Merge pull request #1995 from wiredtiger/skip-checkpoint

Only skip metadata checkpoint if we really skipped it, not force.
(cherry picked from commit 3fe41b427ace35308de552caac67329db0d85c52)

5 files changed, 73 insertions, 8 deletions

diff --git a/src/include/btree.h b/src/include/btree.h
index 44ec40364cc..264ebdefc39 100644
--- a/src/include/btree.h
+++ b/src/include/btree.h
@@ -147,8 +147,9 @@ struct __wt_btree {
 #define	WT_BTREE_IN_MEMORY	0x00200	/* Cache-resident object */
 #define	WT_BTREE_NO_EVICTION	0x00400	/* Disable eviction */
 #define	WT_BTREE_SALVAGE	0x00800	/* Handle is for salvage */
-#define	WT_BTREE_UPGRADE	0x01000	/* Handle is for upgrade */
-#define	WT_BTREE_VERIFY		0x02000	/* Handle is for verify */
+#define	WT_BTREE_SKIP_CKPT	0x01000	/* Handle skipped checkpoint */
+#define	WT_BTREE_UPGRADE	0x02000	/* Handle is for upgrade */
+#define	WT_BTREE_VERIFY		0x04000	/* Handle is for verify */
 	uint32_t flags;
 };
 
diff --git a/src/include/extern.h b/src/include/extern.h
index f42a504e5be..59e795893b5 100644
--- a/src/include/extern.h
+++ b/src/include/extern.h
@@ -316,6 +316,7 @@ extern void __wt_cache_dump(WT_SESSION_IMPL *session);
 extern int __wt_evict(WT_SESSION_IMPL *session, WT_REF *ref, int exclusive);
 extern void __wt_evict_page_clean_update(WT_SESSION_IMPL *session, WT_REF *ref);
 extern int __wt_log_ckpt(WT_SESSION_IMPL *session, WT_LSN *ckp_lsn);
+extern int __wt_log_force_sync(WT_SESSION_IMPL *session, WT_LSN *min_lsn);
 extern int __wt_log_needs_recovery(WT_SESSION_IMPL *session, WT_LSN *ckp_lsn, int *rec);
 extern void __wt_log_written_reset(WT_SESSION_IMPL *session);
 extern int __wt_log_get_all_files(WT_SESSION_IMPL *session, char ***filesp, u_int *countp, uint32_t *maxid, int active_only);
diff --git a/src/log/log.c b/src/log/log.c
index 2c36abd8eab..e6f0ea6d1ab 100644
--- a/src/log/log.c
+++ b/src/log/log.c
@@ -36,6 +36,50 @@ __wt_log_ckpt(WT_SESSION_IMPL *session, WT_LSN *ckp_lsn)
 }
 
 /*
+ * __wt_log_force_sync --
+ *	Force a sync of the log and files.
+ */
+int
+__wt_log_force_sync(WT_SESSION_IMPL *session, WT_LSN *min_lsn)
+{
+	WT_CONNECTION_IMPL *conn;
+	WT_DECL_RET;
+	WT_LOG *log;
+
+	conn = S2C(session);
+	log = conn->log;
+	__wt_spin_lock(session, &log->log_sync_lock);
+	WT_ASSERT(session, log->log_dir_fh != NULL);
+	/*
+	 * Sync the directory if the log file entry hasn't been written
+	 * into the directory.
+	 */
+	if (log->sync_dir_lsn.file < min_lsn->file) {
+		WT_ERR(__wt_verbose(session, WT_VERB_LOG,
+		    "log_force_sync: sync directory %s",
+		    log->log_dir_fh->name));
+		WT_ERR(__wt_directory_sync_fh(session, log->log_dir_fh));
+		log->sync_dir_lsn = *min_lsn;
+		WT_STAT_FAST_CONN_INCR(session, log_sync_dir);
+	}
+	/*
+	 * Sync the log file if needed.
+	 */
+	if (WT_LOG_CMP(&log->sync_lsn, min_lsn) < 0) {
+		WT_ERR(__wt_verbose(session, WT_VERB_LOG,
+		    "log_force_sync: sync to LSN %d/%lu",
+		    min_lsn->file, min_lsn->offset));
+		WT_ERR(__wt_fsync(session, log->log_fh));
+		log->sync_lsn = *min_lsn;
+		WT_STAT_FAST_CONN_INCR(session, log_sync);
+		WT_ERR(__wt_cond_signal(session, log->log_sync_cond));
+	}
+err:
+	__wt_spin_unlock(session, &log->log_sync_lock);
+	return (ret);
+}
+
+/*
  * __wt_log_needs_recovery --
  *	Return 0 if we encounter a clean shutdown and 1 if recovery
  *	must be run in the given variable.
diff --git a/src/txn/txn_ckpt.c b/src/txn/txn_ckpt.c
index 11155134ca7..3954e22583c 100644
--- a/src/txn/txn_ckpt.c
+++ b/src/txn/txn_ckpt.c
@@ -554,11 +554,19 @@ err:	/*
 	txn_global->checkpoint_id = WT_TXN_NONE;
 	txn_global->checkpoint_snap_min = WT_TXN_NONE;
 
-	/* Tell logging that we have finished a database checkpoint. */
-	if (logging)
+	/*
+	 * Tell logging that we have finished a database checkpoint.  Do not
+	 * write a log record if the database was idle.
+	 */
+	if (logging) {
+		if (ret == 0 && full &&
+		    F_ISSET((WT_BTREE *)session->meta_dhandle->handle,
+		    WT_BTREE_SKIP_CKPT))
+			idle = 1;
 		WT_TRET(__wt_txn_checkpoint_log(session, full,
-		    (ret == 0) ? WT_TXN_LOG_CKPT_STOP : WT_TXN_LOG_CKPT_FAIL,
-		    NULL));
+		    (ret == 0 && !idle) ?
+		    WT_TXN_LOG_CKPT_STOP : WT_TXN_LOG_CKPT_CLEANUP, NULL));
+	}
 
 	for (i = 0; i < session->ckpt_handle_next; ++i) {
 		if (session->ckpt_handle[i].dhandle == NULL) {
@@ -789,6 +797,7 @@ __checkpoint_worker(
 	 * means it must exist.
 	 */
 	force = 0;
+	F_CLR(btree, WT_BTREE_SKIP_CKPT);
 	if (!btree->modified && cfg != NULL) {
 		ret = __wt_config_gets(session, cfg, "force", &cval);
 		if (ret != 0 && ret != WT_NOTFOUND)
@@ -797,8 +806,10 @@ __checkpoint_worker(
 			force = 1;
 	}
 	if (!btree->modified && !force) {
-		if (!is_checkpoint)
+		if (!is_checkpoint) {
+			F_SET(btree, WT_BTREE_SKIP_CKPT);
 			goto done;
+		}
 
 		deleted = 0;
 		WT_CKPT_FOREACH(ckptbase, ckpt)
@@ -816,8 +827,10 @@ __checkpoint_worker(
 		    (strcmp(name, (ckpt - 1)->name) == 0 ||
 		    (WT_PREFIX_MATCH(name, WT_CHECKPOINT) &&
 		    WT_PREFIX_MATCH((ckpt - 1)->name, WT_CHECKPOINT))) &&
-		    deleted < 2)
+		    deleted < 2) {
+			F_SET(btree, WT_BTREE_SKIP_CKPT);
 			goto done;
+		}
 	}
 
 	/* Add a new checkpoint entry at the end of the list. */
diff --git a/src/txn/txn_log.c b/src/txn/txn_log.c
index 901785628e1..e87c3cf0e3c 100644
--- a/src/txn/txn_log.c
+++ b/src/txn/txn_log.c
@@ -303,6 +303,12 @@ __wt_txn_checkpoint_log(
 	case WT_TXN_LOG_CKPT_PREPARE:
 		txn->full_ckpt = 1;
 		*ckpt_lsn = S2C(session)->log->write_start_lsn;
+		/*
+		 * We need to make sure that the log records in the checkpoint
+		 * LSN are on disk.  In particular to make sure that the
+		 * current log file exists.
+		 */
+		WT_ERR(__wt_log_force_sync(session, ckpt_lsn));
 		break;
 
 	case WT_TXN_LOG_CKPT_START: