Mail Queue & Retries

When a sender hands a message to an MTA, delivery may not happen immediately. The MTA queues it.

Queue lifecycle:

1. Message accepted with 250 → written to queue (filesystem files in Postfix; database in others).
2. Queue manager picks it up; resolves recipient's MX records.
3. Connects to recipient's MTA; transfers message.
4. Success → remove from queue.
5. Temporary failure (4xx) → retry later (exponential backoff: 5min, 15min, 1h, 4h, ...).
6. Permanent failure (5xx) → bounce (return to sender with DSN).
7. Time-out (default 5 days) → bounce.

incoming/
  <message-id>.envelope (recipients, sender, options)
  <message-id>.message  (headers + body)
deferred/
  <retry queue>
bounced/
  <DSN messages>

Retry policy:

• Standard: 5 days max retention.
• Exponential backoff: ~doubling delay.
• Per-recipient: a message to 10 recipients with 2 failed: queue retries those 2 only.

DNS lookup:

• MX records: example.com. 86400 IN MX 10 mail.example.com.
• Lower priority = preferred. Try lowest first; on failure, try next.
• A/AAAA records for the MX hostname.
• Caching to avoid hammering DNS.

Concurrency:

• Many messages in queue → many parallel connections.
• Limit per destination IP/domain to avoid being rate-limited by recipient.

Failure handling:

• 421 (service not available) → defer.
• 4xx (temporary) → retry.
• 5xx (permanent) → bounce.
• Connection refused / timeout → defer.

Postfix has separate queues: incoming, active (currently delivering), deferred (waiting retry), corrupt, hold.

Hold queue: admin can pause messages (e.g., for review) and release later.