Event Sourcing

TensorDB’s append-only, immutable architecture makes it a natural fit for event sourcing.

What is Event Sourcing?

Instead of storing current state, you store a sequence of events. The current state is derived by replaying events.

TensorDB as an Event Store

use tensordb::Value;

// Append events
db.put("account/acc-001/events/1", Value::Text(
    r#"{"type":"AccountOpened","balance":0}"#.into()
))?;

db.put("account/acc-001/events/2", Value::Text(
    r#"{"type":"Deposited","amount":500}"#.into()
))?;

db.put("account/acc-001/events/3", Value::Text(
    r#"{"type":"Withdrawn","amount":100}"#.into()
))?;

// Replay events to build current state
let events = db.scan_prefix("account/acc-001/events/")?;
let mut balance = 0i64;
for (_key, value) in events {
    // Parse and apply each event
    // balance = apply(balance, event);
}

Benefits

Feature	Benefit for Event Sourcing
Immutability	Events are never modified — append-only by design
Temporal queries	Replay to any point with AS OF
Change feeds	Subscribe to new events in real time
Full history	No data loss — every event is preserved permanently
SQL queries	Materialize views with SQL aggregations

Change Feeds

Subscribe to events in real time:

let mut feed = db.subscribe()?;
while let Some(event) = feed.next() {
    println!("New event: {:?}", event);
    // Update materialized views, notify downstream systems
}

Snapshots

For performance, periodically create snapshots:

// Store a snapshot of the current state
let snapshot = calculate_current_state(&events);
db.put("account/acc-001/snapshot", Value::Text(snapshot.to_json()))?;

// Later: load snapshot + replay events since snapshot