Code review of the Artist/ownership refactor surfaced three gaps:
- SyncUser updated User.Name on every login but never the linked
Artist row, so renames never reached track credits - defeating the
point of centralizing names in Artist.
- YouTubeTrackProvisioner looked up artists by ExternalId only when a
browseId was known, so it could never reuse the ExternalId-less
rows the migration backfilled from legacy Artist strings, creating
duplicate Artist rows for the same person. It now adopts the
backfilled row (assigning it the browseId) when one matches by
normalized name.
- AddArtistsAndTrackOwnership deduped legacy YouTube artist strings
case-sensitively before hashing them into a case-insensitive unique
index, and never guarded against local tracks with no UserHasTrack
row - either would abort the migration on real data. Backfill now
dedupes via DISTINCT ON over the normalized name and drops orphaned
local tracks before the ownership check constraint is added.
Track mixed catalog fields (Title, Artist credit, Duration) with
picture-processing and audio-transcode pipeline state as ten flat
columns plus two computed [NotMapped] flags, making the entity read
like a media-pipeline worksheet.
Introduce TrackPictures and TrackAudio as EF owned types mapped onto
the same existing columns (explicit column names, no schema change -
verified via `dotnet ef migrations has-pending-model-changes`).
Track now reads as Title/Artist/Duration/Pictures/Audio instead of
fourteen loose properties.
Drop the now-redundant Track.Artist string column now that every
track (local and YouTube) links its credited artists through
TrackArtist. TrackApplicationResponse formats the display string by
joining TrackArtists ordered by position, and every read query that
projects a track now includes TrackArtists/Artist so the join isn't
empty at response time.
Track get-or-create for YouTube videos was duplicated between the
old GetYouTubeTrackStream query handler (which also wrote to the
database) and AddYouTubeTrackToPlayList, with frontend-supplied
metadata (title/artist/duration/thumbnail) trusted at add-to-playlist
time instead of resolving it server-side.
Extract YouTubeTrackProvisioner as the single get-or-create path,
resolving structured artist refs (id + name) from YouTube Music so
multi-artist songs map to distinct Artist rows. Rename
GetYouTubeTrackStream to ResolveYouTubeTrackStream (it writes, so it
is a command, not a query) and delegate to the provisioner.
AddYouTubeTrackToPlayList slims down to (UserId, PlayListId, VideoId)
since all metadata now comes from the provisioner/YouTube API.
Track.Artist was a plain string snapshot with no identity, and
UserHasTrack doubled as both "in my library" and "I own this",
conflating collection membership with administrative ownership.
Introduce Artist/TrackArtist to give artist credits their own
identity (needed so YouTube tracks with multiple credited artists,
and future renames, can be modeled), and Track.OwnerUserId to
represent ownership explicitly (local tracks only, enforced by a
check constraint). CreateTrack now provisions the uploader's Artist
row and links it via TrackArtist; DeleteTrack authorizes against
OwnerUserId instead of any UserHasTrack row.
Artist, ExternalId, OriginalPictureName and OriginalAudioName were
non-nullable with '' sentinels, which forced the unique (Source,
ExternalId) index to rely on a filtered "<> ''" clause. Make them
truly nullable and drop the empty-string sentinels, backfilling
existing rows to NULL. Also fixes a formatting regression
(`track != null&&`) introduced alongside the nullability change.
Playing a YouTube song now creates a lightweight track row (without queueing a download) and writes listening history; saving to a playlist queues the download via the new DownloadRequested flag.
GetListeningHistory used AsEnumerable() before DistinctBy/Take, which
forced EF Core to materialize every history row for the user just to
keep the last 15 distinct tracks. Now groups by track and takes the
15 most recently played track ids in SQL, then fetches only those.
TrackApplicationResponse.FromEntity counted track.ListeningHistories,
but none of the queries that built a response ever Included it, so
the collection was always empty and listensCount was always 0. Each
query now projects the count as a correlated subquery before mapping
to the response.
Adds GET /playlists/{id}/cover (mirrors the track cover endpoint) so
uploaded playlist pictures can actually be served. Also drops the
NotEmpty rule on playlist description, which rejected the blank
description the frontend treats as optional; CreatePlayList now falls
back to the domain's default text instead.
- Remove the no-op Worker BackgroundService and its registration; the process
is kept alive and does its work through the MassTransit bus and the upload
intent jobs (their own hosted services)
- Fix the gateway middleware guard to reject on string.IsNullOrEmpty(prefix)
instead of 'is null', which no longer matched TicketValidator returning
string.Empty for invalid tickets (invalid tickets skipped the 401 and an
empty object key could slip through)
Build: 0 warnings, 0 errors.
Tighten the stream-ticket authorization in the gateway middleware:
- Enforce a path-segment boundary on the prefix check instead of a bare
StartsWith, so a ticket for 'users/12' no longer grants 'users/1234/...'
(exact-match still allowed for single-object grants)
- Reject object keys containing '..'/'.' segments before proxying, closing a
path-traversal bypass that SeaweedFS would otherwise resolve outside the
authorized prefix
Also simplify TicketValidator.ValidateTicketAsync from ErrorOr<string> to
string? (null = invalid) since it has a single failure mode; drop the now
unused ErrorOr package reference from the gateway.
Verified: build clean; the authorization predicate passes a table of cases
covering both the escalation and traversal vulnerabilities.
Build: 0 warnings, 0 errors.
- Document every endpoint's responses (Produces<T>/status codes/validation)
so the generated OpenAPI/Scalar docs carry response schemas and status codes
instead of empty responses
- Cache recently-synced user ids (IMemoryCache, 15 min) in the JWT
TokenValidated handler so user provisioning no longer hits the DB on every
authenticated request
- Add ProblemDetails + UseExceptionHandler for consistent RFC7807 errors
- Add ToCreatedResult ErrorOr helper and use it in the create endpoints
- Normalize the playlists route to lowercase (/playlists)
Verified: booted the Api against RabbitMQ + Postgres in Docker and inspected
/openapi/v1.json — operations now expose response schemas/status codes and the
route is /playlists.
Build: 0 warnings, 0 errors.
Single docker-compose stack for local development (Postgres, Zitadel,
RabbitMQ, SeaweedFS + bucket init, Jaeger) plus an optional overlay to run
the .NET apps in Docker. up.ps1 orchestrates the full bootstrap: stack up,
stream-ticket keys, EF Core migrations and Zitadel provisioning.
- Zitadel is provisioned automatically (service-account PAT -> project +
OIDC app with JWT access token); the generated ClientId is wired into
Musify.Api user-secrets and .env. No manual console steps.
- ZITADEL_EXTERNAL_DOMAIN=host.docker.internal so token validation works
both with local apps and full-docker.
- App Dockerfiles copy AppSettings*.json to lowercase so config loads on
case-sensitive Linux; Worker image ships ffmpeg.
- Scripts split by responsibility (common/stream-keys/migrate/provision);
apps overlay de-duped via YAML anchors.
- Remove the superseded standalone seaweedfs compose; point docs/development
at the unified stack.
Register AuditableEntityInterceptor as a singleton and attach it through the
AddDbContext options callback instead of newing it up inside
Database.OnConfiguring. The interceptor is stateless, so a single shared
instance is reused across contexts and its wiring lives in the DI layer.
The design-time factory keeps working (interceptor is runtime-only, irrelevant
for migrations).
Verified: build + ef migrations list (design-time context) OK; the Worker host
builds and validates the full DI graph (reaches endpoint configuration) with no
captive-dependency or interceptor error.
Build: 0 warnings, 0 errors.
- Register IAmazonS3 as a singleton instead of scoped: the AWS SDK client is
thread-safe and manages its own connection pool, so one instance per DI scope
(per request / per job iteration) was wasteful and risked socket exhaustion
- AudioTranscoderService: start reading stdout/stderr before writing stdin to
remove the latent pipe-buffer deadlock shape
- Dispose the resized-picture MemoryStream in ResizePictureActivity
- Background jobs use PeriodicTimer instead of while + Task.Delay
Verified: Worker boots against RabbitMQ + Postgres in Docker with no captive-
dependency error (dev scope validation) and both jobs start.
Build: 0 warnings, 0 errors.
Adds the PlayList picture-processing lifecycle saga (correlated by PlayListId)
mirroring the track one, and generalizes the failure plumbing:
- Rename the track fault translator to ProcessingSlipFaultConsumer and use a
generic Workflow.SubjectId variable + ProcessKind; it now emits typed
Track{Picture,Audio}/PlayListPicture ProcessingFailed events for all slips
- PlayListProcessingStateMachine (+ EF-persisted PlayListProcessingState,
migration): starts on CreatePlayListResourcesEvent, finalizes on
PlayListPictureProcessed, and on failure marks the playlist Failed and
removes the orphaned final picture (PlayListProcessingFailedConsumer +
MarkPlayListAsFailedActivity)
- UpdatePlayListPictureActivity publishes PlayListPictureProcessed on success
- Picture slip carries SubjectId + PlayListPicture kind + fault subscription
Verified end-to-end in Docker (RabbitMQ + migrated Postgres): both saga
endpoints, the fault translator and cleanup consumers configure, and an
addressing cross-check confirms all 21 activity execute queues the builders
address exist (caught + fixed a kebab mismatch on mark-play-list-as-failed).
Build: 0 warnings, 0 errors.
On definitive failure of the picture or audio slip, the saga now reaches a
Failed state and recovers end-to-end:
- Picture/audio track slips carry TrackId + ProcessKind variables and a
subscription to a fault translator (TrackProcessingSlipFaultConsumer) that
turns RoutingSlipFaulted into a typed Track{Picture,Audio}ProcessingFailed
- The saga stores the final object keys from the start event and, on failure,
transitions to Failed and publishes TrackProcessingFailed
- TrackProcessingFailedConsumer runs a cleanup slip that removes the orphaned
final picture/audio copies and marks the track LifeCycleStatus=Failed
(MarkTrackAsFailedActivity) — the track stays as a visible, recoverable record
- Migration for the new saga columns
Verified: Worker boots against RabbitMQ + Postgres in Docker; the new
endpoints (fault translator, failed consumer, mark-track-as-failed) configure
and the bus starts clean.
Build: 0 warnings, 0 errors.
Introduce a MassTransit saga state machine that correlates by TrackId and
tracks the two parallel processing slips (picture + audio):
- TrackProcessingStateMachine + EF-persisted TrackProcessingState
(pessimistic concurrency, EntityFrameworkRepository), migration added
- Starts on CreateTrackResourcesEvent; UpdateTrack{Picture,Audio}Activity now
publish TrackPictureProcessed / TrackAudioProcessed on success; when both
arrive the saga finalizes (Ready). Failure transitions to a Failed state
(cleanup wired in the next phase)
- Add LifeCycleStatus.Failed and the four outcome events
Verified: booted the Worker against RabbitMQ + migrated Postgres in Docker;
the saga endpoint 'track-processing-state' is configured and the bus starts
clean.
Build: 0 warnings, 0 errors.
Replace ~240 lines of hand-written per-endpoint wiring with MassTransit's
ConfigureEndpoints + KebabCaseEndpointNameFormatter, and apply retry + EF
outbox globally through AddConfigureEndpointsCallback. Adding a consumer or
activity is now a single AddConsumer/AddActivity line.
- Normalize the 4 mismatched activity endpoint names to kebab-case so the
routing-slip addressing keeps resolving (update-play-list-picture,
delete/mark/publish play-list variants); cleanup consumer queue is now
routing-slip-clean-up
- Drop the now-unused QueueName consts from event consumers and trim
EndpointHelper / delete the manual endpoint-config partial
Verified end-to-end: booted the Worker against RabbitMQ in Docker and
confirmed all 38 endpoints are configured with the exact names the builders
address (19 execute queues + compensations + consumers + routing-slip-clean-up).
Build: 0 warnings, 0 errors.
Fix (recoverability bug):
- TransferFilesToBucketActivity returned Completed() with no log, so its
compensation could never delete the transferred files (orphaned objects on
fault). IStorageService.TransferFilesAsync now returns the real uploaded
keys; the activity logs them so Compensate can remove them. Also drops the
Path.Combine key recomputation (produced backslash keys on Windows).
Consistency / DRY:
- Normalize all courier activities to internal + folder-based namespaces
(several were public and in the flat .Activities namespace)
- Extract the duplicated exponential retry into UseStandardRetry
- Unify endpoint-name logic into a single EndpointHelper (remove the parallel
MassTransitEndpointNames helper)
Build: 0 warnings, 0 errors.
- Add Musify.Application.AddApplication(services, config) as the layer's DI
entry point: registers Mediator, UploadIntentValidator and all app configs
- Collapse the 5 near-identical AddXConfiguration methods into a single
generic AddValidatedOptions<T>(config, section)
- Move UploadIntentValidator registration out of Infrastructure into the
Application layer (fixes layering leak); Infra config binding now uses the
generic helper
- Api calls AddApplication; Worker uses AddValidatedOptions for its configs
Build: 0 warnings, 0 errors.
One file per use case under the feature folder (Tracks/PlayLists/Users),
dropping the Commands/Queries/Handlers sub-folders and their namespaces.
Each file now holds the record and its handler in namespace
Musify.Application.<Feature>; external usings updated accordingly.
Build: 0 warnings, 0 errors.
Ardalis Result<T> carried incoherent success sub-statuses and a clunky
generic converter; ErrorOr models success-or-errors cleanly.
- Application: all commands/queries/handlers now return ErrorOr<T>
(non-generic Result -> ErrorOr<Success>); errors via Error.NotFound/
Conflict/Validation/Unauthorized/Failure; success via value or new Success()
- Drop ResultExtensions.As<>(): error propagation is now just return .Errors
- Rewrite HTTP mapping (ErrorOrHttpExtensions) off Error.Type; success maps
to 200 (value) or 204 (Success); Created still built in the endpoints
- Endpoints/gateway middleware: !IsSuccess -> IsError
- Swap Ardalis.Result package for ErrorOr 2.1.1 (remove unused ref from Worker)
Build: 0 warnings, 0 errors.
Run autoskills in backend/: detects .NET/C#/ASP.NET Core/Minimal API and
installs 13 skills for universal (.agents) and claude-code (.claude) agents,
with backend/skills-lock.json. Adds ASP.NET Core, Minimal API + OpenAPI,
file-upload and containerization skills on top of the prior C#/.NET set.
Removes the stale root skills from the old layout.
Admin