This is a dumb sort-of-object-store. I blame @glennklockwood.
It uses split gzipped files spread and replicated across a list of nodes, with keyed passwordless scp/ssh for distribution and reading. It's written in bash (though I haven't checked for bashisms, it might be portable to dash/ksh/sh) and a fairly vanilla blend of command-line tools.
- more careful replication, where we actually have multiple replicas on single hosts (or else make sure we don't try to make more replicas than we have hosts: at the moment they just get overwritten on the same host)
- maybe more careful object ID generation
- nodelist syncing
- ability to run in different dirs on different nodes
- any sort of checking that what you get out is what you put in
- bash CGI access
- something to check the number of replicas available and re-replicate if necessary
- something to check replicas haven't diverged through corruption (+ quorum assertion)
- a remote version of the store command
- CI testing
- probably a bunch of other things, I don't know