fix: use file-based job state for bulk discovery progress

In-memory _bulk_jobs dict was per-worker in gunicorn (4 workers),
causing poll requests to miss job state. Now uses /tmp JSON files
visible to all workers.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

blueprints/admin/routes_website_discovery.py

@@ -5,7 +5,9 @@ Admin Website Discovery Routes
 Endpoints for discovering and managing website candidates for companies.
 """
 
+import json
 import logging
+import os
 import threading
 from datetime import datetime
 
@@ -20,8 +22,24 @@ from services.website_discovery_service import WebsiteDiscoveryService
 
 logger = logging.getLogger(__name__)
 
-# Store bulk job progress
-_bulk_jobs = {}
+# File-based job state (shared across gunicorn workers)
+JOB_DIR = '/tmp/nordabiz_discovery_jobs'
+
+
+def _save_job(job_id, data):
+    os.makedirs(JOB_DIR, exist_ok=True)
+    path = os.path.join(JOB_DIR, f'{job_id}.json')
+    with open(path, 'w') as f:
+        json.dump(data, f)
+
+
+def _load_job(job_id):
+    path = os.path.join(JOB_DIR, f'{job_id}.json')
+    try:
+        with open(path) as f:
+            return json.load(f)
+    except (FileNotFoundError, json.JSONDecodeError):
+        return None
 
 
 @bp.route('/discover-website/<int:company_id>', methods=['POST'])
@@ -58,7 +76,8 @@ def discover_websites_bulk():
     job_id = str(uuid.uuid4())[:8]
 
     def run_bulk(job_id):
-        _bulk_jobs[job_id] = {'status': 'running', 'processed': 0, 'total': 0, 'log': []}
+        job = {'status': 'running', 'processed': 0, 'total': 0, 'log': []}
+        _save_job(job_id, job)
         db = SessionLocal()
         try:
             # Skip companies that already have a pending/accepted candidate
@@ -74,7 +93,9 @@ def discover_websites_bulk():
                 ~Company.id.in_(already_have) if already_have else True,
             ).order_by(Company.name).limit(50).all()
 
-            _bulk_jobs[job_id]['total'] = len(companies)
+            job['total'] = len(companies)
+            _save_job(job_id, job)
+
             service = WebsiteDiscoveryService(db=db)
 
             import time
@@ -89,18 +110,20 @@ def discover_websites_bulk():
                 else:
                     status_text += result.get('error', 'brak wyników')
 
-                # Update processed AFTER building status text — atomic from poll perspective
-                _bulk_jobs[job_id]['log'].append(status_text)
-                _bulk_jobs[job_id]['processed'] += 1
+                job['log'].append(status_text)
+                job['processed'] += 1
+                _save_job(job_id, job)
 
-                if _bulk_jobs[job_id]['processed'] < _bulk_jobs[job_id]['total']:
+                if job['processed'] < job['total']:
                     time.sleep(5)
 
-            _bulk_jobs[job_id]['status'] = 'completed'
+            job['status'] = 'completed'
+            _save_job(job_id, job)
         except Exception as e:
             logger.error(f"Bulk discovery error: {e}")
-            _bulk_jobs[job_id]['status'] = 'error'
-            _bulk_jobs[job_id]['log'].append(f"Błąd: {e}")
+            job['status'] = 'error'
+            job['log'].append(f"Błąd: {e}")
+            _save_job(job_id, job)
         finally:
             db.close()
 
@@ -116,10 +139,13 @@ def discover_websites_bulk():
 def discover_websites_status():
     """Poll bulk discovery progress."""
     job_id = request.args.get('job_id')
-    if not job_id or job_id not in _bulk_jobs:
+    if not job_id:
+        return jsonify({'error': 'Job not found'}), 404
+
+    job = _load_job(job_id)
+    if not job:
         return jsonify({'error': 'Job not found'}), 404
 
-    job = _bulk_jobs[job_id]
     log_offset = request.args.get('log_offset', 0, type=int)
     new_entries = job['log'][log_offset:]