Merge pull request #1500 from wiredtiger/evict-trickle

Use the eviction server to write pages with READGEN_OLDEST set.

3 files changed, 32 insertions, 6 deletions

diff --git a/src/evict/evict_lru.c b/src/evict/evict_lru.c
index dbf3a71f222..69fa1f85c93 100644
--- a/src/evict/evict_lru.c
+++ b/src/evict/evict_lru.c
@@ -407,6 +407,15 @@ __evict_has_work(WT_SESSION_IMPL *session, uint32_t *flagsp)
 	    (cache->eviction_dirty_target * bytes_max) / 100)
 		/* Ignore clean pages unless the cache is too large */
 		LF_SET(WT_EVICT_PASS_DIRTY);
+	else if (F_ISSET(cache, WT_EVICT_WOULD_BLOCK)) {
+		/*
+		 * Evict pages with oldest generation (which would otherwise
+		 * block application threads) set regardless of whether we have
+		 * reached the eviction trigger.
+		 */
+		LF_SET(WT_EVICT_PASS_WOULD_BLOCK);
+		F_CLR(cache, WT_EVICT_WOULD_BLOCK);
+	}
 
 	if (F_ISSET(cache, WT_EVICT_STUCK))
 		LF_SET(WT_EVICT_PASS_AGGRESSIVE);
@@ -1076,6 +1085,14 @@ __evict_walk_file(WT_SESSION_IMPL *session, u_int *slotp, uint32_t flags)
 		if (F_ISSET_ATOMIC(page, WT_PAGE_EVICT_LRU))
 			continue;
 
+		/*
+		 * If we are only trickling out pages marked for definite
+		 * eviction, skip anything that isn't marked.
+		 */
+		if (LF_ISSET(WT_EVICT_PASS_WOULD_BLOCK) &&
+		    page->read_gen != WT_READGEN_OLDEST)
+			continue;
+
 		/* Limit internal pages to 50% unless we get aggressive. */
 		if ((page->type == WT_PAGE_COL_INT ||
 		    page->type == WT_PAGE_ROW_INT) &&
diff --git a/src/evict/evict_page.c b/src/evict/evict_page.c
index f7cbb55dbe4..4e96898fd92 100644
--- a/src/evict/evict_page.c
+++ b/src/evict/evict_page.c
@@ -24,9 +24,10 @@ __wt_evict(WT_SESSION_IMPL *session, WT_REF *ref, int exclusive)
 	WT_PAGE *page;
 	WT_PAGE_MODIFY *mod;
 	WT_TXN_STATE *txn_state;
-	int inmem_split, istree;
+	int forced_eviction, inmem_split, istree;
 
 	page = ref->page;
+	forced_eviction = (page->read_gen == WT_READGEN_OLDEST);
 	inmem_split = istree = 0;
 
 	WT_RET(__wt_verbose(session, WT_VERB_EVICT,
@@ -116,6 +117,12 @@ done:	session->excl_next = 0;
 	if (txn_state != NULL)
 		txn_state->snap_min = WT_TXN_NONE;
 
+	if ((inmem_split || (forced_eviction && ret == EBUSY)) &&
+	    !F_ISSET(S2C(session)->cache, WT_EVICT_WOULD_BLOCK)) {
+		F_SET(S2C(session)->cache, WT_EVICT_WOULD_BLOCK);
+		WT_TRET(__wt_evict_server_wake(session));
+	}
+
 	return (ret);
 }
 
@@ -308,7 +315,6 @@ __evict_review(WT_SESSION_IMPL *session, WT_REF *ref,
 	uint32_t flags;
 
 	btree = S2BT(session);
-	page = ref->page;
 
 	/*
 	 * Get exclusive access to the page if our caller doesn't have the tree
@@ -327,6 +333,10 @@ __evict_review(WT_SESSION_IMPL *session, WT_REF *ref,
 		__wt_evict_list_clear_page(session, ref);
 	}
 
+	/* Now that we have exclusive access, review the page. */
+	page = ref->page;
+	mod = page->modify;
+
 	/*
 	 * Recurse through the page's subtree: this happens first because we
 	 * have to write pages in depth-first order, otherwise we'll dirty
@@ -338,8 +348,6 @@ __evict_review(WT_SESSION_IMPL *session, WT_REF *ref,
 		WT_RET(ret);
 	}
 
-	mod = page->modify;
-
 	/*
 	 * If the tree was deepened, there's a requirement that newly created
 	 * internal pages not be evicted until all threads are known to have
@@ -449,7 +457,7 @@ __evict_review(WT_SESSION_IMPL *session, WT_REF *ref,
 		if (exclusive)
 			LF_SET(WT_SKIP_UPDATE_ERR);
 		else if (top && !WT_PAGE_IS_INTERNAL(page) &&
-		    page->memory_footprint > 10 * btree->maxleafpage)
+		    page->read_gen == WT_READGEN_OLDEST)
 			LF_SET(WT_SKIP_UPDATE_RESTORE);
 		WT_RET(__wt_reconcile(session, ref, NULL, flags));
 		WT_ASSERT(session,
diff --git a/src/include/cache.h b/src/include/cache.h
index a6425e97837..cf688b3993f 100644
--- a/src/include/cache.h
+++ b/src/include/cache.h
@@ -19,6 +19,7 @@
 #define	WT_EVICT_PASS_AGGRESSIVE	0x01
 #define	WT_EVICT_PASS_ALL		0x02
 #define	WT_EVICT_PASS_DIRTY		0x04
+#define	WT_EVICT_PASS_WOULD_BLOCK	0x08
 
 /*
  * WT_EVICT_ENTRY --
@@ -33,7 +34,6 @@ struct __wt_evict_entry {
  * WT_EVICT_WORKER --
  *	Encapsulation of an eviction worker thread.
  */
-
 struct __wt_evict_worker {
 	WT_SESSION_IMPL *session;
 	u_int id;
@@ -114,6 +114,7 @@ struct __wt_cache {
 #define	WT_EVICT_ACTIVE		0x04	/* Eviction server is active */
 #define	WT_EVICT_CLEAR_WALKS	0x08	/* Clear eviction walks */
 #define	WT_EVICT_STUCK		0x10	/* Eviction server is stuck */
+#define	WT_EVICT_WOULD_BLOCK	0x20	/* Pages that would block apps */
 	uint32_t flags;
 };