Files
scrabble-game/deploy/caddy/Caddyfile
T
Ilia Denisov c8601c0115
CI / changes (pull_request) Successful in 2s
CI / unit (pull_request) Successful in 10s
CI / integration (pull_request) Successful in 18s
CI / ui (pull_request) Successful in 1m5s
CI / conformance (pull_request) Successful in 9s
CI / gate (pull_request) Successful in 0s
CI / deploy (pull_request) Successful in 1m39s
feat(deploy): prod OOM swap cushion + edge maintenance page
Two prod-deploy hardening measures (owner-requested):

- Swap file (Ansible common role, swap_size=1G, vm.swappiness=10). The per-container
  memory caps enforce that one service can't eat all RAM, but they overcommit the
  1.9 GiB main host (~2.8 GiB of caps), so a simultaneous spike could hit the kernel
  OOM-killer (and it might pick postgres). A small swap absorbs the overshoot.
  Idempotent, builtin-only (no ansible.posix).

- Edge maintenance page. prod-deploy.sh raises a flag around the rolling swap /
  migration window that the caddy edge serves a static 503 "технические работы" page
  from (deploy/caddy/maintenance.html), for the user-facing routes only — /_gm
  (Grafana) stays reachable. Cleared on any exit (success, health failure + rollback,
  or error) by a shell trap so it can never stick on. The 503 carries Retry-After +
  an X-Scrabble-Maintenance marker so a follow-up SPA overlay can tell a planned
  window apart from a transient error (the static page only catches a fresh load; an
  in-session user needs the app-side overlay). Not zero-downtime — the single
  stateful backend still blips — but the window is graceful instead of raw 502s.

Verified: caddy validate; the gate 503s every non-/_gm path incl. the Connect/gRPC
edge and serves the page + markers; /_gm bypasses; toggling needs no reload
(per-request stat). Ansible --syntax-check + compose config (base+prod) pass.
2026-07-03 22:09:37 +02:00

115 lines
5.0 KiB
Caddyfile

# Edge reverse proxy for the Scrabble contour. A single Basic-Auth gate covers
# every operator surface under /_gm (the backend-rendered admin console and the
# Grafana subpath); the game SPA (/app/, /telegram/, /vk/) and the Connect edge go to
# the gateway; the catch-all — notably the public landing at / — goes to the
# static landing container, so stray traffic never reaches the Go edge.
# Mirrors ../galaxy-game's /_gm model.
#
# CADDY_SITE_ADDRESS is ":80" in the test contour (the host caddy terminates TLS
# and forwards); set it to a domain in prod so this caddy does its own
# ACME and the contour is self-contained.
{
admin off
# Trust X-Forwarded-For from private-range upstreams so the real client IP survives
# (chat moderation + per-IP rate limiting in the gateway). Test contour: the host caddy
# (a private IP) is trusted, so its forwarded client IP is preserved. Prod (no host caddy):
# clients connect from public IPs, which are NOT trusted, so Caddy uses the real peer —
# the same config is correct (and spoof-safe) in both contours.
servers {
trusted_proxies static private_ranges
}
}
{$CADDY_SITE_ADDRESS::80} {
# HTTP/3 is advertised by default whenever this caddy terminates TLS (prod:
# CADDY_SITE_ADDRESS is the domain). But UDP/443 is never reachable — the prod
# compose maps only "443:443" (TCP) and ufw opens 443/tcp — so a client that cached
# the `Alt-Svc: h3` advert (sticky for ma=2592000s) stalls on the dead QUIC path
# before falling back to h2, which surfaced as the Telegram Mini App intermittently
# hanging on load. `Alt-Svc: clear` actively drops any cached alternative and pins
# clients to h2/h1; it is applied site-wide so every route is covered. In the test
# contour this caddy serves plain :80 (no h3 to advertise) and the host caddy
# re-stamps its own Alt-Svc, so the live test fix lives in the host caddy — here it
# is the prod fix. Background + alternatives (incl. serving h3 for real): docs/EDGE_HTTP3.md.
header Alt-Svc clear
# Maintenance gate. While the deploy holds the flag file /srv/maint/on (touched by
# prod-deploy.sh around a rolling swap, removed on the script's exit), serve a static
# 503 "works in progress" page instead of proxying to a mid-recreate upstream —
# a graceful signal rather than raw 502s. Operator surfaces (/_gm) are excluded so
# Grafana stays reachable during the window. Caddy stat()s the flag per request, so
# toggling it needs no reload; the page + flag live in the caddy config dir bind-mounted
# read-only at /srv/maint (docker-compose.yml). The test contour never sets the flag
# (only prod-deploy.sh does), so this is inert there.
@maintenance {
not path /_gm /_gm/*
file {
root /srv/maint
try_files on
}
}
handle @maintenance {
error 503
}
handle_errors {
@maint503 expression {err.status_code} == 503
handle @maint503 {
root * /srv/maint
rewrite * /maintenance.html
header Retry-After 120
# Distinctive maintenance marker so the SPA can tell a planned window apart from
# a transient upstream 503 and show its own dimmed overlay (an in-session user
# never sees this static page — it only renders on a fresh load). The header
# rides every gated response, incl. the Connect/gRPC edge the app polls.
header X-Scrabble-Maintenance 1
file_server
}
}
# Operator surfaces under /_gm: a single shared Basic-Auth, then route.
@gm path /_gm /_gm/*
handle @gm {
basic_auth {
{$GM_BASICAUTH_USER:gm} {$GM_BASICAUTH_HASH}
}
# Grafana serves from this sub-path (GF_SERVER_SERVE_FROM_SUB_PATH=true), so
# the prefix is forwarded intact, not stripped.
handle /_gm/grafana* {
reverse_proxy grafana:3000
}
# Everything else under /_gm is the backend-rendered admin console.
handle {
reverse_proxy backend:8080
}
}
# The game SPA and the Connect edge are served by the gateway. Strip any
# client-supplied X-Scrabble-Honeypot here so the gateway only ever honours the
# tag the honeypot block sets below (a client cannot self-tag a real request).
@gateway path /app /app/* /telegram /telegram/* /vk /vk/* /dict/* /dl/* /metrics/* /scrabble.edge.v1.Gateway/*
handle @gateway {
reverse_proxy gateway:8081 {
header_up -X-Scrabble-Honeypot
}
}
# Honeypot decoy paths: classic vulnerability-scanner bait no real client ever
# requests. Route them to the gateway tagged with X-Scrabble-Honeypot — the set
# replaces any client-supplied value — so it logs the scanner hit and (in prod)
# bans the source IP. (A delete + set in one block would not work: Caddy applies
# header_up deletions after sets, which would strip the tag we just set; the real
# endpoints instead strip the header in the @gateway block above.) Keep this list
# disjoint from every legitimate landing/app path.
@honeypot path /.env /.git /.git/* /.aws/* /wp-login.php /wp-admin /wp-admin/* /phpmyadmin /phpmyadmin/*
handle @honeypot {
reverse_proxy gateway:8081 {
header_up X-Scrabble-Honeypot 1
}
}
# Everything else — the public landing at / and any stray path — is static.
handle {
reverse_proxy landing:80
}
}