Admin and dashboard containers crash-loop after their first restart

[marcelo-maciel]

The admin and dashboard containers crash-loop forever after their first restart.

Both clients/admin/docker/docker-entrypoint.sh and clients/dashboard/docker/docker-entrypoint.sh run under set -e, render /usr/share/nginx/html/config.json from /usr/share/nginx/html/config.json.template with envsubst, and then rm the template so it is not served. A container's writable layer survives docker restart, a Docker daemon restart and a host reboot under restart: unless-stopped, so the next start of the same container finds no template. The input redirect fails, set -e exits 1, and the container never comes back.

Repro

docker build -t fsh-dashboard clients/dashboard
docker run -d --name d -e FSH_API_URL=http://api.example fsh-dashboard   # running
docker restart d                                                        # exited, code 1
docker logs d
# /docker-entrypoint.sh: line 9: can't open /usr/share/nginx/html/config.json.template: no such file

The admin image fails the same way, at line 14. A compose stack built from these images sat in this loop with more than 2,000 restarts per container before anyone noticed, because restart: unless-stopped keeps retrying without surfacing anything.

Workaround

Recreate the containers (docker compose up -d --force-recreate admin dashboard). A fresh container gets the template back, but only until its next restart.