Accepted recommendations without a library match now enter a download queue.
A worker downloads the MP3 with yt-dlp into "<music>/AudioManager Imports",
runs a scan, matches the new song, links it to the recommendation and adds it
as a seed. Jobs retry three times and expose queued/downloading/done/failed
state (plus the error) on the recommendation DTO, so clients can show progress
and retry. Downloads can be disabled and the target folder chosen.
Long-form gets resume points: GET/PUT /longform/progress, scoped to longform
folders and to the owner, with automatic completion detection. Longform stays
excluded from moods and radio.
The chart mounts the music hostPath writable for imports and passes the new
DOWNLOAD_* settings; version 0.2.3.
yt-dlp now feeds mood discovery. Queries come from the free-text hint, seed
artists and genres; they run in parallel, are filtered to music-length videos,
matched against library songs by normalized title and artist, scored, and
stored as review candidates. Refresh, accept and reject endpoints are added:
accepting a library match adds a seed immediately, while accepted tracks that
are not owned wait for the future downloader. Decisions survive refreshes and
stale undecided candidates are pruned.
A background pass refreshes stale moods every hour, and the image ships a
pinned yt-dlp binary (YTDLP_PATH, also added to the Nix shell and chart).
Radio now avoids tracks played in the last 7 days when the pool is large
enough, spaces artists and albums across a batch, and weighs skips three times
as heavily as completions. The chart moves to 0.2.2.
The chart repository README was static, so the published version was only
visible in index.yaml. push-chart.sh now regenerates a marked "Newest chart"
section with the version, app version, publish time and source commit, plus a
table of the newest version of every chart in the repository. Plain markdown
only, so the README stays readable on the Gitea front page.
The section is inserted before the first section on the first publish and
replaced in place afterwards.
Helm falls back to the "default" namespace when --namespace is omitted, so a
forgotten flag (or a Kustomize helmCharts entry without namespace) silently
deployed into the wrong place. Render the namespace through one helper that
fails on an empty or implicit "default" namespace, with allowDefaultNamespace
as an explicit opt-in.
scripts/check-chart.sh lints the chart, asserts every namespaced resource
carries the passed namespace, and checks both failure and opt-in paths. CI runs
it instead of the plain template call.
Bump the chart to 0.2.1.
All browsing and mood management now live in the native client. Drop the
embedded HTML page, the WEB_UI setting, the /ui route and the chart env
entry; the server only exposes /api/v1, /health and a small root banner.
Remove the /rest compatibility layer and the playlist model. Moods are now
first-class private collections, and playback is driven by short-lived radio
sessions whose queues are generated by the server.
- Bearer device tokens (hashed) with revocable sessions.
- Library snapshot/delta sync with tombstones and per-folder scoping.
- Mood CRUD, seeds, preview and recommendations.
- Radio batches weighted by seeds, genres, artists, text and playback affinity.
- Playback events (started/completed/skipped) and explicit add-to-mood.
- Stream, download and cover endpoints with query-token support for browsers.
- V5 migration drops playlists and migrates existing moods; scanner writes
tombstones and song update timestamps.
- Built-in UI rebuilt on the native API with a radio player.
- Native API integration tests and a rewritten smoke test (49 checks).
- Chart bumped to 0.2.0.
BREAKING CHANGE: the OpenSubsonic API and static playlists are gone.
Add private server-managed playlist moods with curated seeds and related
library fill. Moods are created or promoted from existing playlists in the
web UI, refresh lazily after their TTL, and are marked stale after a scan.
Symfonium stars pin the active mood: opening a mood restores its context,
and a starred song becomes a permanent seed. Without an active session the
song is pinned to every mood that already contains it.
Includes the V4 migration, custom mood endpoints, management UI, tests and
smoke coverage. Bump the chart to 0.1.4.
Set metadata.namespace from .Release.Namespace on the ConfigMap, Service,
PVC, Deployment and Ingress. This keeps the chart correct when a consumer
does not run Kustomize's namespace transformer, and makes the rendered
manifests explicit about where they belong.
Bump the chart to 0.1.3.
Replace the OCI chart push with a classic Helm repository in the public
TapTap/Charts repository. CI packages the chart, regenerates index.yaml and
pushes both to main, so `helm repo add` and Kustomize helmCharts work.
The publisher authenticates with the AUDIOMANAGER_CHARTS_TOKEN secret and
supports an explicit VERSION for local releases.
Use the product and image name for the chart, release, namespace, secrets,
container and rendered file paths. The application package and local config
paths keep their existing names.
Bump the chart to 0.1.2 so the new chart name and paths are a distinct
release. CI now publishes oci://gitea.tap-tap.win/taptap/charts/audiomanager.
Add a chart-push job that runs after the chart lint on every push and
publishes the packaged chart to oci://gitea.tap-tap.win/taptap/charts.
Main builds use <chart version>-sha.<commit> so parallel commits do not
overwrite each other; v* tags use the tag version. The shared
scripts/push-chart.sh also supports explicit VERSION and local use.
Use the CI-published :latest tag with pullPolicy Always, since the chart
previously defaulted to appVersion 0.1.0, which is not published as an image.
Add ingress annotations with a cert-manager ClusterIssuer so the TLS secret is
issued instead of having to exist beforehand.
Extend release.sh --deploy to create the registry image pull secret from the
local Docker config and pass it to the release, and document both the scripted
and manual deployment paths.
Kaniko unpacks the final stage over /, so tools from the CI image (node in
particular) are gone after the build. The checkout action's post step then
fails with exit code 127 even though the image was pushed successfully.
Use a plain git fetch/checkout in the image job, which has no JavaScript action
and therefore no post step.
A multi-stage kaniko build replaces / with the last stage's root filesystem,
which deletes $HOME/.docker/config.json. Kaniko then resolves credentials at
push time without them, asks the registry for an anonymous token and Gitea
rejects the upload with 401 reqPackageAccess.
Keep the Docker credential store in /kaniko/.docker for the whole image job via
DOCKER_CONFIG and let docker login write there directly. Restore the image job
and document that CI publishes images again.
The Gitea registry rejects kaniko pushes from job containers with
401 reqPackageAccess even though the same credentials work from a normal
machine, so the image job could never go green.
Drop the image job; CI now only runs unit tests, the smoke test and the
Helm chart lint. Add scripts/release.sh, which builds and pushes the
image from a trusted machine and can optionally apply the users secret
and run helm upgrade.
Add ci/Dockerfile (JDK 21, Gradle 8.14.3, ffmpeg, sqlite3, node,
docker CLI and Helm) and build it as taptap/audiomanager-ci. Workflows
now start from that image instead of downloading a JDK and installing
packages on every run.
The registry namespace is the lowercased package owner `taptap`, not
`tap-tap`; update the chart, workflow and documentation accordingly.
The image build uses plain docker build/push, since the job container
starts with no buildx plugin.
Alpine/musl cannot load sqlite-jdbc's bundled native library, so the
container failed at startup. Switch the runtime image to
eclipse-temurin:21-jre-jammy and verify the image end-to-end, including
transcoding, against a read-only root filesystem.
The smoke fixture now uses a 192 kbps source so the maxBitRate check
actually proves transcoding instead of direct play.
- multi-stage Dockerfile with ffmpeg on Alpine JRE 21
- Helm chart for k3s: library ConfigMap, Vault users secret, data PVC,
read-only hostPath media, ingress, probes and security context
- Gitea Actions runs tests, smoke test and chart lint, then builds and
pushes the image on main and version tags
- FfmpegTranscoder with maxBitRate, format fallback to MP3 and timeOffset
- direct play when no transcode is required, preserving HTTP Range support
- transcoded responses are chunked audio/mpeg with process cleanup
- unit tests for policy and real ffmpeg output
- smoke test covers maxBitRate, timeOffset and format fallback
- serve a single-page client at /ui (and redirect / to it)
- login, folder switch, album grid, album songs, search and streaming
- enabled by default, disable with WEB_UI=false
- document the test client and WEB_UI
- getMusicDirectory for folder/album browsing
- server-side play queues with savePlayQueue/getPlayQueue and index variants
- jukebox status and internet radio compatibility responses
- scripts/smoke.sh generates a tagged MP3 library, runs the server and
verifies all endpoints clients use, including Range, download and covers
- getAlbumList2/getAlbumList, getRandomSongs, getGenres, getSongsByGenre,
getStarred2/getStarred, artist/album info and similar/top stubs
- scrobble, star/unstar, setRating, bookmarks and lyrics
- playlist create/get/update/delete with ownership and visibility rules
- persist scanned lyrics via V3 migration
- include last-played timestamps in song and album responses
- library.example.yaml now uses relative ./media paths for local runs
- clearer startup errors for missing roots and config files
- ignore the local config/library.yaml override