Merge pull request #1946 from wiredtiger/checkpoint-rewrite

Fix checkpoint visiting a page that was rewritten in memory
(cherry picked from commit f3e73908a9e19a64648f509c5e0417b05853bca2)

7 files changed, 36 insertions, 23 deletions

diff --git a/src/btree/bt_debug.c b/src/btree/bt_debug.c
index fa7cff35e5f..dba2da223bd 100644
--- a/src/btree/bt_debug.c
+++ b/src/btree/bt_debug.c
@@ -648,6 +648,9 @@ __debug_page_metadata(WT_DBG *ds, WT_PAGE *page)
 		case WT_PM_REC_REPLACE:
 			__dmsg(ds, ", replaced");
 			break;
+		case WT_PM_REC_REWRITE:
+			__dmsg(ds, ", rewrite");
+			break;
 		case 0:
 			break;
 		WT_ILLEGAL_VALUE(session);
diff --git a/src/btree/bt_discard.c b/src/btree/bt_discard.c
index 05a54ad643e..2a0a5e37f98 100644
--- a/src/btree/bt_discard.c
+++ b/src/btree/bt_discard.c
@@ -150,6 +150,7 @@ __free_page_modify(WT_SESSION_IMPL *session, WT_PAGE *page)
 
 	switch (F_ISSET(mod, WT_PM_REC_MASK)) {
 	case WT_PM_REC_MULTIBLOCK:
+	case WT_PM_REC_REWRITE:
 		/* Free list of replacement blocks. */
 		for (multi = mod->mod_multi,
 		    i = 0; i < mod->mod_multi_entries; ++multi, ++i) {
diff --git a/src/btree/bt_split.c b/src/btree/bt_split.c
index 9fc567f02c1..f5c3d5fa331 100644
--- a/src/btree/bt_split.c
+++ b/src/btree/bt_split.c
@@ -646,7 +646,6 @@ __split_multi_inmem(
 	 * when discarding the original page, and our caller will discard the
 	 * allocated page on error, when discarding the allocated WT_REF.
 	 */
-
 	WT_RET(__wt_page_inmem(session, ref,
 	    multi->skip_dsk, ((WT_PAGE_HEADER *)multi->skip_dsk)->mem_size,
 	    WT_PAGE_DISK_ALLOC, &page));
diff --git a/src/btree/bt_sync.c b/src/btree/bt_sync.c
index dae2dd8d480..71b0d0abdb3 100644
--- a/src/btree/bt_sync.c
+++ b/src/btree/bt_sync.c
@@ -150,7 +150,8 @@ __sync_file(WT_SESSION_IMPL *session, int syncop)
 			 */
 			if (!WT_PAGE_IS_INTERNAL(page) &&
 			    F_ISSET(txn, TXN_HAS_SNAPSHOT) &&
-			    TXNID_LT(txn->snap_max, mod->first_dirty_txn)) {
+			    TXNID_LT(txn->snap_max, mod->first_dirty_txn) &&
+			    !F_ISSET(mod, WT_PM_REC_REWRITE)) {
 				__wt_page_modify_set(session, page);
 				continue;
 			}
diff --git a/src/evict/evict_page.c b/src/evict/evict_page.c
index 37612bda7e6..e9d1616149e 100644
--- a/src/evict/evict_page.c
+++ b/src/evict/evict_page.c
@@ -195,25 +195,10 @@ __evict_page_dirty_update(WT_SESSION_IMPL *session, WT_REF *ref, int exclusive)
 		break;
 	case WT_PM_REC_MULTIBLOCK:			/* Multiple blocks */
 		/*
-		 * There are two cases in this code.
-		 *
-		 * First, an in-memory page that got too large, we forcibly
-		 * evicted it, and there wasn't anything to write. (Imagine two
-		 * threads updating a small set keys on a leaf page. The page is
-		 * too large so we try to evict it, but after reconciliation
-		 * there's only a small amount of data (so it's a single page we
-		 * can't split), and because there are two threads, there's some
-		 * data we can't write (so we can't evict it). In that case, we
-		 * take advantage of the fact we have exclusive access to the
-		 * page and rewrite it in memory.)
-		 *
-		 * Second, a real split where we reconciled a page and it turned
-		 * into a lot of pages.
+		 * A real split where we reconciled a page and it turned into a
+		 * lot of pages.
 		 */
-		if (mod->mod_multi_entries == 1)
-			WT_RET(__wt_split_rewrite(session, ref));
-		else
-			WT_RET(__wt_split_multi(session, ref, exclusive));
+		WT_RET(__wt_split_multi(session, ref, exclusive));
 		break;
 	case WT_PM_REC_REPLACE: 			/* 1-for-1 page swap */
 		if (ref->addr != NULL && __wt_off_page(parent, ref->addr)) {
@@ -236,6 +221,20 @@ __evict_page_dirty_update(WT_SESSION_IMPL *session, WT_REF *ref, int exclusive)
 		ref->addr = addr;
 		WT_PUBLISH(ref->state, WT_REF_DISK);
 		break;
+	case WT_PM_REC_REWRITE:
+		/*
+		 * An in-memory page that got too large, we forcibly evicted
+		 * it, and there wasn't anything to write. (Imagine two threads
+		 * updating a small set keys on a leaf page. The page is too
+		 * large so we try to evict it, but after reconciliation
+		 * there's only a small amount of data (so it's a single page
+		 * we can't split), and because there are two threads, there's
+		 * some data we can't write (so we can't evict it). In that
+		 * case, we take advantage of the fact we have exclusive access
+		 * to the page and rewrite it in memory.)
+		 */
+		WT_RET(__wt_split_rewrite(session, ref));
+		break;
 	WT_ILLEGAL_VALUE(session);
 	}
 
diff --git a/src/include/btmem.h b/src/include/btmem.h
index eecbbed176e..303162fcc93 100644
--- a/src/include/btmem.h
+++ b/src/include/btmem.h
@@ -356,8 +356,10 @@ struct __wt_page_modify {
 #define	WT_PM_REC_EMPTY		0x01	/* Reconciliation: no replacement */
 #define	WT_PM_REC_MULTIBLOCK	0x02	/* Reconciliation: multiple blocks */
 #define	WT_PM_REC_REPLACE	0x04	/* Reconciliation: single block */
+#define	WT_PM_REC_REWRITE	0x08	/* Reconciliation: rewrite in place */
 #define	WT_PM_REC_MASK							\
-	(WT_PM_REC_EMPTY | WT_PM_REC_MULTIBLOCK | WT_PM_REC_REPLACE)
+	(WT_PM_REC_EMPTY | WT_PM_REC_MULTIBLOCK |			\
+	 WT_PM_REC_REPLACE | WT_PM_REC_REWRITE)
 	uint8_t flags;			/* Page flags */
 };
 
diff --git a/src/reconcile/rec_write.c b/src/reconcile/rec_write.c
index ba0479bf974..573ea8811f8 100644
--- a/src/reconcile/rec_write.c
+++ b/src/reconcile/rec_write.c
@@ -483,6 +483,7 @@ __rec_root_write(WT_SESSION_IMPL *session, WT_PAGE *page, uint32_t flags)
 	switch (F_ISSET(mod, WT_PM_REC_MASK)) {
 	case WT_PM_REC_EMPTY:				/* Page is empty */
 	case WT_PM_REC_REPLACE:				/* 1-for-1 page swap */
+	case WT_PM_REC_REWRITE:				/* Rewrite */
 		return (0);
 	case WT_PM_REC_MULTIBLOCK:			/* Multiple blocks */
 		break;
@@ -3269,6 +3270,8 @@ __rec_col_int(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_PAGE *page)
 			case WT_PM_REC_REPLACE:
 				addr = &child->modify->mod_replace;
 				break;
+			case WT_PM_REC_REWRITE:
+				break;
 			WT_ILLEGAL_VALUE_ERR(session);
 			}
 		} else
@@ -4819,6 +4822,7 @@ __rec_write_wrapup(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_PAGE *page)
 	case WT_PM_REC_EMPTY:				/* Page deleted */
 		break;
 	case WT_PM_REC_MULTIBLOCK:			/* Multiple blocks */
+	case WT_PM_REC_REWRITE:				/* Rewrite */
 		/*
 		 * Discard the multiple replacement blocks.
 		 */
@@ -4897,7 +4901,7 @@ __rec_write_wrapup(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_PAGE *page)
 			bnd->dsk = NULL;
 			mod->mod_multi_entries = 1;
 
-			F_SET(mod, WT_PM_REC_MULTIBLOCK);
+			F_SET(mod, WT_PM_REC_REWRITE);
 			break;
 		}
 
@@ -5047,10 +5051,14 @@ __rec_write_wrapup_err(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_PAGE *page)
 	 * information (otherwise we might think the backing block is being
 	 * reused on a subsequent reconciliation where we want to free it).
 	 */
-	if (F_ISSET(mod, WT_PM_REC_MASK) == WT_PM_REC_MULTIBLOCK)
+	switch (F_ISSET(mod, WT_PM_REC_MASK)) {
+	case WT_PM_REC_MULTIBLOCK:
+	case WT_PM_REC_REWRITE:
 		for (multi = mod->mod_multi,
 		    i = 0; i < mod->mod_multi_entries; ++multi, ++i)
 			multi->addr.reuse = 0;
+		break;
+	}
 
 	/*
 	 * On error, discard blocks we've written, they're unreferenced by the