| name | verify |
| description | Verify hedwig changes end-to-end by driving real SMTP delivery through the dev DNS + fake-MTA docker harness and observing spool/log state. |
Verifying hedwig end-to-end
Hedwig's outbound path needs MX resolution + a destination MTA on port 25,
so verification runs on the smtp_test docker network from dev/.
net.ipv4.ip_unprivileged_port_start=53 on this host means port 25 can't
be bound locally — everything runs in containers.
Build
Host toolchain differs from the container's; build inside the dev image with
an isolated target dir so the host target/ cache isn't clobbered:
docker run --rm -v "$PWD":/app -v <scratch>/target:/app/target-e2e \
-e CARGO_TARGET_DIR=/app/target-e2e -w /app dev-smtp:latest \
cargo build --release --bin hedwig
(dev-smtp:latest is the compose-built image from dev/Dockerfile.dev;
docker compose -f dev/docker-compose.yml build smtp if missing.)
Harness
- DNS:
docker start dev-dns-1 (or compose up the dns service) —
CoreDNS at 172.30.0.2; *.throttle.test MX → 172.30.0.4.
- Destination MTA: run a python container pinned to 172.30.0.4 on
smtp_test. For retry-lifecycle tests, use a script that keys behavior
on recipient local part at RCPT TO (421 defer / 555 bounce / 250 accept /
hang) with a per-recipient attempt counter, and logs every interaction.
dev/fake-mta/throttle.py is the always-421 original to crib from.
- Hedwig: run the built binary in
dev-smtp:latest with
--network smtp_test --dns 172.30.0.2, publish 2526, and bind-mount a
scratch spool dir to inspect queued//deferred//bounced/ from the
host. Config essentials: outbound_local = true (plaintext SMTP),
disable_outbound = false, storage_type = "fs", format = "json" logs.
Timing constants (not configurable)
- Deferred scan interval: 30s; retry backoff: 60s × 2^attempts.
With
max_retries = 2 a full defer→exhaust→bounce cycle is ~4 min.
- Hung outbound connections time out after ~60s (lettre default) and defer —
a hang recipient freezes a retry in
queued/ for that window (useful for
kill-mid-retry restart tests; use SIGKILL for crash semantics).
What to read
- Delivery outcomes:
docker logs <hedwig> | rg '"email delivery"' —
fields recipient, status (delivered/deferred/bounced/dropped),
attempt, smtp_response.
- Retry schedule ground truth: the fake MTA's own log.
- Spool state: the bind-mounted dir; meta files are
deferred/<msg_id>.meta.json and carry the attempt counter.
Alpine (musl) containers misreport DNS from the CoreDNS templates —
use glibc images (python:3.12-slim) for probe scripts; hedwig itself
(hickory-resolver) is unaffected.
