Reuse approved worker networking and retry only confirmed clean probe failures

examples/revalidation.example.json

@@ -4,7 +4,6 @@
   "project": "otche-local",
   "worker_container": "otche-local-worker-1",
   "postgres_container": "otche-local-postgres-1",
-  "compose_network": "otche-local_private",
   "database_user": "otche",
   "database": "otche",
   "worker_config": "/srv/otche/secrets/worker/config.json",

revalidate.py

@@ -229,7 +229,7 @@ class Renewal:
         # Exec inside the existing approved image; no builds, pulls, migrations or project changes.
         info = json.loads(self.docker('inspect', self.worker))[0]
         image = info['Image']
-        command = ['run', '--rm', '--network', self.c['compose_network'], '--volumes-from', self.worker + ':ro',
+        command = ['run', '--rm', '--network', 'container:' + self.worker, '--volumes-from', self.worker + ':ro',
                    '--env', 'DATABASE_URL_FILE=/run/secrets/database-url', '--env', 'WORKER_CONFIG_FILE=/run/otche/config.json',
                    image, 'bindings-sync']
         self.docker(*command, timeout=120)
@@ -269,6 +269,19 @@ class Renewal:
             self.publish()
             self.current.unlink()
             print('Fresh owner and online proofs published after real probes and cleanup', flush=True)
+        except Exception:
+            if self.work:
+                save(self.state / 'last-failure.json', {'at': dt.datetime.now(dt.timezone.utc).isoformat(), 'work_dir': str(self.work)})
+                owner_journal = self.work / 'owner-resources.json'
+                network_journal = self.work / 'network-resources.json'
+                owner_clean = owner_journal.exists() and any(row['phase'] == 'complete' and row['resource'] == 'cleanup'
+                    for row in load(owner_journal)['events'])
+                network_clean = not network_journal.exists() or load(network_journal).get('cleanup_complete') is True
+                if owner_clean and network_clean and not load('/var/lib/otche-network/state.json')['records']:
+                    self.delete_clone()
+                    self.current.unlink(missing_ok=True)
+                    print('Failed checks left no resources; next timer may retry without renewing old proofs', flush=True)
+            raise
         finally:
             # A failed/ambiguous test retains its journal and cannot be automatically overwritten.
             # Never force-delete a running/changed VM or hide a failed cleanup.