Active-active across datacenters
Run independent Zaris clusters in two or more regions and replicate between them over a resumable, pull-based change feed. Every site takes writes locally and serves in-region reads — and keeps running on its own if the link between them ever drops.
Two clusters, one converging keyspace
This is not a stretched cluster. Each region is a complete, self-coordinating Zaris cluster that owns its own data. A durable change feed carries writes to the other side, both ways.
Global reach without a coordinator in the middle
Low local latency everywhere, independent availability per site, and a keyspace that converges on its own — with the observability to prove it.
Active-active, not failover
Every site serves reads and takes writes locally — no primary region, no promotion step. Users hit the datacenter nearest them and get in-region latency on both the read and the write path.
Independent by design
Each site is its own self-coordinating cluster, not a stretched one. A full inter-site link loss does not take anyone down: both sides keep serving and reconcile automatically when the link returns.
Converges deterministically
Concurrent writes resolve with hybrid-logical-clock last-writer-wins, and counters use a PN-counter so concurrent increments from different regions are all preserved — every region settles on the same value without a coordinator arbitrating in the middle.
Resumable from a durable cursor
Replication is a pull-based change feed with a durable per-site cursor. After an outage or a restart, each site resumes exactly where it stopped — no full resync, no lost window.
Observable end to end
Automatic site.* counters in both directions — applied, skipped, conflicts, bytes, and live rates — plus an RPO reported as the age of the oldest un-applied op: ≈0 when caught up, real backlog-seconds when behind.
Additive and off by default
A store with no geo configuration behaves exactly as before. Turn replication on per store, and manage it live from the web console or PowerShell — no redesign, no new moving parts to operate.
A replication map you can actually read
- Animated topologyThe web console draws the live picture — active vs passive sites, one-way vs bidirectional lanes, and per-site item counts — so you can see the shape of your deployment at a glance.
- Health banner & dashboard calloutEach lane reports healthy, falling-behind, or stalled, surfaced as a banner and a dashboard callout — no log-grepping to know where you stand.
- RPO you can trustReplication lag is reported as the age of the oldest un-applied operation: about zero when caught up, and climbing to real backlog-seconds the moment a site falls behind.
- Managed liveConfigure, switch, suspend, and resume replication from the console or PowerShell cmdlets. A geo-aware clear replicates to peers through versioned tombstones.
Put the data where your users are
Serve users near them
Put a cluster in every region your users live in. Each one answers locally, so a customer in Frankfurt and a customer in Virginia both see cache-speed reads and writes — on the same data.
Disaster recovery & regional failover
Lose a whole region and the others keep serving the same keyspace. When the region comes back, it resumes from its cursor and catches up on its own.
Data locality & sovereignty
Keep a live copy of the data resident in each jurisdiction while still converging to one consistent view, so in-region apps never cross a border to read or write.
Zero-downtime regional maintenance
Drain, patch, or upgrade an entire region on your schedule. Traffic stays on the peers, and the region rejoins and resyncs when you bring it back.
Go multi-region on your own terms
Geo-replication is additive and off by default — add it to the stores that need it, and manage it live. Validated on a two-region cloud deployment over private network peering.