baqueue 0.1.0__tar.gz → 1.0.2__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (38) hide show
  1. {baqueue-0.1.0/baqueue.egg-info → baqueue-1.0.2}/PKG-INFO +18 -1
  2. {baqueue-0.1.0 → baqueue-1.0.2}/README.md +16 -0
  3. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/__init__.py +1 -1
  4. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/cli.py +31 -0
  5. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/config.py +8 -0
  6. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/drivers/base.py +44 -0
  7. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/drivers/redis_driver.py +199 -24
  8. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/pruner.py +18 -9
  9. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/queue.py +2 -0
  10. {baqueue-0.1.0 → baqueue-1.0.2/baqueue.egg-info}/PKG-INFO +18 -1
  11. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue.egg-info/requires.txt +1 -0
  12. {baqueue-0.1.0 → baqueue-1.0.2}/pyproject.toml +1 -0
  13. {baqueue-0.1.0 → baqueue-1.0.2}/LICENSE +0 -0
  14. {baqueue-0.1.0 → baqueue-1.0.2}/MANIFEST.in +0 -0
  15. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/balancer.py +0 -0
  16. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/batch.py +0 -0
  17. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/dashboard/__init__.py +0 -0
  18. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/dashboard/api.py +0 -0
  19. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/dashboard/server.py +0 -0
  20. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/dashboard/static/app.js +0 -0
  21. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/dashboard/static/index.html +0 -0
  22. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/dashboard/static/style.css +0 -0
  23. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/drivers/__init__.py +0 -0
  24. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/drivers/memory_driver.py +0 -0
  25. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/drivers/postgres_driver.py +0 -0
  26. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/drivers/sqlite_driver.py +0 -0
  27. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/events.py +0 -0
  28. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/job.py +0 -0
  29. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/retry.py +0 -0
  30. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/scheduler.py +0 -0
  31. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/serializer.py +0 -0
  32. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/supervisor.py +0 -0
  33. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue/worker.py +0 -0
  34. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue.egg-info/SOURCES.txt +0 -0
  35. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue.egg-info/dependency_links.txt +0 -0
  36. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue.egg-info/entry_points.txt +0 -0
  37. {baqueue-0.1.0 → baqueue-1.0.2}/baqueue.egg-info/top_level.txt +0 -0
  38. {baqueue-0.1.0 → baqueue-1.0.2}/setup.cfg +0 -0
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: baqueue
3
- Version: 0.1.0
3
+ Version: 1.0.2
4
4
  Summary: A powerful Python queue management package inspired by Laravel Horizon
5
5
  Author: Basalam, BaQueue Contributors
6
6
  License: MIT
@@ -45,6 +45,7 @@ Provides-Extra: dev
45
45
  Requires-Dist: baqueue[all]; extra == "dev"
46
46
  Requires-Dist: pytest>=8.0; extra == "dev"
47
47
  Requires-Dist: pytest-asyncio>=0.23; extra == "dev"
48
+ Requires-Dist: fakeredis>=2.21; extra == "dev"
48
49
  Requires-Dist: build>=1.0; extra == "dev"
49
50
  Requires-Dist: twine>=5.0; extra == "dev"
50
51
  Dynamic: license-file
@@ -274,6 +275,21 @@ await Queue.prune(status="completed", hours=24)
274
275
  await Queue.prune(tag="batch:newsletter")
275
276
  ```
276
277
 
278
+ #### Redis index health
279
+
280
+ The Redis driver keeps secondary indexes (sorted sets) so the dashboard can list and
281
+ count jobs by queue/status efficiently. All deletes go through an index-consistent path
282
+ that removes the job hash *and* every index entry in one atomic step, so the indexes stay
283
+ bounded. If entries are ever orphaned out-of-band (e.g. job hashes deleted directly via
284
+ `redis-cli`), pruning reaps them automatically, and you can force a full repair:
285
+
286
+ ```bash
287
+ baqueue reconcile-indexes -d redis --driver-url redis://localhost:6379/0
288
+ ```
289
+
290
+ Set `reconcile_on_connect=True` to run that repair once on every startup (off by default
291
+ to keep connect fast on large datasets).
292
+
277
293
  ### Retry Failed Jobs
278
294
 
279
295
  Bulk-retry failed jobs from the CLI, from Python, or from the dashboard.
@@ -508,6 +524,7 @@ baqueue schedule Start the job scheduler
508
524
  baqueue dashboard Launch the monitoring dashboard
509
525
  baqueue prune Prune old jobs
510
526
  baqueue retry-failed Retry all failed jobs (filter by queue/tag/age)
527
+ baqueue reconcile-indexes Repair Redis secondary indexes (drop stale entries)
511
528
  baqueue status Show queue status
512
529
  baqueue test Run the test suite
513
530
  ```
@@ -223,6 +223,21 @@ await Queue.prune(status="completed", hours=24)
223
223
  await Queue.prune(tag="batch:newsletter")
224
224
  ```
225
225
 
226
+ #### Redis index health
227
+
228
+ The Redis driver keeps secondary indexes (sorted sets) so the dashboard can list and
229
+ count jobs by queue/status efficiently. All deletes go through an index-consistent path
230
+ that removes the job hash *and* every index entry in one atomic step, so the indexes stay
231
+ bounded. If entries are ever orphaned out-of-band (e.g. job hashes deleted directly via
232
+ `redis-cli`), pruning reaps them automatically, and you can force a full repair:
233
+
234
+ ```bash
235
+ baqueue reconcile-indexes -d redis --driver-url redis://localhost:6379/0
236
+ ```
237
+
238
+ Set `reconcile_on_connect=True` to run that repair once on every startup (off by default
239
+ to keep connect fast on large datasets).
240
+
226
241
  ### Retry Failed Jobs
227
242
 
228
243
  Bulk-retry failed jobs from the CLI, from Python, or from the dashboard.
@@ -457,6 +472,7 @@ baqueue schedule Start the job scheduler
457
472
  baqueue dashboard Launch the monitoring dashboard
458
473
  baqueue prune Prune old jobs
459
474
  baqueue retry-failed Retry all failed jobs (filter by queue/tag/age)
475
+ baqueue reconcile-indexes Repair Redis secondary indexes (drop stale entries)
460
476
  baqueue status Show queue status
461
477
  baqueue test Run the test suite
462
478
  ```
@@ -7,7 +7,7 @@ from baqueue.batch import Batch
7
7
  from baqueue.events import EventBus
8
8
  from baqueue.retry import BackoffStrategy
9
9
 
10
- __version__ = "0.1.0"
10
+ __version__ = "1.0.2"
11
11
 
12
12
  __all__ = [
13
13
  "BaQueueConfig",
@@ -365,6 +365,37 @@ async def _run_retry_failed(
365
365
  await Queue.disconnect()
366
366
 
367
367
 
368
+ @cli.command(name="reconcile-indexes")
369
+ @click.option("--batch", default=500, type=int, help="Index entries scanned per batch.")
370
+ @click.option("--driver", "-d", default="redis", help="Driver name (sqlite, memory, redis, postgres).")
371
+ @click.option("--driver-url", default=None, help="Driver connection URL.")
372
+ @click.pass_context
373
+ def reconcile_indexes(
374
+ ctx: click.Context,
375
+ batch: int,
376
+ driver: str,
377
+ driver_url: str | None,
378
+ ) -> None:
379
+ """Repair secondary indexes: remove entries pointing at jobs that no longer exist.
380
+
381
+ Only the Redis driver maintains secondary indexes; this is a no-op elsewhere."""
382
+ _validate_driver(driver)
383
+ config: BaQueueConfig = ctx.obj["config"]
384
+ config.driver = DriverConfig(name=driver, url=driver_url or "")
385
+
386
+ removed = _run_async(_run_reconcile_indexes, config, batch)
387
+ click.echo(f"Removed {removed or 0} stale index entr{'y' if removed == 1 else 'ies'}.")
388
+
389
+
390
+ async def _run_reconcile_indexes(config: BaQueueConfig, batch: int) -> int:
391
+ Queue.configure(config)
392
+ await Queue.connect()
393
+ try:
394
+ return await Queue.get_driver().reconcile_indexes(batch=batch)
395
+ finally:
396
+ await Queue.disconnect()
397
+
398
+
368
399
  @cli.command()
369
400
  @click.option("--driver", "-d", default="sqlite", help="Driver name (sqlite, memory, redis, postgres).")
370
401
  @click.option("--driver-url", default=None, help="Driver connection URL.")
@@ -59,6 +59,14 @@ class BaQueueConfig(BaseModel):
59
59
  prune_completed_seconds: int = 5 # delete completed jobs ~5s after completion
60
60
  prune_other_seconds: int = 86400 # 1 day — applies to failed + cancelled
61
61
  prune_metrics_seconds: int = 604800 # 7 days
62
+ # Per-call cap for index-consistent bulk deletes; the pruner loops to drain.
63
+ prune_batch_size: int = 1000
64
+
65
+ # ── Secondary-index reconciliation (Redis) ─────────────────
66
+ # When True, connect() runs a one-shot reconcile pass that removes index
67
+ # entries pointing at jobs that no longer exist. Off by default — run on
68
+ # demand via `baqueue reconcile-indexes` to keep startup fast.
69
+ reconcile_on_connect: bool = False
62
70
 
63
71
  # ── Legacy hour-based overrides (kept for back-compat) ──────
64
72
  # When > 0, these take precedence over the seconds fields above for the
@@ -10,6 +10,10 @@ from baqueue.serializer import JobPayload
10
10
 
11
11
  logger = logging.getLogger("baqueue.driver")
12
12
 
13
+ # Default per-call cap for batched bulk-delete / prune operations. Keeps a single
14
+ # call from blocking the backend on very large datasets; callers loop to drain.
15
+ DEFAULT_PRUNE_BATCH = 1000
16
+
13
17
 
14
18
  class BaseDriver(ABC):
15
19
  """Every BaQueue driver must implement this interface."""
@@ -18,6 +22,11 @@ class BaseDriver(ABC):
18
22
  # an emergency cleanup and one retry. Wired from BaQueueConfig in queue.py.
19
23
  auto_cleanup_on_disk_full: bool = True
20
24
 
25
+ # When True, connect() runs a one-shot reconcile_indexes() pass to heal any
26
+ # secondary-index drift accumulated while offline. Off by default so connect
27
+ # stays fast on large datasets. Wired from BaQueueConfig in queue.py.
28
+ reconcile_on_connect: bool = False
29
+
21
30
  # Re-entrancy guard so emergency_cleanup() doesn't recurse if its own
22
31
  # prune calls also hit disk-full.
23
32
  _in_emergency_cleanup: bool = False
@@ -193,6 +202,41 @@ class BaseDriver(ABC):
193
202
  """Delete matching jobs. Returns count of pruned jobs."""
194
203
  ...
195
204
 
205
+ async def bulk_delete_jobs(self, job_ids: list[str], *, limit: int | None = None) -> int:
206
+ """Delete an explicit list of jobs, keeping any secondary indexes consistent.
207
+
208
+ Default implementation deletes one id at a time via ``delete``; drivers with
209
+ secondary indexes (Redis) override this with an atomic, batched version that
210
+ also reaps orphaned index entries. Returns the count of ids processed."""
211
+ if limit is not None:
212
+ job_ids = job_ids[:limit]
213
+ for job_id in job_ids:
214
+ await self.delete(job_id)
215
+ return len(job_ids)
216
+
217
+ async def prune_terminal_jobs(
218
+ self,
219
+ queue: str | None = None,
220
+ status: str | None = None,
221
+ *,
222
+ older_than: float | None = None,
223
+ limit: int = DEFAULT_PRUNE_BATCH,
224
+ ) -> int:
225
+ """Index-consistent bulk delete of terminal jobs, capped at ``limit`` per call.
226
+
227
+ Default implementation delegates to ``prune``; the Redis driver overrides it to
228
+ use its status index as the work source, reap orphaned index entries, and bound
229
+ the per-call cost. Callers loop until a pass returns fewer than ``limit``."""
230
+ return await self.prune(status=status, queue=queue, older_than_seconds=older_than)
231
+
232
+ async def reconcile_indexes(self, batch: int = 500) -> int:
233
+ """Repair secondary indexes by removing entries whose job no longer exists.
234
+
235
+ No-op for drivers without secondary indexes (memory/sqlite/postgres). The Redis
236
+ driver overrides this to walk its index ZSETs and ZREM orphaned ids. Returns the
237
+ number of stale index entries removed."""
238
+ return 0
239
+
196
240
  @abstractmethod
197
241
  async def flush(self, queue: str | None = None) -> None:
198
242
  """Remove all jobs (optionally for a specific queue)."""
@@ -6,11 +6,16 @@ import json
6
6
  import logging
7
7
  from typing import Any
8
8
 
9
- from baqueue.drivers.base import BaseDriver
9
+ from baqueue.drivers.base import DEFAULT_PRUNE_BATCH, BaseDriver
10
10
  from baqueue.serializer import JobPayload, _now_ts
11
11
 
12
12
  logger = logging.getLogger("baqueue.redis")
13
13
 
14
+ # Every status a job hash can carry. Used when reaping orphaned index entries:
15
+ # the job hash is gone, so we can't read its status — we ZREM from every global
16
+ # status index to be sure the stale id is cleared.
17
+ _ALL_STATUSES = ("pending", "processing", "completed", "failed", "cancelled")
18
+
14
19
 
15
20
  class RedisDriver(BaseDriver):
16
21
  """Redis-backed driver using sorted sets for indexed pagination.
@@ -88,6 +93,10 @@ class RedisDriver(BaseDriver):
88
93
  self._redis = aioredis.from_url(self._url, decode_responses=True, **self._kwargs)
89
94
  await self._redis.ping()
90
95
  await self._backfill_indexes_if_needed()
96
+ if self.reconcile_on_connect:
97
+ removed = await self.reconcile_indexes()
98
+ if removed:
99
+ logger.info("reconcile_on_connect removed %d stale index entr(ies)", removed)
91
100
 
92
101
  async def disconnect(self) -> None:
93
102
  if self._redis:
@@ -97,7 +106,11 @@ class RedisDriver(BaseDriver):
97
106
  async def _backfill_indexes_if_needed(self) -> None:
98
107
  """One-time rebuild of secondary ZSETs for upgrades from a version
99
108
  that didn't maintain them. Safe to call on every connect — exits fast
100
- when the global index is non-empty."""
109
+ when the global index is non-empty.
110
+
111
+ This is *add-only*: it inserts index entries for existing job hashes. It
112
+ cannot remove drift (index entries whose hash is gone) — that is the job
113
+ of reconcile_indexes(). Together they fully heal the indexes."""
101
114
  if await self._redis.exists(self._idx_all()):
102
115
  return
103
116
  cursor: Any = "0"
@@ -518,20 +531,41 @@ class RedisDriver(BaseDriver):
518
531
 
519
532
  # ── Pruning ─────────────────────────────────────────────────
520
533
 
521
- async def prune(
522
- self,
523
- status: str | None = None,
524
- tag: str | None = None,
525
- older_than_seconds: float | None = None,
526
- queue: str | None = None,
527
- ) -> int:
528
- if not (status or tag or older_than_seconds or queue):
529
- return 0
534
+ def _index_remove_orphan(self, pipe: Any, job_id: str, queue: str | None, status: str | None) -> None:
535
+ """ZREM a stale id whose job hash is gone. We can't read the job's real
536
+ queue/status, so we clear every index family we can infer from the call:
537
+ always jobs:all + every global status index, plus the queue-scoped families
538
+ when the caller knows the queue/status it was iterating."""
539
+ pipe.zrem(self._idx_all(), job_id)
540
+ for st in _ALL_STATUSES:
541
+ pipe.zrem(self._idx_status(st), job_id)
542
+ if queue:
543
+ pipe.zrem(self._idx_queue(queue), job_id)
544
+ for st in _ALL_STATUSES:
545
+ pipe.zrem(self._idx_queue_status(queue, st), job_id)
530
546
 
531
- index = self._index_key(queue, status)
532
- candidate_ids: list[str] = await self._redis.zrange(index, 0, -1)
547
+ async def _prune_index_batch(
548
+ self,
549
+ index: str,
550
+ queue: str | None,
551
+ status: str | None,
552
+ tag: str | None,
553
+ older_than_seconds: float | None,
554
+ offset: int,
555
+ limit: int,
556
+ ) -> tuple[int, int, int]:
557
+ """Process one window ``[offset, offset+limit)`` of an index in a single
558
+ atomic pass.
559
+
560
+ Live jobs matching the filters are fully deleted (hash + all four index
561
+ families). Orphaned ids (hash already gone) are reaped from the indexes so
562
+ they can never accumulate. Non-matching live jobs are left in place. Returns
563
+ ``(removed, scanned, skipped)``: removed = deleted + reaped, scanned = window
564
+ size actually read, skipped = live jobs left in place (so the caller can step
565
+ its offset past them)."""
566
+ candidate_ids: list[str] = await self._redis.zrange(index, offset, offset + limit - 1)
533
567
  if not candidate_ids:
534
- return 0
568
+ return 0, 0, 0
535
569
 
536
570
  pipe = self._redis.pipeline()
537
571
  for jid in candidate_ids:
@@ -540,29 +574,170 @@ class RedisDriver(BaseDriver):
540
574
 
541
575
  now = _now_ts()
542
576
  to_delete: list[JobPayload] = []
543
- for raw in raws:
577
+ orphans: list[str] = []
578
+ skipped = 0
579
+ for jid, raw in zip(candidate_ids, raws):
544
580
  if not raw:
581
+ orphans.append(jid)
545
582
  continue
546
583
  job = JobPayload.from_json(raw)
547
584
  if tag and tag not in job.tags:
585
+ skipped += 1
548
586
  continue
549
587
  if older_than_seconds and (now - job.updated_at) < older_than_seconds:
588
+ skipped += 1
550
589
  continue
551
590
  to_delete.append(job)
552
591
 
553
- if not to_delete:
592
+ if to_delete or orphans:
593
+ async def _do():
594
+ pipe = self._redis.pipeline()
595
+ for job in to_delete:
596
+ pipe.lrem(self._key("queue", job.queue), 0, job.id)
597
+ pipe.zrem(self._key("delayed"), job.id)
598
+ pipe.unlink(self._key("job", job.id))
599
+ self._index_remove(pipe, job.id, job.queue, job.status)
600
+ for jid in orphans:
601
+ self._index_remove_orphan(pipe, jid, queue, status)
602
+ await pipe.execute()
603
+ await self._with_disk_full_recovery(_do)
604
+
605
+ return len(to_delete) + len(orphans), len(candidate_ids), skipped
606
+
607
+ async def _drain_index(
608
+ self,
609
+ index: str,
610
+ queue: str | None,
611
+ status: str | None,
612
+ tag: str | None,
613
+ older_than_seconds: float | None,
614
+ batch: int,
615
+ ) -> int:
616
+ """Page through an index in ``batch``-sized windows, deleting matches and
617
+ reaping orphans, until the whole index has been scanned.
618
+
619
+ Each Redis round-trip handles at most ``batch`` ids, so a huge (possibly
620
+ orphan-laden) index never blocks the server on one giant zrange + delete —
621
+ while every entry is still examined. Entries a filter skips stay in the index,
622
+ so the offset is advanced past them; that is what keeps matches deeper than
623
+ the first window from being missed (re-reading ``zrange(0, batch)`` forever
624
+ would stop early)."""
625
+ batch = max(1, batch)
626
+ offset = 0
627
+ total = 0
628
+ while True:
629
+ removed, scanned, skipped = await self._prune_index_batch(
630
+ index, queue, status, tag, older_than_seconds, offset, batch,
631
+ )
632
+ total += removed
633
+ offset += skipped # kept entries remain; step past them next round
634
+ if scanned < batch:
635
+ break
636
+ return total
637
+
638
+ async def prune(
639
+ self,
640
+ status: str | None = None,
641
+ tag: str | None = None,
642
+ older_than_seconds: float | None = None,
643
+ queue: str | None = None,
644
+ ) -> int:
645
+ if not (status or tag or older_than_seconds or queue):
554
646
  return 0
647
+ index = self._index_key(queue, status)
648
+ return await self._drain_index(
649
+ index, queue, status, tag, older_than_seconds, DEFAULT_PRUNE_BATCH,
650
+ )
651
+
652
+ async def prune_terminal_jobs(
653
+ self,
654
+ queue: str | None = None,
655
+ status: str | None = None,
656
+ *,
657
+ older_than: float | None = None,
658
+ limit: int = DEFAULT_PRUNE_BATCH,
659
+ ) -> int:
660
+ """Index-consistent bulk delete from a status index, draining fully in
661
+ ``limit``-sized batches (each Redis round-trip handles at most ``limit`` ids).
662
+
663
+ Uses the secondary index itself as the work source — no SCAN of every job
664
+ hash — and reaps orphaned index entries in the same pass."""
665
+ index = self._index_key(queue, status)
666
+ return await self._drain_index(index, queue, status, None, older_than, limit)
667
+
668
+ async def bulk_delete_jobs(self, job_ids: list[str], *, limit: int | None = None) -> int:
669
+ """Delete an explicit list of jobs atomically, keeping all four index
670
+ families consistent. Live jobs are removed precisely (real queue/status from
671
+ the hash); ids whose hash is already gone are reaped from jobs:all and every
672
+ global status index (per-queue orphans are caught by reconcile_indexes)."""
673
+ if limit is not None:
674
+ job_ids = job_ids[:limit]
675
+ if not job_ids:
676
+ return 0
677
+
678
+ pipe = self._redis.pipeline()
679
+ for jid in job_ids:
680
+ pipe.hget(self._key("job", jid), "data")
681
+ raws = await pipe.execute()
555
682
 
556
683
  async def _do():
557
684
  pipe = self._redis.pipeline()
558
- for job in to_delete:
559
- pipe.lrem(self._key("queue", job.queue), 0, job.id)
560
- pipe.zrem(self._key("delayed"), job.id)
561
- pipe.delete(self._key("job", job.id))
562
- self._index_remove(pipe, job.id, job.queue, job.status)
685
+ for jid, raw in zip(job_ids, raws):
686
+ if raw:
687
+ job = JobPayload.from_json(raw)
688
+ pipe.lrem(self._key("queue", job.queue), 0, jid)
689
+ pipe.zrem(self._key("delayed"), jid)
690
+ pipe.unlink(self._key("job", jid))
691
+ self._index_remove(pipe, jid, job.queue, job.status)
692
+ else:
693
+ self._index_remove_orphan(pipe, jid, None, None)
563
694
  await pipe.execute()
564
695
  await self._with_disk_full_recovery(_do)
565
- return len(to_delete)
696
+ return len(job_ids)
697
+
698
+ async def reconcile_indexes(self, batch: int = 500) -> int:
699
+ """Walk every secondary-index ZSET and ZREM ids whose job hash is gone.
700
+
701
+ Self-healing repair for index drift (e.g. job hashes deleted out-of-band).
702
+ Index keys are discovered by SCAN (every ``baqueue:jobs:*`` key — jobs:all,
703
+ jobs:status:*, jobs:queue:* and jobs:queue:*:status:*) so the repair reaches
704
+ families for queues no longer in the queues set, and never wastes a round-trip
705
+ on an index combination that does not exist. Each index is then walked with
706
+ ZSCAN — never loading a huge set at once — checking hash existence in pipelined
707
+ batches. Returns the number of stale entries removed."""
708
+ # Job hashes are baqueue:job:* (singular); the index ZSETs are baqueue:jobs:*.
709
+ index_keys: list[str] = []
710
+ cursor: Any = "0"
711
+ pattern = self._key("jobs", "*")
712
+ while True:
713
+ cursor, keys = await self._redis.scan(cursor=cursor, match=pattern, count=batch)
714
+ index_keys.extend(keys)
715
+ if cursor == "0" or cursor == 0:
716
+ break
717
+
718
+ removed = 0
719
+ for index in index_keys:
720
+ zcursor: Any = 0
721
+ while True:
722
+ zcursor, members = await self._redis.zscan(index, cursor=zcursor, count=batch)
723
+ ids = [m[0] if isinstance(m, (tuple, list)) else m for m in members]
724
+ if ids:
725
+ pipe = self._redis.pipeline()
726
+ for jid in ids:
727
+ pipe.exists(self._key("job", jid))
728
+ exists_flags = await pipe.execute()
729
+ stale = [jid for jid, ok in zip(ids, exists_flags) if not ok]
730
+ if stale:
731
+ async def _do(index=index, stale=stale):
732
+ pipe = self._redis.pipeline()
733
+ for jid in stale:
734
+ pipe.zrem(index, jid)
735
+ await pipe.execute()
736
+ await self._with_disk_full_recovery(_do)
737
+ removed += len(stale)
738
+ if zcursor == 0 or zcursor == "0":
739
+ break
740
+ return removed
566
741
 
567
742
  async def prune_metrics(self, older_than_seconds: float) -> int:
568
743
  cutoff = _now_ts() - older_than_seconds
@@ -635,11 +810,11 @@ class RedisDriver(BaseDriver):
635
810
  pipe.delete(self._key("job", jid))
636
811
  pipe.zrem(self._idx_all(), jid)
637
812
  pipe.zrem(self._key("delayed"), jid)
638
- for st in ("pending", "processing", "completed", "failed"):
813
+ for st in _ALL_STATUSES:
639
814
  pipe.zrem(self._idx_status(st), jid)
640
815
  # Drop all per-queue and per-(queue,status) indexes
641
816
  pipe.delete(self._idx_queue(queue))
642
- for st in ("pending", "processing", "completed", "failed"):
817
+ for st in _ALL_STATUSES:
643
818
  pipe.delete(self._idx_queue_status(queue, st))
644
819
  pipe.srem(self._key("queues"), queue)
645
820
  await pipe.execute()
@@ -57,26 +57,35 @@ class Pruner:
57
57
  return self.config.prune_metrics_hours * 3600
58
58
  return float(self.config.prune_metrics_seconds)
59
59
 
60
+ async def _prune_terminal(self, status: str, older_than: float) -> int:
61
+ """Prune a terminal status via the driver's index-consistent bulk delete.
62
+
63
+ The driver drains the whole backlog in capped batches (so a large or
64
+ orphan-laden index never blocks the backend) and reaps orphaned index entries
65
+ in the same pass, returning the total removed."""
66
+ return await self.driver.prune_terminal_jobs(
67
+ status=status,
68
+ older_than=older_than,
69
+ limit=max(1, int(self.config.prune_batch_size)),
70
+ )
71
+
60
72
  async def prune_once(self) -> dict[str, int]:
61
73
  """Run a single prune pass based on config."""
62
74
  results: dict[str, int] = {}
63
75
 
64
76
  if self.completed_threshold > 0:
65
- results["completed"] = await self.driver.prune(
66
- status="completed",
67
- older_than_seconds=self.completed_threshold,
77
+ results["completed"] = await self._prune_terminal(
78
+ "completed", self.completed_threshold,
68
79
  )
69
80
 
70
81
  if self.failed_threshold > 0:
71
- results["failed"] = await self.driver.prune(
72
- status="failed",
73
- older_than_seconds=self.failed_threshold,
82
+ results["failed"] = await self._prune_terminal(
83
+ "failed", self.failed_threshold,
74
84
  )
75
85
 
76
86
  if self.cancelled_threshold > 0:
77
- results["cancelled"] = await self.driver.prune(
78
- status="cancelled",
79
- older_than_seconds=self.cancelled_threshold,
87
+ results["cancelled"] = await self._prune_terminal(
88
+ "cancelled", self.cancelled_threshold,
80
89
  )
81
90
 
82
91
  if self.metrics_threshold > 0:
@@ -30,6 +30,7 @@ class Queue:
30
30
  cls._config = config or BaQueueConfig()
31
31
  if driver is not None:
32
32
  driver.auto_cleanup_on_disk_full = cls._config.auto_cleanup_on_disk_full
33
+ driver.reconcile_on_connect = cls._config.reconcile_on_connect
33
34
  cls._driver = driver
34
35
  cls._events = EventBus.default()
35
36
 
@@ -222,4 +223,5 @@ def _create_driver(config: BaQueueConfig) -> BaseDriver:
222
223
  else:
223
224
  raise ValueError(f"Unknown driver: {name}")
224
225
  driver.auto_cleanup_on_disk_full = config.auto_cleanup_on_disk_full
226
+ driver.reconcile_on_connect = config.reconcile_on_connect
225
227
  return driver
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: baqueue
3
- Version: 0.1.0
3
+ Version: 1.0.2
4
4
  Summary: A powerful Python queue management package inspired by Laravel Horizon
5
5
  Author: Basalam, BaQueue Contributors
6
6
  License: MIT
@@ -45,6 +45,7 @@ Provides-Extra: dev
45
45
  Requires-Dist: baqueue[all]; extra == "dev"
46
46
  Requires-Dist: pytest>=8.0; extra == "dev"
47
47
  Requires-Dist: pytest-asyncio>=0.23; extra == "dev"
48
+ Requires-Dist: fakeredis>=2.21; extra == "dev"
48
49
  Requires-Dist: build>=1.0; extra == "dev"
49
50
  Requires-Dist: twine>=5.0; extra == "dev"
50
51
  Dynamic: license-file
@@ -274,6 +275,21 @@ await Queue.prune(status="completed", hours=24)
274
275
  await Queue.prune(tag="batch:newsletter")
275
276
  ```
276
277
 
278
+ #### Redis index health
279
+
280
+ The Redis driver keeps secondary indexes (sorted sets) so the dashboard can list and
281
+ count jobs by queue/status efficiently. All deletes go through an index-consistent path
282
+ that removes the job hash *and* every index entry in one atomic step, so the indexes stay
283
+ bounded. If entries are ever orphaned out-of-band (e.g. job hashes deleted directly via
284
+ `redis-cli`), pruning reaps them automatically, and you can force a full repair:
285
+
286
+ ```bash
287
+ baqueue reconcile-indexes -d redis --driver-url redis://localhost:6379/0
288
+ ```
289
+
290
+ Set `reconcile_on_connect=True` to run that repair once on every startup (off by default
291
+ to keep connect fast on large datasets).
292
+
277
293
  ### Retry Failed Jobs
278
294
 
279
295
  Bulk-retry failed jobs from the CLI, from Python, or from the dashboard.
@@ -508,6 +524,7 @@ baqueue schedule Start the job scheduler
508
524
  baqueue dashboard Launch the monitoring dashboard
509
525
  baqueue prune Prune old jobs
510
526
  baqueue retry-failed Retry all failed jobs (filter by queue/tag/age)
527
+ baqueue reconcile-indexes Repair Redis secondary indexes (drop stale entries)
511
528
  baqueue status Show queue status
512
529
  baqueue test Run the test suite
513
530
  ```
@@ -18,6 +18,7 @@ websockets>=12.0
18
18
  baqueue[all]
19
19
  pytest>=8.0
20
20
  pytest-asyncio>=0.23
21
+ fakeredis>=2.21
21
22
  build>=1.0
22
23
  twine>=5.0
23
24
 
@@ -53,6 +53,7 @@ dev = [
53
53
  "baqueue[all]",
54
54
  "pytest>=8.0",
55
55
  "pytest-asyncio>=0.23",
56
+ "fakeredis>=2.21",
56
57
  "build>=1.0",
57
58
  "twine>=5.0",
58
59
  ]
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes