Delivery and queues

Outbound federation never happens inside a user’s request. Publishing a post enqueues a job; the request returns immediately.

The queue abstraction

Application code calls one thing:

queue.add("federate_post", { postId });

Two backends sit behind that call:

Backend	When	Behaviour
In-process	No REDIS_URL	Jobs live in memory; lost on restart
Redis	REDIS_URL set (the Compose default)	Jobs persist; survives restarts; shared across processes

The call sites and signatures are identical either way, so switching is a matter of setting one environment variable.

What Redis backs

When REDIS_URL is set, three things move out of process memory:

1. The application job queue (federation deliveries and other background work).
2. Fedify’s key-value store and message queue — its own delivery retry state.
3. The rate limiter, so counters are shared across backend processes rather than per-process.

Delivery mechanics

For each recipient inbox, Fedify signs the request with the actor’s private key and POSTs the activity. It handles:

• HTTP signatures on every outbound request.
• Retries with backoff for transient failures.
• Shared-inbox optimisation — one delivery to a remote server serving many of your followers, instead of one per follower.

Failure modes

Symptom	Cause	Action
One remote server never receives anything	It has defederated your domain	Nothing to fix on your side
Deliveries stall after a restart	In-process queue, jobs lost	Set REDIS_URL
Persistent signature errors	Domain or scheme mismatch in generated URIs	Check the public domain setting and that Caddy sets x-forwarded-proto
Everything is slow to arrive	Large follower set, normal backoff	Watch the backend logs; this is usually just retry pacing

Federated reach

The writer dashboard’s “reach” figure counts successful deliveries to remote inboxes. It measures your instance’s sending, not anyone’s reading — there is no mechanism, and no intention, to observe reads on other servers. See Writer dashboard.

Inbound throttling

The receiving side is throttled too, so a hostile peer cannot flood you:

Variable	Default	Meaning
RL_INBOX_MAX	300	inbox POSTs per minute per source IP
INBOX_MAX_BODY_BYTES	1000000	maximum accepted payload (bytes)

Both are generous on purpose — a busy instance legitimately delivers many activities in a burst.

The redis_data volume

It holds only transient queue state and is deliberately excluded from backups — Redis re-drains it after a restart, and no durable content lives there. See Backups and restore.