# Priorities, Delays, Length Limits and Single Active Consumer — RabbitMQ

Source: https://www.skillbyai.com/en/rabbitmq/d-features

> Use RabbitMQ's special features where they fit.

## Useful features, used carefully

**Priority queues** let urgent messages overtake others: classic queues support up to 255 levels through `x-max-priority` (use a small number such as 5), and quorum queues in 4.x support a simplified two-level priority (normal and high). Priorities only help when there is a backlog. **Delayed delivery** can be built with TTL plus dead-lettering (see the retry pattern) or with the **delayed message exchange** plugin; for long or high-volume delays, a scheduler or database is often better. **Queue length limits** (`x-max-length`, `x-max-length-bytes`) cap backlogs; the **overflow** behaviour decides whether to drop the oldest messages (`drop-head`, the default) or reject new publishes (`reject-publish`, which publishers see as nacks with confirms). **Single active consumer** (`x-single-active-consumer`) lets several consumers attach to a queue while only one receives messages at a time, with automatic failover: useful when strict ordering matters but you still want a standby. The **consistent hash exchange** plugin spreads messages over several queues by a hash of the routing key, keeping each key's messages in one queue.

## Ordered processing with a standby consumer

Two instances attach; one is active and the other takes over if it dies.

```python
channel.queue_declare(
    queue="ledger.entries",
    durable=True,
    arguments={
        "x-queue-type": "quorum",
        "x-single-active-consumer": True,       # strict order, automatic failover
        "x-max-length": 1_000_000,
        "x-overflow": "reject-publish",          # publishers get nacks instead of silent drops
    },
)

# both instances run the same code; RabbitMQ delivers to one at a time
channel.basic_qos(prefetch_count=1)
channel.basic_consume("ledger.entries", on_message_callback=apply_ledger_entry)
```

## One cashier with a relief cashier waiting

Single active consumer is one cashier serving customers in strict order, with a relief cashier sitting behind who steps in the moment the first one leaves. Customers are never served by both at once.

**Quiz:** What does the `reject-publish` overflow setting do when a queue reaches its length limit?

- [ ] Deletes the oldest messages silently
- [ ] Doubles the limit automatically
- [ ] Moves the queue to another node
- [x] Rejects new publishes, which publishers see as nacks when confirms are enabled

*Answer:* Rejects new publishes, which publishers see as nacks when confirms are enabled. reject-publish refuses new messages instead of dropping old ones, signalling back-pressure to publishers.