Correcting a wrong ACME challenge backend was impossible without a shell, and
even with one the correction did not reach the nodes.
The mint gate only fired when `acme_enabled` flipped. `acme_backend_url` was
written to the DB and minted nothing, so Apply answered "No pending changes to
apply" and the nodes kept the old address forever. It is now decided by
comparing the rendered `server _acme_mgmt` line against the active version —
the one line that answers "would the nodes talk to a different address?".
Comparing whole configs would flag every unrelated pending edit.
The field had no UI at all. Added to the cluster form with validation that
mirrors the backend rules, and keyed on `model_fields_set` so clearing it
reverts to the global setting — with a plain `is not None` test an empty box is
indistinguishable from "not submitted", so a value could never be removed.
Validation is asymmetric on purpose (utils/acme_backend_url):
- at the write boundary, reject what cannot express a reachable target —
including the two silent traps: a scheme-less value became `localhost`, and
an out-of-range port raised inside the generator and destroyed the config
- at render time, never reject. The shipped defaults are themselves loopback,
so refusing to render would make every acme_enabled cluster unappliable,
including for changes unrelated to ACME. Problems are logged and surfaced.
The port-less default stays 8080 rather than moving to HTTP's 80: the bundled
compose publishes nginx on 8080, so installs relying on it work today and the
first sign of breaking them would be the unattended renewal loop months later.
The omission is warned about instead.
RFC1918 is allowed and is usually the right answer here, and no DNS resolution
is performed — both deliberate departures from utils/ssrf_guard, whose policy
is the opposite of what this address needs. What the management host can
resolve says nothing about what the HAProxy node can reach.
Diagnostics stop reporting success on a dead path:
- check_port80 uses GET instead of HEAD and classifies the body. A proxy that
has lost its /.well-known/acme-challenge/ location serves its SPA with HTTP
200, which `status in (200, 404)` accepted as healthy. Warnings also surface
when other domains pass, which previously hid the most diagnostic outcome.
- check_routing filters `mode`, joins `acme_enabled` and reads the APPLIED
config instead of counting database rows, and reports a loopback target.
- every new condition is `warn`, never `fail`: the site wizard blocks submit on
any fail, so a new failing condition would lock every install on upgrade day.
Also: normalise `frontends.mode` once per frontend. It is nullable, and the
raw value was interpolated into `mode {}`, emitting a literal `mode None` that
HAProxy rejects — taking down the whole cluster config. The ACME gate and the
backend-mode check now read the same normalised value.
And stop hardcoding PUBLIC_URL / MANAGEMENT_BASE_URL in docker-compose, which
silently ignored the operator's .env and made the wrong default load-bearing.
Infra-only change: image tag bump in docker-compose and k8s manifest.
Backend redis-py client (redis>=5.0.0, resolves to 8.x) verified compatible
against Redis 8.8.0 for all commands in use (get/setex/incr/expire/delete/ping).
No application code or version change.
A user running the API on a 2-core/4GB host reported slow-feeling API
responses (Issue #35 follow-up). Review of the hot paths found no
pathological defect; the dominant factors are the single uvicorn worker
(one core serves all requests) and the constant agent-poll baseline
(4 requests per agent every 30s). Two zero-risk improvements:
- backend/Dockerfile: CMD now honors UVICORN_WORKERS, falling back to
WEB_CONCURRENCY and then 1. Flagless uvicorn natively honors
WEB_CONCURRENCY, so the fallback keeps any deployment that relied on it
byte-for-byte compatible; with the final default of 1 worker uvicorn runs
in-process exactly as before. >1 enables the multiprocess supervisor so
multi-core hosts can use all cores. Background tasks are already
multi-replica safe (FOR UPDATE SKIP LOCKED / advisory locks), as exercised
by the k8s HPA deployment (2-10 replicas). `exec` keeps uvicorn as PID 1
(clean SIGTERM, verified ~1s docker stop with 2 workers).
docker-compose.yml passes UVICORN_WORKERS through as empty-when-unset so a
user-set WEB_CONCURRENCY is never overridden; .env.template documents it.
- routers/agent.py heartbeat (by-name endpoint): the agent's
status/version/upgrade_status were read with three separate single-column
SELECTs against the same row; now one SELECT. Identical values and None
semantics (single consistent snapshot instead of three reads); saves two
round-trips per heartbeat per agent every 30s. The legacy by-id heartbeat
endpoint is untouched; the heartbeat API contract is unchanged for agents
of every version.
- README: new "Performance Tuning" section (worker/replica scaling, and how
to use the X-Response-Time header plus "Slow request detected" logs to
pinpoint slow endpoints).
Verification: full backend suite in docker green (1063 passed, 151 skipped;
also re-run by the runtime image build); worker-count expansion matrix
(unset->1, UVICORN_WORKERS=2->2, WEB_CONCURRENCY=3->3, both->UVICORN_WORKERS,
empty->fallback) all correct; default run confirmed single-process with
uvicorn as PID 1 and healthy API; UVICORN_WORKERS=2 confirmed parent + 2
workers, healthy API, clean shutdown; live heartbeats verified for
register + existing-agent paths AND degraded agents (no stats socket /
haproxy stopped / garbage stats CSV / unknown backend in server_statuses):
all return 200, agent row updates correctly, zero backend errors.
No schema, API, or agent changes.
Supersedes the v1.6.4 stable pin (1.30.2-alpine) with the mainline
patched release. No config, schema, or behavior changes.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The bundled nginx reverse proxy was flagged for the nginx 'poolslip' advisory
(affected: mainline <=1.31.0; fixed: stable 1.30.2+ / mainline 1.31.1+). The
config-level mitigation (named capture groups instead of $1/$2 in rewrite) does
not apply — the product's nginx config (nginx/nginx.conf and the k8s configmap)
has no rewrite capture-group directives, only prefix locations + proxy_pass. So
the fix is the version: pin nginx:alpine -> nginx:1.30.2-alpine in
docker-compose.yml and k8s/manifests/10-nginx.yaml.
No config/schema/behavior change (nginx only reverse-proxies). Version bumped to
1.6.4 across all layers. Verified in Docker: nginx -v=1.30.2; nginx -t OK on both
the compose and production configmap configs; all proxied routes work through
nginx; no nginx errors.
Adds opt-in TOTP-based Multi-Factor Authentication that is fully
backwards compatible with existing logins. Operators choose to enable
MFA per account; nothing changes for users who do not opt in.
Highlights
==========
* RFC 6238 TOTP (6 digits, 30s period, SHA1) with ±30s skew tolerance,
compatible with Microsoft / Google Authenticator, Authy, Duo, 1Password.
* Per-step replay protection (`mfa_last_used_totp_step`) so a captured
code cannot be reused inside the same window.
* Fernet-encrypted TOTP secrets at rest, key resolution via
`MFA_ENCRYPTION_KEY` env (HKDF-derived from `SECRET_KEY` as fallback).
* 10 single-use, bcrypt-hashed backup codes per user, formatted
`XXXX-YYYY` from a confusion-free alphabet (no 0/O/1/I/L).
* Two-step login flow: `POST /api/auth/login` returns `mfa_required`
+ `mfa_token`, then `POST /api/auth/login/mfa-verify` accepts a TOTP
code OR a backup code. JWT is minted only after MFA succeeds.
* Self-service: users enable / disable MFA from their own row in the
Users page; admins reset (single user or bulk) but never enable on
behalf of someone else (matches AWS IAM / GitHub / Google Workspace).
* Bulk emergency reset CLI: `scripts/admin-mfa-reset-all.sh`.
Security hardening
==================
* Atomic transactions with `SELECT … FOR UPDATE` on `mfa_pending_logins`
and `users` rows so concurrent verify / enroll calls cannot race.
* `/api/mfa/enroll/start` refuses re-enrollment when MFA is already on
(prevents silent secret rotation via a stolen JWT).
* Pydantic `ValidationError` messages are sanitized before reaching the
audit log so request bodies (TOTP / backup codes in flight) never
appear in plaintext.
* Slowapi rate limits are per-USER, not per-IP, with a trusted-proxy
XFF strategy so a single ingress address cannot exhaust the bucket
for thousands of operators (`MFA_TRUSTED_PROXY_CIDRS`,
`MFA_RATE_LIMIT_*` env-overridable).
* Login query now scopes to `is_active = TRUE` so a soft-deleted row
with the same username can no longer occlude the active user
(also closes a small account-enumeration side channel).
Database
========
Additive migrations (idempotent `ADD COLUMN IF NOT EXISTS`,
`CREATE TABLE IF NOT EXISTS`):
- users: mfa_enabled, mfa_method, mfa_secret_encrypted,
mfa_enrolled_at, mfa_last_used_at, mfa_last_used_totp_step
- mfa_backup_codes (user_id ON DELETE CASCADE)
- mfa_pending_logins (user_id ON DELETE CASCADE, challenge_token,
attempts, expires_at)
- mfa_pending_enrollments (user_id ON DELETE CASCADE)
Frontend
========
* Login page becomes a 3-phase state machine
(credentials → MFA → submitting); legacy single-step login is
preserved for users who haven't enrolled.
* New MFAEnrollModal (3-step wizard: QR + secret → verify → backup
codes) using `qrcode.react`.
* Users page shows MFA column + per-row enable/disable/reset actions.
Admins viewing other users with MFA off see a non-actionable info
icon explaining that only the user themselves can enable MFA.
Deployment
==========
* `MFA_ENCRYPTION_KEY` is added to `k8s/manifests/03-secrets.yaml` as
a placeholder; `SECRET_KEY` is also placeholder-ized so both are
injected by the existing pipeline pattern (sed-replace + apply).
* No new build-time env vars are required for the frontend. The SPA
uses `window.location.host` for `/api/*` and is routed by the
existing nginx ingress configuration.
* `frontend/.dockerignore` ensures host `.env*` files cannot bleed
into the production bundle.
Tests
=====
* New unit suites:
- `test_mfa_service.py` (TOTP, encryption, backup codes)
- `test_mfa_backwards_compat.py` (regression — non-MFA flow unchanged)
- `test_mfa_rate_limits.py` (env override + dataclass immutability)
- `test_mfa_rate_limit_key.py` (JWT key, trusted-proxy XFF, fallbacks)
* All existing 1000+ unit tests continue to pass.
Documentation
=============
* README MFA section (overview, day-to-day operations, emergency
reset CLI, env variables, rate-limit tuning).
* `scripts/README.md` documents the bulk reset script.
Issue: #18
- Backend/frontend now pull pre-built images from Docker Hub by default
- Build sections commented out (can be uncommented for local builds)
- Removed dev volume mount (./backend:/app) for production use
- Backend healthcheck changed from curl to python urllib (curl not in slim image)
- Added nginx /health endpoint for compose healthcheck
Made-with: Cursor
Mustafa ULUKAYA