What's new in RELAY

Private Beta launch. Cron scheduling, Go SDK, and real-time log streaming.

• Cron scheduling for jobs — define schedules with standard cron syntax or our fluent builder API.
• Go SDK (v1.0.0) — full parity with the Node.js and Python SDKs.
• Real-time log streaming in the dashboard — watch job execution logs as they happen.
• Bulk job enqueue — enqueue up to 1,000 jobs in a single API call.
• Edge routing algorithm now considers real-time queue depth per region for smarter load balancing.
• Dead-letter queue payload display was truncating JSON payloads over 16 KB.

Stability fixes and improved retry jitter.

• Exponential backoff retry was not applying jitter correctly on the 4th and subsequent attempts.
• Webhook delivery sometimes duplicated events when the target responded with a 2xx after a timeout.
• CLI output is now more concise with a `--verbose` flag for full detail.
• Updated all transitive dependencies. No known CVEs affected.

Metrics export, Python SDK GA, and dead-letter queue replay.

• Metrics export to Datadog and Prometheus — queue depth, execution latency, error rate, and throughput.
• Python SDK is now generally available (was previously in alpha).
• One-click replay for any job in the dead-letter queue from the dashboard.
• Queue pausing — pause and resume queues without losing jobs.
• Dashboard load time reduced from ~1.8s to ~320ms by caching queue stats at the CDN edge.

Alpha launch. Node.js SDK, 5 edge regions, and the first dashboard.

• Node.js SDK (TypeScript-first) with full type inference from payload schema.
• Job execution in 5 edge regions: US-East, US-West, EU-West, AP-Southeast, AP-Northeast.
• Web dashboard with live queue stats, job browser, and log viewer.
• Dead-letter queue with basic inspection UI.
• CLI for project init, job deploy, and status check.

Internal beta. Core runtime stabilization.

• Job handler cold-start reduced to under 50ms for all supported runtimes.
• REST API stabilized — all endpoints versioned under /v1/.
• Race condition in job state machine could cause a job to run twice in rare network-split scenarios.