It fits the estate you already run.
AnantState runs in your own cloud account or network, on a standard relational database, and reads events from the systems you have. Your warehouse stays yours, and a new domain is a definition your team can read and change.
In plain termsYour cloud, your network, your warehouse.
At a glance
Three topologies
Single-host containers, Kubernetes, or your cloud account with a managed database.
Events from where they live
Streams, REST sources, scheduled pulls or bulk import.
A definition, not a fork
A new area of the business is configuration your team can read and change.
Observable
Health, throughput and latency exposed as metrics for your operations team.
The questions architecture asks
- Where does it run, and what does it need from us to run?
- How do events get in, and what happens to the ones that fail?
- What does it replace, and what does it sit beside?
- How do we customize it without forking it?
Three supported topologies, the same services
Three supported topologies, the same services
- DatabaseIn: Single-host containers
- API serviceIn: Single-host containers
- Web applicationIn: Single-host containers
- Identity provider: Optional, for testingIn: Single-host containers
- The same services: Deployed to a cluster you already runIn: Kubernetes
- Application servicesIn: Cloud with a managed database
- Managed database: An instance your team already operatesIn: Cloud with a managed database
- Provisioning and retention scriptsIn: Cloud with a managed database
- Keycylinder: A standard relational database; stack: The same services; external: Optional
However an event arrives, it lands the same way
Every route lands as an event
- Streaming: A live stream; state advances in real timeLeads to: An event
- REST source: A system can push, or be polled, over HTTPLeads to: An event
- Scheduled pull: A warehouse or batch system holds the historyLeads to: An event
- File or bulk import: Backfilling a population for a new domainLeads to: An event
- An event: Attributed to a domain, mapped to an entityLeads to: State update; Dead-letter view (could not be mapped)
- State update
- Dead-letter view: What could not be mapped, with the reason
- Keyqueue: Where every route converges; cylinder: Stored state; external: Kept in view, not lost
Concerns architecture is likely to raise
| The concern | The answer |
|---|---|
| Another platform to run. | In every topology the store is a standard relational database, and a vanilla install is a fully supported target. Migrations run at boot. |
| Lock-in. | Your warehouse stays yours, and a domain is a definition in your own vocabulary rather than code in ours. |
| An integration project. | A connector is a convenience and nothing depends on one. If a system can emit an event or answer an HTTP call, it can be a source. |
| Upgrades. | An upgrade does not require your data to be migrated first. |
Bring your topology
Bring:
- Your target topology
- Two or three event sources
You leave with how events would land, and where it would run. We don’t publish a deployment timeline, so we will say what a first domain would need.
Bring your topology and your event sources
We will map how events would land and where it would run.