- Replace ProgramData.id-based dedup with stable (tvg_id, start_time, end_time) key from Recording.custom_properties.program. ProgramData IDs change on every EPG refresh; the old dedup always passed after refresh, creating duplicates.
- Fix secondary timeslot guard to compare original program times (stored in custom_properties) instead of offset-adjusted Recording times. With DVR pre/post offsets configured, the old guard never matched.
- Add acquire_task_lock/release_task_lock to serialize concurrent evaluate_series_rules calls. Multiple EPG source refreshes firing evaluate_series_rules.delay() simultaneously raced to create duplicate recordings.
- Rewrite ghost client tests to call real ChannelStatus and ClientManager code paths instead of reimplementing detection logic in test helpers
- Rewrite initializing state tests to invoke StreamManager.run() so the real finally block executes under test
- Replace hardcoded state list tests with assertions against ChannelState.PRE_ACTIVE frozenset
- Add missing SOURCE_BITRATE and FFMPEG_BITRATE constants to ChannelMetadataField (referenced by channel_status.py but never defined, causing AttributeError on detailed stats)
- Change ChannelState.PRE_ACTIVE from list to frozenset (immutable,
O(1) membership tests, no risk of silent mutation)
- Add optional client_ids parameter to ClientManager.remove_ghost_clients()
so callers that have already fetched SMEMBERS can pass it through,
eliminating a redundant Redis round-trip
- Pass pre-fetched client_ids from get_basic_channel_info() into
remove_ghost_clients() to remove the duplicate SMEMBERS call
- Add MAX_KEEPALIVE_DURATION = 300s to TSConfig
- Track keepalive_start_time in _stream_data_generator; disconnect after cap is exceeded with a warning log
- Reset timer on real data so independent stalls don't accumulate
- Fix pre-existing test helper missing throttle attributes in NonOwnerWorkerKeepaliveTests
Previously only UnicodeDecodeError/UnicodeEncodeError were caught. Unexpected errors like disk full or permission denied would propagate
and crash the EPG refresh, which would have succeeded before the preprocessing step existed. Now all exceptions are caught and logged, leaving the original file untouched for lxml to handle as before.
lxml 6.0.2 with recover=True silently drops HTML named entities (e.g. é, î) that are not valid XML, causing channel names and program data to lose characters during EPG ingestion.
- Add preprocessing step in fetch_xmltv() that converts HTML named entities to Unicode while preserving XML-predefined entities
- Detect encoding from XML declaration, fall back to UTF-8 for unknown codecs, skip files that fail to decode cleanly
- Guard against entities resolving to XML-special characters and html.unescape partial matching edge cases
- Add 28 unit tests covering entity resolution, encoding detection, and file-level processing
- Write ERROR via ownership check OR state guard fallback when ownership TTL expired during retries
- Add INITIALIZING to cleanup task grace period monitoring
- Validate client SET entries against metadata hashes in orphaned channel cleanup; remove ghosts and clean up when no real clients remain
- Stats page self-heals by removing ghost SET entries on read
- Extract remove_ghost_clients() helper to ClientManager
- Add ChannelState.PRE_ACTIVE constant
- Fix 6 pre-existing NonOwnerWorkerKeepalive test failures
Replace DRF ModelSerializer with .values() + plain dict construction in
EPGGridAPIView, eliminating per-row serializer instantiation, field
object creation, and OrderedDict overhead across ~12k programs.
Pre-compute season/episode at import time instead of on every API request:
- Extract S/E from onscreen episode-num inline in extract_custom_properties()
immediately after storing the raw value, rather than in a separate pass
- Add missing description-based S/E fallback to parse_programs_for_tvg_id(),
bringing it in line with parse_programs_for_source()
Remove runtime extract_season_episode() call from ProgramDataSerializer and
EPGGridAPIView — values are now guaranteed to be in custom_properties after
import, so season/episode resolution is a plain dict lookup at request time.
Also remove unnecessary select_related("epg") JOIN and .count() query from
the grid endpoint.
Replace expanded inline cards with a ProgramDetailModal for viewing full program
details. Add real-time progress bars to currently-airing programs using direct DOM manipulation. Extract and display season/episode info from EPG data with a 3-tier fallback (custom_properties → onscreen_episode → description text).
Backend:
- Add ProgramDetailSerializer for rich EPG data (categories, credits, ratings, images, external IDs) served on the detail endpoint
- Add infer_is_live() for status badges (live, new, premiere, finale)
- Add shared S/E extraction utilities in apps/epg/utils.py
- Optimize querysets with select_related("epg")
Frontend:
- ProgramDetailModal with conditional field rendering and image fallback chain
- Progress bars via scaleX transform updated every second
- Parallel data fetching with Promise.all for channels + programs
- Browser tab visibility recovery for progress bars
- Now-line with triangle marker and sub-millisecond precision
- Card layout: title, season/episode + subtitle, time + badges
- Shared formatSeasonEpisode() and external URL helpers
P1 (Critical):
- Replace undefined @swagger_auto_schema with @extend_schema
- Add custom_properties validation with size limits
- Add merge semantics in serializer update() to prevent data loss
P2 (Important):
- Add debouncing (800ms) for drag operations in NavOrderForm
- Add error handling with notifications for visibility toggle
- Use store getters in Sidebar instead of direct property access
- Cache hiddenNav before render loop
P3 (Nice-to-have):
- Add canHide: false to settings nav item
- Fix admin check to use >= instead of ==
- Fix ESLint unused variable warnings
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Allow users to update their own profile via PATCH /api/accounts/users/me/
- Supports partial updates to custom_properties (including navOrder)
- Add comprehensive test coverage for the endpoint
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Stream.get_stream() was never called in the preview path of
generate_stream_url(), so the channel_stream and stream_profile
Redis keys were never written. This caused the stream_id to be
missing from channel metadata, which silently broke two things:
- Stats page showed no stream name or logo
- stream_stats were never saved to the database
Fix: replace the manual profile-selection loop in the preview path
with a direct call to Stream.get_stream(), matching the channel path
exactly. This also fixes a pre-existing non-atomic slot reservation
(read-then-check vs INCR-first) that could over-allocate connections
on concurrent previews.
Regression introduced in 49b7b9e2.
