AD sign-in over LDAPS (search with the service account, bind as the user, nested groups through LDAP_MATCHING_RULE_IN_CHAIN, disabled accounts refused), roles as the union of PS-* groups mapped by a table derived from the seeded departments. Server-side sessions with idle, lock and absolute timeouts, Origin and X-Requested-With check, step-up unlock. Optional Argon2id PIN set with the domain password; PIN sign-in by employee ID on registered devices only; five wrong attempts disable the PIN, three wrong at unlock or signature end the session. Rate limits per account and per address before any bind. Devices, signatures table and sign helper, audit rows for every identity event, SSE over LISTEN/NOTIFY with presence. Account settings and Admin APIs and screens (AD view, devices, PIN clearing, group map with missing-group flags, refresh from AD now). CLI ad-groups --list and ad-refresh. Migration 0002. Frontend: real sign-in (password tab, PIN keypad on registered devices), lock screen, role switcher, Account settings, Admin shell, screen placeholders. CI: Samba AD DC container (PSTEST.LOCAL) with a generated CA, 49 of the 50 groups plus one extra, 17 test users; wired into the CI compose overlay and the Gitea workflow; Playwright sign-in scenario. Verified on docker-1: 171 server tests green inside the CI stack against PostgreSQL and the Samba DC, 48 frontend tests, 8 Playwright scenarios against the live stack (password and PIN sign-in, lock and unlock, Admin group map). Findings recorded in the docs: the DC needs CAP_SYS_ADMIN, a hidden tab locks only after a minute as the old build did, and the app uses the alias ps-app for Playwright because "app" is on Chromium's HSTS preload list. Co-Authored-By: Claude Fable 5.1 <[email protected]>
Samba AD DC for CI
A throwaway Active Directory domain controller for CI and local end-to-end runs. It lets
sign-in run over real LDAPS against a real directory without touching PESCO's AD.
It is only ever started by deploy/compose.ci.yaml. Nothing here is used by the test or
production stacks.
What it is
- Domain
PSTEST.LOCAL, NetBIOSPSTEST, baseDC=pstest,DC=local, hostdc1.pstest.local. - Debian bookworm with the distribution's Samba. Built from
Dockerfilein this folder. - No published port. The app reaches it on the internal
backendnetwork asldaps://dc1.pstest.local:636. Port 389 also answers, for STARTTLS tests. - Simple binds work only over TLS (
ldap server require strong auth = allow_sasl_over_tls). A plain LDAP simple bind is refused, as on PESCO's DCs.
Files
| File | Job |
|---|---|
Dockerfile |
Debian bookworm-slim, samba, ldb-tools, openssl, tini. |
entrypoint.sh |
Provisions on first start, makes the TLS certificates, seeds once, refreshes /export, runs samba in the foreground. |
seed.sh |
OU tree, service account, groups, test users, users.json. |
healthcheck.sh |
Healthy once seeded, exported, and LDAPS verifies against the exported CA. |
groups.txt |
The 50 group names, a copy of packetstation ad-groups --list. A server test keeps them equal. |
First start
samba-tool domain provisionwith--use-rfc2307 --dns-backend=NONE. The Administrator password comes from the compose secretsamba_admin_password. Setting the sysvol NT ACLs needssecurity.*xattrs, so the service runs withcap_add: [SYS_ADMIN](verified on docker-1 with Samba 4.17; without it the provision step fails with access denied).- openssl makes a CA (
PSTEST CI CA) and a server certificate fordc1.pstest.localanddc1, valid 10 years, under/var/lib/samba/private/tls/. smb.conf points at them. - The password policy is relaxed: no complexity, no age, no history.
seed.shruns once against the localsam.ldb, before samba starts.- smb.conf is kept on the data volume (
/var/lib/samba/etc/smb.conf), so a recreated container starts from the same domain.
Provisioning and seeding take about one to two minutes. The healthcheck allows 120 seconds of start period and 30 retries after that.
Later starts skip provisioning and seeding but always refresh the export.
What seed.sh creates
OU=PESCO_FARMINGTONwithOU=Groups,OU=Groups->OU=PacketStation, andOU=Users.OU=Admin->OU=System_Accounts.CN=svc-packetstation,OU=System_Accounts,OU=Admin,DC=pstest,DC=local. Its password is theldap_bind_passwordcompose secret, the same one the app reads. It never expires.- In
OU=PacketStation: the 50 groups fromgroups.txtexceptPS-14-Electrical-Lead(missing on purpose), plusPS-Extra-Legacy(extra on purpose). Global security groups. - In
OU=Groups:Shop-Welders-Nested, a member ofPS-10-WeldAssembly-Worker. - In
OU=Users: the 17 test users from the phase 1 contract. CN is the account name. employeeID and displayName are set withldbmodify.nested.nedis only in the nested group.disabled.danis disabled.noemp.norahas no employeeID.dup.daveanddup.dinashare 9001.norole.nickis in no group.
Every user gets a new random password on each seed. Nothing secret is committed.
The export volume
The DC writes to the ci-dc-export volume, mounted at /export:
ca.pem: the CI CA. migrate, app and worker mount the volume at/etc/ssl/pstest, andci.envsetsLDAP_CA_BUNDLE=/etc/ssl/pstest/ca.pem.users.json: the test accounts, mode 0444.{"domain": "pstest.local", "users": {"<sam>": {"password", "emp_no", "display_name"}}}.emp_noisnullfornoemp.nora.seeded: the time of the seed. The healthcheck waits for it.
The Playwright service mounts the same volume read-only and reads E2E_DC_USERS_FILE.
Running the CI stack on docker-1
From the repository root:
bash deploy/ci/make-secrets.sh
CI="docker compose -p packetstation-ci --env-file deploy/ci/ci.env -f deploy/compose.yaml -f deploy/compose.ci.yaml"
$CI up -d --build --wait --wait-timeout 300
$CI --profile e2e run --rm playwright # report in tests/e2e/playwright-report
$CI logs samba-dc # provisioning and seeding log, no secrets
$CI --profile e2e down -v
To run the server's DC-backed tests from docker-1 against this DC, copy the export out:
mkdir -p tests/fixtures/local/dc
docker run --rm -v packetstation-ci_ci-dc-export:/export:ro -v "$PWD/tests/fixtures/local/dc:/out" \
debian:bookworm-slim cp /export/ca.pem /export/users.json /out/
tests/fixtures/local/ is ignored by git. The DC publishes no port, so those tests must run in
a container on the stack's backend network (for example $CI run --rm --entrypoint pytest app
with the tests mounted), with TEST_LDAP_URL=ldaps://dc1.pstest.local:636,
TEST_DC_USERS_FILE and the other TEST_LDAP_* values set.
What CI does today
- The
e2ejob in.gitea/workflows/ci.ymlbuilds this DC with the stack, runs the Playwright scenarios intests/e2e/, uploads the report and tears down withdown -v. - The
serverjob does not start the DC. Its DC-backed tests skip becauseTEST_LDAP_URLis unset. Follow-up: run them against the DC, either as a job service or inside the e2e stack.
Running the server suite against the DC by hand
On docker-1, with the CI stack up, run the suite from a container on the stack's backend
network (it is internal, so bake pytest into a derived image first):
printf "FROM packetstation:ci
USER root
RUN pip install -q pytest==9.1.1 anyio==4.15.1
" | docker build -q -t packetstation:ci-test -
docker run --rm --user root --network packetstation-ci_backend -v "$PWD/tests:/work/tests:ro" -v "$PWD/pyproject.toml:/work/pyproject.toml:ro" -v "$PWD/server:/work/server:ro" -v "$PWD/deploy/samba/groups.txt:/work/deploy/samba/groups.txt:ro" -v packetstation-ci_ci-dc-export:/export:ro -v "$PWD/deploy/ci/ci-secrets:/secrets:ro" -w /work -e STACK=ci -e TEST_LDAP_URL=ldaps://dc1.pstest.local:636 -e TEST_LDAP_BASE_DN=DC=pstest,DC=local -e TEST_LDAP_BIND_DN=CN=svc-packetstation,OU=System_Accounts,OU=Admin,DC=pstest,DC=local -e TEST_LDAP_PASSWORD_FILE=/secrets/ldap_bind_password -e TEST_LDAP_GROUP_BASE_DN=OU=PacketStation,OU=Groups,OU=PESCO_FARMINGTON,DC=pstest,DC=local -e TEST_LDAP_CA_BUNDLE=/export/ca.pem -e TEST_DC_USERS_FILE=/export/users.json -e TEST_PG_SUPERUSER_URL=postgresql://postgres:PW@some-other-postgres:5432/postgres --entrypoint python packetstation:ci-test -m pytest -q
Point TEST_PG_SUPERUSER_URL at a throwaway PostgreSQL, never at the stack's own. The
pg_settings fixture resets the ps_app and ps_migrate passwords on the server it is given,
which breaks the running app and worker (seen on docker-1 on 2026-10-16; down -v and up
repair it). Omit the variable to run only the DC tests that need no database.
Open question 15
Whether the Gitea runner can run this container. Status: verified on docker-1 (Docker 29, Ubuntu 24.04) on 2026-10-16; not yet tried on the Gitea runner.
What it needs:
- A Docker daemon that can run a root container with
cap_add: [SYS_ADMIN]. Provisioning sets NT ACLs on sysvol, which needssecurity.*xattrs; without the capability Samba 4.17 fails with access denied (tried on docker-1). The volume's filesystem must support xattrs and POSIX ACLs (ext4 and xfs under/var/lib/docker/volumesdo). If provisioning fails, the provision log tail is printed in$CI logs samba-dc. - Debian bookworm's
samba-ad-provisionpackage, which the Dockerfile installs (the schema files live there, not insamba). - Bind mounts that resolve. The base stack bind-mounts
deploy/ci/ci-secrets/(secrets),deploy/postgres/initdbandtests/e2e. If act_runner runs the job inside a container that talks to the host daemon through its socket, those paths must exist on the host at the same place. If they do not, use a runner on docker-1 in host mode, or one whose workdir is mounted at the same path on the host. - Outbound HTTPS from the playwright container to the npm registry and from the builder to the Debian mirrors.