fix self heal

iaai_scraper/worker/self_heal.py

@@ -108,6 +108,31 @@ def _read_last_db_progress_ts(redis_client: Redis) -> int | None:
 	return max_ts or None
 
 
+def _has_active_recent_progress(redis_client: Redis, now_ts: int, stall_seconds: int) -> bool:
+	"""Return True when a task is still reporting non-DB progress.
+
+	Hourly only_new runs can legitimately skip every listed vehicle as existing.
+	Those runs may not emit fast_db_progress for a long time, but they still emit
+	regular listing/segment progress. Treating old DB timestamps as fatal caused
+	the watchdog to kill healthy production workers during such scans.
+	"""
+	for key in redis_client.scan_iter(match="iaai:state:task_progress:*"):
+		try:
+			payload = redis_client.get(key)
+			if not payload:
+				continue
+			data = json.loads(payload)
+			stage = str(data.get("stage") or "")
+			if stage in {"failed", "sync_done", "segment_task_failed", "segment_task_soft_timeout"}:
+				continue
+			ts = _safe_int(data.get("ts"), 0)
+			if ts > 0 and now_ts - ts <= max(60, int(stall_seconds)):
+				return True
+		except Exception:
+			continue
+	return False
+
+
 def _reset_bootstrap_checkpoint_for_db_idle(redis_client: Redis) -> None:
 	pipe = redis_client.pipeline()
 	pipe.delete(SYNC_LISTING_CHECKPOINT_KEY)
@@ -219,6 +244,12 @@ def main() -> None:
 			restart_reason = f"progress_age={age}s > {stall_seconds}s"
 			db_idle_restart = False
 			if age <= stall_seconds:
+				# If the task is actively reporting progress, do not require DB writes.
+				# Hourly only_new listing scans often skip already-known cars and can
+				# legitimately have no DB writes while still moving through segments.
+				if _has_active_recent_progress(redis_client, now_ts, stall_seconds):
+					time.sleep(check_interval)
+					continue
 				last_db_ts = _read_last_db_progress_ts(redis_client)
 				db_age = None if last_db_ts is None else max(0, now_ts - int(last_db_ts))
 				if db_age is None: