Skip to content

Performance · Published 26 August 2026

Benchmarking recipient mail infrastructure without hiding queue behaviour

A reproducible test plan for SMTP acceptance, LMTP handoff, durable write, webhook delivery, duplicates, and recovery.

Measure each acknowledgement boundary

Record SMTP connection, greeting, recipient validation, end-of-data response, LMTP handoff, durable raw write, normalized write, and webhook enqueue separately. The public acceptance response must correspond to a durable state the system can recover.

Publish message size, recipient count, TLS mode, attachment mix, concurrency, and storage condition with every latency result.

Queue pressure is the real workload

Slow storage, a failing webhook consumer, DNS delays, and repeated remote delivery attempts are normal mail conditions. Apply bounded pressure until each queue reaches its designed limit and confirm that backpressure does not become data loss.

Measure oldest-message age and recovery time, not only average throughput.

Duplicate and crash testing

Retry the same SMTP transaction where protocol semantics permit, duplicate LMTP delivery, replay a signed webhook, kill the process between raw and normalized writes, and restart with queued work.

The expected outcome is one canonical message identity, durable raw source, idempotent projections, and an operator-visible failure state.

Compare the same product boundary

Inbound parsing services and authoritative mailbox systems solve adjacent jobs. Compare whether the provider is the MX, what it durably stores, how webhooks authenticate, how retries work, and how owners export data.

Do not compress those rows into one generic deliverability score.

Production checklist

← Engineering