From e7f58da2acc157b02ceb88467b5920fd041958b8 Mon Sep 17 00:00:00 2001
From: Leo Yuriev <leo@yuriev.ru>
Date: Tue, 1 Sep 2015 12:04:08 +0300
Subject: [PATCH] lmdb: support for a utterly no-sync mode, by combination of
 MDB_NOSYNC and MDB_MAPASYNC.

Change-Id: I3e2d6c8f044a7522436dab4d0818d4d59b78b78d
---
 lmdb.h    |  2 ++
 mdb.c     | 27 +++++++++++++++++++++++----
 mdb_chk.c |  2 +-
 3 files changed, 26 insertions(+), 5 deletions(-)

diff --git a/lmdb.h b/lmdb.h
index 0a833971..0e8199e3 100644
--- a/lmdb.h
+++ b/lmdb.h
@@ -288,6 +288,8 @@ typedef void (MDB_rel_func)(MDB_val *item, void *oldptr, void *newptr, void *rel
 #define MDB_COALESCE	0x2000000
 	/** LIFO policy for reclaiming FreeDB records */
 #define MDB_LIFORECLAIM	0x4000000
+	/** make a steady-sync only on close and explicit env-sync */
+#define MDB_UTTERLY_NOSYNC (MDB_NOSYNC|MDB_MAPASYNC)
 /** @} */
 
 /**	@defgroup	mdb_dbi_open	Database Flags
diff --git a/mdb.c b/mdb.c
index 83e47f51..a976d445 100644
--- a/mdb.c
+++ b/mdb.c
@@ -2341,10 +2341,24 @@ mdb_page_alloc(MDB_cursor *mc, int num, MDB_page **mp, int flags)
 			MDB_meta* head = mdb_meta_head_w(env);
 			MDB_meta* tail = mdb_env_meta_flipflop(env, head);
 
-			if (META_IS_WEAK(head) && oldest == tail->mm_txnid) {
+			if (oldest == tail->mm_txnid
+					&& META_IS_WEAK(head) && !META_IS_WEAK(tail)) {
 				MDB_meta meta = *head;
+				/* LY: Here an oom was happened:
+				 *  - all pages had allocated;
+				 *  - reclaiming was stopped at the last steady-sync;
+				 *  - the head-sync is weak.
+				 * Now we need make a sync to resume reclaiming. If both
+				 * MDB_NOSYNC and MDB_MAPASYNC flags are set, then assume that
+				 * utterly no-sync write mode was requested. In such case
+				 * don't make a steady-sync, but only a legacy-mode checkpoint,
+				 * just for resume reclaiming only, not for data consistency. */
+				int flags = env->me_flags & MDB_WRITEMAP;
+				if ((env->me_flags & MDB_UTTERLY_NOSYNC) == MDB_UTTERLY_NOSYNC)
+					flags |= MDB_UTTERLY_NOSYNC;
+
 				mdb_assert(env, env->me_sync_pending > 0);
-				if (mdb_env_sync0(env, env->me_flags & MDB_WRITEMAP, &meta) == MDB_SUCCESS) {
+				if (mdb_env_sync0(env, flags, &meta) == MDB_SUCCESS) {
 					txnid_t snap = mdb_find_oldest(env, NULL);
 					if (snap > oldest)
 						continue;
@@ -4068,8 +4082,13 @@ mdb_env_sync0(MDB_env *env, unsigned flags, MDB_meta *pending)
 	}
 
 	/* LY: step#2 - update meta-page. */
-	pending->mm_datasync_sign = env->me_sync_pending
-			? MDB_DATASIGN_WEAK : mdb_meta_sign(pending);
+	if (env->me_sync_pending == 0) {
+		pending->mm_datasync_sign = mdb_meta_sign(pending);
+	} else {
+		pending->mm_datasync_sign =
+			(flags & MDB_UTTERLY_NOSYNC) == MDB_UTTERLY_NOSYNC
+				? MDB_DATASIGN_NONE : MDB_DATASIGN_WEAK;
+	}
 	mdb_debug("writing meta page %d for root page %zu",
 		offset >= env->me_psize, pending->mm_dbs[MAIN_DBI].md_root);
 	if (env->me_flags & MDB_WRITEMAP) {
diff --git a/mdb_chk.c b/mdb_chk.c
index 42734c8d..16b3c7da 100644
--- a/mdb_chk.c
+++ b/mdb_chk.c
@@ -490,7 +490,7 @@ static void usage(char *prog)
 const char* meta_synctype(size_t sign) {
 	switch(sign) {
 	case 0:
-		return "legacy/unknown";
+		return "no-sync/legacy";
 	case 1:
 		return "weak";
 	default: