The healthcheck was written in single quotes, which shellcheck flags as
"expressions don't expand in single quotes" even though deferring the
expansion to the container is exactly what is wanted. Keep the same value
with escaped double quotes: the backslash stops the host shell from touching
${RS3_PORT}, and the container expands it at run time.
Verification:
- bash -n scripts/run_rs3.sh
- docker run with the rendered --health-cmd resolves RS3_PORT inside the container
The rs3 server binary never reads RS3_ENDPOINT; it is a client-only setting
for rs3-cli, which only the old healthcheck used. Switch the rs3 healthcheck
to `curl --fail .../readyz` with a start period (matching scripts/run_rs3.sh
and CI) and remove the now-dead RS3_ENDPOINT from compose.yml, the demo-server
compose, run_rs3.sh, and the test workflow.
Verification:
- git grep RS3_ENDPOINT -> no matches
- docker compose config for compose.yml and tools/demo-server/compose.yml
Run the self-hosted rs3 service instead of Minio in test.yml and swap the
local helper script, Makefile helm values, and AGENTS.md notes accordingly.
Fix the E2E workflow (refs #426): it only triggered on the `main` branch,
which does not exist in this repository, so the job never ran. Trigger it
on `dev` for both push and pull_request like test.yml, build the image via
the consolidated `make docker-build`, bump the checkout/setup-go actions,
drop the hosted-toolcache cleanup, and refresh the E2E tooling to k6
v2.3.0 and skopeo v1.24.1.
Repair the image startup path as well: start nginx with an explicit
/etc/nginx/nginx.conf and a /dev/stderr error log so the non-root user does
not touch the compiled-in log path, create /etc/sigma with mode 0755 before
copying config.yaml (COPY --chmod made the directory non-traversable), and
disable redis in the shipped default config since cache, locker and work
queue are inmemory.
Tosone