The previous regression case selected `--openresty-global-hsts`, but `hsts` is rejected by `openresty:set --global` because it is not in the plugin's GLOBAL_KEYS list. Switch to `--openresty-letsencrypt-server`, which the bats setup already pins to a known staging URL, so the test exercises the same `--global --<info-flag>` parsing path without needing a separate set step.
The argument-handling block in each plugin's report command overwrote APP with --global before checking whether $1 actually held an info flag, so `dokku <plugin>:report --global --<plugin>-foo` always emitted the full report instead of the requested value. Reorder the conditionals across the sixteen plugins that share the pattern so the info-flag check runs first, and add a regression bats case for each one.
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 default nginx template rendered four near-identical server blocks per app, so the same listen, access_log, error_log, ssl_*, error_page, and proxy chain had to be maintained across http, https, grpc, and grpcs branches. Merging http and https into a single branch keyed on an `is_ssl` boolean, and likewise grpc and grpcs, removes the duplicate proxy_set_header chain and error-page locations and brings the structure in line with the openresty proxy template. Output is preserved up to whitespace and the existing `cat -s` pass already squashes the leftover blank lines. Adds plugins/nginx-vhosts/template_test.go exercising the rendered output via sigil as a Go library across HTTP-only, HTTPS, HTTP-to-HTTPS redirect, no-listeners 502 fallback, gRPC, gRPCs, gRPC-without-listeners skip, IPv4 bind, upstream blocks with and without keepalive, the access_log `off` short-circuit, the X-Forwarded-Ssl toggle, the http2 listen-parameter vs directive split, http2_push_preload conditional emission, and the nginx.conf.d/*.conf include in every code path; and tests/unit/nginx-vhosts_16.bats covering deploy plus `nginx -t` end-to-end for both HTTP and HTTPS apps.
Adds plugins/storage/migrate_test.go exercising migrateApp and MigrateLegacyMounts: single deploy phase, cross-phase grouping into one attachment, :ro and free-form volume options preserved on the attachment, idempotency on a second pass, the per-app flag-file fast path that skips already-migrated apps, and the name-collision refusal that aborts before draining. tests/unit/storage.bats gains an end-to-end case that stages a -v line via docker-options:add, runs storage:migrate, and asserts the synthesized colon form shows up in storage:list, the legacy-<hash> entry shows up in storage:list-entries, and the original -v line is gone from docker-options:report on both phases. To make that test (and operators with restored backups) practical, storage gains a storage:migrate <app>|--all subcommand backed by a new MigrateApp helper that skips the per-app flag file so a re-run actually re-scans.
Adds typed JSON build records under data/builds/<app>/<build-id>.{json,log} keyed on a stable base36 ULID-style DOKKU_BUILD_ID generated for every deploy. The new commands surface that history (builds:list, builds:info, builds:prune) and an operator-configurable retention via builds:set retention. The existing builds:cancel and builds:output now key on the build-id (with safe handling for already-finalized and abandoned records), and the per-build log file replaces journalctl as the durable source of truth for builds:output.
Moves the actual exec out of the storage plugin and into a new scheduler-storage-exec plugn trigger. scheduler-docker-local does docker run with TTY-aware -it/-i selection and --user derived from entry.Chown; scheduler-k3s creates a throwaway Pod via the kubernetes API, waits for it to reach Running with structured error reporting (ImagePullBackOff and friends are surfaced from the container status verbatim, no kubectl involvement), execs the user command via the existing SPDY plumbing in k8s.go, and deletes the Pod on the way out. (Entry).Validate now accepts either an absolute path or a docker named-volume token for docker-local entries so the migration synthesizer's named-volume legacy entries work cleanly. storage:exec gains --as-user for one-off uid overrides, propagates the underlying tool's exit code via os.Exit, and detects TTY/interactive mode from os.Stdin so non-interactive scripted use no longer trips over docker's input-device-is-not-a-tty error.
storage:list was calling the storage-list plugn trigger which read -v lines from docker-options. After the install-time migration drains those lines into the attachment store, that source is empty for every migrated app and for any app that only ever used storage:create + storage:mount. The fix moves CommandList to call a new in-process ListAppMountEntries helper that reads attachments directly, surfaces the entry name in JSON output via a new entry_name field, and falls back to the entry name as the host token for k3s entries with no host path so the colon form remains well-formed. The storage-list plugn trigger is kept for back-compat with external callers but emits a deprecation warning and now reads from the same attachment-driven source.
Updates persistent-storage.md to lead with the named storage entry workflow while keeping the legacy colon-form documentation intact, adds a Persistent storage section to the k3s scheduler doc, documents the storage-app-mounts, storage-create, storage-destroy, and storage-status triggers in plugin-triggers, and adds an entry to the 0.38.0 migration guide explaining the install-time migration of legacy mounts and the new DNS-1123 name validation. Bats coverage in tests/unit/storage.bats now exercises storage:create / list-entries / destroy, name validation rejections, multi-entry attachment, the destroy-while-mounted error, and the ensure-directory deprecation warning.
Multi-flag inputs (e.g. `--build-arg X=Y --link a --link b`) used to be stored as a single line, which bypassed the per-line filter that drops `--link` and similar flags for dockerfile-based builders. Each `--flag [value]` group is now stored as its own entry, and a `--process` typed after the app name is lifted into the subcommand flag instead of being stored as a docker option.
The new config and pull secret helm releases need to honor user-set annotations and labels for `--resource-type secret` so existing scheduler-k3s annotation tests keep passing. The dedicated charts now render `.Values.global.annotations` and `.Values.global.labels` onto their Secret manifest, and the deploy trigger plumbs `SecretAnnotations` and `SecretLabels` from the global annotation/label config. The rollback regression bats test now uses `dokku ps:rebuild` for its second deploy because git push of an unchanged ref is rejected by the dokku remote.
Bundling these Secrets in the app helm chart caused two bugs in the scheduler-k3s plugin: a chart rollback could delete Secrets that older ReplicaSets still referenced by exact timestamped name (`env-{app}.{ts}` and `ims-{app}.{ts}`), hard-crashing pods until manual intervention; and the strategic-merge `patchMergeKey` on `imagePullSecrets` let stale entries leak into the live Deployment until the list pointed at many nonexistent Secrets. Each Secret now lives in its own helm release with a stable name (`config-{app}` and `pull-secret-{app}`), installed before the app chart on every deploy. The deployment trigger also prunes any leaked `imagePullSecrets` entries from the live Deployment so the next deploy lands on a clean list, and the rename and destroy paths uninstall the new releases (and the previously-leaked TLS release on rename) under the old app name.
Every `:report` subcommand now recognizes `--global` as a scope selector that limits the report to globally-configured properties, including in JSON form via `--global --format json`. Previously this combination was rejected because `--global` was treated as an info flag, conflicting with `--format`. The shared `common.ParseReportArgs` helper now returns a `ReportArgs` struct exposing the parsed scope; each Go and bash report selects a global-only flag map when scope is global, and skips per-app verification.
The previous assertion piped grep -c through `|| echo 0`, which printed an extra `0` when the file existed without matches because grep -c emits the count of `0` and exits 1, triggering the fallback.
The scheduler-run script classified CNB-based images as `herokuish` because
`is_image_herokuish_based` returns true for them, which caused the
`docker-args-process-run` trigger for builder-pack to skip injecting
`--entrypoint launcher`. Without that flag the container fell back to the
image entrypoint (`/cnb/process/web`) and dropped the user-supplied
arguments. Mirror the deploy-side detection so CNB images set
`IMAGE_SOURCE_TYPE=pack`, allowing the launcher entrypoint to be added.
When ps:rebuild runs against an image-based deploy via git:from-image, the resulting image often shares the same SHA as the previous deployment, so retiring the old container's image would target the live image of the new container. The retirement is now skipped when another running container of the same app still references the image, and the cron retire loop self-heals previously stuck entries the next time it encounters them.
Renders the user-supplied nginx.conf.sigil via sigil into a tmp file and runs `nginx -t` against a wrapped copy as soon as the template is extracted from the source tree, so syntactically invalid templates abort the deploy before the build phase runs. Skipped when `proxy-type` is not `nginx`, when `disable-custom-config=true`, or when no custom template was extracted. Closes#7827.
Adds a `docker-args-process-build` trigger to the resource plugin so
limits set via `dokku resource:limit --process-type build APP` are
applied during the build phase. Only `build.limit.*` properties are
read - defaults do not inherit, since builds typically need more memory
than runtime and a leaked tiny default would cause confusing OOM
failures. Reservations are never applied at build time. Allowed flags
are filtered per builder: herokuish gets cpu, memory, memory-swap, and
nvidia-gpu; dockerfile gets memory and memory-swap; pack, nixpacks,
railpack, lambda, and null emit nothing because their underlying CLIs
do not accept docker run resource flags. The dockerfile builder
whitelists the new memory flags and corrects pre-existing typos where
`--ssh` mapped to `--platform` and `--ulimit` mapped to `--tag`.
The bats test for migration idempotency invoked the install binary directly via sudo, which kept hitting fresh missing env vars on each iteration (DOKKU_LIB_ROOT, then PLUGIN_PATH, etc). The dokku launcher script exports a chain of vars that the install path reads via common.MustGetEnv, and mirroring that chain in a bats test is fragile. Migration is pure file IO plus a property marker - perfect for a Go unit test that isolates itself with t.TempDir and t.Setenv. The new test covers parsing comments and blank lines, the marker-based no-op on re-run, and the rule that a manually re-created legacy file is left untouched after the marker is set.
The ps plugin auto-adds --restart=on-failure:10 to the default deploy scope on app create, so docker-options:list --phase deploy is never empty for a freshly-created app. Switch the default-scope assertions from exact-match to contains/not-contains so the auto-injected restart line stops causing false failures. The migration test invoked the install binary directly via sudo, but DOKKU_LIB_ROOT is normally exported by the dokku launcher script and isn't set in that subprocess; pass it (and DOKKU_ROOT) explicitly.
Keda 2.17+ rejects ScaledObjects whose spec.fallback is set unless at least one trigger is not a cpu or memory scaler, so unconditionally emitting fallback broke deploys for apps autoscaled on cpu or memory alone. The chart now skips the fallback block when every configured trigger is cpu or memory and keeps the existing behavior otherwise.
The :list subcommand had `SetInterspersed(false)` which forced flags to come before positional arguments; in practice users invoke it as `docker-options:list <app> --process X --phase Y`. Drop the non-interspersed setting so flags can appear after the app name. Two existing buildpacks/dockerfile tests appended directly to the legacy `DOCKER_OPTIONS_DEPLOY` file - that path is no longer the source of truth, so switch them to `docker-options:add`. The new migration test invoked `dokku plugin:trigger install` which fans out to every plugin and trips on unrelated permission errors; call the docker-options install binary directly instead. Tighten the report and JSON assertions to match the actual output format (`.` becomes a space in the display key, and JSON values may concatenate options).
The docker-local scheduler now sends `SIGTERM` to old containers immediately after a successful deploy via `docker container kill --signal=SIGTERM`, rather than waiting `wait-to-retire` seconds before signaling. This matches Heroku's graceful-shutdown contract and lets applications begin draining in-flight work as soon as proxy traffic switches. The existing `wait-to-retire` grace period and `stop-timeout-seconds` hard-stop continue to apply unchanged as the authoritative cleanup path.
Adds a `--process` flag (repeatable) to docker-options:add/remove/clear/list and the new docker-options:list subcommand for querying a single process+phase pair. Process scoping is supported only for the deploy phase since build runs once per app and run covers ad-hoc commands and cron tasks where no Procfile process type is available. Storage moves from `$DOKKU_ROOT/$APP/DOCKER_OPTIONS_*` files to property lists under `/var/lib/dokku/config/docker-options/$APP/{processType}.{phase}`, with `_default_` as the sentinel for app-wide options. The install trigger migrates pre-existing DOCKER_OPTIONS_* files into property lists once and renames them to `.migrated`; a global marker makes re-runs strictly no-op. The legacy docker-args-{build,deploy,run} bash triggers are reimplemented in Go alongside a new docker-args-process-deploy trigger that surfaces per-process options to the scheduler. The :report command exposes one dynamic flag per configured `process.deploy` pair (e.g. `--docker-options-deploy.web`) and supports `--format json`. There is no `--global` flag; omitting `--process` keeps the historical default behaviour, since `--global` elsewhere in dokku means "across all apps". Closes#2441.
The single-udp network test was calling `docker-options:add $TEST_APP -p 1194:1194` without a phase argument. The bash subcommand silently no-op'd because `-p` is not a recognized phase, so the test happened to pass even though it was not actually setting the option. The go port surfaces the validation error correctly, so the test now needs the proper `deploy` phase to keep working.
Re-enabling vhost (via `proxy:enable`, `domains:enable`, or `domains:add` with no global VHOST) was leaving the app pinned to the random high port and the empty VHOST file from the previous disabled state, so the new catch-all vhost intercepted requests for the app's domain. The `domains-enable` trigger now restores default vhosts via `domains_enable`, and `pre-enable-vhost` clears any stored proxy ports so they can be recomputed to 80/443. Also fixes the netrc lookup in the `git:auth` test to read dokku's netrc file directly.
Renaming `/etc/nginx/sites-enabled/default` to `.dokku-disabled` left a broken symlink that nginx's `include /etc/nginx/sites-enabled/*;` glob still picked up, breaking config validation. Files inside `sites-enabled/` are now removed; the real config in `sites-available/default` is still preserved via rename.
Fresh apt installs now drop a catch-all server block at `/etc/nginx/conf.d/00-default-vhost.conf` that uses `ssl_reject_handshake on` and `return 444` to drop requests with unknown Host headers. Conflicting upstream nginx default vhosts are renamed to `*.dokku-disabled` rather than deleted, preserving any local edits. The new `dokku/install_default_site` debconf flag opts out of the install. Upgrades leave existing nginx config untouched.
The intermediate container count check raced with the existing
zero-downtime retire of the previous container, so docker ps still
saw the renamed-but-pending container until wait-to-retire elapsed.
When scaling a process type to a smaller count, the indices above the
new count had their `CONTAINER.<proctype>.<idx>` state files deleted but
the underlying Docker containers were never registered for retirement,
so they remained running indefinitely. The cleanup block now reads each
orphaned state file and registers its container id with the existing
`scheduler-register-retired` trigger before the file is removed.
The `post-app-rename-setup` trigger redirected `sed` output to the new
app's VHOST file inside a loop over global vhosts, so each iteration
clobbered the previous one and only the last global subdomain was
rewritten. When no global vhost was set, the loop never ran and the
new app's VHOST file was never written, leaving the renamed app with
no domains.
Copy the old app's VHOST file once as a baseline, then apply `sed -i`
in place for each global vhost so substitutions accumulate. Guard
against a missing global VHOST file and skip blank lines.
