Yet another rearrangement of the forced eviction code. Now check the first time we read a page in a transaction. That way, it shouldn't contain updates from the current transaction that prevent eviction.

5 files changed, 28 insertions, 16 deletions

diff --git a/src/btree/bt_page.c b/src/btree/bt_page.c
index ace833f8e26..5e43c53d6cd 100644
--- a/src/btree/bt_page.c
+++ b/src/btree/bt_page.c
@@ -30,8 +30,12 @@ __wt_page_in_func(
 {
 	WT_DECL_RET;
 	WT_PAGE *page;
+	WT_PAGE_MODIFY *mod;
+	WT_TXN *txn;
 	int busy, oldgen;
 
+	txn = &session->txn;
+
 	for (oldgen = 0;;) {
 		switch (ref->state) {
 		case WT_REF_DISK:
@@ -79,9 +83,17 @@ __wt_page_in_func(
 			 * updates.  This should be extremely unlikely in real
 			 * applications, wait for eviction of the page to avoid
 			 * the issue.
+			 *
+			 * Also, make sure the page isn't too big.  Only do
+			 * this check once per transaction: it is not a common
+			 * case, and we don't want to get stuck if it isn't
+			 * possible to evict the page.
 			 */
-			if (page->modify != NULL &&
-			    __wt_txn_ancient(session, page->modify->first_id)) {
+			if ((mod = page->modify) != NULL &&
+			    (__wt_txn_ancient(session, mod->first_id) ||
+			    (!F_ISSET(txn, TXN_FORCE_EVICT) &&
+			    __wt_eviction_page_force(session, page)))) {
+				F_SET(txn, TXN_FORCE_EVICT);
 				page->read_gen = WT_READ_GEN_OLDEST;
 				WT_RET(__wt_page_release(session, page));
 				break;
diff --git a/src/include/btree.i b/src/include/btree.i
index ed3031471bc..3d49972043f 100644
--- a/src/include/btree.i
+++ b/src/include/btree.i
@@ -445,11 +445,10 @@ __wt_page_release(WT_SESSION_IMPL *session, WT_PAGE *page)
 		return (0);
 
 	/*
-	 * Try to immediately evict pages if they require forced eviction or
-	 * have the special "oldest" read generation.
+	 * Try to immediately evict pages if they have the special "oldest"
+	 * read generation.
 	 */
-	if ((page->read_gen == WT_READ_GEN_OLDEST ||
-	    __wt_eviction_page_force(session, page)) &&
+	if (page->read_gen == WT_READ_GEN_OLDEST &&
 	    WT_ATOMIC_CAS(page->ref->state, WT_REF_MEM, WT_REF_LOCKED)) {
 		if ((ret = __wt_hazard_clear(session, page)) != 0) {
 			page->ref->state = WT_REF_MEM;
diff --git a/src/include/extern.h b/src/include/extern.h
index ce7a96efb76..f9d4d5d3bbe 100644
--- a/src/include/extern.h
+++ b/src/include/extern.h
@@ -1239,10 +1239,10 @@ extern void __wt_stat_clear_connection_stats(void *stats_arg);
 extern int __wt_txnid_cmp(const void *v1, const void *v2);
 extern void __wt_txn_release_snapshot(WT_SESSION_IMPL *session);
 extern void __wt_txn_get_oldest(WT_SESSION_IMPL *session);
-extern void __wt_txn_get_snapshot( WT_SESSION_IMPL *session,
+extern void __wt_txn_get_snapshot(WT_SESSION_IMPL *session,
     wt_txnid_t my_id,
     wt_txnid_t max_id,
-    int force);
+    int committing);
 extern void __wt_txn_get_evict_snapshot(WT_SESSION_IMPL *session);
 extern int __wt_txn_begin(WT_SESSION_IMPL *session, const char *cfg[]);
 extern void __wt_txn_release(WT_SESSION_IMPL *session);
diff --git a/src/include/txn.h b/src/include/txn.h
index 80e0a05668a..6d022c2d078 100644
--- a/src/include/txn.h
+++ b/src/include/txn.h
@@ -98,7 +98,8 @@ struct __wt_txn {
 
 #define	TXN_AUTOCOMMIT	0x01
 #define	TXN_ERROR	0x02
-#define	TXN_OLDEST	0x04
-#define	TXN_RUNNING	0x08
+#define	TXN_FORCE_EVICT	0x04
+#define	TXN_OLDEST	0x08
+#define	TXN_RUNNING	0x10
 	uint32_t flags;
 };
diff --git a/src/txn/txn.c b/src/txn/txn.c
index 70c7bd2bdfb..6bc860d7329 100644
--- a/src/txn/txn.c
+++ b/src/txn/txn.c
@@ -116,8 +116,8 @@ __wt_txn_get_oldest(WT_SESSION_IMPL *session)
  *	Set up a snapshot in the current transaction, without allocating an ID.
  */
 void
-__wt_txn_get_snapshot(
-    WT_SESSION_IMPL *session, wt_txnid_t my_id, wt_txnid_t max_id, int force)
+__wt_txn_get_snapshot(WT_SESSION_IMPL *session,
+   wt_txnid_t my_id, wt_txnid_t max_id, int committing)
 {
 	WT_CONNECTION_IMPL *conn;
 	WT_TXN *txn;
@@ -132,7 +132,7 @@ __wt_txn_get_snapshot(
 	txn_state = &txn_global->states[session->id];
 
 	/* If nothing has changed since last time, we're done. */
-	if (!force && txn->last_id == txn_global->current &&
+	if (!committing && txn->last_id == txn_global->current &&
 	    txn->last_gen == txn_global->gen) {
 		WT_ASSERT(session,
 		    TXNID_LE(txn->oldest_snap_min, txn->snap_min));
@@ -151,7 +151,7 @@ __wt_txn_get_snapshot(
 		for (i = n = 0, s = txn_global->states;
 		    i < session_cnt;
 		    i++, s++) {
-			if ((id = s->id) != WT_TXN_NONE &&
+			if (!committing && (id = s->id) != WT_TXN_NONE &&
 			    TXNID_LT(id, oldest_snap_min))
 				oldest_snap_min = id;
 			/*
@@ -182,7 +182,7 @@ __wt_txn_get_snapshot(
 	    oldest_snap_min);
 	id = (my_id == WT_TXN_NONE || TXNID_LT(txn->snap_min, my_id)) ?
 	    txn->snap_min : my_id;
-	WT_ASSERT(session, TXNID_LE(oldest_snap_min, id));
+	WT_ASSERT(session, committing || TXNID_LE(oldest_snap_min, id));
 	txn_state->snap_min = id;
 }
 
@@ -340,7 +340,7 @@ __wt_txn_release(WT_SESSION_IMPL *session)
 	if (session->ncursors == 0)
 		__wt_txn_release_snapshot(session);
 	txn->isolation = session->isolation;
-	F_CLR(txn, TXN_ERROR | TXN_OLDEST | TXN_RUNNING);
+	F_CLR(txn, TXN_ERROR | TXN_FORCE_EVICT | TXN_OLDEST | TXN_RUNNING);
 
 	/* Update the global generation number. */
 	++txn_global->gen;