Deterministic, seeded streams of operations for Jepsen-style workloads.

A generator emits a sequence of opaque "ops" — tuples or atoms describing what a worker should do next. Workers pull from generators in a loop; the operations themselves are recorded via Lockstep.History.

Built-in generators

• random/2 — uniform random over a pool of ops, seeded so iterations are reproducible.
• cycle/1 — fixed sequence repeated forever.
• weighted/2 — biased random sampling.
• mix/2 — interleave outputs from multiple generators round-robin.
• take/2 — limit any generator to N ops total.

Quick example

gen =
  Lockstep.Generator.random(
    [:read, {:write, 1}, {:write, 2}, {:cas, 1, 2}],
    seed: 42
  )
  |> Lockstep.Generator.take(50)

Lockstep.Generator.each(gen, fn op ->
  Lockstep.History.op(history, op_kind(op), op_args(op), fn ->
    apply_op(reg, op)
  end)
end)

Generators are immutable; next/1 returns {op, new_gen} or :done. They DON'T allocate processes — pass a generator into a closure and the worker advances it locally with no controller involvement.

Seeding

When a generator is created without an explicit :seed, it derives one from :erlang.unique_integer/1 — fine for one-off use, but iterations may diverge. For Jepsen-style reproducibility, pass an explicit seed (typically derived from the iteration's seed via phash2({iter_seed, worker_id})).