If one deletes a locker file while sush is running, it currently wedges closed. But it could heal around it, though some caution is required, because we want to distinguish between "slot directory vanished" and "dataset unmounted". The safe path is probably to capture an open dirfd on the cluster dataset root at startup and recreate the wiped directory with mkdirat through that fd.
If one deletes a locker file while sush is running, it currently wedges closed. But it could heal around it, though some caution is required, because we want to distinguish between "slot directory vanished" and "dataset unmounted". The safe path is probably to capture an open
dirfdon the cluster dataset root at startup and recreate the wiped directory withmkdiratthrough that fd.