Drops Slack's retried event deliveries so a handler never fires twice.

Slack re-sends an event (same event_id) when it doesn't see a timely ACK — even though both transports ACK before the handler runs, a slow network or a restart can still produce duplicates. This keeps a short-lived set of seen keys ({handler_module, event_id}) in ETS; the dispatcher consults it before dispatching.

It's a set ETS table owned by this process but read/written directly by callers (so the dedup check never blocks on a GenServer). A periodic sweep evicts entries past their TTL.

The table is node-local, so dedup is per-instance. Across a multi-node fleet a retry can land on a different instance than the original delivery (Slack load-balances Socket Mode connections, and an HTTP retry can hit a different pod), and that instance won't recognise it. Retries are rare — both transports ACK before the handler runs — so this is a best-effort safety net; hard cross-instance dedup would need a shared store.

Configuration

• config :slink, :dedup, true — master switch (default true). When false, seen?/1 always returns false and nothing is tracked.
• config :slink, :dedup_ttl_ms, 660_000 — how long a key is remembered (default 11 minutes). Slack retries a failed delivery immediately, after ~1 minute, and after ~5 minutes, so the default covers the whole schedule with margin.