Skip to content

docker.md

Latest commit

 

History

History
167 lines (132 loc) · 8.54 KB

docker.md

File metadata and controls

167 lines (132 loc) · 8.54 KB
Raw

📖 This documentation is also published, web-native, at https://valthon.github.io/zigbase/docs/docker — the site is the canonical reading experience.

Docker

ZigBase has no native Windows build (the embedded HTTP server depends on facil.io/zap, Linux/macOS only). The official Docker image, ghcr.io/valthon/zigbase, is the supported path for Windows-hardware users and for anyone who prefers a Docker-first self-host workflow on Linux/macOS too. It ships the same static-musl binary as the release tarballs — no in-image compilation, no extra runtime deps.

Quick start

docker run -d --name zigbase -p 8090:8090 -v zigbase_data:/data ghcr.io/valthon/zigbase:latest

Zero arguments beyond the standard docker run flags — the image's default command is serve. Open http://localhost:8090/_/ and create a superuser:

docker exec zigbase /zigbase superuser create --email you@example.com --password 'change-me' --data-dir /data

A docker-compose.yml quick-start lives at the repo root; docker compose up brings up the same thing with a named volume pre-wired.

Image details

  • Base: gcr.io/distroless/static-debian12:nonroot — no shell, no package manager, just the CA bundle and a numeric non-root user (uid/gid 65532). The binary is already statically linked (musl); distroless is chosen over scratch specifically because ZigBase's outbound TLS clients (SMTP/STARTTLS mail delivery, a Postgres backend with sslmode=require, OAuth2 token exchange, Sentry error reporting) load the system CA bundle from disk at connect time — scratch has none, and every one of those integrations would fail TLS verification. If you don't use any of those integrations this doesn't affect you either way, but distroless costs nothing over scratch here.
  • User: runs as non-root uid 65532 always. There is no way to run the container as root.
  • Data directory: /data, baked in as ZIGBASE_DATA_DIR=/data and declared as a VOLUME. Mount a named volume (-v zigbase_data:/data, auto-created and owned correctly by Docker) or a bind mount — a bind mount from the host must be pre-created and chown 65532:65532'd, or the container will fail to write data.db (this is the most common non-root-container support question; see below).
  • Network bind: the image sets ZIGBASE_HTTP_HOST=0.0.0.0, overriding the binary's own secure-by-default loopback bind (127.0.0.1 on bare metal). Inside a container this override is necessary — otherwise nothing outside the container's network namespace could reach it — and it does not weaken security: the container boundary plus whatever you pass to -p/--network is the real access-control surface here, same as it would be for any other containerized server.

Custom application images

A framework-mode application that builds its own non-root image must reproduce the runtime contract the official image supplies. Before switching to uid/gid 65532, create /data with that ownership; otherwise a new named volume is mounted as root-owned and ZigBase cannot create data.db:

Zig does not document an official registry image. Do not guess a builder such as ghcr.io/ziglang/zig:<version>: a syntactically plausible registry name can still be private or nonexistent. Start from a pinned general-purpose Linux builder and download the matching Zig release archive from https://ziglang.org/download/<version>/, verifying the SHA-256 published in Zig's official download/index.json, or use an independently maintained builder image only when its provenance and immutable digest are explicitly accepted. For Zig 0.16.0 on x86_64 Linux, the official archive SHA-256 is 70e49664a74374b48b51e6f3fdfbf437f6395d42509050588bd49abe52ba3d00.

The final runtime base and the build target must agree. In particular, gcr.io/distroless/static-debian12:nonroot has no glibc dynamic loader, so build a static musl artifact (for example, zig build -Dtarget=x86_64-linux-musl -Doptimize=ReleaseSafe) before copying it there. A native glibc build copied into that image exits as exec /path/to/app: no such file or directory even though the file exists. If you intentionally build for glibc, choose a runtime image containing the matching loader instead. Smoke-test the assembled image's real serve command and HTTP readiness; a version healthcheck alone does not exercise its loader, writable data directory, or listener.

RUN install -d -o 65532 -g 65532 /data
ENV ZIGBASE_DATA_DIR=/data \
    ZIGBASE_HTTP_HOST=0.0.0.0 \
    ZIGBASE_SERVE_BACKGROUND=0
USER 65532:65532
VOLUME ["/data"]
ENTRYPOINT ["/usr/local/bin/my-app"]
CMD ["serve"]

The explicit foreground setting prevents an inherited coding-agent environment from detaching the process inside Docker. Verify the assembled image—not only the build stage—with a named volume, an HTTP readiness probe, and doctor --production.

Bind-mount ownership

mkdir -p ./zb_data && sudo chown 65532:65532 ./zb_data
docker run -d -p 8090:8090 -v "$(pwd)/zb_data:/data" ghcr.io/valthon/zigbase:latest

Skip this and use a named volume (-v zigbase_data:/data) instead if you don't need the data directory visible on the host filesystem — Docker creates and owns named volumes correctly without a manual chown.

Plain-HTTP / local / LAN use

Auth cookies are Secure by default (HTTPS-only), same as the bare-metal binary. For a plain-HTTP local or LAN deployment, set ZIGBASE_COOKIE_SECURE=false (equivalent to the bare-metal --insecure-cookies flag):

docker run -d -p 8090:8090 -v zigbase_data:/data \
  -e ZIGBASE_COOKIE_SECURE=false \
  ghcr.io/valthon/zigbase:latest

Put a TLS-terminating reverse proxy (Caddy, Traefik, nginx) in front for anything beyond local/LAN use, and drop ZIGBASE_COOKIE_SECURE=false once you do.

Configuration

Every environment variable in the Configuration reference applies unchanged — pass them with -e / compose's environment:. The two the image itself sets (ZIGBASE_HTTP_HOST=0.0.0.0, ZIGBASE_DATA_DIR=/data) can be overridden the same way if you have a reason to (e.g. mounting the data dir somewhere else inside the container).

Healthchecks

GET /api/health (see API reference) is the right endpoint to probe, but the distroless base has no shell/curl/wget to run a classic HEALTHCHECK CMD curl … instruction from inside the image. Two supported options:

  • Orchestrator-native HTTP probes (Kubernetes httpGet readiness/liveness probes, ECS/Nomad HTTP health checks) — point them at GET /api/health on the published port directly; they run outside the container and don't need a shell inside it.
  • A reverse proxy in front (Caddy/Traefik) that itself health-checks the upstream via HTTP and can expose its own HEALTHCHECK.

The bundled docker-compose.yml example uses /zigbase version (an in-image, no-shell liveness check) as a lightweight default — it confirms the binary starts and runs, but it is not an HTTP readiness check. Prefer an orchestrator-native HTTP probe against /api/health for production.

Tags

  • latest — newest tagged release.
  • X.Y.Z — an exact release, e.g. ghcr.io/valthon/zigbase:0.9.0.
  • X.Y / X — rolling aliases that track the newest patch/minor within that line.

Pin to X.Y.Z for reproducible deployments. Images are published to ghcr.io/valthon/zigbase (GitHub Container Registry) on each tagged release. Maintainer note: GHCR creates a newly pushed package as private by default — after the first release push, a repo owner must flip its visibility to Public once (package Settings → Change visibility) before docker pull works for anonymous/public users.

What's NOT in the image

No shell, no package manager, no Zig toolchain, no build tools — you cannot docker exec … sh into it. Use docker exec … /zigbase <subcommand> for superuser create, migrate, etc. If you need a debugging shell, docker cp files out or run the same binary from a release tarball locally instead.

See also