Replicated before acknowledgement
Normal writes never wait for object storage.
Open source · Apache 2.0 · Built for the cloud
One coordination backbone. From laptop to cloud scale.
Orbita is a strongly consistent store for locks, leases, catalogs, epochs, and control-plane state. Start with one process. Seamlessly grow with your system.
MODELRANGE / KV
CONSISTENCYLINEARIZABLE
TOPOLOGYMULTITENANT
OWNERP–017epoch 42 · live
REPLICA 01readable
REPLICA 02readable
KEYSPACE / CATALOG
WAL 18,442 · QUORUM 2/3
STATUS / ACTIVE DEVELOPMENT
Orbita is under active development and is not production-ready.
See the roadmap ↓Why Orbita
Small state becomes a large responsibility.
Coordination state starts with a few locks, leases, and ownership records. Then more teams depend on it, the working set grows, and one small cluster becomes several.
The other direction is a broader distributed database with an operating model larger than the problem. Orbita keeps the coordination surface focused while capacity grows underneath it.
Read the thesis01Focused surface
02Object-backed data
03Independent capacity
04Replayable failures
Why Orbita can grow
The data outlives the nodes.
Orbita separates the fast coordination path from bulk durability. Workers can move, rebuild, and multiply without becoming permanent owners of the durable dataset.
Immutable segments, one atomic manifest
Bulk durability is independent from any worker.
Rebuild directly from the bucket
No full database copy from a busy peer.
BEFOREP–017 / A—Zworker 01
→AFTERP–017 / A—Mworker 01
AFTERP–018 / N—Zworker 04
The WAL keeps writes fast. Object storage stays off the acknowledgement path.
Object storage makes workers replaceable. Durable data is not trapped on one machine.
Range partitions create horizontal capacity. Worker count grows independently from the voter group.
The product model
Easy to start. Built to expand.
The operating model changes in capacity, not in kind. Start locally, add workers when the system asks for them, then give each workload its own keyspace.
Easy to start
Run a complete development cluster as one process, with one command and a default keyspace.
Seamlessly grow with your system
Add workers as partitions, indexes, and hot data expand. Clients keep the same small KV surface.
Share safely
Keyspaces isolate credentials, quotas, limits, and workloads inside one coordination backbone.
A focused system
Built for coordination, not every database problem.
A smaller surface is easier to understand, test, operate, and trust. Orbita keeps the primitives that control-plane state needs.
Ordered keys
GET, SET, DELETE, and ordered prefix LIST keep the data model direct.
Conditional writes
Compare-and-swap and IF NOT PRESENT build locks, ownership records, and fencing patterns.
Absolute TTLs
Deadlines are part of the record, so expiry has one meaning across retries and failover.
Tenant boundaries
Keyspaces carry their own credentials, quotas, limits, and partition growth.
Correctness is a product surface
Inspect the failure. Replay the proof.
Network, disk, clock, and node failures run under a deterministic simulator. A failing seed becomes a reproducible trace and a permanent regression test.
SIMULATION / REPLAYABLE
- SEED
- emitted on failure
- FAULT
- partition owner unreachable
- CHECK
- acknowledged writes survive
- TRACE
- same schedule, every replay
- REGRESSION
- pinned when fixed
ORBITA_SIM_SEED=<seed> moon run orbita-sim:simOne process. Two commands.
Start small without starting over.
The development cluster is one process, one partition, and a default keyspace on 127.0.0.1:7100.
$ cargo run --bin orbita -- dev
orbita dev · 127.0.0.1:7100
$ orbita set default greeting hello
version 1
$ orbita get default greeting
helloRoadmap
Make the architecture real. Then publish the evidence.
The sequence matters: finish the storage and control-plane model, strengthen the test system, then widen the product surface.
Make the architecture real
Open partition format, object-backed storage and hydration, Raft control plane, combined nodes, automatic voter management, safe upgrades, and resource visibility.
Build the evidence
Broader fault injection, seeded regressions, real-cluster scale tests, published correctness results, backup and restore, an offline reader, and the operator kit.
Complete the coordination surface
Watch streams, the transaction ladder, official client support, tested coordination recipes, and migration tooling.
Harden the boundary
TLS, peer mTLS, prefix-level authorization, certificate rotation, and an administrative audit log.
From the journal
Design in public.
The coordination substrate you can verify
Why the layer that holds your locks and epochs deserves evidence, and why deterministic simulation has to start on day one.
↗One log, two quorums
Durability asks whether a write survives. Coherence asks whether anyone can still serve the old value. They are not the same vote.
↗Start with one process
Start with the system you can understand.
Run Orbita locally, inspect the architecture, and follow the work toward a production-ready coordination backbone.