Files
OwnCord/Server/Dockerfile
T
J3vbandClaude Fable 5 fb04a579c4 fix(docker): ship /app owned by the runtime uid; run the boot-smoke in CI too (#1378)
* fix(release): give the Docker boot-smoke a writable /app, and run it in CI

The v1.2.0-alpha.3 release run died at "Boot-smoke Docker image": a bare
`docker run` of the distroless image has nowhere the uid-65532 server can
write — /app is root-owned, and the VOLUME /app/data anonymous volume is
created root-owned too — so config.Load failed on "writing default config:
open config.yaml: permission denied" and the container exited. Real
deployments bind-mount config.yaml and data/, which is why the image itself
is fine.

Move the smoke into Server/scripts/docker-smoke.sh, run the container with
`--tmpfs /app --tmpfs /app/data` (Docker's tmpfs default mode is 1777, so
the non-root server can write both), and call the same script from ci.yml's
docker-build job — loading the image it already builds — so the smoke is
exercised on every PR to main instead of for the first time at tag time.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(docker): ship /app and /app/data owned by the runtime uid so a bare run boots

The tmpfs approach did not survive CI: runc re-applies the underlying
directory's mode to a tmpfs mounted over an existing path, so /app stayed
root:755 and the write still failed. Fix the image instead of the harness:
stage /app/data in the builder, chown it to 65532, COPY --chown it into the
distroless stage before WORKDIR. Docker seeds the VOLUME's anonymous volume
from that image dir, ownership included, so `docker run <image>` with no
mounts now boots and answers /health — which is also the contract the smoke
should be testing, so it goes back to a bare `docker run`.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
2026-08-15 22:17:21 +02:00

52 lines
2.0 KiB
Docker

# syntax=docker/dockerfile:1
# ─── Build stage ────────────────────────────────────────────────────────────
FROM golang:1.26-bookworm AS builder
ARG VERSION=dev
WORKDIR /src
COPY go.mod go.sum ./
RUN go mod download
COPY . .
RUN CGO_ENABLED=0 GOOS=linux go build \
-o /chatserver \
-ldflags "-s -w -X main.version=${VERSION}" \
.
# Skeleton for the runtime image's working dir. The server writes its default
# config.yaml to the cwd and the SQLite DB, cert and uploads under data/, and
# it runs as uid 65532 — so both must be owned by that uid or a bare
# `docker run` dies on boot with "writing default config: permission denied"
# (a WORKDIR created by the final stage is root-owned, and Docker seeds the
# VOLUME's anonymous volume from the image dir, ownership included). Staged
# here because distroless has no shell to chown with.
RUN mkdir -p /skel/app/data && chown -R 65532:65532 /skel/app
# ─── Final stage ─────────────────────────────────────────────────────────────
# gcr.io/distroless/static-debian12: no shell, includes CA certs (needed for
# TLS/ACME) and timezone data. Attack surface is minimal.
FROM gcr.io/distroless/static-debian12
COPY --from=builder --chown=65532:65532 /skel/app /app
WORKDIR /app
COPY --from=builder /chatserver /chatserver
# /app/data is the default data_dir (SQLite DB + uploads).
# Mount a named volume here to persist data across container restarts.
VOLUME ["/app/data"]
# Server listens on this port by default (configurable via config.yaml).
EXPOSE 8443
# Refuses the in-place self-update endpoint (the binary is image content;
# upgrades are image pulls). See Server/updater/container.go.
ENV OWNCORD_CONTAINER=1
# Run as non-root (distroless provides uid 65532 = "nonroot").
USER 65532:65532
ENTRYPOINT ["/chatserver"]