Ceph » CephFS

As discussed briefly in #805, there is a race condition due to MDS failover where an old MDS could have written log data that the new MDS is not aware of. To deal with this, we need to truncate log objects before we start writing to them.

I think I'll set this up so that it can truncate an arbitrary (configurable) number ahead (and just does the minimum required number on startup) so that we can tune the number of blocks to prevent this impacting performance by blocking logging, which would be sad.