docker-compose.yml

# OpenWA - Docker Compose Configuration
# Smart Orchestration with Profiles

services:
  # ===== CORE: Docker Socket Proxy (sole container with /var/run/docker.sock access) =====
  docker-proxy:
    # Pinned by release tag and digest: this container holds the Docker socket, so a re-pushed tag
    # must not change what runs here.
    image: tecnativa/docker-socket-proxy:v0.5.0@sha256:1f5038b54f06c3e18422902cf00ba21803d1c97805aae032e5e6673d532d3459
    container_name: openwa-docker-proxy
    restart: unless-stopped
    # Only on the isolated internal network, reachable solely by openwa-api (including any
    # plugin loaded into it), NOT the dashboard or any other peer. `internal: true` also denies the proxy
    # outbound access; it only needs the locally-mounted docker socket.
    networks:
      - internal-docker
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock:ro
    environment:
      # Minimal set that keeps DockerService (src/modules/docker/docker.service.ts) working:
      # PING (health), CONTAINERS (list/inspect/create/start/stop),
      # IMAGES (pull), VOLUMES (createVolume). POST is the proxy's single all-or-nothing
      # method gate for non-GET requests (haproxy.cfg: `deny unless METH_GET || env(POST)`).
      # Two consequences of the pinned proxy config, accepted and documented in SECURITY.md:
      #  - its `DELETE` env flag is dead config (never read), so it is not set here; DELETE is
      #    admitted to enabled paths as a side effect of POST — OpenWA itself never issues
      #    deletes (profile teardown is stop-only, see DockerService.stopManagedService).
      #  - the proxy cannot scope container-create payloads, so a compromised API container
      #    could create containers with host bind-mounts. If you don't use the built-in
      #    datastore orchestration (Dashboard > Infrastructure built-in toggles), disable
      #    this service entirely, e.g. via a docker-compose.override.yml:
      #      services:
      #        docker-proxy:
      #          profiles: ['disabled']
      #    DockerService then reports Docker unavailable and orchestration degrades gracefully.
      CONTAINERS: 1
      IMAGES: 1
      VOLUMES: 1
      PING: 1
      POST: 1
      # Everything else is denied by default (AUTH, SECRETS, NETWORKS, PLUGINS, SWARM, TASKS, SERVICES, CONFIGS, NODES, DISTRIBUTIONS)
    labels:
      - 'com.openwa.service=docker-proxy'
      - 'com.openwa.core=true'

  # ===== CORE: OpenWA Backend API =====
  openwa-api:
    build:
      context: .
      dockerfile: Dockerfile
    container_name: openwa-api
    restart: unless-stopped
    # Give the ordered teardown room to finish before Docker SIGKILLs the container, which would cut
    # Chromium, the usage flush and the database close short. Bounded stages in order: drain 3s
    # (SHUTDOWN_DELAY_MS) + ingress reconciler 5s + engines 10s + auto-start wait 5s + webhook drain 5s
    # (WEBHOOK_SHUTDOWN_DRAIN_MS) + usage flush 5s = 33s. An enabled plugin adds 10s. QUEUE_ENABLED and
    # REDIS_ENABLED add 20s (worker close 10s, producer queue close 2s for each of two queues, three
    # Redis quits 2s each): 63s worst case. The process force-exits 67s after the drain and 5s before
    # this deadline. Raising SHUTDOWN_DELAY_MS or WEBHOOK_SHUTDOWN_DRAIN_MS moves that exit later by the
    # same amount, so raise this by the same amount too. The session count does not add: engines stop
    # in parallel.
    stop_grace_period: 75s
    # App network for datastores + the isolated network to reach
    # docker-proxy (DOCKER_HOST). createService() in docker.service.ts attaches
    # orchestrated containers to the literal `openwa-network`, so its name is fixed.
    networks:
      - openwa-network
      - internal-docker
    # Container hardening: Chromium runs with --no-sandbox, so the
    # container itself is the confinement boundary. cap_drop ALL + a minimal re-add
    # for root entrypoint ownership, the gosu drop and init signal forwarding. Once gosu setuids to
    # the openwa user the node/Chromium process keeps NO effective capabilities.
    # read_only rootfs (Chromium's profile lives on the writable /app/data volume;
    # HOME=/tmp + tmpfs absorb stray writes). no-new-privileges blocks setuid escalation.
    security_opt:
      - 'no-new-privileges:true'
    cap_drop:
      - ALL
    cap_add:
      - CHOWN # entrypoint: chown openwa /app/data on the named volume
      - DAC_OVERRIDE # entrypoint: chown across pre-existing files
      - FOWNER
      - KILL # dumb-init: forward signals to the openwa process after gosu drops uid
      - SETGID # gosu: drop to the openwa group
      - SETUID # gosu: drop to the openwa user
    read_only: true
    tmpfs:
      - /tmp
    # Per-container PID ceiling (writes the cgroup pids.max). A fork-bomb guard, NOT an allocation —
    # the kernel only rejects new forks once the count is reached, so a higher limit is free for
    # light containers. The old default (512, from the #243 hardening pass) was picked without
    # accounting for Chromium's multi-process model: whatsapp-web.js runs a full Chromium instance
    # per session (browser + renderer + GPU + zygote + utilities), and WhatsApp Web is process-heavy
    # (service workers/iframes), so ~4 concurrent sessions already approach 512 and a new session's
    # Chromium gets killed mid-spawn — surfacing in the API as `Code: null`. 2048 fits ~8-10 wwjs
    # sessions with startup-spike headroom; Baileys is single-process (no Chromium) and uses far
    # less, so the higher ceiling is a no-op there. Raise via OPENWA_PIDS_LIMIT for larger fleets;
    # do NOT set -1 (unlimited) — that drops the fork-bomb guard. See #636.
    pids_limit: ${OPENWA_PIDS_LIMIT:-2048}
    mem_limit: ${OPENWA_MEM_LIMIT:-2g} # tune up for many concurrent sessions
    ports:
      - '127.0.0.1:${API_PORT:-2785}:2785'
    expose:
      - '2785'
    environment:
      # Core
      - NODE_ENV=${NODE_ENV:-production}
      # The image already runs on UTC; stated here so it is a decision rather than an accident. Stored
      # timestamps are UTC on either database whatever this says (the Postgres data connection pins its
      # own session), so overriding it only moves log stamps and the local day the "today" stats count.
      - TZ=${TZ:-UTC}
      # Writable HOME on tmpfs so Chromium's HOME-relative writes don't hit the read_only rootfs
      - HOME=/tmp
      # Chromium resolves its home from the passwd entry (no /home/openwa), ignoring $HOME, so without
      # writable, existing config/cache dirs it hard-crashes at launch on the read_only rootfs. Pin XDG
      # to the tmpfs; the entrypoint pre-creates these owned by openwa. (#254)
      - XDG_CONFIG_HOME=/tmp/.config
      - XDG_CACHE_HOME=/tmp/.cache
      - PORT=2785
      - LOG_LEVEL=${LOG_LEVEL:-info}
      # Database. Forwarded blank by default (`${VAR:-}`) so a dashboard switch saved to
      # data/.env.generated actually applies at runtime — main.ts clears the blank (BLANK_SHADOWED_ENV_KEYS)
      # and the file wins. Set any of these in your .env/host to pin it (a real value keeps top precedence
      # and the dashboard control then shows "managed by environment"). First-run defaults (sqlite) are
      # written to data/.env.generated by the app, so a blank stack still boots SQLite.
      # NOTE: if you run the built-in Postgres MANUALLY (`docker compose --profile postgres up`) instead
      # of via the dashboard, set DATABASE_TYPE=postgres, DATABASE_HOST=postgres, DATABASE_PORT=5432,
      # DATABASE_USERNAME=openwa and DATABASE_PASSWORD in your .env/host — these are no longer defaulted here.
      - DATABASE_TYPE=${DATABASE_TYPE:-}
      - DATABASE_NAME=${DATABASE_NAME:-}
      - DATABASE_HOST=${DATABASE_HOST:-}
      - DATABASE_PORT=${DATABASE_PORT:-}
      - DATABASE_USERNAME=${DATABASE_USERNAME:-}
      # No committed default secret; blank unless the operator sets it (built-in Postgres is provisioned
      # with its own credentials by the orchestrator, not via this line).
      - DATABASE_PASSWORD=${DATABASE_PASSWORD:-}
      # PostgreSQL schema (blank-forwarded like the other DATABASE_* keys so a dashboard-saved
      # POSTGRES_SCHEMA in data/.env.generated applies; a real host value pins). Ignored for sqlite.
      - POSTGRES_SCHEMA=${POSTGRES_SCHEMA:-}
      # TLS to a managed PostgreSQL, blank-forwarded the same way.
      - DATABASE_SSL=${DATABASE_SSL:-}
      - DATABASE_SSL_REJECT_UNAUTHORIZED=${DATABASE_SSL_REJECT_UNAUTHORIZED:-}
      - DATABASE_SYNCHRONIZE=${DATABASE_SYNCHRONIZE:-false}
      # Engine. Forwarded empty by default so the dashboard (Infrastructure > Engine) selects the
      # active engine via data/.env.generated (defaults to whatsapp-web.js); main.ts treats a blank
      # ENGINE_TYPE as unset, so .env.generated wins. Set ENGINE_TYPE in your .env/host to pin an
      # engine (e.g. baileys) — a real value flows through here and keeps top precedence.
      - ENGINE_TYPE=${ENGINE_TYPE:-}
      # Engine launch options (session data path, headless, browser args). Forwarded EMPTY by default
      # so the dashboard (Infrastructure > Engine) selections saved to data/.env.generated apply — like
      # ENGINE_TYPE above. main.ts treats a blank value as unset (.env.generated wins), and the app
      # layer (configuration.ts) supplies the sane container default (headless, ./data/sessions, the
      # sandbox + no-dev-shm flag set) when nothing is set anywhere. Set one on the host/.env to pin it.
      - SESSION_DATA_PATH=${SESSION_DATA_PATH:-}
      - PUPPETEER_HEADLESS=${PUPPETEER_HEADLESS:-}
      - PUPPETEER_ARGS=${PUPPETEER_ARGS:-}
      - PUPPETEER_PROTOCOL_TIMEOUT_MS=${PUPPETEER_PROTOCOL_TIMEOUT_MS:-}
      # Optional WhatsApp Web version override. DEFAULT (empty, "latest", or "auto"): OpenWA auto-resolves a
      # settled build from the third-party wppconnect-team/wa-version registry and pins its remote HTML
      # (fetched into the web.whatsapp.com origin without an integrity check). Set "off" to disable pinning
      # and use the first-party build served by WhatsApp, or an exact version to pin a specific build.
      - WWEBJS_WEB_VERSION=${WWEBJS_WEB_VERSION:-}
      - WWEBJS_WEB_VERSION_REMOTE_PATH=${WWEBJS_WEB_VERSION_REMOTE_PATH:-}
      # Raise whatsapp-web.js's first-boot init wait (default 30000ms) on slow boots — e.g. WSL2 or
      # low-resource hosts where the QR can time out before WA Web loads. Empty = default; see
      # docs/12-troubleshooting-faq.md. (Without this line the var in .env never reaches the container.)
      - WWEBJS_AUTH_TIMEOUT_MS=${WWEBJS_AUTH_TIMEOUT_MS:-}
      # Re-start previously authenticated sessions at boot. Opt-in: the app reads the exact string
      # "true", so a blank forward (nothing set) keeps the default off. Without this line the value in
      # .env never reaches the container and the flag silently does nothing.
      - AUTO_START_SESSIONS=${AUTO_START_SESSIONS:-}
      # Fetching status@broadcast immediately after a fresh pairing can make affected accounts lose
      # the companion at WhatsApp Web's first scheduled reload. Disabled by default; live statuses
      # still arrive normally. Opt in only after validating the account.
      - STATUS_SEED_ON_READY=${STATUS_SEED_ON_READY:-false}
      # Session ownership / multi-node routing (see the "Session ownership" section of .env.example
      # and docs/13). NODE_ID must be STABLE across restarts: the app default (the container
      # hostname) changes on every recreate, making the new container a "new node" that has to wait
      # out its previous identity's lease before re-adopting sessions.
      - NODE_ID=${NODE_ID:-}
      - NODE_URL=${NODE_URL:-}
      - SESSION_LEASE_TTL_MS=${SESSION_LEASE_TTL_MS:-}
      - SESSION_LEASE_HEARTBEAT_MS=${SESSION_LEASE_HEARTBEAT_MS:-}
      - SESSION_TAKEOVER_SWEEP_MS=${SESSION_TAKEOVER_SWEEP_MS:-}
      - SESSION_PROXY_TIMEOUT_MS=${SESSION_PROXY_TIMEOUT_MS:-}
      # Autoreply rules per session (0 = unlimited). Blank forwards to the app default (32); an
      # unforwarded key could not be set from .env at all, which is the whole point of this list.
      - AUTOMATION_MAX_PER_SESSION=${AUTOMATION_MAX_PER_SESSION:-}
      # Storage. Blank-forwarded (see Database note) so a dashboard local↔S3 switch applies; a real
      # host value pins. Credentials use the canonical S3_ACCESS_KEY_ID / S3_SECRET_ACCESS_KEY names the
      # app and dashboard write; the legacy S3_ACCESS_KEY / S3_SECRET_KEY are ALSO forwarded (and read
      # as a fallback by the storage layer) so existing .env files keep working unchanged. S3_REGION is
      # forwarded so external-S3 works.
      - STORAGE_TYPE=${STORAGE_TYPE:-}
      - STORAGE_LOCAL_PATH=${STORAGE_LOCAL_PATH:-}
      - S3_ENDPOINT=${S3_ENDPOINT:-}
      - S3_ACCESS_KEY_ID=${S3_ACCESS_KEY_ID:-}
      - S3_SECRET_ACCESS_KEY=${S3_SECRET_ACCESS_KEY:-}
      - S3_ACCESS_KEY=${S3_ACCESS_KEY:-}
      - S3_SECRET_KEY=${S3_SECRET_KEY:-}
      - S3_REGION=${S3_REGION:-}
      - S3_BUCKET=${S3_BUCKET:-}
      - S3_KEY_PREFIX=${S3_KEY_PREFIX:-}
      # Chat-media archiving (opt-in). Forwarded explicitly: this file has no env_file, so a
      # variable absent from this list simply never reaches the container and the feature stays off
      # no matter what the operator's .env says.
      - CHAT_MEDIA_ARCHIVE_ENABLED=${CHAT_MEDIA_ARCHIVE_ENABLED:-}
      - CHAT_MEDIA_ARCHIVE_OUTBOUND=${CHAT_MEDIA_ARCHIVE_OUTBOUND:-}
      - CHAT_MEDIA_ARCHIVE_MAX_BYTES=${CHAT_MEDIA_ARCHIVE_MAX_BYTES:-}
      - CHAT_MEDIA_ARCHIVE_TTL_DAYS=${CHAT_MEDIA_ARCHIVE_TTL_DAYS:-}
      - MESSAGE_INLINE_MEDIA=${MESSAGE_INLINE_MEDIA:-}
      - CHAT_MEDIA_ORPHAN_SWEEP_INTERVAL_MS=${CHAT_MEDIA_ORPHAN_SWEEP_INTERVAL_MS:-}
      - CHAT_MEDIA_ORPHAN_GRACE_MS=${CHAT_MEDIA_ORPHAN_GRACE_MS:-}
      # Send pacing (opt-in anti-ban governor). Forwarded explicitly for the same reason as the
      # chat-media block above: without these lines SEND_PACING_ENABLED=true in the operator's .env
      # never reaches the container and every send stays unpaced with no signal anywhere.
      - SEND_PACING_ENABLED=${SEND_PACING_ENABLED:-}
      - SEND_PACING_WARMUP_SCHEDULE=${SEND_PACING_WARMUP_SCHEDULE:-}
      - SEND_PACING_COLD_DAILY_CAP=${SEND_PACING_COLD_DAILY_CAP:-}
      - SEND_PACING_BREAKER_THRESHOLD=${SEND_PACING_BREAKER_THRESHOLD:-}
      - SEND_PACING_BREAKER_COOLDOWN_MS=${SEND_PACING_BREAKER_COOLDOWN_MS:-}
      # Inbound media: the response budget for inline media on the message list, and the download
      # limits. Forwarded so they can be tuned from .env without editing this file.
      - MESSAGE_LIST_INLINE_MEDIA_BUDGET_BYTES=${MESSAGE_LIST_INLINE_MEDIA_BUDGET_BYTES:-}
      - MEDIA_DOWNLOAD_ENABLED=${MEDIA_DOWNLOAD_ENABLED:-}
      - MEDIA_DOWNLOAD_MAX_BYTES=${MEDIA_DOWNLOAD_MAX_BYTES:-}
      - MEDIA_DOWNLOAD_TIMEOUT_MS=${MEDIA_DOWNLOAD_TIMEOUT_MS:-}
      - INBOUND_MEDIA_CONCURRENCY=${INBOUND_MEDIA_CONCURRENCY:-}
      - INBOUND_MEDIA_GLOBAL_CONCURRENCY=${INBOUND_MEDIA_GLOBAL_CONCURRENCY:-}
      - SESSION_PROXY_URL_FETCH=${SESSION_PROXY_URL_FETCH:-}
      # Server-side media conversion (opt-in). The image ships ffmpeg, but without these forwards
      # the feature could not be switched on from .env at all.
      - MEDIA_CONVERSION_ENABLED=${MEDIA_CONVERSION_ENABLED:-}
      - FFMPEG_PATH=${FFMPEG_PATH:-}
      - MEDIA_CONVERSION_TIMEOUT_MS=${MEDIA_CONVERSION_TIMEOUT_MS:-}
      - MEDIA_CONVERSION_MAX_OUTPUT_BYTES=${MEDIA_CONVERSION_MAX_OUTPUT_BYTES:-}
      - MEDIA_CONVERSION_CONCURRENCY=${MEDIA_CONVERSION_CONCURRENCY:-}
      # Redis. Blank-forwarded (see Database note) so a dashboard enable/disable applies; a real host
      # value pins.
      - REDIS_ENABLED=${REDIS_ENABLED:-}
      - REDIS_HOST=${REDIS_HOST:-}
      - REDIS_PORT=${REDIS_PORT:-}
      # Webhook
      - WEBHOOK_TIMEOUT=${WEBHOOK_TIMEOUT:-10000}
      - WEBHOOK_RETRY_DELAY=${WEBHOOK_RETRY_DELAY:-5000}
      - WEBHOOK_DISPATCH_CONCURRENCY=${WEBHOOK_DISPATCH_CONCURRENCY:-16}
      - WEBHOOK_DISPATCH_MAX_QUEUED=${WEBHOOK_DISPATCH_MAX_QUEUED:-1000}
      # Flags the app reads with a bare `=== 'true'` / `!== 'false'` and that have no dashboard
      # route, so without these forwards setting them in .env does nothing under compose.
      # BAILEYS_MARK_ONLINE_ON_CONNECT is the one that bites: its default silences push
      # notifications on the paired phone for as long as the gateway is connected (#871).
      - WEBHOOK_CONTACT_DETAILS=${WEBHOOK_CONTACT_DETAILS:-}
      - BAILEYS_MARK_ONLINE_ON_CONNECT=${BAILEYS_MARK_ONLINE_ON_CONNECT:-}
      - BAILEYS_SYNC_FULL_HISTORY=${BAILEYS_SYNC_FULL_HISTORY:-}
      - BAILEYS_WA_VERSION=${BAILEYS_WA_VERSION:-}
      - ALLOW_UNSIGNED_INGRESS=${ALLOW_UNSIGNED_INGRESS:-}
      - STORE_EPHEMERAL_MESSAGES=${STORE_EPHEMERAL_MESSAGES:-}
      - RESOLVE_LID_TO_PHONE=${RESOLVE_LID_TO_PHONE:-}
      - SIMULATE_TYPING=${SIMULATE_TYPING:-}
      - MCP_ENABLED=${MCP_ENABLED:-}
      - SEARCH_ENABLED=${SEARCH_ENABLED:-}
      - SEARCH_CONTAINS_ENABLED=${SEARCH_CONTAINS_ENABLED:-}
      - SERVE_DASHBOARD=${SERVE_DASHBOARD:-}
      - CACHE_ENABLED=${CACHE_ENABLED:-}
      - DATABASE_LOGGING=${DATABASE_LOGGING:-}
      - MAIN_DATABASE_SYNCHRONIZE=${MAIN_DATABASE_SYNCHRONIZE:-}
      # Rate limits. Blank values are cleared before lower-priority generated config is loaded.
      - RATE_LIMIT_SHORT_TTL=${RATE_LIMIT_SHORT_TTL:-}
      - RATE_LIMIT_SHORT_LIMIT=${RATE_LIMIT_SHORT_LIMIT:-}
      - RATE_LIMIT_MEDIUM_TTL=${RATE_LIMIT_MEDIUM_TTL:-}
      - RATE_LIMIT_MEDIUM_LIMIT=${RATE_LIMIT_MEDIUM_LIMIT:-}
      - RATE_LIMIT_LONG_TTL=${RATE_LIMIT_LONG_TTL:-}
      - RATE_LIMIT_LONG_LIMIT=${RATE_LIMIT_LONG_LIMIT:-}
      # Max request body size. Base64 media sends ride in the JSON body, so the default is generous;
      # raise this for large documents (e.g. BODY_SIZE_LIMIT=50mb). Blank keeps the app default of 25mb.
      - BODY_SIZE_LIMIT=${BODY_SIZE_LIMIT:-}
      # Plugins
      - PLUGINS_DIR=${PLUGINS_DIR:-/app/data/plugins}
      - PLUGIN_STATE_DIR=${PLUGIN_STATE_DIR:-}
      # Plugin install-from-URL redirect policy (ssrf-guard): off unless the exact string 'true'.
      - PLUGIN_DOWNLOAD_ALLOW_INSECURE_REDIRECTS=${PLUGIN_DOWNLOAD_ALLOW_INSECURE_REDIRECTS:-false}
      # Security
      - API_MASTER_KEY=${API_MASTER_KEY:-}
      - TRUSTED_PROXIES=${TRUSTED_PROXIES:-}
      # Forward the plain-HTTP dashboard CSP override; unset preserves the application default.
      - CSP_UPGRADE_INSECURE_REQUESTS=${CSP_UPGRADE_INSECURE_REQUESTS:-}
      # Security / boot knobs the app reads. This list is the only path into the container, so a
      # knob without a line here is inert no matter what the operator's .env says; compose-parity
      # .spec.ts derives the required set from .env.example so a new knob cannot ship unreachable.
      - API_KEY_PEPPER=${API_KEY_PEPPER:-}
      - BOOTSTRAP_KEY_FILE=${BOOTSTRAP_KEY_FILE:-}
      - ENABLE_SWAGGER=${ENABLE_SWAGGER:-}
      - VALIDATION_ERROR_DETAIL=${VALIDATION_ERROR_DETAIL:-}
      - MCP_READONLY=${MCP_READONLY:-}
      - MCP_RATE_LIMIT_MAX=${MCP_RATE_LIMIT_MAX:-}
      - MCP_RATE_LIMIT_WINDOW_MS=${MCP_RATE_LIMIT_WINDOW_MS:-}
      - MCP_IP_RATE_LIMIT_MAX=${MCP_IP_RATE_LIMIT_MAX:-}
      - MCP_IP_RATE_LIMIT_WINDOW_MS=${MCP_IP_RATE_LIMIT_WINDOW_MS:-}
      - METRICS_TOKEN=${METRICS_TOKEN:-}
      - CORS_ORIGINS=${CORS_ORIGINS:-}
      - SSRF_ALLOWED_HOSTS=${SSRF_ALLOWED_HOSTS:-}
      - SSRF_DNS_TIMEOUT_MS=${SSRF_DNS_TIMEOUT_MS:-}
      - WEBHOOK_SSRF_REDIRECTS=${WEBHOOK_SSRF_REDIRECTS:-}
      # HTTP server tuning (timeouts are relevant behind a proxy that holds connections).
      - REQUEST_TIMEOUT_MS=${REQUEST_TIMEOUT_MS:-}
      - HEADERS_TIMEOUT_MS=${HEADERS_TIMEOUT_MS:-}
      - KEEPALIVE_TIMEOUT_MS=${KEEPALIVE_TIMEOUT_MS:-}
      - INFLIGHT_BODY_BUDGET_BYTES=${INFLIGHT_BODY_BUDGET_BYTES:-}
      # Webhook delivery tuning and retention.
      - WEBHOOK_SHUTDOWN_DRAIN_MS=${WEBHOOK_SHUTDOWN_DRAIN_MS:-}
      - WEBHOOK_MEDIA_INLINE_MAX_BYTES=${WEBHOOK_MEDIA_INLINE_MAX_BYTES:-}
      - WEBHOOK_MAX_PAYLOAD_BYTES=${WEBHOOK_MAX_PAYLOAD_BYTES:-}
      - WEBHOOK_MAX_PER_SESSION=${WEBHOOK_MAX_PER_SESSION:-}
      - WEBHOOK_FAILURE_RETENTION_DAYS=${WEBHOOK_FAILURE_RETENTION_DAYS:-}
      - WEBHOOK_FAILURE_PAYLOAD_RETENTION_HOURS=${WEBHOOK_FAILURE_PAYLOAD_RETENTION_HOURS:-}
      - WEBHOOK_WORKER_CONCURRENCY=${WEBHOOK_WORKER_CONCURRENCY:-}
      - WEBHOOK_DEGRADED_SESSION_CONCURRENCY=${WEBHOOK_DEGRADED_SESSION_CONCURRENCY:-}
      # Integration Fabric ingress: per-instance fairness, retries, retention, reconciler, workers.
      - INGRESS_INSTANCE_LIMIT=${INGRESS_INSTANCE_LIMIT:-}
      - INGRESS_INSTANCE_TTL=${INGRESS_INSTANCE_TTL:-}
      - INGRESS_IP_LIMIT=${INGRESS_IP_LIMIT:-}
      - INGRESS_MAX_ATTEMPTS=${INGRESS_MAX_ATTEMPTS:-}
      - INGRESS_RETRY_DELAY_MS=${INGRESS_RETRY_DELAY_MS:-}
      - INGRESS_DEDUP_RETENTION_DAYS=${INGRESS_DEDUP_RETENTION_DAYS:-}
      - INGRESS_RETENTION_DAYS=${INGRESS_RETENTION_DAYS:-}
      - INGRESS_TIMESTAMP_TOLERANCE_SEC=${INGRESS_TIMESTAMP_TOLERANCE_SEC:-}
      - INGRESS_RECONCILE_INTERVAL_MS=${INGRESS_RECONCILE_INTERVAL_MS:-}
      - INGRESS_RECONCILE_GRACE_MS=${INGRESS_RECONCILE_GRACE_MS:-}
      - INGRESS_RECONCILE_BATCH_SIZE=${INGRESS_RECONCILE_BATCH_SIZE:-}
      - INGRESS_RECONCILE_MAX_ATTEMPTS=${INGRESS_RECONCILE_MAX_ATTEMPTS:-}
      - WEBHOOK_RECONCILE_INTERVAL_MS=${WEBHOOK_RECONCILE_INTERVAL_MS:-}
      - WEBHOOK_RECONCILE_GRACE_MS=${WEBHOOK_RECONCILE_GRACE_MS:-}
      - WEBHOOK_RECONCILE_BATCH_SIZE=${WEBHOOK_RECONCILE_BATCH_SIZE:-}
      - WEBHOOK_RECONCILE_MAX_ATTEMPTS=${WEBHOOK_RECONCILE_MAX_ATTEMPTS:-}
      - WEBHOOK_OUTBOX_RETENTION_DAYS=${WEBHOOK_OUTBOX_RETENTION_DAYS:-}
      - INGRESS_WORKER_CONCURRENCY=${INGRESS_WORKER_CONCURRENCY:-}
      # WebSocket gateway limits (frames bypass the HTTP guard pipeline).
      - WS_RATE_LIMIT_FRAME_PER_SECOND=${WS_RATE_LIMIT_FRAME_PER_SECOND:-}
      - WS_RATE_LIMIT_FRAME_BURST=${WS_RATE_LIMIT_FRAME_BURST:-}
      - WS_RATE_LIMIT_HANDSHAKE_MAX=${WS_RATE_LIMIT_HANDSHAKE_MAX:-}
      - WS_RATE_LIMIT_HANDSHAKE_WINDOW_MS=${WS_RATE_LIMIT_HANDSHAKE_WINDOW_MS:-}
      - WS_MAX_SOCKETS_PER_KEY=${WS_MAX_SOCKETS_PER_KEY:-}
      # Database / Redis connection tuning (selection keys are forwarded above).
      - DATABASE_POOL_SIZE=${DATABASE_POOL_SIZE:-}
      - DATABASE_STATEMENT_TIMEOUT_MS=${DATABASE_STATEMENT_TIMEOUT_MS:-}
      - DATABASE_IDLE_TIMEOUT_MS=${DATABASE_IDLE_TIMEOUT_MS:-}
      - DATABASE_CONNECTION_TIMEOUT_MS=${DATABASE_CONNECTION_TIMEOUT_MS:-}
      - MAIN_DATABASE_NAME=${MAIN_DATABASE_NAME:-}
      - REDIS_USERNAME=${REDIS_USERNAME:-}
      - REDIS_PASSWORD=${REDIS_PASSWORD:-}
      - REDIS_CONNECT_TIMEOUT_MS=${REDIS_CONNECT_TIMEOUT_MS:-}
      - REDIS_TLS=${REDIS_TLS:-}
      - REDIS_CACHE_DB=${REDIS_CACHE_DB:-}
      - QUEUE_PREFIX=${QUEUE_PREFIX:-}
      # Engine / session tuning.
      - MAX_CONCURRENT_SESSIONS=${MAX_CONCURRENT_SESSIONS:-}
      - BAILEYS_AUTH_DIR=${BAILEYS_AUTH_DIR:-}
      - BAILEYS_BROWSER_NAME=${BAILEYS_BROWSER_NAME:-}
      - BAILEYS_LOG_LEVEL=${BAILEYS_LOG_LEVEL:-}
      - BAILEYS_MESSAGE_STORE_LIMIT=${BAILEYS_MESSAGE_STORE_LIMIT:-}
      - BAILEYS_SESSION_STORE_MAX_ENTRIES=${BAILEYS_SESSION_STORE_MAX_ENTRIES:-}
      - BAILEYS_CHAT_STATE_CACHE_MAX=${BAILEYS_CHAT_STATE_CACHE_MAX:-}
      - BAILEYS_GROUP_METADATA_CACHE_TTL_MS=${BAILEYS_GROUP_METADATA_CACHE_TTL_MS:-}
      - WWEBJS_ONBOARDING_CONTINUE_LABELS=${WWEBJS_ONBOARDING_CONTINUE_LABELS:-}
      # Default matches the image's ENV (Dockerfile); a blank forward would shadow it and make
      # puppeteer fall back to a bundled Chromium the image does not ship. compose-parity.spec.ts
      # asserts the default stays in sync with the Dockerfile value.
      - PUPPETEER_EXECUTABLE_PATH=${PUPPETEER_EXECUTABLE_PATH:-/usr/local/bin/puppeteer-chrome}
      - SHUTDOWN_DELAY_MS=${SHUTDOWN_DELAY_MS:-}
      - SIMULATE_TYPING_MAX_MS=${SIMULATE_TYPING_MAX_MS:-}
      - LID_MAPPING_CACHE_MAX=${LID_MAPPING_CACHE_MAX:-}
      # Media / storage budgets and sweeps.
      - EXPORT_INLINE_MEDIA_BUDGET_BYTES=${EXPORT_INLINE_MEDIA_BUDGET_BYTES:-}
      - CHAT_HISTORY_MEDIA_BUDGET_BYTES=${CHAT_HISTORY_MEDIA_BUDGET_BYTES:-}
      - STATUS_MEDIA_MAX_BYTES=${STATUS_MEDIA_MAX_BYTES:-}
      - STATUS_ORPHAN_SWEEP_INTERVAL_MS=${STATUS_ORPHAN_SWEEP_INTERVAL_MS:-}
      - STATUS_ORPHAN_GRACE_MS=${STATUS_ORPHAN_GRACE_MS:-}
      - STORAGE_IMPORT_MAX_BYTES=${STORAGE_IMPORT_MAX_BYTES:-}
      - STORAGE_IMPORT_MAX_ENTRIES=${STORAGE_IMPORT_MAX_ENTRIES:-}
      - STORAGE_IMPORT_MAX_TOTAL_BYTES=${STORAGE_IMPORT_MAX_TOTAL_BYTES:-}
      - STORAGE_EXPORT_TTL_MS=${STORAGE_EXPORT_TTL_MS:-}
      - STORAGE_EXPORT_SWEEP_MAX_AGE_MS=${STORAGE_EXPORT_SWEEP_MAX_AGE_MS:-}
      - S3_REPROBE_INTERVAL_MS=${S3_REPROBE_INTERVAL_MS:-}
      - MESSAGE_REAPER_INTERVAL_MS=${MESSAGE_REAPER_INTERVAL_MS:-}
      - MESSAGE_REAPER_GRACE_MS=${MESSAGE_REAPER_GRACE_MS:-}
      - MESSAGE_REAPER_BATCH_SIZE=${MESSAGE_REAPER_BATCH_SIZE:-}
      # Plugin sandbox limits and the install trust knobs.
      - PLUGIN_DOWNLOAD_MAX_BYTES=${PLUGIN_DOWNLOAD_MAX_BYTES:-}
      - PLUGIN_INSTALL_REQUIRE_PIN=${PLUGIN_INSTALL_REQUIRE_PIN:-}
      - PLUGIN_CATALOG_URL=${PLUGIN_CATALOG_URL:-}
      - PLUGIN_CAP_TIMEOUT_MS=${PLUGIN_CAP_TIMEOUT_MS:-}
      - PLUGIN_STORAGE_MAX_BYTES=${PLUGIN_STORAGE_MAX_BYTES:-}
      # The dashboard's update notice asks GitHub for the latest release; false turns that request off.
      - UPDATE_CHECK_ENABLED=${UPDATE_CHECK_ENABLED:-}
      # Misc application tuning.
      - AUDIT_RETENTION_DAYS=${AUDIT_RETENTION_DAYS:-}
      - MESSAGE_RETENTION_DAYS=${MESSAGE_RETENTION_DAYS:-}
      - BULK_MAX_CONCURRENT_BATCHES=${BULK_MAX_CONCURRENT_BATCHES:-}
      - TEMPLATE_RENDER_MAX_CHARS=${TEMPLATE_RENDER_MAX_CHARS:-}
      - STATS_CACHE_TTL_MS=${STATS_CACHE_TTL_MS:-}
      - SEARCH_PROVIDER=${SEARCH_PROVIDER:-}
      - SEARCH_LIMIT_MAX=${SEARCH_LIMIT_MAX:-}
      - LOG_FORMAT=${LOG_FORMAT:-}
      - BASE_URL=${BASE_URL:-}
      - DASHBOARD_URL=${DASHBOARD_URL:-}
      # Docker socket proxy (openwa-api never touches /var/run/docker.sock directly)
      - DOCKER_HOST=tcp://docker-proxy:2375
    volumes:
      - openwa-data:/app/data
    depends_on:
      docker-proxy:
        condition: service_started
        # Not required: lets operators disable the proxy (see the docker-proxy env comment)
        # without breaking `docker compose up`. DockerService degrades gracefully without it.
        required: false
      postgres:
        condition: service_healthy
        required: false
      redis:
        condition: service_healthy
        required: false
    healthcheck:
      test: ['CMD', 'curl', '-f', 'http://localhost:2785/api/health/ready']
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 30s
    labels:
      - 'com.openwa.service=api'
      - 'com.openwa.core=true'

  # The dashboard SPA is now bundled into the openwa-api image and served by NestJS on the
  # same port (2785) — there is no separate dashboard container. Reach it at the openwa-api
  # port directly; put your own TLS reverse proxy (nginx/Caddy/cloud LB) in front if needed.

  # ===== OPTIONAL: Built-in PostgreSQL =====
  postgres:
    image: postgres:16-alpine
    container_name: openwa-postgres
    profiles: ['postgres', 'full']
    restart: unless-stopped
    networks:
      - openwa-network
    security_opt:
      - 'no-new-privileges:true'
    environment:
      POSTGRES_USER: ${DATABASE_USERNAME:-openwa}
      # No committed default secret. Empty unless the operator sets it; the postgres
      # image refuses to initialize with an empty password (clear fail-fast), and the app's
      # production boot guard rejects empty/placeholder secrets before startup.
      POSTGRES_PASSWORD: ${DATABASE_PASSWORD:-}
      POSTGRES_DB: ${DATABASE_NAME:-openwa}
      # PostgreSQL schema for OpenWA's tables + migration ledger. Default 'public'. The init
      # script below creates the schema (if non-public) and sets the database default search_path,
      # so the built-in container works zero-config with a custom schema. Ignored when 'public'.
      POSTGRES_SCHEMA: ${POSTGRES_SCHEMA:-public}
    volumes:
      - postgres-data:/var/lib/postgresql/data
      # Runs only on first init (postgres image docker-entrypoint-initdb.d semantics): creates the
      # configured schema and sets the DB default search_path. No-op for 'public'; does not re-run
      # on an already-initialized volume (see the note in scripts/postgres-init-schema.sh).
      - ./scripts/postgres-init-schema.sh:/docker-entrypoint-initdb.d/01-create-schema.sh:ro
    healthcheck:
      test: ['CMD-SHELL', 'pg_isready -U ${DATABASE_USERNAME:-openwa}']
      interval: 5s
      timeout: 3s
      retries: 5
    labels:
      - 'com.openwa.service=database'
      - 'com.openwa.builtin=true'

  # ===== OPTIONAL: Built-in Redis =====
  redis:
    image: redis:7-alpine
    container_name: openwa-redis
    profiles: ['redis', 'full']
    restart: unless-stopped
    networks:
      - openwa-network
    security_opt:
      - 'no-new-privileges:true'
    # noeviction is required by BullMQ: under any other policy Redis may evict queue keys once
    # maxmemory is reached, silently dropping queued jobs.
    command: redis-server --appendonly yes --maxmemory-policy noeviction
    volumes:
      - redis-data:/data
    healthcheck:
      test: ['CMD', 'redis-cli', 'ping']
      interval: 5s
      timeout: 3s
      retries: 5
    labels:
      - 'com.openwa.service=cache'
      - 'com.openwa.builtin=true'

  # ===== OPTIONAL: Built-in S3-compatible storage (pgsty/silo, a maintained MinIO fork) =====
  # Pinned by release tag and digest. Same server command, MINIO_ROOT_* variables, /data layout and
  # health route as MinIO, so an existing openwa_minio-data volume is used as-is. The service,
  # container and host name stay `minio`.
  minio:
    image: pgsty/silo:RELEASE.2026-09-16T00-00-00Z@sha256:635197cb9f36d01bee221d34d1c7d7960f6a95c48b0b6c01d99cd13bdae51a46
    container_name: openwa-minio
    profiles: ['minio', 'full']
    restart: unless-stopped
    networks:
      - openwa-network
    security_opt:
      - 'no-new-privileges:true'
    command: server /data --console-address ":9001"
    environment:
      # No committed default credentials. With both left empty the server would start on its own
      # built-in default pair, so the password falls back to a placeholder shorter than the
      # 8-character minimum: every combination without a real secret stops at startup with a fatal
      # credential error (with a user set and no secret it reads "Invalid credentials"). The user
      # default stays empty on purpose: a non-empty placeholder would be a valid access key.
      # Canonical names, matching what the app/dashboard read (legacy S3_ACCESS_KEY/S3_SECRET_KEY
      # still accepted as a fallback so older .env files keep working).
      MINIO_ROOT_USER: ${S3_ACCESS_KEY_ID:-${S3_ACCESS_KEY:-}}
      MINIO_ROOT_PASSWORD: ${S3_SECRET_ACCESS_KEY:-${S3_SECRET_KEY:-unset}}
    volumes:
      - minio-data:/data
    ports:
      - '127.0.0.1:9000:9000'
      - '127.0.0.1:9001:9001'
    healthcheck:
      test: ['CMD', 'curl', '-f', 'http://localhost:9000/minio/health/live']
      interval: 10s
      timeout: 5s
      retries: 3
    labels:
      - 'com.openwa.service=storage'
      - 'com.openwa.builtin=true'

volumes:
  # Pin explicit volume names so the compose path and the Docker-API orchestration path (which uses
  # the literal `openwa_<svc>-data` names in docker.service.ts) bind the SAME volume regardless of the
  # compose project name — otherwise re-enabling a built-in service could mount a fresh empty volume.
  openwa-data:
    name: openwa_openwa-data
    driver: local
  postgres-data:
    name: openwa_postgres-data
    driver: local
  redis-data:
    name: openwa_redis-data
    driver: local
  minio-data:
    name: openwa_minio-data
    driver: local

networks:
  # Application network — datastores and openwa-api. Keep the name `openwa-network`:
  # docker.service.ts attaches orchestrated containers to it by literal name.
  openwa-network:
    name: openwa-network
  # Isolated network for the Docker socket proxy. `internal: true` means no external
  # connectivity; only openwa-api joins it, so nothing else can reach docker-proxy:2375.
  # Plugins run inside openwa-api, so a loaded plugin can reach it too (see SECURITY.md).
  internal-docker:
    name: openwa-internal-docker
    internal: true