RELAY
A distributed task queue in Go for systems that cannot afford to lose work. Durable by default. Simple enough to read end to end.
PROBLEM
Requests look instant but they are not. The real work happens after you return the response: emails, payments, image processing. This is exactly where failures hide, in the background, where nobody is looking. Most queues make it worse: they lock you into a vendor or bury the behavior you need to trust behind abstractions you cannot inspect.
SYSTEM
Relay strips it down to the essentials. One Redis instance. Deterministic state. Workers that do one thing well. No coordination layer, no hidden magic. Relay is a small, scalable system you can read in an afternoon and know exactly how it behaves when things go wrong.
Every job runs. Nothing is silently dropped.
Sorted sets keep execution order exact.
Exponential backoff with jitter, out of the box.
Failed jobs are preserved for inspection, never erased.
Add workers, throughput follows. Linearly.
Prometheus endpoint exposed by default.
8K+
JOBS / SEC
<12ms
P50 LATENCY
25
WORKERS
POSITIONING
Not a managed service you rent by the month. Relay is for teams that want to own their queue with readable code, predictable behavior, and no black boxes. Run it on your own infrastructure. Understand every line of it.