Cap credential-scoped group counters at profile.max_streams (not group.max_streams),
skip credential counters when fingerprint resolution fails, and always swap profile
counters on update_stream_profile. Restore per-profile-only selection for VOD/live
switches so a second stream can use a different login without invalid HTTP errors.
Co-authored-by: Cursor <cursoragent@cursor.com>
Drop auto-fingerprint migration and restore per-profile selection for live/VOD.
Enforce shared limits on reserve using login-scoped group counters, and add
Server Groups UI for manual account assignment per maintainer feedback (#1137).
Co-authored-by: Cursor <cursoragent@cursor.com>
Group M3U/XC accounts and profiles that share the same provider login into auto-assigned ServerGroups keyed by credential fingerprint. Enforce combined Redis limits for live TV and VOD via apps/m3u/connection_pool.py.
- Per-profile fingerprinting (XC transforms and STD stream URLs)
- VOD profile selection tries alternates when default credential pool is full (fixes live then VOD failure)
- Stats UI shows provider login from active stream URL
- Tests: apps/m3u/tests/test_connection_pool.py (11 tests, all passing)
Co-authored-by: Cursor <cursoragent@cursor.com>
gevent registers a pthread_atfork handler that never yields. Any
subprocess.Popen/run call in a uWSGI greenlet hangs indefinitely at
fork() before the child process even starts.
- dispatcharr/gevent_patch.py: new; monkey.patch_all() + psycogreen,
loaded via uWSGI import= in all four ini files
- live_proxy/input/manager.py: posix_spawn + _SpawnedProcess; removed
dead forwarder code
- live_proxy/input/http_streamer.py: O_NONBLOCK on relay pipe + EAGAIN
retry (blocking write to a full pipe stalls the gevent hub)
- live_proxy/utils.py: new posix_spawn_proc() helper with O_NONBLOCK
stdin, shared by both output managers
- live_proxy/output/fmp4/manager.py, output/profile/manager.py:
posix_spawn_proc(); _write_all() treats EAGAIN (None) as cooperative
select.select wait instead of fatal error
- core/views.py (stream_view): posix_spawn; fixed pre-existing bug
where return StreamingHttpResponse(...) was indented inside
stream_generator, making the success path always return None
- connect/handlers/script.py: _posix_run() with posix_spawn +
cooperative select reads + non-blocking waitpid; fixes deadlock when
a script integration fires on a uWSGI-context event (client_connect
fires in the live proxy, not Celery)
- Enable gevent cooperative multitasking in all uWSGI worker configs
(gevent-early-monkey-patch + import dispatcharr.gevent_patch)
- Rewrite WebSocket group sends to bypass asyncio in gevent workers:
_gevent_ws_send() replicates the channels_redis 4.x wire format
directly via synchronous Redis so send_websocket_update() and
_send_async() no longer fail silently after epoll is patched out
- Fix PostgreSQL connection exhaustion: CONN_MAX_AGE=0 + explicit
close_old_connections() in stream manager and cleanup watchdog loops
- Fix stream proxy race: register client before the connect-wait loop
so the cleanup watchdog never sees zero clients on a live channel
- Channel list/logo/profile queryset optimisations: conditional DISTINCT,
EXISTS semi-joins for filter-options, channel_count annotation to
eliminate N+1 in LogoSerializer, prefetched memberships in
ChannelProfileSerializer
- JsonResponse for channel ID list and summary endpoints
Plugins live in /data/plugins/<slug>/plugin.py, outside INSTALLED_APPS,
so celery's autodiscover_tasks() never imports them. Any plugin using
module-level @shared_task for cron-scheduled work therefore has its
task unregistered with the worker after every worker restart, until
something else (a connect/utils.trigger_event call, etc.) lazily
imports the plugin module via PluginManager.discover_plugins.
In the meantime, beat fires the periodic task on time and the worker
rejects it:
ERROR celery.worker.consumer.consumer Received unregistered task of
type 'telegram_alerts.send_daily_report'.
PeriodicTask.last_run_at advances anyway, so the failure is silent at
the default INFO log level.
Fix: hook celery's worker_ready signal in dispatcharr/celery.py to
call PluginManager.discover_plugins(sync_db=False) on every worker
boot. Importing the plugin modules runs their @shared_task decorators,
registering the tasks before beat starts firing.
Exception handling is intentionally broad: if one plugin's plugin.py
has an import error, the worker must still come up — discovery failure
should not be a fatal error for the whole worker process.
Tests in tests/test_celery_plugin_discovery.py verify:
- the handler calls discover_plugins(sync_db=False)
- the handler swallows PluginManager.get() exceptions
- the handler swallows discover_plugins() exceptions
- the handler is actually connected to the worker_ready signal
Confirmed the test fails against current dev with the exact
'cannot import name discover_plugins_on_worker_ready' that motivates
this change, and passes with the fix.
- Implemented HDHR output profile URL support for specific transcode profiles.
- Introduced a default output profile setting in Stream Settings.
- Updated API views to handle channel profiles and output profiles.
- Migrated preferred region and auto-import settings to system settings.
- Enhanced frontend forms to include output profile selection and descriptions.
trigger_event in apps/connect/utils.py iterates pm.list_plugins() and
on disabled plugins emits a debug log that accesses dict items as if
they were attributes:
logger.debug(f"Skipping disabled plugin id={plugin.key} name={plugin.name}")
Python evaluates f-string arguments eagerly even when the logger
discards the message at INFO level, so this raises AttributeError on
the first disabled plugin encountered. The exception bubbles out of
trigger_event with no try/except in the loop, aborting dispatch for
every plugin sorted after the disabled one.
Effect for users: any plugin subscribed to events via `events: [...]`
on an action silently never receives events whenever any
alphabetically-earlier plugin is disabled. Manual button actions still
work, masking the failure as a plugin bug rather than a dispatch one.
Fix: replace the two attribute accesses with dict access. Add a
regression test under apps/connect/tests/ that mocks PluginManager to
return [disabled, enabled-with-events], asserts the enabled plugin's
action is dispatched, and a sanity check that non-matching events
aren't dispatched. Verified the test fails against the original code
with the expected AttributeError and passes with the fix.
