Restarting worker with directory cache gives "Destination directory already exists" / "Failed to hardlink newly cached directory"

I see:

^[^[  2025-12-12T19:58:54.780199Z  WARN nativelink_worker::running_actions_manager: Directory cache failed, falling back to traditional download, digest: DigestInfo("2c6786e19c74725b23307422bc0d9bfab5ea04b130e56d31e726df03776ea082-166"), e: Error { code: InvalidArgument, messages: ["Destination directory already exists: /Users/user/Work/nativelink/work/df72a46e-f96f-4641-b17c-d8f8fa2bb0b5/work", "Failed to hardlink newly cached directory"] }

Most likely this is due to workers crashing with state still on disk.

Start by reproducing a worker restart while the directory cache state remains on disk, then trace the directory-cache handling that reports “Destination directory already exists” and “Failed to hardlink newly cached directory.” Done means restarting a worker no longer produces these errors for existing on-disk state and the directory cache behavior remains correct.