RES consistently increasing while using leveldb

We use leveldb as a recovery store in Hadoop YARN nodemanager process. We are seeing incessant increase in resident size for nodemanager process if recovery is enabled.
While the heap is under control, RES increases to upto 8GB with default leveldb options and never comes down. We restart the process after it reaches 8GB.

Level DB is also used as a backend store for another Hadoop process named timeline server. Similar issues were encountered there as well. JIRA for reference (https://issues.apache.org/jira/browse/YARN-5368)

Can somebody explain what can cause this and if there is anything to worry i.e. if it keeps on increasing, can it potentially lead to lack of memory ?
Also which options can be tuned to get rid of this issue ?
Please note we call compactRange(NULL, NULL) every 1 hour. And as we use leveldb as a recovery store in nodemanager, hence it only issues writes (reads happen only on startup).