Getting Started

Microservice Orchestration & Distributed Systems


Coordinating multiple microservices into a reliable end-to-end transaction is one of the hardest problems in distributed systems. A payment service might succeed while the inventory service fails, leaving your data inconsistent and your customers confused. Orcher solves this by letting you write distributed workflows as regular code — with every cross-service call journaled to PostgreSQL.

The transaction state, made explicit

The classic distributed failure isn't that a service goes down — it's that afterwards nobody knows where the transaction stands. When every service call runs through a journaled task, that question always has an answer.

One Transaction, Three Services
✓
payment-serviceCharge accepted — journaled
✗
inventory-serviceCall fails — service is down
=
the journal knowsPayment done, inventory not, shipping never called
→
resume or roll backRetry inventory when it recovers — or compensate the payment
No reconciliation jobs, no "did it happen?" — both recovery paths start from a known state.

Either recovery path is durable: retries pick up exactly the calls that never completed, and rollback runs saga compensations in reverse order — themselves journaled, so an interrupted rollback resumes instead of restarting.

Compose along service boundaries

Child workflows let the orchestration mirror your architecture: one child per service, each owning its service's steps, retries, and failure handling — independently testable, reusable across parent workflows.

order_fulfillment
parent — owns the end-to-end transaction
start_child_workflow × 3
payment_flow
own journal · retries · tests
gRPC
payment-service
inventory_flow
own journal · retries · tests
gRPC
inventory-service
shipping_flow
own journal · retries · tests
gRPC
shipping-service

The services stay stateless — the transaction's state lives in the workflows that call them.

See Child Workflows for Service Composition for the composition mechanics, and Multi-Service Deployment for running the children in parallel across services or regions.

Waiting on other systems

Distributed transactions rarely stay inside your infrastructure. A step may need a webhook from a third party, a human approval, or a callback that arrives hours later — event gates pause the workflow at zero resource cost until the answer arrives or a timeout decides for it.


Patterns