Skip to content

Introduction

View as Markdown

Durable Runner is an open-source, self-hosted background-job runner. You submit a versioned job — a name for the work plus a list of named steps — and it is acknowledged only after the record is durable on disk. A job the server said yes to survives the process dying: one framed append-only journal, written by a single writer under an exclusive lock, is the only authority. No database queue, no Redis.

Each step calls an HTTP handler you register and allowlist, and every attempt is at-least-once under one idempotency key — so a reply that vanished is reconciled, never repeated blindly. When a run finishes, its history expires on a schedule; the deduplication record stays so a resubmitted operation key returns the same run.

  • Developers submit, inspect, list and cancel runs with the durable CLI or plain curl against the HTTP API, and write step handlers in any language — a handler is just an endpoint that takes a POST and honours an idempotency key.
  • Operators run one Rust binary next to a filesystem: backups are consistent and verified, restores are checked before a byte is trusted, compaction keeps history bounded, and every queue, payload, retry and connection has a limit the server admits to instead of queueing silently.
Piece What it is Who uses it
durable-server The coordinator: the HTTP API, the engine and the one journal writer Your deploy
HTTP API POST /jobs, GET /jobs/{id}, cancel, compact, backup, diagnostics The CLI, your scripts, the inspector
durable CLI Submit, inspect, list, cancel, compact, backup and restore from a shell Developers, operators
Step handlers Your own HTTP endpoints, allowlisted at startup The work itself
Inspector A small read-mostly UI served by the server at /inspector/ Operators

Was this page helpful?

Either answer opens a short issue on GitHub, so you can say what worked or what was missing.