A fork of chef-bcpc but targeted directly at a fast and robust object store. After running chef-bcpc object store in production for 18months, for fun I decided to rip out just the ceph object store.

Currently this differs from BCPC in the following ways:

• Removal of all openstack components, this is just object store.
• Rewrite of the networking system to remove much of the complexity that was required to make openstack go.
• Dedicated roles. chef-bcpc has "headnodes" and "workernodes". In BOS each node has a dedicated purpose, e.g. OSD storage or ceph monitors, reducing burden on mons and allowing the specalization of hardware.
• Collectd based monitoring, in painful detail, of the health of ceph.
• Changes in the RGW configurations, e.g. civetweb rather than apache, varnish rather than haproxy.

BOS uses collectd to monitor the state of the system. Currently, all perf counters are monitored and stats about each bucket. Per bucket stats are labeled like so, <username>.<bucket_name>.<stat>.

Stats are reported every 5 mins by default.