SuperWorker
An Elixir library for supervised background work. Declare workers as plain functions and run them — in long-lived supervised processes, in a partitioned job pool, or as in-process pipelines — without designing a supervision tree upfront.
Note: the library is still unstable, please don't use it in production.
Which primitive do I need?
| Need | Use | Guide |
|---|---|---|
| Long-lived processes kept alive, messaged by id, restarted on crash | Supervisor (groups, chains, standalone workers) | Supervisor Guide |
| Short jobs submitted to workers, with retries, backoff and backpressure | Pool | Pool Guide |
| Data-transformation steps with per-step retry/skip/branch/resume, no processes | FunctionChain | FunctionChain Guide |
Supervisors declared in config files: Config Guide (Quickstart).
60-second quickstart
alias SuperWorker.Supervisor, as: Sup
# A supervisor holding one group worker (see the Supervisor Guide).
{:ok, _} = Sup.start_with_config([id: :sup1, num_partitions: 2, link: false])
{:ok, _} = Sup.add_group(:sup1, [id: :group1, restart_strategy: :one_for_one])
{:ok, _} = Sup.add_group_worker(:sup1, :group1, fn -> do_work() end, [id: :w1])
# Or a pool running jobs (see the Pool Guide).
{:ok, _} = SuperWorker.Pool.start_link(name: MyPool, task: fn job -> do_work(job) end, size: 20)
{:ok, result} = SuperWorker.Pool.run(MyPool, job)Guides
Start with the chooser above, then read in any order:
- Supervisor Guide — groups, chains and standalone workers, fault tolerance, introspection.
- Pool Guide — a partitioned job pool: run/cast, retries with backoff, middleware, telemetry.
- FunctionChain Guide — composable in-process pipelines with retry, branches and checkpoint/resume.
- Config Guide — declarative supervisor trees in config files (Quickstart).
Full API reference ships with the docs (mix docs) and in the Hex package.
Features
- Supervisor — groups (broadcast + direct messaging), ordered chains with backpressure, and standalone workers, each with its own restart strategy.
- Pool — partitioned workers and bounded queues, retries with backoff, crash isolation with poison-pill safety, and pluggable middleware (telemetry, circuit breaker).
- FunctionChain — in-process pipelines with per-step error policies, parallel/branch steps, runtime argument overrides, and checkpoint/resume.
- ConfigLoader — declare supervisors, groups, chains and workers in
config.exsand boot them automatically. - Introspection — list supervisors, groups, chains and workers at runtime.
Planned features
- Multiprocess per chain node.
- Auto scale for chain.
- Distributed in cluster.