fix: propagate start() to backends in DistributedObjectStore

DistributedObjectStore inherited the no-op base ObjectStore.start(),
so IRODSObjectStore.start() was never called after fork. This meant
the ConnectionPoolMonitorThread never started, causing stale iRODS
connections to accumulate. When get_connection() found them, it called
conn.disconnect() -> ssl.unwrap() on a dead socket, blocking for the
full socket timeout before raising TimeoutError.

Fixes #22219

lib/galaxy/objectstore/__init__.py

@@ -1482,6 +1482,10 @@ class DistributedObjectStore(NestedObjectStore):
         as_dict["backends"] = backends
         return as_dict
 
+    def start(self):
+        for backend in self.backends.values():
+            backend.start()
+
     def shutdown(self):
         """Shut down. Kill the free space monitor if there is one."""
         super().shutdown()

test/unit/objectstore/test_objectstore.py

@@ -6,7 +6,10 @@ from tempfile import (
     mkdtemp,
     mkstemp,
 )
-from unittest.mock import patch
+from unittest.mock import (
+    MagicMock,
+    patch,
+)
 from uuid import uuid4
 
 import pytest
@@ -514,6 +517,15 @@ def test_distributed_store_with_cache_targets():
             assert len(object_store.cache_targets()) == 2
 
 
+def test_distributed_store_start_propagates_to_backends():
+    with TestConfig(DISTRIBUTED_TEST_CONFIG) as (_, object_store):
+        for backend in object_store.backends.values():
+            backend.start = MagicMock()
+        object_store.start()
+        for backend in object_store.backends.values():
+            backend.start.assert_called_once()
+
+
 HIERARCHICAL_MUST_HAVE_UNIFIED_QUOTA_SOURCE = """<?xml version="1.0"?>
 <object_store type="hierarchical" private="true">
     <backends>