306362de57f..f2784890ab7 100644
--- a/src/third_party/wiredtiger/src/evict/evict_lru.c
+++ b/src/third_party/wiredtiger/src/evict/evict_lru.c
@@ -159,7 +159,8 @@ __evict_server(void *arg)
 	WT_DECL_RET;
 	WT_SESSION_IMPL *session;
 #ifdef HAVE_DIAGNOSTIC
-	struct timespec now, stuck_ts = { 0, 0 };
+	struct timespec now, stuck_ts;
+	uint64_t pages_evicted = 0;
 #endif
 	u_int spins;
 
@@ -204,10 +205,11 @@ __evict_server(void *arg)
 			/* Next time we wake up, reverse the sweep direction. */
 			cache->flags ^= WT_CACHE_WALK_REVERSE;
 #ifdef HAVE_DIAGNOSTIC
-			stuck_ts.tv_sec = 0;
-		} else if (stuck_ts.tv_sec == 0)
+			pages_evicted = 0;
+		} else if (pages_evicted != cache->pages_evict) {
 			WT_ERR(__wt_epoch(session, &stuck_ts));
-		else {
+			pages_evicted = cache->pages_evict;
+		} else {
 			/* After being stuck for 5 minutes, give up. */
 			WT_ERR(__wt_epoch(session, &now));
 			if (WT_TIMEDIFF_SEC(now, stuck_ts) > 300) {
@@ -481,6 +483,13 @@ __evict_update_work(WT_SESSION_IMPL *session)
 		goto done;
 	}
 
+	/*
+	 * If the cache has been stuck and is now under control, clear the
+	 * stuck flag.
+	 */
+	if (bytes_inuse < bytes_max)
+		F_CLR(cache, WT_CACHE_STUCK);
+
 	dirty_inuse = __wt_cache_dirty_inuse(cache);
 	if (dirty_inuse > (cache->eviction_dirty_target * bytes_max) / 100) {
 		FLD_SET(cache->state, WT_EVICT_PASS_DIRTY);
@@ -498,6 +507,7 @@ __evict_update_work(WT_SESSION_IMPL *session)
 		F_CLR(cache, WT_CACHE_WOULD_BLOCK);
 		goto done;
 	}
+
 	return (false);
 
 done:	if (F_ISSET(cache, WT_CACHE_STUCK))
@@ -1284,19 +1294,6 @@ fast:		/* If the page can't be evicted, give up. */
 				continue;
 		}
 
-		/*
-		 * If the oldest transaction hasn't changed since the last time
-		 * this page was written, it's unlikely we can make progress.
-		 * Similarly, if the most recent update on the page is not yet
-		 * globally visible, eviction will fail.  These heuristics
-		 * attempt to avoid repeated attempts to evict the same page.
-		 */
-		if (modified && !would_split &&
-		    !FLD_ISSET(cache->state, WT_CACHE_STUCK) &&
-		    (mod->last_oldest_id == __wt_txn_oldest_id(session) ||
-		    !__wt_txn_visible_all(session, mod->update_txn)))
-			continue;
-
 		WT_ASSERT(session, evict->ref == NULL);
 		__evict_init_candidate(session, evict, ref);
 		++evict;
@@ -1419,7 +1416,6 @@ static int
 __evict_page(WT_SESSION_IMPL *session, bool is_server)
 {
 	WT_BTREE *btree;
-	WT_CACHE *cache;
 	WT_DECL_RET;
 	WT_PAGE *page;
 	WT_REF *ref;
@@ -1458,12 +1454,6 @@ __evict_page(WT_SESSION_IMPL *session, bool is_server)
 
 	(void)__wt_atomic_subv32(&btree->evict_busy, 1);
 
-	WT_RET(ret);
-
-	cache = S2C(session)->cache;
-	if (F_ISSET(cache, WT_CACHE_STUCK))
-		F_CLR(cache, WT_CACHE_STUCK);
-
 	return (ret);
 }
 
@@ -1607,8 +1597,8 @@ __wt_cache_dump(WT_SESSION_IMPL *session, const char *ofile)
 
 		next_walk = NULL;
 		session->dhandle = dhandle;
-		while (__wt_tree_walk(session,
-		    &next_walk, NULL, WT_READ_CACHE | WT_READ_NO_WAIT) == 0 &&
+		while (__wt_tree_walk(session, &next_walk, NULL,
+		    WT_READ_CACHE | WT_READ_NO_EVICT | WT_READ_NO_WAIT) == 0 &&
 		    next_walk != NULL) {
 			page = next_walk->page;
 			size = page->memory_footprint;
diff --git a/src/third_party/wiredtiger/src/evict/evict_page.c b/src/third_party/wiredtiger/src/evict/evict_page.c
index 26ea9117fae..9281e7cdb2d 100644
--- a/src/third_party/wiredtiger/src/evict/evict_page.c
+++ b/src/third_party/wiredtiger/src/evict/evict_page.c
@@ -241,19 +241,14 @@ __evict_page_dirty_update(WT_SESSION_IMPL *session, WT_REF *ref, bool closing)
 {
 	WT_ADDR *addr;
 	WT_DECL_RET;
-	WT_PAGE *parent;
 	WT_PAGE_MODIFY *mod;
 
-	parent = ref->home;
 	mod = ref->page->modify;
 
 	switch (mod->rec_result) {
 	case WT_PM_REC_EMPTY:				/* Page is empty */
 		/* Discard the parent's address. */
-		if (ref->addr != NULL && __wt_off_page(parent, ref->addr)) {
-			__wt_free(session, ((WT_ADDR *)ref->addr)->addr);
-			__wt_free(session, ref->addr);
-		}
+		__wt_ref_free_addr(session, ref);
 
 		/*
 		 * Update the parent to reference a deleted page.  The fact that
@@ -308,10 +303,7 @@ __evict_page_dirty_update(WT_SESSION_IMPL *session, WT_REF *ref, bool closing)
 			return (EBUSY);
 
 		/* Discard the parent's address. */
-		if (ref->addr != NULL && __wt_off_page(parent, ref->addr)) {
-			__wt_free(session, ((WT_ADDR *)ref->addr)->addr);
-			__wt_free(session, ref->addr);
-		}
+		__wt_ref_free_addr(session, ref);
 
 		/*
 		 * Update the parent to reference the replacement page.
diff --git a/src/third_party/wiredtiger/src/include/btree.i b/src/third_party/wiredtiger/src/include/btree.i
index 3f8dc08a1da..02633c3206a 100644
--- a/src/third_party/wiredtiger/src/include/btree.i
+++ b/src/third_party/wiredtiger/src/include/btree.i
@@ -466,6 +466,22 @@ __wt_off_page(WT_PAGE *page, const void *p)
 }
 
 /*
+ * __wt_ref_free_addr --
+ *	Free the address in a reference, if necessary.
+ */
+static inline void
+__wt_ref_free_addr(WT_SESSION_IMPL *session, WT_REF *ref)
+{
+	if (ref->addr != NULL) {
+		if (ref->home == NULL || __wt_off_page(ref->home, ref->addr)) {
+			__wt_free(session, ((WT_ADDR *)ref->addr)->addr);
+			__wt_free(session, ref->addr);
+		} else
+			ref->addr = NULL;
+	}
+}
+
+/*
  * __wt_ref_key --
  *	Return a reference to a row-store internal page key as cheaply as
  * possible.
@@ -1046,6 +1062,7 @@ __wt_page_can_evict(WT_SESSION_IMPL *session, WT_REF *ref, bool *inmem_splitp)
 	WT_BTREE *btree;
 	WT_PAGE *page;
 	WT_PAGE_MODIFY *mod;
+	bool modified;
 
 	if (inmem_splitp != NULL)
 		*inmem_splitp = false;
@@ -1070,14 +1087,15 @@ __wt_page_can_evict(WT_SESSION_IMPL *session, WT_REF *ref, bool *inmem_splitp)
 		return (true);
 	}
 
+	modified = __wt_page_is_modified(page);
+
 	/*
 	 * If the file is being checkpointed, we can't evict dirty pages:
 	 * if we write a page and free the previous version of the page, that
 	 * previous version might be referenced by an internal page already
 	 * been written in the checkpoint, leaving the checkpoint inconsistent.
 	 */
-	if (btree->checkpointing != WT_CKPT_OFF &&
-	    __wt_page_is_modified(page)) {
+	if (btree->checkpointing != WT_CKPT_OFF && modified) {
 		WT_STAT_FAST_CONN_INCR(session, cache_eviction_checkpoint);
 		WT_STAT_FAST_DATA_INCR(session, cache_eviction_checkpoint);
 		return (false);
@@ -1104,6 +1122,19 @@ __wt_page_can_evict(WT_SESSION_IMPL *session, WT_REF *ref, bool *inmem_splitp)
 	    F_ISSET_ATOMIC(page, WT_PAGE_SPLIT_BLOCK))
 		return (false);
 
+	/*
+	 * If the oldest transaction hasn't changed since the last time
+	 * this page was written, it's unlikely we can make progress.
+	 * Similarly, if the most recent update on the page is not yet
+	 * globally visible, eviction will fail.  These heuristics
+	 * attempt to avoid repeated attempts to evict the same page.
+	 */
+	if (modified &&
+	    !F_ISSET(S2C(session)->cache, WT_CACHE_STUCK) &&
+	    (mod->last_oldest_id == __wt_txn_oldest_id(session) ||
+	    !__wt_txn_visible_all(session, mod->update_txn)))
+		return (false);
+
 	return (true);
 }
 
diff --git a/src/third_party/wiredtiger/src/log/log_slot.c b/src/third_party/wiredtiger/src/log/log_slot.c
index 255551f99a4..8155397d823 100644
--- a/src/third_party/wiredtiger/src/log/log_slot.c
+++ b/src/third_party/wiredtiger/src/log/log_slot.c
@@ -293,7 +293,7 @@ __wt_log_slot_init(WT_SESSION_IMPL *session)
 		    &log->slot_pool[i].slot_buf, log->slot_buf_size));
 		F_SET(&log->slot_pool[i], WT_SLOT_INIT_FLAGS);
 	}
-	WT_STAT_FAST_CONN_INCRV(session,
+	WT_STAT_FAST_CONN_SET(session,
 	    log_buffer_size, log->slot_buf_size * WT_SLOT_POOL);
 	/*
 	 * Set up the available slot from the pool the first time.
diff --git a/src/third_party/wiredtiger/src/reconcile/rec_write.c b/src/third_party/wiredtiger/src/reconcile/rec_write.c
index 8bf81eafac2..0e1e7498568 100644
--- a/src/third_party/wiredtiger/src/reconcile/rec_write.c
+++ b/src/third_party/wiredtiger/src/reconcile/rec_write.c
@@ -1395,12 +1395,7 @@ __rec_child_deleted(WT_SESSION_IMPL *session,
 	    __wt_txn_visible_all(session, page_del->txnid))) {
 		WT_RET(__wt_ref_info(session, ref, &addr, &addr_size, NULL));
 		WT_RET(__rec_block_free(session, addr, addr_size));
-
-		if (__wt_off_page(ref->home, ref->addr)) {
-			__wt_free(session, ((WT_ADDR *)ref->addr)->addr);
-			__wt_free(session, ref->addr);
-		}
-		ref->addr = NULL;
+		__wt_ref_free_addr(session, ref);
 	}
 
 	/*
@@ -5434,12 +5429,7 @@ __rec_write_wrapup(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_PAGE *page)
 			WT_RET(__wt_ref_info(
 			    session, ref, &addr, &addr_size, NULL));
 			WT_RET(__rec_block_free(session, addr, addr_size));
-			if (__wt_off_page(ref->home, ref->addr)) {
-				__wt_free(
-				    session, ((WT_ADDR *)ref->addr)->addr);
-				__wt_free(session, ref->addr);
-			}
-			ref->addr = NULL;
+			__wt_ref_free_addr(session, ref);
 		}
 		break;
 	case WT_PM_REC_EMPTY:				/* Page deleted */