The server node created by `scheduler-k3s:initialize` had no way to receive kubelet arguments, unlike nodes joined through `scheduler-k3s:cluster:add` or configured via `scheduler-k3s:profiles:add`. This meant settings such as `allowed-unsafe-sysctls` were unreachable on a single-node install.
The dokku entrypoint already exports DOKKU_ROOT, PLUGIN_PATH, PLUGIN_AVAILABLE_PATH, PLUGIN_ENABLED_PATH, PLUGIN_CORE_AVAILABLE_PATH and DOKKU_LIB_ROOT before invoking plugn. Delegating to `dokku plugin:trigger` keeps the helper at env-var parity with the production trigger path without enumerating every var the helper must forward.
Triggers that call verify_app_name or otherwise reference $DOKKU_ROOT fail under bats because the helper did not forward DOKKU_ROOT to the plugn subshell. The production dokku entrypoint exports DOKKU_ROOT before invoking plugn, so bring the helper to env-var parity.
The dokku entrypoint exports both vars before invoking plugn, so triggers that source `$PLUGIN_AVAILABLE_PATH/...` work in production. The helper only forwarded PLUGIN_PATH / PLUGIN_CORE_AVAILABLE_PATH / DOKKU_LIB_ROOT, so triggers that follow the same pattern fail under bats.
Parallel `unit.scheduler-k3s-*` matrix jobs all pushed to the same `savant/rdmtestapp:1` tag on Docker Hub, so a herokuish run pod could pull a CNB or dockerfile image that another job had just overwritten and fail with `exec: "/exec": stat /exec: no such file or directory`. The image-repo-template now embeds the bats file basename so each job owns its own tag namespace.
Bats runs tests under `set -eo pipefail`, so when `grep -F -o` finds nothing inside the count pipe it exits 1, the whole pipe fails, errexit fires, and the function aborts before reaching the count comparison. Wrap grep in `{ ... || true; }` so the pipe stays zero when the pattern is absent and the helper falls through to the flunk message.
Replace the bash pattern-substitution loop with grep -F -o piped to wc -l so the helper counts literal substring occurrences instead of treating the expected value as a glob pattern. The old implementation interpreted `[`, `]`, `*`, `?`, and `\` as pattern syntax, which made `assert_output_contains "['task.py', 'test']"` report 17 matches against an output that contained the string exactly once - the inner characters were being matched as a character class. assert_output_not_contains delegates to assert_output_contains and is fixed transitively.
Both helpers wrap `run /bin/bash -c "..."` with the env vars dokku plugin scripts and `plugn` need, replacing the long inline boilerplate that was duplicated across `tests/unit/resource_3.bats` and the new `core-post-extract` regression tests in the builder bats files.
The ppa:cncf-buildpacks/pack-cli launchpad source is currently down, breaking the runtime image build and any bats run that calls install_pack. Pack is now tracked in contrib/dependencies.json and pulled from the buildpacks/pack GitHub release tarball, matching how other binary deps are managed. update-deb-dependencies gains hyphen-aware word-boundary matching so single-word names like pack cannot accidentally rewrite unrelated debian/control lines, keeping pack out of the package's hard requirements.
The byjg/easy-haproxy image polls Docker for label changes every 10
seconds by default, which races with the haproxy bats suite and
intermittently produces curl exit 7. Expose `refresh-conf` as a
global-only haproxy property that maps to `EASYHAPROXY_REFRESH_CONF`,
lower it to 2 seconds in the bats setup, and wrap the localhost HTTP
assertions in a retry loop so checks wait for haproxy to converge
rather than failing on the first attempt.
The jq binary doesn't support jsonc, so we can't modify it for testing purposes. Instead, just create an app.json with the correct heroku.postdeploy task.
This plugin is mostly compatible with the nginx plugin, but runs the proxy within a docker container. Users do not have direct access to add custom openresty configuration at this time, but instead receive the ability to setup automatic ssl on first request via letsencrypt integration.
When developing Dokku, it is useful to use a real domain to test actual functionality. This is impossible if the domain is hardcoded to dokku.me in testing.
Previously, we would always set the port mapping during a dockerfile build, making it difficult for users to override mappings. We also only _sometimes_ updated the detected port mapping, further confusing issues when users were migrating from Dockerfile to Buildpacks for builds.
Now, we always detect the port mapping during the build process, and only use that detected port mapping if an override is not specified. This greatly simplifies the experience around port mapping, as now a user can create an app, set a port mapping, and that first deploy will respect the port mapping without an additional deploy.
The builder always has the best context for what the app should be listening on, and thus we can always specify a "default" port mapping at this stage. Users can override this map as desired later.
This change also results in the removal of a ton of internal code that is now centralized in the ports plugin.
Closes#4067
Due to how the variable was assigned, there were cases where the network was specified twice with a := as a delimiter. This would result in no network being removed.
During an app build, we now auto-detect ports based on the source code. This is usually http:80:5000, with Dockerfile-based deploys having their ports extracted from the docker image or Dockerfile. Additionally, we add an https:443 mapping for any detected http:80 mapping when there is an ssl certificate, and all http port mappings are transformed to https mappings for Dockerfile-based deploys.
While the ports aren't currently consumed, a future refactor will provide the ability to fallback to the new detected ports when there is no user-specified port mapping.
The shorthand is more prevalent in this codebase, and is something that bash supports, so we should just use the same thing everywhere.
Note that we do not use shorthand redirect in Makefile as shell parsing is a bit different in Make and the shorthand redirect doesn't seem to be properly supported, causing CI errors.
This makes standard use of shellcheck work without needing to provide extra configuration anywhere.
Also remove use of inline 'shellcheck disable' calls that are already defined in the .shellcheckrc and don't need to be set inline.
