1# App Manifest
2
3Application workloads are declared in standalone TOML files — one file per app — not in `gordon.toml`. Apply with `gordon apps apply --file <app>.toml`, activate with `gordon apps deploy <app>`.
4
5The file is intended for Git: it must never contain secret values. Confidential values use `secrets`, whose values stay in pass; public values use `[env]` or `[services.<name>.env]`.
6
7## Minimal Example
8
9```toml
10name = "blog"
11
12[services.web]
13image = "gordon.mydomain.com/blog:1.4.2"
14
15[[services.web.http]]
16host = "blog.mydomain.com"
17port = 3000
18```
19
20Each `[services.<name>]` table declares one service, and the table key is the service name. A name may contain dots, so quote the key when it does: `[services."web.api"]`, `[[services."web.api".http]]`.
21
22## Full Example
23
24```toml
25name = "blog"
26
27[env]                          # optional, app-wide public env for all services
28APP_ENV = "production"
29
30[services.web]                    # one table per service; the key is the name
31image = "gordon.mydomain.com/blog:1.4.2"
32command = ["node", "server.js"]  # optional override
33stop_grace = "30s"               # optional, default 30s
34devices = ["transcode-gpu"]      # 0..n logical names from [app_devices.<name>]
35
36[services.web.readiness]         # optional explicit readiness
37type = "http"
38path = "/healthz"
39port = 3000
40timeout = "30s"
41
42[[services.web.http]]            # 0..n HTTP interfaces
43host = "blog.mydomain.com"
44port = 3000
45tls = "auto"                     # auto | always | never
46
47[[services.web.http]]            # optional private interface
48visibility = "internal"          # public (default) | internal
49port = 8080                      # required; no host or tls
50
51[services.web.env]               # optional, non-secret env for this service only
52LOG_LEVEL = "debug"              # overrides an app-wide [env] key of the same name
53
54[services.web.secrets]           # ENV name -> secret name (values in pass)
55API_KEY = "api-key"                                     # env delivery
56DATABASE_URL = { name = "database-url", as = "file" }   # /run/secrets/DATABASE_URL
57DB_PASS = { name = "db-pass", as = "env" }              # explicit env delivery
58
59[[services.web.volume]]          # 0..n named volumes only
60name = "web-data"
61path = "/data"
62readonly = false
63
64[[services.web.bind]]            # 0..n; name references [app_mounts.<name>]
65name = "app-logs"                # policy name, never a host path
66path = "/var/log/app"            # absolute container destination
67readonly = true
68
69[[services.web.database]]        # explicit database declarations
70name = "main"
71type = "postgres"
72schedule = "daily"
73
74[services.web.backup]            # backup targets by reference
75postgres = ["main"]
76volume = ["web-data"]
77
78[[network.shared]]               # optional shared-network memberships
79network = "backend"
80services = ["web"]
81```
82
83## Identity Rules
84
85- App name: DNS label (lowercase alphanumerics and hyphens, max 63), must not contain `--`, reserved: `gordon`, `registry`, `admin`, `localhost`. Case-insensitive uniqueness.
86- Service name: the `[services.<name>]` table key — `[a-z0-9_.-]`, max 63, quoted when it contains a dot. TOML keys are unique, so one table exists per name; names that share a runtime identifier after normalization (`web.api` and `web-api`) are rejected.
87- Bind name: `[a-z0-9_.-]`, max 63, must not contain `--`, unique within its service.
88- Device name: `[a-z0-9_.-]`, max 63, must not contain `--`, unique within its service.
89- Removing an app ends its incarnation: the name is freed, volumes and secrets are archived as retained under the old internal UUID, desired/active/intent state is cleared, and the next apply allocates a new UUID. A new app reusing the name never adopts the old secrets or volumes.
90
91## Services
92
93One or more explicitly named image-backed services per app, each in its own `[services.<name>]` table. Exactly one container per service — no replicas field. A service may expose HTTP, TCP, UDP interfaces, or none.
94
95```toml
96[[services.web.tcp]]
97entrypoint = "game"     # names installation entrypoints.<name>
98port = 25565
99publish = "0.0.0.0:25565"
100
101[[services.web.udp]]
102entrypoint = "game"
103port = 28015
104publish = "0.0.0.0:28015"
105```
106
107RCON is ordinary TCP: use TCP interfaces, no special RCON kind.
108
109`publish` must match the entrypoint listener it attaches to exactly: the same host, port, and transport. The traffic manager binds the entrypoint address, so a declaration that differs (for example a loopback bind on a wildcard entrypoint, or a different port) is rejected at apply time instead of being silently widened. `0.0.0.0` and an omitted host are the same wildcard.
110
111Image registry names and digest syntax are validated during manifest apply, resolution, deployment preflight, and immediately before every pull, including boot, restart, and recovery. Docker Hub (`docker.io` and canonical pull host `registry-1.docker.io`), `ghcr.io`, `quay.io`, and Gordon's configured registry are allowed by default. Add other exact registry hostname+port entries with `images.allowed_registries`; this does not configure credentials, and external resolution and pulls use anonymous access. When `images.require_digest` is enabled, every image reference, including Gordon registry images, must use `@sha256:<64 hex chars>`. See [Images](./images.md). This is a hostname allowlist, not DNS/IP validation or runtime egress enforcement.
112
113## Replacement on Deploy
114
115A deploy recreates a service whose image digest, spec, environment, or secret values changed and keeps an unchanged one.
116
117- A stateless TCP or HTTP service (no volume, bind, device, or UDP interface) is replaced without downtime. The new container starts and passes its readiness probe while the old one keeps serving; traffic then switches to the new container; only afterwards does the old container receive SIGTERM, and it is killed once its `stop_grace` (default `30s`) runs out. If the new container is not ready, it is removed and the old one is left untouched.
118- During the overlap both containers answer to the same network alias, so other services of the app may reach either generation for a short time. Database migrations must be backward compatible with the previous version of the service.
119- A service with a volume, bind, device, or UDP interface, or whose currently running generation has one, is replaced sequentially: traffic is withdrawn, the old container is stopped and removed, then the new one starts. It may be briefly interrupted, and two generations of it never run at once. UDP services are excluded from the overlap because UDP sessions stay pinned to the backend they started on.
120
121## Environment and Secrets
122
123- `[env]` is app-wide public env injected into all services. Each key must be disjoint from every `[services.<name>.secrets]` key in the app. A change to `[env]` changes every service, so the next deploy recreates all of them, databases included. A change shows as `env/<KEY>` in diffs, one entry per added, removed, or changed key (names only, never values), and blocks `apps deploy --service`. Prefer `[services.<name>.env]` for values only one service needs.
124- `[services.<name>.env]` is public env for that service only. A key here overrides the same key in `[env]` for this service. Each key must be disjoint from the same service's `[services.<name>.secrets]` keys. Values follow the `[env]` rules: no newlines, 1 byte to 64 KiB, no secret references. Values are stored in revisions and shown in plain text, so use secrets for anything sensitive. A change shows as `service/<name>/env/<KEY>` in diffs and applies on the next deploy.
125- `[services.<name>.secrets]` maps ENV var name to service-local secret name. Values are written with `gordon apps secrets set` and stay in pass under `gordon/apps/<uuid>/<service>/<name>`. Running containers keep the values they were created with; `gordon apps deploy` applies new values. `restart` does not.
126- A secret entry is either a secret name string (env delivery) or a table `{ name = "...", as = "env" | "file" }`. `as` defaults to `env`; any other key or value is rejected.
127- `as = "file"` delivers the value as a read-only file (mode `0444`) at `/run/secrets/<ENV>` and sets `<ENV>_FILE=/run/secrets/<ENV>` instead of `<ENV>`. The value never appears in the container environment, `docker inspect`, or `/proc/1/environ`. Use it for images that support the `*_FILE` convention.
128- With file delivery, `<ENV>_FILE` must not be another secret key or a key in `[env]` or `[services.<name>.env]`, and no bind or volume of the service may mount at or under `/run/secrets`.
129- Switching a secret between `env` and `file` shows as `service/<name>/secret/<secret>` in diffs and applies on the next deploy. See [Secrets](./secrets.md#secret-files).
130
131## Volumes and Databases
132
133- Named volumes only, and no service-shared volumes. Replacement reuses volumes; removed services leave volumes retained and visible.
134- Manifests never carry host paths. A `[[services.<name>.bind]]` references an `[app_mounts.<name>]` policy the operator declares in `gordon.toml`; a bind whose name has no matching policy is rejected at apply time. See [Volumes](./volumes.md) and [Security Hardening](./security-hardening.md).
135- Named volumes are Gordon-owned app data: created, labeled, retained, backed up, and pruned by Gordon. Administrative binds are operator-owned host locations: Gordon mounts them and never creates, deletes, owns, backs up, or prunes them.
136- `[[services.<name>.bind]]` requires an absolute, normalized container `path` and an optional `readonly`. The policy's `read_only` and the bind's `readonly` force read-only together: either side wins and a manifest can never weaken its policy. Reserved destinations (`/`, `/proc`, `/sys`, `/dev`, `/boot` and their children) and paths colliding with a declared volume or another bind are rejected at apply time.
137- `devices` lists logical device names granted by `[app_devices.<name>]` policy the operator declares in `gordon.toml`; a device whose name has no matching policy, or whose policy does not allow the app+service pair, is rejected at apply time. Gordon resolves each name to explicit CDI IDs at activation time and encodes them as one native CDI `DeviceRequest`. Revisions persist the logical names, never the host resolution. A device add/remove shows as `service/<name>/devices` in diffs; reorder-only input is a no-op. Changing a mapping never recreates a running container: the next deploy serves the new resolution. Revoking a grant fails subsequent deploys closed while the running service is untouched. Device-bearing creates require Podman 5.4+ or Docker 28.3+ with native CDI configured; older or unrecognized engines return a structured `runtime-unsupported` error and never run without devices. Gordon installs no drivers, manages no quotas, and injects no `NVIDIA_*` environment: images carry their own runtime expectations.
138- A service with any bind runs as a single writer, like a volume-backed service: replacements never run two generations at once. `gordon apps status` and `apps show` mark such services `single-writer` (`restart_unsafe` in JSON); the mark is informational. See [apps status](../cli/apps.md#gordon-apps-status).
139- A volume declared `readonly = true` is mounted read-only in the container; the service cannot modify protected data.
140- `[[services.<name>.database]]` declares databases explicitly (no image inference). Only PostgreSQL is supported, and each database declares its own backup `schedule` (`hourly`, `daily`, `weekly`, or `monthly`).
141- `[services.<name>.backup]` lists the declared databases (`postgres`) and volumes (`volume`) that are backup targets. A declared database or volume that is not referenced here is never backed up. Schedules follow the declaration through deploys; stored backups are never deleted when declarations change.
142
143## HTTP Interfaces
144
145`[[services.<name>.http]]` declares 0..n HTTP interfaces per service. `visibility` is optional and defaults to `public` when the key is absent or empty.
146
147```toml
148[[services.web.http]]            # public: proxied by host
149host = "blog.mydomain.com"
150port = 3000
151tls = "auto"                     # auto | always | never
152
153[[services.web.http]]            # internal: not published (app private network only)
154visibility = "internal"
155port = 8080
156```
157
158- `public`: `host` is required and must be a valid public hostname, and `tls` is `auto` (default when absent), `always`, or `never`. A public interface gets a proxy route, a global host reservation, a certificate target when TLS applies, and a `127.0.0.1` loopback backend publication.
159- `internal`: creates no proxy route, no host reservation, no certificate target, and no host port publication, so it is never reachable through the host or proxy plane. `port` is required, `host` must be absent, and `tls` must be absent — any declared TLS value is rejected. It is still a declared TCP-capable container port for readiness metadata. Reachability follows network membership, not visibility: any container attached to a network this service joins can reach the port — sibling services on the app's own private network, and peers on any `[[network.shared]]` network the service is enrolled in.
160
161There is no `.internal` pseudo-domain: internal interfaces carry no hostname at all.
162
163A container port declared by both an internal HTTP interface and an externally backed interface (public HTTP or TCP) is rejected at apply time, because publication is socket-level. Duplicate internal HTTP ports within a service are rejected the same way.
164
165## Readiness
166
167`type = "http"` requires an origin-form `path` beginning with a single `/`. Absolute URLs, authority forms such as `@host:port`, scheme-relative paths, and control characters are rejected at apply time. The probe always dials the declared loopback backend, never follows redirects, ignores environment proxy settings, and is bounded per request and for the whole operation. A public or otherwise published interface keeps this loopback probe. An internal HTTP port is instead probed over the app private network by one bounded, short-lived helper, using HTTP or TCP according to the interface's `[services.<name>.readiness]` type; the internal port is never temporarily published to the host to probe it.
168
169`type = "tcp"` on a published port is ready when a connection to the loopback backend is accepted and then either receives data or stays open for 300ms. A connection closed right after accept counts as not ready: the host port relay of rootless Docker accepts on behalf of a container that does not listen yet. A workload that accepts and closes immediately by design should use `log` readiness instead.
170
171When a service declares no readiness (or `none`), deploy and restart still probe it in two cases: a stateless service whose interfaces are all public HTTP gets an `http` probe on `/`, and a stateless service with exactly one TCP interface and no HTTP or UDP interface gets a `tcp` probe on that port. Other services without a declared probe are not checked.
172
173## TLS
174
175TLS applies to public interfaces only; an internal interface never declares `tls`.
176
177- `auto` keeps the host eligible for HTTP and HTTPS; plaintext is redirected when an HTTPS endpoint exists and redirects are enabled.
178- `always` is enforced: a plaintext request to an `always` host is redirected whenever an HTTPS endpoint exists, and refused with `421 Misdirected Request` when none does. The backend is never reached over plaintext.
179- `never` stays plaintext and is never issued an app certificate.
180
181## Networks
182
183Each app gets a private network automatically. `[[network.shared]]` adds services to named shared networks, created/reused only within verified Gordon ownership. Deploy adds AND removes memberships without disconnecting unrelated services.
184
185Services of the same app communicate over that private network and resolve each other by service alias. Different apps are isolated by default; cross-app traffic requires both services to declare the same `[[network.shared]]` membership. See [Network Isolation](./network-isolation.md).
186
187## Telemetry
188
189When [telemetry log export](./telemetry.md#logs) is enabled, the stdout and stderr of every service are exported by default under `service.name = "<app>.<service>"`. `logs = false` opts out; a service setting overrides the app setting:
190
191```toml
192[telemetry]
193logs = false         # no service of this app exports...
194
195[services.web.telemetry]
196logs = true          # ...except web
197```
198
199Changing either setting is a service change: it applies on the next deploy.
200
201## Strictness
202
203Unknown fields, duplicate service names, unresolved service/entrypoint/backup refs, secret/env collisions (app-wide or per-service), unsafe identities, and forbidden mounts are hard errors at apply time. Canonical host conflicts (including system domains and external routes) fail before persistence.
204
205The keyed `[services.<name>]` form is the only accepted app schema: manifests written for an earlier v3 alpha using `[[service]]` with a `name` key and `[service.*]` tables are rejected with a keyed-schema diagnostic, not converted. See [Migrating a v3 alpha app manifest](../upgrading.md#migrating-a-v3-alpha-app-manifest).
206
207## Related
208
209- [Apps CLI](../cli/apps.md)
210- [Secrets](./secrets.md)
211- [Volumes](./volumes.md)
212- [External Routes](./external-routes.md)