RelayStack blog

Engineering notes from the RelayStack team on webhooks, retries, and running orchestration in production.

Designing webhook handlers that survive a bad day

Webhooks fail in predictable ways, including duplicates, out-of-order delivery, slow consumers, and long outages. This post describes the handler patterns that hold up, including signature checks, durable queues, idempotency keys, and replay. It shows how each pattern is expressed in a RelayStack workflow, and it explains how to test the failure cases locally before they ever appear in your production traffic.

Retry policies that do not take your partners down

A retry policy that is too aggressive turns a partner's small outage into a large one. This post explains backoff, jitter, and attempt budgets, and shows how per-destination rate limits stop a burst from overwhelming a downstream API. It also covers how to decide which errors are worth retrying and how to alert when a workflow keeps hitting the same failure.

Build or adopt, an engineering view of orchestration

Most engineering teams can build a queue and a scheduler in a few weeks, but few want to maintain them for years. This post lists what a production orchestration layer needs, including state, retries, tracing, replay, and access control, and offers a way to estimate the ongoing cost. It helps teams decide whether to build on a platform or continue to maintain their own.

Putting workflows through your continuous integration pipeline

When workflows are code, they can be tested like code. This post shows how to run workflows against recorded events in a pipeline, how to assert on traces, and how to deploy to a staging environment before promoting a change. It also describes how scoped API tokens let a deployment job do its work without being able to read production payloads.