soulsync/core/download_engine/worker.py
Broque Thomas 2c19d7d1f2 Per-source lock sharding on the engine
Per JohnBaumb: the single state_lock serialized progress callbacks
across every source. Pre-refactor each client owned its own download
lock, so Deezer / YouTube / Tidal workers never blocked each other.
Multi-source concurrent downloads under the unified lock fought for
the same RLock on every progress update.

Replaced the engine-wide state_lock with per-source RLocks. Each
source gets its own lock, lazily created via _source_lock() on first
use (meta-lock guards the create-race). All record mutations
(add/update/update_unless_state/remove/get/iter) take only that
source's lock — Deezer progress updates no longer block Tidal writes.

Cancelled-preserve semantics still hold because cancel + worker
terminal write target the same source, so they share that source's
lock. New test pins lock independence: holding source-A's lock from
one thread does not block a write on source-B from another.
2026-05-05 11:56:09 -07:00

315 lines
13 KiB
Python
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

"""BackgroundDownloadWorker — engine-owned thread spawning + state
lifecycle for downloads.
Today every streaming download client (YouTube, Tidal, Qobuz, HiFi,
Deezer, SoundCloud) hand-rolls the same thread-spawn pattern:
```python
async def download(self, ...):
download_id = str(uuid.uuid4())
with self._download_lock:
self.active_downloads[download_id] = {...initial state...}
threading.Thread(
target=self._download_thread_worker,
args=(download_id, target_id, display_name, ...),
daemon=True,
).start()
return download_id
def _download_thread_worker(self, download_id, target_id, display_name, ...):
with self._download_semaphore:
# rate-limit sleep
# update state to 'InProgress, Downloading'
file_path = self._download_sync(...) # the source-specific atomic op
# update state to 'Completed, Succeeded' / 'Errored'
```
That pattern is duplicated 6+ times across the codebase (~70 LOC
each, ~490 total). The worker class lifts it into the engine — each
plugin only has to provide the atomic op (``impl_callable``) and
declare its rate-limit policy. Adding a new download source becomes
a much smaller patch.
Phase C1 scope: introduce the worker. No client migrated yet — the
worker just exists for C2C7 to migrate sources one at a time, each
under a passing pinning test.
"""
from __future__ import annotations
import threading
import time
import uuid
from typing import Any, Callable, Dict, Optional
from utils.logging_config import get_logger
logger = get_logger("download_engine.worker")
# Type aliases for clarity. ``ImplCallable`` is the per-plugin
# atomic download operation — synchronous, returns a file path on
# success or raises (or returns None) on failure.
ImplCallable = Callable[[str, Any, str], Optional[str]]
class BackgroundDownloadWorker:
"""Engine-owned thread spawner for per-source downloads.
State-machine semantics (preserved verbatim from the legacy
per-client workers so consumers reading these fields keep
working):
- ``Initializing`` — set on dispatch, before the thread starts.
- ``InProgress, Downloading`` — set when the worker thread
acquires the semaphore and is about to call the impl.
- ``Completed, Succeeded`` — set when impl returns a non-None
file path. ``progress=100.0`` and ``file_path=<the path>``
also written.
- ``Errored`` — set when impl returns None OR raises. The
record is left in place so downstream consumers can inspect
what failed.
Per-source serialization: each source gets a ``threading.Semaphore``
(default size 1, configurable per-source via ``set_concurrency``).
Same shape the existing clients use today (each source defines
its own semaphore). Engine owning them centrally lets a future
Phase E rate-limiter swap the semaphore for a smarter pool.
Per-source delay-between-downloads: default 0 seconds (most
sources don't need it). YouTube currently uses 3s, Qobuz uses
1s — the legacy values get configured in via ``set_delay``
when the source registers.
"""
def __init__(self, engine: Any) -> None:
self._engine = engine
# Per-source semaphores + delay state. The first dispatch
# for a source auto-creates a semaphore with concurrency=1
# if the source hasn't been configured explicitly.
self._semaphores: Dict[str, threading.Semaphore] = {}
self._delays: Dict[str, float] = {}
self._last_download_at: Dict[str, float] = {}
self._config_lock = threading.Lock()
# ------------------------------------------------------------------
# Per-source rate-limit configuration
# ------------------------------------------------------------------
def set_concurrency(self, source_name: str, max_concurrent: int) -> None:
"""Set the max number of concurrent downloads for a source.
Default is 1 (serial). Most sources will keep the default —
the streaming APIs all rate-limit at the API gateway level
anyway, parallel downloads just trade rate-limit errors for
thread overhead."""
with self._config_lock:
self._semaphores[source_name] = threading.Semaphore(max_concurrent)
def set_delay(self, source_name: str, seconds: float) -> None:
"""Set a minimum delay between successive downloads from the
same source. YouTube uses 3s today (avoid yt-dlp 429s),
Qobuz uses 1s. Other sources use 0 (no delay)."""
with self._config_lock:
self._delays[source_name] = float(seconds)
def _get_semaphore(self, source_name: str) -> threading.Semaphore:
with self._config_lock:
sem = self._semaphores.get(source_name)
if sem is None:
sem = threading.Semaphore(1)
self._semaphores[source_name] = sem
return sem
def _get_delay(self, source_name: str) -> float:
with self._config_lock:
return self._delays.get(source_name, 0.0)
# ------------------------------------------------------------------
# Dispatch — public API
# ------------------------------------------------------------------
def dispatch(
self,
source_name: str,
target_id: Any,
display_name: str,
original_filename: str,
impl_callable: ImplCallable,
extra_record_fields: Optional[Dict[str, Any]] = None,
username_override: Optional[str] = None,
thread_name: Optional[str] = None,
) -> str:
"""Kick off a background download.
Args:
source_name: Canonical source name (e.g. 'youtube',
'tidal'). Used as the engine state key + the
username slot in the record (unless overridden).
target_id: Source-specific identifier (track_id, video_id,
permalink_url, album_foreign_id, etc.). Passed
verbatim to ``impl_callable``.
display_name: Human-readable label for logs / UI.
original_filename: The encoded filename the orchestrator
received (e.g. ``'12345||Song Title'``). Stored in
the record's ``filename`` slot for context-key lookups.
impl_callable: Synchronous function that performs the
actual download. Signature:
``impl_callable(download_id, target_id, display_name) -> Optional[str]``.
Returns the final file path on success or None /
raises on failure.
extra_record_fields: Per-source extras to merge into the
initial record (e.g. ``{'video_id': '...', 'url':
'...', 'title': '...'}`` for YouTube). Used to
preserve source-specific slots that downstream
consumers + status APIs read.
username_override: Use this instead of ``source_name``
in the record's ``username`` slot. Required for
Deezer (legacy ``'deezer_dl'``) — every other source
uses the canonical name.
thread_name: Optional thread name for diagnostics. Deezer
uses ``'deezer-dl-<track_id>'`` — Phase A pinning
tests catch any drift in this convention.
Returns:
download_id (UUID4 string). The orchestrator polls via
``engine.get_download_status(download_id)`` for progress.
"""
download_id = str(uuid.uuid4())
record: Dict[str, Any] = {
'id': download_id,
'filename': original_filename,
'username': username_override or source_name,
'state': 'Initializing',
'progress': 0.0,
'size': 0,
'transferred': 0,
'speed': 0,
'time_remaining': None,
'file_path': None,
}
if extra_record_fields:
record.update(extra_record_fields)
self._engine.add_record(source_name, download_id, record)
thread = threading.Thread(
target=self._worker_loop,
args=(source_name, download_id, target_id, display_name, impl_callable),
daemon=True,
name=thread_name,
)
thread.start()
return download_id
# ------------------------------------------------------------------
# Worker thread — the lifted boilerplate
# ------------------------------------------------------------------
def _worker_loop(
self,
source_name: str,
download_id: str,
target_id: Any,
display_name: str,
impl_callable: ImplCallable,
) -> None:
"""Runs on the spawned daemon thread. Handles semaphore
acquisition, rate-limit sleep, state lifecycle, exception
capture. The plugin-specific work happens entirely inside
``impl_callable``."""
try:
with self._get_semaphore(source_name):
# Rate-limit delay against the LAST download from
# this source (not just this worker — semaphore
# ensures serial access while delay is configured).
delay = self._get_delay(source_name)
if delay > 0:
last_at = self._last_download_at.get(source_name, 0.0)
elapsed = time.time() - last_at
if last_at > 0 and elapsed < delay:
wait_time = delay - elapsed
logger.info(
"Rate-limit delay for %s: waiting %.1fs before next download",
source_name, wait_time,
)
time.sleep(wait_time)
self._engine.update_record(source_name, download_id, {
'state': 'InProgress, Downloading',
})
try:
file_path = impl_callable(download_id, target_id, display_name)
except Exception as exc:
logger.error(
"%s download %s failed (impl raised): %s",
source_name, download_id, exc,
)
self._mark_terminal(
source_name, download_id,
success=False, error=str(exc),
)
return
self._last_download_at[source_name] = time.time()
if file_path:
# Atomic write — preserve Cancelled if user cancelled
# between impl returning and this write. Same guard
# _mark_terminal uses; Cin flagged both split sites.
self._engine.update_record_unless_state(
source_name, download_id,
{
'state': 'Completed, Succeeded',
'progress': 100.0,
'file_path': file_path,
},
skip_if_state_in=('Cancelled',),
)
logger.info(
"%s download %s completed: %s",
source_name, download_id, file_path,
)
else:
self._mark_terminal(source_name, download_id, success=False)
logger.error(
"%s download %s failed (impl returned None)",
source_name, download_id,
)
except Exception as exc:
# Defensive — semaphore / sleep shouldn't blow up the
# thread, but if they do the record needs SOME terminal
# state or it sits at 'Initializing' forever.
logger.exception(
"%s worker_loop crashed for download %s: %s",
source_name, download_id, exc,
)
self._mark_terminal(
source_name, download_id,
success=False, error=f'worker crash: {exc}',
)
def _mark_terminal(self, source_name: str, download_id: str,
success: bool, error: Optional[str] = None) -> None:
"""Write a terminal state, but DON'T clobber an explicit
'Cancelled' state set by the user via cancel_download.
Mirrors the legacy per-client guard
(``if state != 'Cancelled': state = 'Errored'``) every
client used to hand-roll inside its thread worker.
Uses ``update_record_unless_state`` so the check + write are
atomic under the engine's per-source lock. Cin caught a race
where a cancel landing between the read-snapshot + write
could overwrite Cancelled back to Errored / Completed.
"""
patch: Dict[str, Any] = {
'state': 'Completed, Succeeded' if success else 'Errored',
}
if error is not None:
patch['error'] = error
self._engine.update_record_unless_state(
source_name, download_id, patch, skip_if_state_in=('Cancelled',),
)