In early alpha testing.
The data formats in Bazil are still undergoing major changes, and significant functionality is missing or has not been tested well enough.
While we don't expect it to literally lose data, we will not put extra effort into data format compatibility until the first formally released version.
As it is, Bazil is intended primarily for other developers and power users exploring future options. Do not stare into laser with remaining eye.
The current version has not been tuned for performance and should not be taken as an indication of what will be possible later.
Bazil is provided “as is” and with no warranty. See the file LICENSE in the source distribution for more.
With all that said, we are very enthusiastic about the possible uses of Bazil, and think that the architecture can be very resilient against data loss.
Bazil is still missing features, and has intentional limits to keep it simple and focused on specific aspects of the design. These limitation may be removed later.
See also anti-goals which are less likely to be “fixed”.
See also roadmap for planned features.
At this point in time, Bazil is a single-node system. It'll store your
data, take snapshots of it, and you could use for example rsync to
backup the object store, but Bazil itself is not yet ready to transfer
data between computers, or synchronize file changes between two peers.
The design for peer synchronization exists and is pretty solid, and current work focuses on a good RPC mechanism; most existing frameworks assume small messages, can't prioritize data transfers, or are constrained by to strict client-server request-response messaging that is not the best match for two-way synchronization traffic.
See roadmap for sync, remote KV access via SFTP and S3, background synchronization, and so on.
Current implementation does not write file contents to disk until it
sees fsync or close. That means creating a large file consumes a
lot of RAM temporarily.
Right now, objects added to the CAS remain there permanently. A plan for garbage collection exists (though distributed, weakly connected GC is tricky!), but implementing it is not yet a priority.