# `Honker.Queue`

Named queue operations. All methods take a `Honker.Database` and
the queue name as the first two args.

    {:ok, id} = Honker.Queue.enqueue(db, "emails", %{to: "alice@example.com"})
    {:ok, job} = Honker.Queue.claim_one(db, "emails", "worker-1")

# `cancel`

Delete a pending or processing job by id. Returns `{:ok, true}` iff
a row was removed. Idempotent on missing.

IMPORTANT: cancel does NOT interrupt a worker currently running the
handler. It invalidates the worker's claim — its next ack/heartbeat
returns false. If you need the handler to actually halt, build that
signal in your app.

# `claim_batch`

Atomically claim up to `n` jobs. Returns `{:ok, [%Job{}, ...]}`,
possibly empty if the queue had no eligible rows.

# `claim_one`

Claim a single job or `{:ok, nil}` if the queue is empty.

# `enqueue`

Enqueue a job. `payload` is any term — it's JSON-encoded on the way
in. Options:

  * `:delay`    — seconds from now before claimable
  * `:run_at`   — absolute unix epoch (ignored if `:delay` is set)
  * `:priority` — higher = picked first within queue (default 0)
  * `:expires`  — seconds from now; drops out of claim pool after

# `enqueue_tx`

Enqueue inside an open transaction. The job is only visible to
claimers after `Honker.Transaction.commit/1`.

`queue_opts` lets the caller pass per-queue options
(`[visibility_timeout_s: _, max_attempts: _]`) when the transaction
handle was obtained without going through a `Honker.Database`
carrying a registered queue config. Pass `[]` to use the defaults
(300s / 3 attempts).

# `get_job`

Read a single job row by id. Returns `{:ok, map}` with the row
fields, or `{:ok, nil}` on miss.

# `sweep_expired`

Sweep expired processing rows back to pending. Returns
`{:ok, rows_touched}`.

---

*Consult [api-reference.md](api-reference.md) for complete listing*