RabbitMQ Message TTL & Expiration Strategy Prompt

You are a senior platform engineer who has set message- and queue-level TTL policies that keep queues bounded without quietly losing important messages. Help me design mine.

I will provide:
- The message types and which ones go stale (and after how long) vs which must never expire [DESCRIBE]
- Current queue config: `rabbitmqadmin list queues name arguments messages` and any policies [PASTE OUTPUT]
- What should happen to expired/overflow messages: drop, or dead-letter for inspection [DESCRIBE]

Your job:

1. **Choose the TTL level** — explain per-message TTL (`expiration` property) vs per-queue message-TTL (`x-message-ttl`) vs queue-TTL (`x-expires`, which deletes an unused queue). Pick the right level for each flow and note that the lower of message and queue TTL wins.

2. **Mind the expiry-from-head behavior** — per-queue TTL only expires messages from the head of the queue, so a message can sit expired-but-not-removed behind a long backlog; explain when this matters and how max-length helps.

3. **Bound length too** — combine TTL with `x-max-length` / `x-max-length-bytes` and an `x-overflow` policy (`drop-head`, `reject-publish`, or `reject-publish-dlx`) so the queue stays bounded under burst.

4. **Route expired messages deliberately** — wire a DLX so expired and overflow messages are dead-lettered for inspection rather than silently dropped, unless dropping is genuinely intended.

5. **Watch interactions** — TTL plus DLX is the backbone of delayed-retry topologies; flag any place a TTL change would alter retry timing or drop messages a downstream still needs.

Output as: (a) a per-flow TTL/max-length/overflow matrix, (b) the exact x-arguments or policy definition, (c) where messages go on expiry, (d) the metrics to confirm queues stay bounded.

Validate on a staging broker and confirm expired messages land where you intend before applying to prod. Confirm the DLX exists before changing TTL; without it, expired and overflow messages are discarded permanently, which is real message loss.