Merge pull request #747 from Nezreka/fix/soulseek-album-poll-stall

Fix/soulseek album poll stall
This commit is contained in:
BoulderBadgeDad 2026-05-29 22:52:58 -07:00 committed by GitHub
commit 0e1f07433a
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 333 additions and 4 deletions

View file

@ -1601,7 +1601,15 @@ class SoulseekClient(DownloadSourcePlugin):
result['fallback'] = False
completed = self._poll_album_bundle_downloads(transfer_keys, _emit)
if not completed:
result['error'] = 'Soulseek album download failed or timed out'
# The selected folder yielded ZERO usable files — every transfer
# failed / aborted / stalled (a dead or unwilling peer). Don't hard-
# fail the batch: fall back to the per-track flow, which searches ALL
# sources per track and can pull each from a live peer. We reuse that
# proven multi-source robustness instead of looping candidate folders
# here. (Per-track only fires for a genuinely-missing album anyway.)
result['error'] = ('Soulseek album folder produced no usable files '
'(peer failed/aborted/stalled) — falling back to per-track')
result['fallback'] = True
return result
_emit('staging', release=getattr(picked, 'album_title', folder_path) if picked else folder_path)
@ -1700,6 +1708,22 @@ class SoulseekClient(DownloadSourcePlugin):
# Seconds an "slskd Completed but locally unresolved" key
# has to stay stuck before we give up on it.
_unresolved_grace = 45.0
# Bundle-level stall guard. The #715 grace above only covers
# "slskd says Completed but the file isn't on disk yet". It does NOT
# cover a transfer the peer stalls on — stuck InProgress / Queued, or
# dropped by slskd entirely — which is never failed, never completed,
# and never marked unresolved, so it blocks BOTH the all-terminal
# finish check AND the grace exit, and the poll spun to the full
# ``get_poll_timeout()`` deadline (the Slipknot hang). If NOTHING
# progresses — no transfer completes/fails and no pending transfer's
# byte count moves — for this long, the folder has stalled: mark the
# stuck transfers failed so the bundle resolves with whatever
# completed (the per-track matcher then handles the missing tracks).
# Conservative on purpose: only trips when EVERYTHING is frozen, so a
# slow-but-progressing or still-queued-then-starting transfer is safe.
_stall_grace = 180.0
_last_progress_marker = None
_last_progress_at = time.monotonic()
while time.monotonic() < deadline:
try:
downloads = run_async(self.get_all_downloads())
@ -1724,7 +1748,14 @@ class SoulseekClient(DownloadSourcePlugin):
os.path.basename((key[1] or '').replace('\\', '/')),
))
state = (getattr(dl, 'state', '') or '') if dl else ''
if any(token in state for token in ('Errored', 'Failed', 'Rejected', 'TimedOut')):
# NOTE: check failure tokens BEFORE the 'Completed' branch — slskd
# reports terminal failures as "Completed, <reason>" (e.g.
# "Completed, Aborted" / "Completed, Cancelled" when a peer accepts
# then drops every transfer at 0 bytes). Those contain "Completed",
# so without catching the failure reason first they'd be misread as
# "completed but file missing" (the #715 download_path path).
if any(token in state for token in
('Errored', 'Failed', 'Rejected', 'TimedOut', 'Aborted', 'Cancelled')):
failed_states[key] = state or 'Failed'
logger.warning(
"[Soulseek album] Transfer failed from selected folder: %s (%s)",
@ -1755,6 +1786,40 @@ class SoulseekClient(DownloadSourcePlugin):
count=len(completed_paths),
failed=len(failed_states),
)
# Bundle-level stall detection (see ``_stall_grace`` above). Advance
# the progress marker on ANY forward motion — a transfer reaching a
# terminal state, or a still-pending transfer downloading more bytes.
# If the marker is frozen for ``_stall_grace``, the peer has stalled;
# mark the stuck transfers failed so the finish/all-failed checks
# below resolve the bundle instead of spinning to the deadline.
now = time.monotonic()
stall_pending = [
k for k in transfer_keys
if k not in completed_paths and k not in failed_states
]
pending_bytes = 0
for k in stall_pending:
dl = by_key.get(k) or by_key.get((
k[0], os.path.basename((k[1] or '').replace('\\', '/')),
))
pending_bytes += (getattr(dl, 'transferred', 0) or 0) if dl else 0
marker = (len(completed_paths) + len(failed_states), pending_bytes)
if marker != _last_progress_marker:
_last_progress_marker = marker
_last_progress_at = now
elif stall_pending and (now - _last_progress_at) >= _stall_grace:
logger.warning(
"[Soulseek album] No progress for %.0fs — peer stalled on %d "
"transfer(s) (stuck / queued / dropped). Marking them failed and "
"resolving with what completed; missing tracks fall back to "
"per-track. Stalled: %s",
_stall_grace, len(stall_pending),
[transfer_keys[k].filename for k in stall_pending[:5]],
)
for k in stall_pending:
failed_states[k] = 'Stalled'
if completed_paths and len(completed_paths) + len(failed_states) == len(transfer_keys):
logger.warning(
"[Soulseek album] Selected folder finished with %d completed and %d failed transfer(s)",

View file

@ -0,0 +1,100 @@
"""Regression: a Soulseek album folder that yields ZERO usable files must fall
back to the per-track flow rather than hard-failing the whole batch.
The Slipknot case the preflight-selected peer aborts/stalls every transfer (all
tracks reported "Completed, Aborted" at 0 bytes) makes
``_poll_album_bundle_downloads`` return ``[]``. ``download_album_to_staging`` used
to return that empty result with ``fallback=False``, so the dispatcher
(``core/downloads/album_bundle_dispatch.try_dispatch``) hit its ``mark_failed``
branch and the batch died with nothing tried elsewhere.
The fix flips that branch to ``fallback=True`` so the existing, proven per-track
flow takes over and re-searches every missing track across ALL sources/peers
reusing that robustness instead of looping candidate folders inside the bundle.
The downstream "fallback=True -> per-track" routing is covered by
``tests/test_album_bundle_dispatch.py``
(``test_dispatch_fallback_failure_returns_false_for_per_track_flow``); these tests
prove the empty-folder branch actually sets the flag, and that a healthy folder
does NOT fall back.
"""
from __future__ import annotations
from types import SimpleNamespace
from unittest.mock import patch
from core.soulseek_client import SoulseekClient
class _Stub:
"""Stand-in exposing only what download_album_to_staging touches on the
preflight-reuse path (preferred_source + preferred_tracks skip search/browse)."""
def __init__(self, poll_result):
self._poll_result = poll_result
def is_configured(self):
return True
def filter_results_by_quality_preference(self, tracks):
return tracks
def download(self, username, filename, size):
return f"dl-{filename}" # truthy id; run_async patched to identity
def _poll_album_bundle_downloads(self, transfer_keys, emit):
return self._poll_result
def _track(name):
return SimpleNamespace(username="deadpeer", filename=name, size=100)
def _run(poll_result):
stub = _Stub(poll_result)
tracks = [_track("01.flac"), _track("02.flac")]
with patch("core.soulseek_client.run_async", lambda x: x):
result = SoulseekClient.download_album_to_staging(
stub,
album_name="All Hope Is Gone",
artist_name="Slipknot",
staging_dir="/tmp/staging-does-not-matter",
preferred_source={
"username": "deadpeer",
"folder_path": "music/Slipknot/All Hope Is Gone",
},
preferred_tracks=tracks,
)
return result, tracks
def test_empty_folder_falls_back_to_per_track():
# Poll returns [] — every transfer aborted / stalled (dead peer).
result, _ = _run([])
assert result['success'] is False
assert result['fallback'] is True # <-- the fix: hand off to per-track
assert result['files'] == []
assert 'per-track' in (result['error'] or '')
def test_healthy_folder_does_not_fall_back():
# Positive control: the poll yields staged files and the copy succeeds, so we
# must NOT flip fallback — this proves the empty-branch isn't blanket-returning
# True. Patch the atomic copy to echo the completed paths through.
from pathlib import Path
completed = [Path("/staged/01.flac"), Path("/staged/02.flac")]
with patch("core.soulseek_client.copy_audio_files_atomically", lambda files, dest: list(files)):
stub = _Stub(completed)
with patch("core.soulseek_client.run_async", lambda x: x):
result = SoulseekClient.download_album_to_staging(
stub,
album_name="All Hope Is Gone",
artist_name="Slipknot",
staging_dir="/tmp/staging-does-not-matter",
preferred_source={"username": "goodpeer", "folder_path": "music/Slipknot/AHIG"},
preferred_tracks=[_track("01.flac"), _track("02.flac")],
)
assert result['success'] is True
assert result['fallback'] is False
assert result['partial'] is False
assert result['files'] == completed

View file

@ -0,0 +1,162 @@
"""Regression for the Soulseek album-bundle poll hanging when the peer stalls.
`_poll_album_bundle_downloads` waits for every transfer in the selected folder
to reach a terminal state (completed or failed). A transfer the peer stalls on
stuck InProgress/Queued, or dropped by slskd is never failed, never completed,
and never marked "completed-but-unresolved", so it used to block both the
all-terminal finish check AND the #715 grace exit, and the poll spun to the full
~6h timeout (the Slipknot hang).
The fix adds a bundle-level stall guard: if NOTHING progresses (no transfer
reaches terminal AND no pending transfer's byte count moves) for `_stall_grace`
seconds, the stuck transfers are marked failed so the bundle resolves with what
completed. These tests drive the real poll with a fake clock + scripted slskd
states.
"""
from __future__ import annotations
import os
from pathlib import Path
from types import SimpleNamespace
from unittest.mock import patch
from core.soulseek_client import SoulseekClient
class _Clock:
def __init__(self):
self.now = 0.0
def monotonic(self):
return self.now
def sleep(self, seconds):
self.now += seconds
def _dl(username, filename, state, *, size=100, transferred=100):
return SimpleNamespace(
username=username, filename=filename, state=state,
size=size, transferred=transferred,
)
class _StubClient:
"""Minimal stand-in for SoulseekClient with just what the poll touches."""
def __init__(self, states, resolvable):
self._states = states # list of download-lists, one per poll; last repeats
self._call = 0
self._resolvable = set(resolvable)
def get_all_downloads(self):
i = min(self._call, len(self._states) - 1)
self._call += 1
return self._states[i]
def _resolve_downloaded_album_file(self, filename):
base = os.path.basename((filename or "").replace("\\", "/"))
if base in self._resolvable or filename in self._resolvable:
return Path(f"/staged/{base}")
return None
def _run_poll(stub, transfer_keys, *, timeout=7200.0, interval=2.0):
clock = _Clock()
emits = []
with patch("core.soulseek_client.time", clock), \
patch("core.soulseek_client.run_async", lambda x: x), \
patch("core.soulseek_client.get_poll_timeout", lambda: timeout), \
patch("core.soulseek_client.get_poll_interval", lambda: interval):
result = SoulseekClient._poll_album_bundle_downloads(
stub, transfer_keys, lambda phase, **kw: emits.append((phase, kw))
)
return result, clock, emits
def _keys(*names, user="peer"):
"""Build {(user, name): TrackResult-ish} preserving order."""
return {(user, n): SimpleNamespace(filename=n) for n in names}
def test_stalled_peer_gives_up_and_returns_completed_subset():
tk = _keys("01.flac", "02.flac", "03.flac")
# 01 completes (resolvable); 02/03 stuck InProgress with FROZEN byte counts.
frozen = [
_dl("peer", "01.flac", "Completed", transferred=100, size=100),
_dl("peer", "02.flac", "InProgress", transferred=50, size=100),
_dl("peer", "03.flac", "InProgress", transferred=30, size=100),
]
stub = _StubClient([frozen], resolvable={"01.flac"})
result, clock, _ = _run_poll(stub, tk, timeout=7200.0, interval=2.0)
# Resolved with the one completed track instead of hanging to the deadline.
assert result == [Path("/staged/01.flac")]
# Gave up around the stall window (~180s), nowhere near the 7200s timeout.
assert clock.now < 600.0
def test_progressing_bundle_is_not_falsely_stalled():
tk = _keys("01.flac", "02.flac")
# 02 keeps downloading more bytes each poll, then completes — must NOT trip
# the stall guard even though it takes a while.
states = [
[_dl("peer", "01.flac", "Completed"), _dl("peer", "02.flac", "InProgress", transferred=10, size=100)],
[_dl("peer", "01.flac", "Completed"), _dl("peer", "02.flac", "InProgress", transferred=40, size=100)],
[_dl("peer", "01.flac", "Completed"), _dl("peer", "02.flac", "InProgress", transferred=80, size=100)],
[_dl("peer", "01.flac", "Completed"), _dl("peer", "02.flac", "Completed", transferred=100, size=100)],
]
stub = _StubClient(states, resolvable={"01.flac", "02.flac"})
result, _clock, _ = _run_poll(stub, tk, timeout=7200.0, interval=2.0)
assert set(result) == {Path("/staged/01.flac"), Path("/staged/02.flac")}
def test_all_transfers_stalled_returns_empty():
tk = _keys("01.flac", "02.flac")
frozen = [
_dl("peer", "01.flac", "InProgress", transferred=10, size=100),
_dl("peer", "02.flac", "Queued", transferred=0, size=100),
]
stub = _StubClient([frozen], resolvable=set())
result, clock, _ = _run_poll(stub, tk, timeout=7200.0, interval=2.0)
assert result == [] # nothing completed → empty (caller falls back)
assert clock.now < 600.0 # didn't spin to the deadline
def test_dropped_transfers_also_stall_out():
"""slskd dropping the transfers entirely (dl=None) must also trip the guard,
not hang there's no byte progress and nothing terminal."""
tk = _keys("01.flac", "02.flac")
stub = _StubClient([[]], resolvable=set()) # get_all_downloads returns nothing
result, clock, _ = _run_poll(stub, tk, timeout=7200.0, interval=2.0)
assert result == []
assert clock.now < 600.0
def test_completed_aborted_classified_as_failed_not_unresolved():
"""slskd reports a peer-side abort as 'Completed, Aborted' at 0 bytes. Because
that string contains 'Completed', it was misread as 'completed but file
missing' (#715 path). It must be classified as FAILED — so an all-aborted
folder resolves immediately, not after the unresolved/stall grace."""
tk = _keys("01.flac", "02.flac")
aborted = [
_dl("peer", "01.flac", "Completed, Aborted", transferred=0, size=3_600_000),
_dl("peer", "02.flac", "Completed, Aborted", transferred=0, size=24_700_000),
]
stub = _StubClient([aborted], resolvable=set())
result, clock, _ = _run_poll(stub, tk, timeout=7200.0, interval=2.0)
assert result == [] # all failed → empty (caller falls back)
assert clock.now < 30.0 # resolved on the first poll, no 45s/180s wait
def test_clean_finish_unaffected():
tk = _keys("01.flac", "02.flac")
done = [_dl("peer", "01.flac", "Completed"), _dl("peer", "02.flac", "Succeeded")]
stub = _StubClient([done], resolvable={"01.flac", "02.flac"})
result, clock, _ = _run_poll(stub, tk)
assert set(result) == {Path("/staged/01.flac"), Path("/staged/02.flac")}
assert clock.now < 10.0 # resolves on the first couple polls

View file

@ -1418,8 +1418,10 @@ def validate_and_heal_batch_states():
queue = batch_data.get('queue', [])
phase = batch_data.get('phase', 'unknown')
# AUTO-CLEANUP: Remove completed batches after 5 minutes to prevent stale state
if phase in ['complete', 'error', 'cancelled']:
# AUTO-CLEANUP: Remove terminal batches after 5 minutes to prevent stale state.
# 'failed' (e.g. an album-bundle hard failure) was missing here, so a failed
# batch lingered in the UI forever ("No tracks loaded") and never cleared.
if phase in ['complete', 'error', 'cancelled', 'failed']:
# Check if batch has a completion timestamp
completion_time = batch_data.get('completion_time')
if not completion_time: