SERVER-18875 Don't keep deleted pages during a checkpoint. Sync transaction up with WiredTiger 2.6 to ease back-porting, including picking up WT-1912.

3 files changed, 51 insertions, 12 deletions

diff --git a/src/include/extern.h b/src/include/extern.h
index 87433ae6a9e..b6f0869999f 100644
--- a/src/include/extern.h
+++ b/src/include/extern.h
@@ -662,7 +662,7 @@ extern int WT_CDECL __wt_txnid_cmp(const void *v1, const void *v2);
 extern void __wt_txn_release_snapshot(WT_SESSION_IMPL *session);
 extern void __wt_txn_get_snapshot(WT_SESSION_IMPL *session);
 extern void __wt_txn_update_oldest(WT_SESSION_IMPL *session, int force);
-extern int __wt_txn_begin(WT_SESSION_IMPL *session, const char *cfg[]);
+extern int __wt_txn_config(WT_SESSION_IMPL *session, const char *cfg[]);
 extern void __wt_txn_release(WT_SESSION_IMPL *session);
 extern int __wt_txn_commit(WT_SESSION_IMPL *session, const char *cfg[]);
 extern int __wt_txn_rollback(WT_SESSION_IMPL *session, const char *cfg[]);
diff --git a/src/include/txn.h b/src/include/txn.h
index 62f565c0535..dbaa11309ab 100644
--- a/src/include/txn.h
+++ b/src/include/txn.h
@@ -46,14 +46,14 @@ struct __wt_txn_global {
 	volatile int32_t scan_count;
 
 	/*
-	 * Track information about the running checkpoint. The transaction IDs
-	 * used when checkpointing are special. Checkpoints can run for a long
-	 * time so we keep them out of regular visibility checks. Eviction and
-	 * checkpoint operations know when they need to be aware of
-	 * checkpoint IDs.
+	 * Track information about the running checkpoint. The transaction
+	 * snapshot used when checkpointing are special. Checkpoints can run
+	 * for a long time so we keep them out of regular visibility checks.
+	 * Eviction and checkpoint operations know when they need to be aware
+	 * of checkpoint transactions.
 	 */
+	volatile uint32_t checkpoint_id;	/* Checkpoint's session ID */
 	volatile uint64_t checkpoint_gen;
-	volatile uint64_t checkpoint_id;
 	volatile uint64_t checkpoint_snap_min;
 
 	WT_TXN_STATE *states;		/* Per-session transaction states */
diff --git a/src/include/txn.i b/src/include/txn.i
index 9e8d9cee748..f0b0534ff4a 100644
--- a/src/include/txn.i
+++ b/src/include/txn.i
@@ -99,23 +99,31 @@ __wt_txn_oldest_id(WT_SESSION_IMPL *session)
 	WT_BTREE *btree;
 	WT_TXN_GLOBAL *txn_global;
 	uint64_t checkpoint_snap_min, oldest_id;
+	uint32_t checkpoint_id;
 
 	txn_global = &S2C(session)->txn_global;
 	btree = S2BT_SAFE(session);
 
 	/*
-	 * Take a local copy of ID in case they are updated while we are
+	 * Take a local copy of these IDs in case they are updated while we are
 	 * checking visibility.
 	 */
+	checkpoint_id = txn_global->checkpoint_id;
 	checkpoint_snap_min = txn_global->checkpoint_snap_min;
 	oldest_id = txn_global->oldest_id;
 
 	/*
-	 * If there is no active checkpoint or this handle is up to date with
-	 * the active checkpoint it's safe to ignore the checkpoint ID in the
-	 * visibility check.
+	 * Checkpoint transactions often fall behind ordinary application
+	 * threads.  Take special effort to not keep changes pinned in cache
+	 * if they are only required for the checkpoint and it has already
+	 * seen them.
+	 *
+	 * If there is no active checkpoint, this session is doing the
+	 * checkpoint, or this handle is up to date with the active checkpoint
+	 * then it's safe to ignore the checkpoint ID in the visibility check.
 	 */
-	if (checkpoint_snap_min != WT_TXN_NONE && (btree == NULL ||
+	if (checkpoint_snap_min != WT_TXN_NONE &&
+	    checkpoint_id != session->id && (btree == NULL ||
 	    btree->checkpoint_gen != txn_global->checkpoint_gen) &&
 	    TXNID_LT(checkpoint_snap_min, oldest_id))
 		/*
@@ -206,6 +214,37 @@ __wt_txn_visible(WT_SESSION_IMPL *session, uint64_t id)
 }
 
 /*
+ * __wt_txn_begin --
+ *	Begin a transaction.
+ */
+static int
+__wt_txn_begin(WT_SESSION_IMPL *session, const char *cfg[])
+{
+	WT_TXN *txn;
+
+	txn = &session->txn;
+	txn->isolation = session->isolation;
+	txn->txn_logsync = S2C(session)->txn_logsync;
+
+        if (cfg != NULL)
+                WT_RET(__wt_txn_config(session, cfg));
+
+	F_SET(txn, TXN_RUNNING);
+	if (txn->isolation == TXN_ISO_SNAPSHOT) {
+		if (session->ncursors > 0)
+			WT_RET(__wt_session_copy_values(session));
+
+                /*
+                 * We're about to allocate a snapshot: if we need to block for
+                 * eviction, it's better to do it beforehand.
+                 */
+                WT_RET(__wt_cache_full_check(session));
+		__wt_txn_get_snapshot(session);
+	}
+	return (0);
+}
+
+/*
  * __wt_txn_read --
  *	Get the first visible update in a list (or NULL if none are visible).
  */