Import wiredtiger: d9ec69f9111b036ee0b19b47368e15bff8d4818d from branch mongodb-4.2

ref: 617a81369c..d9ec69f911
for: 4.1.11

WT-4278       Clarify in docs that schema operations are not transactional
WT-4309       Increase the maximum expected LAS test reads
WT-4352       Resolve birthmarks during eviction in more cases
WT-4582       Respect prepared updates in lookaside
WT-4642       Store transaction IDs durably
WT-4670       Remove support for WiredTiger LevelDB APIs
WT-4688       Fix test and docs for correct log cursor usage
WT-4690       Make sure eviction does not split during checkpoints
WT-4697       Python3: change workgen to work with Python3
WT-4706       Add a statistic to track the lookaside table size
WT-4723       Restructure the reconciliation code
WT-4725       Python3: change miscellaneous scripts to run under Python3
WT-4726       Python3: set up pull request tester to run with Python3
WT-4728       Remove the stop timestamp from the WT_UPDATE structure
WT-4738       Tighten rules around use of timestamps
WT-4745       Update upgrading documentation before 3.2.0 WiredTiger release
WT-4746       Document that transactions without timestamps clear history
WT-4749       Fix error message formatting (possible SEGV) for prepare_transaction
WT-4751       Change __wt_timestamp_to_string() to return a pointer to simplify verbose
WT-4752       WiredTiger autoconfig check for the compiler incorrectly quoted
WT-4753       Strip the timing_stress configuration options from the base configuration
WT-4754       Change the checkpoint code to not require a list order
WT-4755       Switch to a better timestamp and transaction ID ordering convention
WT-4756       Fix comment in python integer encoder
WT-4759       Save a copy when an old overflow value is discarded
WT-4764       Avoid choosing a transaction frequency of 0 in test/format

1 files changed, 33 insertions, 21 deletions

diff --git a/src/third_party/wiredtiger/src/evict/evict_page.c b/src/third_party/wiredtiger/src/evict/evict_page.c
index 34e138a7a48..4f472b9c9c6 100644
--- a/src/third_party/wiredtiger/src/evict/evict_page.c
+++ b/src/third_party/wiredtiger/src/evict/evict_page.c
@@ -8,9 +8,9 @@
 
 #include "wt_internal.h"
 
-static int __evict_page_clean_update(WT_SESSION_IMPL *, WT_REF *, bool);
-static int __evict_page_dirty_update(WT_SESSION_IMPL *, WT_REF *, bool);
-static int __evict_review(WT_SESSION_IMPL *, WT_REF *, bool, bool *);
+static int __evict_page_clean_update(WT_SESSION_IMPL *, WT_REF *, uint32_t);
+static int __evict_page_dirty_update(WT_SESSION_IMPL *, WT_REF *, uint32_t);
+static int __evict_review(WT_SESSION_IMPL *, WT_REF *, uint32_t, bool *);
 
 /*
  * __evict_exclusive_clear --
@@ -51,19 +51,20 @@ __evict_exclusive(WT_SESSION_IMPL *session, WT_REF *ref)
  *	Release a reference to a page, and attempt to immediately evict it.
  */
 int
-__wt_page_release_evict(WT_SESSION_IMPL *session, WT_REF *ref)
+__wt_page_release_evict(WT_SESSION_IMPL *session, WT_REF *ref, uint32_t flags)
 {
 	WT_BTREE *btree;
 	WT_DECL_RET;
 	WT_PAGE *page;
 	uint64_t time_start, time_stop;
-	uint32_t previous_state;
+	uint32_t evict_flags, previous_state;
 	bool locked, too_big;
 
 	btree = S2BT(session);
 	locked = false;
 	page = ref->page;
 	time_start = __wt_clock(session);
+	evict_flags = LF_ISSET(WT_READ_NO_SPLIT) ? WT_EVICT_CALL_NO_SPLIT : 0;
 
 	/*
 	 * This function always releases the hazard pointer - ensure that's
@@ -89,7 +90,7 @@ __wt_page_release_evict(WT_SESSION_IMPL *session, WT_REF *ref)
 	 * Track how long the call to evict took. If eviction is successful then
 	 * we have one of two pairs of stats to increment.
 	 */
-	ret = __wt_evict(session, ref, false, previous_state);
+	ret = __wt_evict(session, ref, previous_state, evict_flags);
 	time_stop = __wt_clock(session);
 	if (ret == 0) {
 		if (too_big) {
@@ -124,20 +125,25 @@ __wt_page_release_evict(WT_SESSION_IMPL *session, WT_REF *ref)
  */
 int
 __wt_evict(WT_SESSION_IMPL *session,
-    WT_REF *ref, bool closing, uint32_t previous_state)
+    WT_REF *ref, uint32_t previous_state, uint32_t flags)
 {
 	WT_CONNECTION_IMPL *conn;
 	WT_DECL_RET;
 	WT_PAGE *page;
-	bool clean_page, inmem_split, local_gen, tree_dead;
+	bool clean_page, closing, inmem_split, local_gen, tree_dead;
 
 	conn = S2C(session);
 	page = ref->page;
+	closing = LF_ISSET(WT_EVICT_CALL_CLOSING);
 	local_gen = false;
 
 	__wt_verbose(session, WT_VERB_EVICT,
 	    "page %p (%s)", (void *)page, __wt_page_type_string(page->type));
 
+	tree_dead = F_ISSET(session->dhandle, WT_DHANDLE_DEAD);
+	if (tree_dead)
+		LF_SET(WT_EVICT_CALL_NO_SPLIT);
+
 	/*
 	 * Enter the eviction generation. If we re-enter eviction, leave the
 	 * previous eviction generation (which must be as low as the current
@@ -171,7 +177,7 @@ __wt_evict(WT_SESSION_IMPL *session,
 	 * Make this check for clean pages, too: while unlikely eviction would
 	 * choose an internal page with children, it's not disallowed.
 	 */
-	WT_ERR(__evict_review(session, ref, closing, &inmem_split));
+	WT_ERR(__evict_review(session, ref, flags, &inmem_split));
 
 	/*
 	 * If there was an in-memory split, the tree has been left in the state
@@ -208,7 +214,6 @@ __wt_evict(WT_SESSION_IMPL *session,
 	}
 
 	/* Update the reference and discard the page. */
-	tree_dead = F_ISSET(session->dhandle, WT_DHANDLE_DEAD);
 	if (__wt_ref_is_root(ref))
 		__wt_ref_out(session, ref);
 	else if ((clean_page && !F_ISSET(conn, WT_CONN_IN_MEMORY)) || tree_dead)
@@ -216,10 +221,9 @@ __wt_evict(WT_SESSION_IMPL *session,
 		 * Pages that belong to dead trees never write back to disk
 		 * and can't support page splits.
 		 */
-		WT_ERR(__evict_page_clean_update(
-		    session, ref, tree_dead || closing));
+		WT_ERR(__evict_page_clean_update(session, ref, flags));
 	else
-		WT_ERR(__evict_page_dirty_update(session, ref, closing));
+		WT_ERR(__evict_page_dirty_update(session, ref, flags));
 
 	if (clean_page) {
 		WT_STAT_CONN_INCR(session, cache_eviction_clean);
@@ -250,7 +254,7 @@ done:	/* Leave any local eviction generation. */
  *	split.
  */
 static int
-__evict_delete_ref(WT_SESSION_IMPL *session, WT_REF *ref, bool closing)
+__evict_delete_ref(WT_SESSION_IMPL *session, WT_REF *ref, uint32_t flags)
 {
 	WT_DECL_RET;
 	WT_PAGE *parent;
@@ -264,7 +268,7 @@ __evict_delete_ref(WT_SESSION_IMPL *session, WT_REF *ref, bool closing)
 	 * Avoid doing reverse splits when closing the file, it is wasted work
 	 * and some structures may have already been freed.
 	 */
-	if (!closing) {
+	if (!LF_ISSET(WT_EVICT_CALL_NO_SPLIT | WT_EVICT_CALL_CLOSING)) {
 		parent = ref->home;
 		WT_INTL_INDEX_GET(session, parent, pindex);
 		ndeleted = __wt_atomic_addv32(&pindex->deleted_entries, 1);
@@ -302,9 +306,12 @@ __evict_delete_ref(WT_SESSION_IMPL *session, WT_REF *ref, bool closing)
  *	Update a clean page's reference on eviction.
  */
 static int
-__evict_page_clean_update(WT_SESSION_IMPL *session, WT_REF *ref, bool closing)
+__evict_page_clean_update(WT_SESSION_IMPL *session, WT_REF *ref, uint32_t flags)
 {
 	WT_DECL_RET;
+	bool closing;
+
+	closing = LF_ISSET(WT_EVICT_CALL_CLOSING);
 
 	/*
 	 * Before discarding a page, assert that all updates are globally
@@ -334,7 +341,7 @@ __evict_page_clean_update(WT_SESSION_IMPL *session, WT_REF *ref, bool closing)
 		WT_REF_SET_STATE(ref, WT_REF_LOOKASIDE);
 	} else if (ref->addr == NULL) {
 		WT_WITH_PAGE_INDEX(session,
-		    ret = __evict_delete_ref(session, ref, closing));
+		    ret = __evict_delete_ref(session, ref, flags));
 		WT_RET_BUSY_OK(ret);
 	} else
 		WT_REF_SET_STATE(ref, WT_REF_DISK);
@@ -347,14 +354,17 @@ __evict_page_clean_update(WT_SESSION_IMPL *session, WT_REF *ref, bool closing)
  *	Update a dirty page's reference on eviction.
  */
 static int
-__evict_page_dirty_update(WT_SESSION_IMPL *session, WT_REF *ref, bool closing)
+__evict_page_dirty_update(WT_SESSION_IMPL *session, WT_REF *ref,
+    uint32_t evict_flags)
 {
 	WT_ADDR *addr;
 	WT_DECL_RET;
 	WT_MULTI multi;
 	WT_PAGE_MODIFY *mod;
+	bool closing;
 
 	mod = ref->page->modify;
+	closing = FLD_ISSET(evict_flags, WT_EVICT_CALL_CLOSING);
 
 	WT_ASSERT(session, ref->addr == NULL);
 
@@ -370,7 +380,7 @@ __evict_page_dirty_update(WT_SESSION_IMPL *session, WT_REF *ref, bool closing)
 		 */
 		__wt_ref_out(session, ref);
 		WT_WITH_PAGE_INDEX(session,
-		    ret = __evict_delete_ref(session, ref, closing));
+		    ret = __evict_delete_ref(session, ref, evict_flags));
 		WT_RET_BUSY_OK(ret);
 		break;
 	case WT_PM_REC_MULTIBLOCK:			/* Multiple blocks */
@@ -511,20 +521,22 @@ __evict_child_check(WT_SESSION_IMPL *session, WT_REF *parent)
  */
 static int
 __evict_review(
-    WT_SESSION_IMPL *session, WT_REF *ref, bool closing, bool *inmem_splitp)
+    WT_SESSION_IMPL *session, WT_REF *ref, uint32_t evict_flags,
+	bool *inmem_splitp)
 {
 	WT_CACHE *cache;
 	WT_CONNECTION_IMPL *conn;
 	WT_DECL_RET;
 	WT_PAGE *page;
 	uint32_t flags;
-	bool lookaside_retry, *lookaside_retryp, modified;
+	bool closing, lookaside_retry, *lookaside_retryp, modified;
 
 	*inmem_splitp = false;
 
 	conn = S2C(session);
 	page = ref->page;
 	flags = WT_REC_EVICT;
+	closing = FLD_ISSET(evict_flags, WT_EVICT_CALL_CLOSING);
 	if (!WT_SESSION_BTREE_SYNC(session))
 		LF_SET(WT_REC_VISIBLE_ALL);