Class RabbitMqBroker

Creates a new RabbitMQ broker.

Connects to RabbitMQ.

Disconnects from RabbitMQ.

Boolean port member (M70c): false only when positively unreachable.

A probe that could not answer inside the bound reports true — "not known down" — because the port contract is boolean and the honest tri-state is what RabbitMqBroker.reachability carries. This is the ServiceBusBroker.isHealthy() shape exactly.

Checks if the broker is connected (lifecycle — M70c).

true while connect() has run and disconnect() has not, even during a reconnect window: the lifecycle is intact, the backend is what is down. Reachability is isHealthy/reachability.

Publishes a message to a topic.

Publishes with framework-owned headers under the configured acknowledgement bound.

Backend reachability (M70c; a real round trip since M95b §3.6).

false while the supervisor is in a fault window (the connection dropped and the drive-mode reconnect has not yet succeeded) — that short-circuit performs no round trip and is read ahead of the cache — and false with no open connection. Otherwise the answer comes from RabbitMqBroker.#openThrowawayChannel, cached 5 s and bounded 2 s, resolving undefined — could not determine — for a probe that does not answer inside the bound.

The M70c probe was Promise.resolve(!this.#supervisor.faulted) — a flag read, which a HUNG broker never trips (a paused broker keeps its socket; only a stopped one drops it), so a paused broker reported reachable: true while no AMQP handshake could complete.

The cache and the bound live here rather than only at the health indicator, because the indicator is not the only caller: the realtime backplane's 'messaging' transport delegates to isHealthy() directly and documents that it keeps no cache of its own. Without this, a real round trip fired once per health poll and a paused broker left that caller waiting on a promise that never settled.

One bounded-window cost is named rather than hidden: against a PAUSED broker the channel open itself never settles, so each abandoned poll leaves one pending open that the finally cannot yet run for; the pending opens settle and close on recovery (or with the connection at disconnect()), and amqplib buffers them, so the window is transient rather than a leak. The 5 s cache also bounds how many such pending opens a polling endpoint can accumulate.

Sends a request and awaits a single correlated reply.

Sends request-reply traffic with framework-owned headers.

Registers a responder whose result is returned to the requesting caller.

Subscribes to a topic.

Subscribes through the header-aware internal path.