andysturrock / cassettej Goto Github PK

There is a race condition in ContentAddressableStoreFileImpl.write() near line 153 where if two stores are sharing the same filesystem and are both trying to write the same content then after the first checks whether the file exists, the other writes it and then when doing the rename of the tmp file you get a java.nio.file.FileAlreadyExistsException. The move call should simply overwrite the file if it already exists.

Some filesystems, eg gluster when used under Openshift, don't support atomic moves, and a AtomicMoveNotSupportedException is thrown.