fix: prevent duplicate queue entries with Redis SET markers

Recovery sweeps and the retry endpoint now check a per-document Redis
key (SET NX, 1h TTL) before pushing to the queue. If the marker exists,
the doc is already enqueued and gets skipped. This prevents the
scheduler from re-enqueuing the same parsed docs every 5 minutes.

services/api/app.py

@@ -1891,6 +1891,7 @@ async def retry_failed_extractions_endpoint():
         return {"retried": 0, "message": "No extraction-failed documents to retry"}
 
     doc_ids = []
+    enqueued_set_prefix = f"{QUEUE_PREFIX}:enqueued"
     for row in rows:
         doc_type = row["document_type"]
         if doc_type == "macro_event":
@@ -1898,10 +1899,14 @@ async def retry_failed_extractions_endpoint():
         else:
             target = queue_key("extraction")
 
-        await rds.rpush(target, json.dumps({
+        doc_id = str(row["id"])
-            "document_id": str(row["id"]),
+        marker = f"{enqueued_set_prefix}:{doc_id}"
-            "ticker": row["ticker"] or "",
+        added = await rds.set(marker, "1", nx=True, ex=3600)
-        }))
+        if added:
+            await rds.rpush(target, json.dumps({
+                "document_id": doc_id,
+                "ticker": row["ticker"] or "",
+            }))
         doc_ids.append(row["id"])
 
     # Delete failed intelligence rows so extractor starts fresh

services/scheduler/app.py

@@ -22,6 +22,7 @@ from services.shared.redis_keys import (
     QUEUE_EXTRACTION,
     QUEUE_INGESTION,
     QUEUE_MACRO_CLASSIFICATION,
+    QUEUE_PREFIX,
     lock_key,
     queue_key,
     rate_limit_key,
@@ -538,6 +539,35 @@ STALE_PARSED_THRESHOLD_MINUTES: int = 30
 # How long after an extraction failure before we retry
 EXTRACTION_FAILED_RETRY_MINUTES: int = 60
 
+# Redis set key for tracking enqueued doc IDs (prevents duplicate enqueuing)
+_ENQUEUED_SET = f"{QUEUE_PREFIX}:enqueued"
+# How long an enqueued marker lives before it can be re-enqueued (seconds)
+_ENQUEUED_TTL = 3600
+
+
+async def _enqueue_if_new(
+    rds: aioredis.Redis,
+    queue: str,
+    document_id: str,
+    ticker: str,
+) -> bool:
+    """Push a job onto *queue* only if *document_id* isn't already tracked.
+
+    Uses a Redis SET with per-member expiry (via a separate key) to prevent
+    the same document from being enqueued multiple times by recovery sweeps.
+    Returns True if enqueued, False if skipped as duplicate.
+    """
+    marker_key = f"{_ENQUEUED_SET}:{document_id}"
+    # SET NX returns True only if the key was created (not already present)
+    added = await rds.set(marker_key, "1", nx=True, ex=_ENQUEUED_TTL)
+    if not added:
+        return False
+    await rds.rpush(queue, json.dumps({
+        "document_id": document_id,
+        "ticker": ticker,
+    }))
+    return True
+
 
 async def recover_stale_documents(pool: asyncpg.Pool, rds: aioredis.Redis) -> int:
     """Re-enqueue documents stuck in 'parsed' status for extraction.
@@ -575,12 +605,10 @@ async def recover_stale_documents(pool: asyncpg.Pool, rds: aioredis.Redis) -> in
         else:
             target = queue_key(QUEUE_EXTRACTION)
 
-        await rds.rpush(target, json.dumps({
+        added = await _enqueue_if_new(rds, target, str(row["id"]), row["ticker"] or "")
-            "document_id": str(row["id"]),
+        if added:
-            "ticker": row["ticker"] or "",
+            doc_ids.append(row["id"])
-        }))
+            enqueued += 1
-        doc_ids.append(row["id"])
-        enqueued += 1
 
     # Touch updated_at so these docs won't be re-enqueued until the threshold passes again
     if doc_ids:
@@ -627,12 +655,10 @@ async def retry_failed_extractions(pool: asyncpg.Pool, rds: aioredis.Redis) -> i
         else:
             target = queue_key(QUEUE_EXTRACTION)
 
-        await rds.rpush(target, json.dumps({
+        added = await _enqueue_if_new(rds, target, str(row["id"]), row["ticker"] or "")
-            "document_id": str(row["id"]),
+        if added:
-            "ticker": row["ticker"] or "",
+            doc_ids.append(row["id"])
-        }))
+            enqueued += 1
-        doc_ids.append(row["id"])
-        enqueued += 1
 
     if doc_ids:
         # Delete failed intelligence rows so extractor starts fresh