CernVM-FS Known Issues

Mounting might take very long time

This issue occurs if access to /dev/fuse is restricted to the fuse group and users and groups are maintained in large LDAP directories. As part of the mounting, CernVM-FS mount helpers drop privileges from root to the cvmfs user and, if necessary, to the fuse group. In order to do so, the system has to search for the group membership of root in the fuse group. On lxplus, for instance, this takes up to 30 seconds.

As work-around, you can add the root user explicitly to the fuse group in /etc/group.  In that case, the system stops sucessfully in the local search and does not query LDAP.

Thanks to Steve Traylen for pointing out the issue and the work-arounds. See also Savannah: https://savannah.cern.ch/bugs/?79512

Exporting CernVM-FS mount point via NFS

A mounted CernVM-FS repository cannot be exported via NFS. This is related to the Fuse "high-level interface" CernVM-FS uses.

As NFS is stateless, file system objects are identified by "handles" that are passed with every RPC call.  In Fuse that has to be translated into inode lookups.  However, the Fuse high-level interface works on paths rather than inodes. In result, exporting works as long as all inodes are cached in the NFS server's Linux kernel buffers.  Once those kernel caches are flushed, clients are stuck with stale handles. Further details are can be found in README.NFS in the libfuse sources.

A suggested solution is to use the noforget mount option for the Fuse file system. Besides resulting in an ever-growing memory consumption, stale handles still occur with CernVM-FS after a file catalog update has been applied.