Actor-Web

Appearance

Production operations ​

Actor-Web ships a local-first runtime plus the pieces needed to run nodes in production — transport, gateway, telemetry, and discovery. This page is the map; the full procedures live in the Production Operations Runbook.

Deployment lanes ​

The runbook defines progressive lanes — local demo, staging, production — so you adopt only the hardening you need. Each lane is explicit about what is proven vs. what an app must still provide (durable storage, secrets, TLS).

Ownership boundary ​

Actor-Web owns runtime behavior, transport, gateway projection, and telemetry surfaces. Your deployment owns process supervision, TLS termination, secret rotation, durable storage, and alerting. The runbook draws this line precisely so nothing falls between the two.

Runtime status & telemetry ​

  • Status — each served node exposes a runtime status surface (peers, transport health, gateway streams) for liveness checks and dashboards.
  • Telemetry — transport events can be exported to a durable sink (JSONL by default) for inspection; OpenTelemetry wiring is left to the app.

Incident triage ​

The runbook includes focused runbooks for the failure modes the transport layer surfaces:

  • Stale-peer and rejected-peer triage (membership/incarnation issues).
  • Replay recovery (gateway sequence gaps).
  • Duplicate-drop and backpressure incidents (delivery and queue limits).

Each is a short, observable checklist tied to the telemetry the runtime emits.

Delivery guarantees in production ​

Actor send is at-most-once by default. Transport adds idempotency and retry for control traffic, but design application behaviors to tolerate a dropped message rather than assume exactly-once. See Transport.

Not yet production-hardened

Some surfaces (durable replay providers, full OTel integration, hardened node transport) are intentionally incomplete and tracked as follow-ups in the runbook. Treat the lanes above as a maturity gradient, not a finished SRE platform.