memdb: take out read-lock during iteration

CHANGELOG.md

@@ -6,6 +6,8 @@
 
 - [\#71](https://github.com/tendermint/tm-db/pull/71) Closed or written batches can no longer be reused, all non-`Close()` calls will panic
 
+- [memdb] [\#74](https://github.com/tendermint/tm-db/pull/74) `Iterator()` and `ReverseIterator()` now take out database read locks for the duration of the iteration
+
 - [memdb] [\#56](https://github.com/tendermint/tm-db/pull/56) Removed some exported methods that were mainly meant for internal use: `Mutex()`, `SetNoLock()`, `SetNoLockSync()`, `DeleteNoLock()`, and `DeleteNoLockSync()`
 
 ### Improvements

memdb.go

@@ -154,15 +154,13 @@ func (db *MemDB) NewBatch() Batch {
 }
 
 // Iterator implements DB.
+// Takes out a read-lock on the database until the iterator is closed.
 func (db *MemDB) Iterator(start, end []byte) (Iterator, error) {
-	db.mtx.RLock()
-	defer db.mtx.RUnlock()
-	return newMemDBIterator(db.btree, start, end, false), nil
+	return newMemDBIterator(db, start, end, false), nil
 }
 
 // ReverseIterator implements DB.
+// Takes out a read-lock on the database until the iterator is closed.
 func (db *MemDB) ReverseIterator(start, end []byte) (Iterator, error) {
-	db.mtx.RLock()
-	defer db.mtx.RUnlock()
-	return newMemDBIterator(db.btree, start, end, true), nil
+	return newMemDBIterator(db, start, end, true), nil
 }

memdb_iterator.go

@@ -26,7 +26,7 @@ type memDBIterator struct {
 var _ Iterator = (*memDBIterator)(nil)
 
 // newMemDBIterator creates a new memDBIterator.
-func newMemDBIterator(bt *btree.BTree, start []byte, end []byte, reverse bool) *memDBIterator {
+func newMemDBIterator(db *MemDB, start []byte, end []byte, reverse bool) *memDBIterator {
 	ctx, cancel := context.WithCancel(context.Background())
 	ch := make(chan *item, chBufferSize)
 	iter := &memDBIterator{
@@ -36,7 +36,9 @@ func newMemDBIterator(bt *btree.BTree, start []byte, end []byte, reverse bool) *
 		end:    end,
 	}
 
+	db.mtx.RLock()
 	go func() {
+		defer db.mtx.RUnlock()
 		// Because we use [start, end) for reverse ranges, while btree uses (start, end], we need
 		// the following variables to handle some reverse iteration conditions ourselves.
 		var (
@@ -61,23 +63,23 @@ func newMemDBIterator(bt *btree.BTree, start []byte, end []byte, reverse bool) *
 		}
 		switch {
 		case start == nil && end == nil && !reverse:
-			bt.Ascend(visitor)
+			db.btree.Ascend(visitor)
 		case start == nil && end == nil && reverse:
-			bt.Descend(visitor)
+			db.btree.Descend(visitor)
 		case end == nil && !reverse:
 			// must handle this specially, since nil is considered less than anything else
-			bt.AscendGreaterOrEqual(newKey(start), visitor)
+			db.btree.AscendGreaterOrEqual(newKey(start), visitor)
 		case !reverse:
-			bt.AscendRange(newKey(start), newKey(end), visitor)
+			db.btree.AscendRange(newKey(start), newKey(end), visitor)
 		case end == nil:
 			// abort after start, since we use [start, end) while btree uses (start, end]
 			abortLessThan = start
-			bt.Descend(visitor)
+			db.btree.Descend(visitor)
 		default:
 			// skip end and abort after start, since we use [start, end) while btree uses (start, end]
 			skipEqual = end
 			abortLessThan = start
-			bt.DescendLessOrEqual(newKey(end), visitor)
+			db.btree.DescendLessOrEqual(newKey(end), visitor)
 		}
 		close(ch)
 	}()