Roadmap

This document serves as a big-picture roadmap of where Bazil is headed. This is all speculative and may change at any time.

Near future

Synchronize file changes between peers

Work in progress, slowed by lacking RPC infrastructure.

Concurrent multi-backend KV storage

Current version is sequential and naive.

Introduce concept of cancel/interrupt to CAS and KV APIs: cancel <-chan struct{}, tie into fuse.Intr when serving FS request.

Bandwidth/latency estimation to prioritize backend attempts. Staggered time delays before starting too many concurrent requests, but move forward faster if too few requests in flight.

Circuit breaker for better behavior under faults.

Remote KV store access over SFTP

Shuffle data in/out of KV stores interactively

Sketch: bazil get PATH, bazil push PATH KVSTORE

Synchronize KV stores on the background

Support reflink for instant file copies

Further out

Pin file content

Persistently remember and respect pins, strive to keep pinned data on local disk.

Sketch: bazil pin PATH, bazil unpin PATH

Arena storage for CAS objects

Once an arena is sealed, make an index with perfect hashing.

Support object deletion by hole punching.

Remote KV store access over S3

Review crypto usage

Review convergencent encryption against ideas in Cryptosphere.

Garbage collection

Challenging because it's distributed garbage collection in a weakly-connected system.

Recursive lazy accounting of subtree size etc

Make directory reflect the size, file counts, modification time etc possible aggregates of the whole subtree.

Update lazily in the background.

Report the non-standard aggregations via xattrs?

See Ceph for ideas on the user experience.

Blue Sky

These may never happen, but are still worth writing down.

Remote client access through SFTP/CIFS/NFS