PostgreSQL persistence backend for the Sayiir workflow engine.

This crate provides [PostgresBackend], a production-grade implementation of SnapshotStore, SignalStore, and TaskClaimStore backed by PostgreSQL via [sqlx].

Features

• Codec-generic: Serialise snapshots with any codec (JSON for debuggability, rkyv/bincode for speed). The data column is always BYTEA.
• ACID transactions: Composite signal operations (check_and_cancel, check_and_pause, unpause) use single Postgres transactions with SELECT … FOR UPDATE for true atomicity.
• Snapshot history: Every checkpoint is appended to an immutable history table for debugging, auditing, and future replay.
• Observability-ready: Indexed metadata columns (status, current_task_id, completed_task_count, error, timestamps) plus a denormalised sayiir_workflow_tasks table enable monitoring without deserialising blobs.
• Distributed task claiming: TaskClaimStore with TTL-based claims, expired-claim replacement, and soft worker bias.

PostgreSQL version support

Minimum supported version: PostgreSQL 13. The schema uses INSERT … ON CONFLICT DO UPDATE (9.5+) and ALTER TABLE … ADD COLUMN IF NOT EXISTS (9.6+); 13 is the floor because it is the oldest major release still receiving security patches. Integration tests run against both 13 and 17.

Quick start

use sayiir_postgres::PostgresBackend;
use sayiir_runtime::serialization::JsonCodec;
use sayiir_persistence::SnapshotStore;
use sayiir_core::snapshot::WorkflowSnapshot;

# async fn example() -> Result<(), Box<dyn std::error::Error>> {
let backend = PostgresBackend::<JsonCodec>::connect("postgresql://localhost/sayiir").await?;

let snapshot = WorkflowSnapshot::new("order-123".to_string(), "hash-abc".to_string());
backend.save_snapshot(&snapshot).await?;

let loaded = backend.load_snapshot("order-123").await?;
# Ok(())
# }