WT-2801 Prevent eviction of metadata updates by a running checkpoint. (#2922)

2 files changed, 40 insertions, 38 deletions

diff --git a/src/btree/bt_sync.c b/src/btree/bt_sync.c
index 412358fed05..64650d6336d 100644
--- a/src/btree/bt_sync.c
+++ b/src/btree/bt_sync.c
@@ -26,14 +26,12 @@ __sync_file(WT_SESSION_IMPL *session, WT_CACHE_OP syncop)
 	uint64_t internal_bytes, internal_pages, leaf_bytes, leaf_pages;
 	uint64_t oldest_id, saved_snap_min;
 	uint32_t flags;
-	u_int saved_evict_walk_period;
 
 	conn = S2C(session);
 	btree = S2BT(session);
 	walk = NULL;
 	txn = &session->txn;
 	saved_snap_min = WT_SESSION_TXN_STATE(session)->snap_min;
-	saved_evict_walk_period = btree->evict_walk_period;
 	flags = WT_READ_CACHE | WT_READ_NO_GEN;
 
 	internal_bytes = leaf_bytes = 0;
@@ -231,41 +229,9 @@ err:	/* On error, clear any left-over tree walk. */
 	    saved_snap_min == WT_TXN_NONE)
 		__wt_txn_release_snapshot(session);
 
-	if (btree->checkpointing != WT_CKPT_OFF) {
-		/*
-		 * Update the checkpoint generation for this handle so visible
-		 * updates newer than the checkpoint can be evicted.
-		 *
-		 * This has to be published before eviction is enabled again,
-		 * so that eviction knows that the checkpoint has completed.
-		 */
-		WT_PUBLISH(btree->checkpoint_gen,
-		    conn->txn_global.checkpoint_gen);
-		WT_STAT_FAST_DATA_SET(session,
-		    btree_checkpoint_generation, btree->checkpoint_gen);
-
-		/*
-		 * Clear the checkpoint flag and push the change; not required,
-		 * but publishing the change means stalled eviction gets moving
-		 * as soon as possible.
-		 */
-		btree->checkpointing = WT_CKPT_OFF;
-		WT_FULL_BARRIER();
-
-		/*
-		 * In case this tree was being skipped by the eviction server
-		 * during the checkpoint, restore the previous state.
-		 */
-		btree->evict_walk_period = saved_evict_walk_period;
-
-		/*
-		 * Wake the eviction server, in case application threads have
-		 * stalled while the eviction server decided it couldn't make
-		 * progress.  Without this, application threads will be stalled
-		 * until the eviction server next wakes.
-		 */
-		WT_TRET(__wt_evict_server_wake(session));
-	}
+	/* Clear the checkpoint flag and push the change. */
+	if (btree->checkpointing != WT_CKPT_OFF)
+		WT_PUBLISH(btree->checkpointing, WT_CKPT_OFF);
 
 	__wt_spin_unlock(session, &btree->flush_lock);
 
diff --git a/src/txn/txn_ckpt.c b/src/txn/txn_ckpt.c
index 62f6d099d0f..1efb56dccfe 100644
--- a/src/txn/txn_ckpt.c
+++ b/src/txn/txn_ckpt.c
@@ -1225,7 +1225,43 @@ err:	/*
 static int
 __checkpoint_tree_helper(WT_SESSION_IMPL *session, const char *cfg[])
 {
-	return (__checkpoint_tree(session, true, cfg));
+	WT_BTREE *btree;
+	WT_CONNECTION_IMPL *conn;
+	WT_DECL_RET;
+	u_int saved_evict_walk_period;
+
+	btree = S2BT(session);
+	conn = S2C(session);
+	saved_evict_walk_period = btree->evict_walk_period;
+
+	ret = __checkpoint_tree(session, true, cfg);
+
+	/*
+	 * Update the checkpoint generation for this handle so visible
+	 * updates newer than the checkpoint can be evicted.
+	 *
+	 * This has to be published before eviction is enabled again,
+	 * so that eviction knows that the checkpoint has completed.
+	 */
+	WT_PUBLISH(btree->checkpoint_gen, conn->txn_global.checkpoint_gen);
+	WT_STAT_FAST_DATA_SET(session,
+	    btree_checkpoint_generation, btree->checkpoint_gen);
+
+	/*
+	 * In case this tree was being skipped by the eviction server
+	 * during the checkpoint, restore the previous state.
+	 */
+	btree->evict_walk_period = saved_evict_walk_period;
+
+	/*
+	 * Wake the eviction server, in case application threads have
+	 * stalled while the eviction server decided it couldn't make
+	 * progress.  Without this, application threads will be stalled
+	 * until the eviction server next wakes.
+	 */
+	WT_TRET(__wt_evict_server_wake(session));
+
+	return (ret);
 }
 
 /*