Fix error in init command

Author: Saif Al-Din Ali

src/migrate/HISTORY.rst

@@ -2,6 +2,9 @@
 
 Release History
 ===============
+3.0.0b4
++++++++++++++++
+* Fix edge case bugs in az migrate local replication init & new commands.
 
 3.0.0b3
 +++++++++++++++

src/migrate/azext_migrate/helpers/replication/init/_setup_policy.py

@@ -170,6 +170,7 @@ def find_fabric(all_fabrics, appliance_name, fabric_instance_type,
 def get_fabric_agent(cmd, replication_fabrics_uri, fabric, appliance_name,
                      fabric_instance_type):
     """Get and validate fabric agent (DRA) for the given fabric."""
+    logger = get_logger(__name__)
     fabric_name = fabric.get('name')
     dras_uri = (
         f"{replication_fabrics_uri}/{fabric_name}"
@@ -180,18 +181,60 @@ def get_fabric_agent(cmd, replication_fabrics_uri, fabric, appliance_name,
     dras = dras_response.json().get('value', [])
 
     dra = None
+    found_but_not_responsive = None
     for candidate in dras:
         props = candidate.get('properties', {})
         custom_props = props.get('customProperties', {})
-        if (props.get('machineName') == appliance_name and
-                custom_props.get('instanceType') == fabric_instance_type and
-                bool(props.get('isResponsive'))):
-            dra = candidate
-            break
+        machine_name = props.get('machineName', '')
+        if (machine_name.lower() == appliance_name.lower() and
+                custom_props.get('instanceType') == fabric_instance_type):
+            if bool(props.get('isResponsive')):
+                dra = candidate
+                break
+            else:
+                found_but_not_responsive = candidate
+
+    # Accept a non-responsive DRA if it's the only match and is provisioned
+    if not dra and found_but_not_responsive:
+        nr_props = found_but_not_responsive.get('properties', {})
+        last_heartbeat = nr_props.get('lastHeartbeat', 'unknown')
+        if (nr_props.get('provisioningState') ==
+                ProvisioningState.Succeeded.value):
+            logger.warning(
+                "The appliance '%s' DRA is not responsive "
+                "(last heartbeat: %s). Proceeding since provisioning "
+                "state is 'Succeeded'.",
+                appliance_name, last_heartbeat)
+            dra = found_but_not_responsive
+        else:
+            raise CLIError(
+                f"The appliance '{appliance_name}' is in a "
+                f"disconnected state (last heartbeat: {last_heartbeat}, "
+                f"provisioningState: "
+                f"{nr_props.get('provisioningState')})."
+            )
 
     if not dra:
+        # Log available DRAs for diagnostics
+        if dras:
+            logger.warning(
+                "No matching fabric agent found for appliance '%s' "
+                "(expected instanceType '%s'). Available agents:",
+                appliance_name, fabric_instance_type)
+            for candidate in dras:
+                props = candidate.get('properties', {})
+                custom_props = props.get('customProperties', {})
+                logger.warning(
+                    "  - machineName: '%s', instanceType: '%s', "
+                    "isResponsive: %s",
+                    props.get('machineName'),
+                    custom_props.get('instanceType'),
+                    props.get('isResponsive'))
+
         raise CLIError(
-            f"The appliance '{appliance_name}' is in a disconnected state."
+            f"No fabric agent found for appliance '{appliance_name}' "
+            f"on fabric '{fabric_name}'. Verify that the appliance is "
+            f"properly registered and connected."
         )
 
     return dra

src/migrate/azext_migrate/helpers/replication/new/_process_inputs.py

@@ -740,19 +740,41 @@ def process_target_fabric(cmd,
     source_dras = source_dras_response.json().get('value', [])
 
     source_dra = None
+    source_found_not_responsive = None
     for dra in source_dras:
         props = dra.get('properties', {})
         custom_props = props.get('customProperties', {})
-        if (props.get('machineName') == source_appliance_name and
-                custom_props.get('instanceType') == fabric_instance_type and
-                bool(props.get('isResponsive'))):
-            source_dra = dra
-            break
+        machine_name = props.get('machineName', '')
+        if (machine_name.lower() == source_appliance_name.lower() and
+                custom_props.get('instanceType') == fabric_instance_type):
+            if bool(props.get('isResponsive')):
+                source_dra = dra
+                break
+            else:
+                source_found_not_responsive = dra
+
+    if not source_dra and source_found_not_responsive:
+        nr_props = source_found_not_responsive.get('properties', {})
+        last_hb = nr_props.get('lastHeartbeat', 'unknown')
+        if (nr_props.get('provisioningState') ==
+                ProvisioningState.Succeeded.value):
+            logger.warning(
+                "The source appliance '%s' DRA is not responsive "
+                "(last heartbeat: %s). Proceeding since provisioning "
+                "state is 'Succeeded'.",
+                source_appliance_name, last_hb)
+            source_dra = source_found_not_responsive
+        else:
+            raise CLIError(
+                f"The source appliance '{source_appliance_name}' is in a "
+                f"disconnected state (last heartbeat: {last_hb}).")
 
     if not source_dra:
         raise CLIError(
-            f"The source appliance '{source_appliance_name}' is in a "
-            f"disconnected state.")
+            f"No fabric agent found for source appliance "
+            f"'{source_appliance_name}' on fabric "
+            f"'{source_fabric_name}'. Verify that the appliance is "
+            f"properly registered and connected.")
 
     target_fabric, target_fabric_candidates, \
         target_fabric_instance_type = _process_target_fabrics(
@@ -778,19 +800,41 @@ def process_target_fabric(cmd,
     target_dras = target_dras_response.json().get('value', [])
 
     target_dra = None
+    target_found_not_responsive = None
     for dra in target_dras:
         props = dra.get('properties', {})
         custom_props = props.get('customProperties', {})
-        if (props.get('machineName') == target_appliance_name and
+        machine_name = props.get('machineName', '')
+        if (machine_name.lower() == target_appliance_name.lower() and
                 custom_props.get('instanceType') ==
-                target_fabric_instance_type and
-                bool(props.get('isResponsive'))):
-            target_dra = dra
-            break
+                target_fabric_instance_type):
+            if bool(props.get('isResponsive')):
+                target_dra = dra
+                break
+            else:
+                target_found_not_responsive = dra
+
+    if not target_dra and target_found_not_responsive:
+        nr_props = target_found_not_responsive.get('properties', {})
+        last_hb = nr_props.get('lastHeartbeat', 'unknown')
+        if (nr_props.get('provisioningState') ==
+                ProvisioningState.Succeeded.value):
+            logger.warning(
+                "The target appliance '%s' DRA is not responsive "
+                "(last heartbeat: %s). Proceeding since provisioning "
+                "state is 'Succeeded'.",
+                target_appliance_name, last_hb)
+            target_dra = target_found_not_responsive
+        else:
+            raise CLIError(
+                f"The target appliance '{target_appliance_name}' is in a "
+                f"disconnected state (last heartbeat: {last_hb}).")
 
     if not target_dra:
         raise CLIError(
-            f"The target appliance '{target_appliance_name}' is in a "
-            f"disconnected state.")
+            f"No fabric agent found for target appliance "
+            f"'{target_appliance_name}' on fabric "
+            f"'{target_fabric_name}'. Verify that the appliance is "
+            f"properly registered and connected.")
 
     return target_fabric, source_dra, target_dra

src/migrate/azext_migrate/tests/latest/test_migrate_commands.py

@@ -3385,7 +3385,7 @@ def test_find_fabric_matching_succeeded(self):
 
     @mock.patch('azext_migrate.helpers.replication.init._setup_policy.send_get_request')
     def test_get_fabric_agent_not_responsive(self, mock_get_request):
-        """Test get_fabric_agent when agent is not responsive."""
+        """Test get_fabric_agent when agent is not responsive and not Succeeded."""
         from azext_migrate.helpers.replication.init._setup_policy import get_fabric_agent
         from knack.util import CLIError
 
@@ -3397,6 +3397,8 @@ def test_get_fabric_agent_not_responsive(self, mock_get_request):
                     'properties': {
                         'machineName': 'appliance1',
                         'isResponsive': False,
+                        'provisioningState': 'Failed',
+                        'lastHeartbeat': '2026-01-01T00:00:00Z',
                         'customProperties': {
                             'instanceType': 'HyperV'
                         }
@@ -3413,6 +3415,99 @@ def test_get_fabric_agent_not_responsive(self, mock_get_request):
 
         self.assertIn('disconnected state', str(context.exception))
 
+    @mock.patch('azext_migrate.helpers.replication.init._setup_policy.send_get_request')
+    def test_get_fabric_agent_not_responsive_but_succeeded(self, mock_get_request):
+        """Test get_fabric_agent proceeds when DRA is not responsive but provisioning succeeded."""
+        from azext_migrate.helpers.replication.init._setup_policy import get_fabric_agent
+
+        mock_cmd = mock.Mock()
+        mock_response = mock.Mock()
+        mock_response.json.return_value = {
+            'value': [
+                {
+                    'properties': {
+                        'machineName': 'appliance1',
+                        'isResponsive': False,
+                        'provisioningState': 'Succeeded',
+                        'lastHeartbeat': '2026-03-05T21:46:47Z',
+                        'customProperties': {
+                            'instanceType': 'HyperV'
+                        }
+                    }
+                }
+            ]
+        }
+        mock_get_request.return_value = mock_response
+
+        fabric = {'name': 'fabric1'}
+
+        # Should succeed with a warning, not raise
+        result = get_fabric_agent(
+            mock_cmd, '/fabrics', fabric, 'appliance1', 'HyperV')
+        self.assertEqual(
+            result['properties']['machineName'], 'appliance1')
+
+    @mock.patch('azext_migrate.helpers.replication.init._setup_policy.send_get_request')
+    def test_get_fabric_agent_case_insensitive_match(self, mock_get_request):
+        """Test get_fabric_agent matches machineName case-insensitively."""
+        from azext_migrate.helpers.replication.init._setup_policy import get_fabric_agent
+
+        mock_cmd = mock.Mock()
+        mock_response = mock.Mock()
+        mock_response.json.return_value = {
+            'value': [
+                {
+                    'properties': {
+                        'machineName': 'Appliance1',
+                        'isResponsive': True,
+                        'customProperties': {
+                            'instanceType': 'HyperV'
+                        }
+                    }
+                }
+            ]
+        }
+        mock_get_request.return_value = mock_response
+
+        fabric = {'name': 'fabric1'}
+
+        # Should succeed despite case difference
+        result = get_fabric_agent(
+            mock_cmd, '/fabrics', fabric, 'appliance1', 'HyperV')
+        self.assertEqual(
+            result['properties']['machineName'], 'Appliance1')
+
+    @mock.patch('azext_migrate.helpers.replication.init._setup_policy.send_get_request')
+    def test_get_fabric_agent_no_matching_agent(self, mock_get_request):
+        """Test get_fabric_agent when no agent matches the appliance name."""
+        from azext_migrate.helpers.replication.init._setup_policy import get_fabric_agent
+        from knack.util import CLIError
+
+        mock_cmd = mock.Mock()
+        mock_response = mock.Mock()
+        mock_response.json.return_value = {
+            'value': [
+                {
+                    'properties': {
+                        'machineName': 'other-appliance',
+                        'isResponsive': True,
+                        'customProperties': {
+                            'instanceType': 'HyperV'
+                        }
+                    }
+                }
+            ]
+        }
+        mock_get_request.return_value = mock_response
+
+        fabric = {'name': 'fabric1'}
+
+        with self.assertRaises(CLIError) as context:
+            get_fabric_agent(
+                mock_cmd, '/fabrics', fabric, 'appliance1', 'HyperV')
+
+        self.assertIn('No fabric agent found', str(context.exception))
+
 
 class MigrateNewExecuteTests2(unittest.TestCase):
     """Additional test class for new/_execute_new.py functions."""

src/migrate/setup.py

@@ -7,7 +7,7 @@
 
 from setuptools import setup, find_packages
 
-VERSION = "3.0.0b3"
+VERSION = "3.0.0b4"
 
 CLASSIFIERS = [
     'Development Status :: 4 - Beta',