Durable workflow examples

Two implementations of the same order-fulfilment app — python/ (SDK runr-workflows, source in sdk/python) and node/ (SDK @runr/workflows, source in sdk/node):

Workflow	Trigger	Shows
`process-order`	event `order.created`	steps with retries and `RetryAfterError`, a 10-minute sleep, `waitForEvent` with `match` and a timeout, compensation, `sendEvent`, keyed concurrency, idempotency, `cancel_on`
`nightly-report`	cron `0 2 * * *` (Europe/Istanbul)	`invoke` of another workflow, handling `StepError`
`render-report`	— (invoked)	a child workflow
`resize-image`	queue `images`	the queue bridge (`queue/images` events), a workflow-wide concurrency limit

Deploy#

Create a FaaS project (runtime Python or Node.js) with the trigger Workflow; for resize-image set the queue stream images (the region needs Redis).
Add the files of python/ or node/ (handler main.handler) and deploy. RunR injects RUNR_WORKFLOW_SIGNING_KEY; the SDK answers only requests signed with it.
Open the project's Workflows tab: the definitions sync automatically once the deployment is active (or press Sync definitions). runr.workflows.json declares the same definitions for the case the function cannot be reached during a sync.

Try it#

TOKEN=...   # an API token (Settings → API Tokens)
API=https://runr.example.com/api

# Start a run through an event
curl -sS -X POST "$API/events" -H "Authorization: Bearer $TOKEN" -H 'Content-Type: application/json' \
  -d '{"name":"order.created","id":"order-o-1","data":{"order_id":"o-1","customer_id":"c-1","total":120,"items":["book"]}}'

# 10 minutes later the run waits for the shipment; resume it
curl -sS -X POST "$API/events" -H "Authorization: Bearer $TOKEN" -H 'Content-Type: application/json' \
  -d '{"name":"order.shipped","data":{"order_id":"o-1","tracking":"T-1"}}'

# Start a run directly
curl -sS -X POST "$API/projects/<project-id>/workflows/nightly-report/runs" -H "Authorization: Bearer $TOKEN" \
  -H 'Content-Type: application/json' -d '{"data":{}}'

The events above fit the Python example. The node example reads camelCase fields: send {"orderId": "o-1", "customerId": "c-1", …} as the event data there.

Without a cluster, python/local_run.py plays the engine against main.py through the real protocol (pip install -e sdk/python first).