conductor(checkpoint): Checkpoint end of Phase 1

scripts/cluster_status/tests/test_aggregator.py

@@ -6,8 +6,10 @@ import cluster_aggregator
 @patch("litefs_client.get_node_status")
 @patch("nomad_client.get_allocation_id")
 @patch("nomad_client.get_allocation_logs")
-def test_aggregate_cluster_status(mock_nomad_logs, mock_nomad_id, mock_litefs, mock_consul):
+@patch("nomad_client.get_node_map")
+def test_aggregate_cluster_status(mock_node_map, mock_nomad_logs, mock_nomad_id, mock_litefs, mock_consul):
     """Test aggregating Consul and LiteFS data."""
+    mock_node_map.return_value = {"id": "name"}
     # Mock Consul data
     mock_consul.return_value = [
         {"node": "node1", "address": "1.1.1.1", "role": "primary", "status": "passing"},
@@ -15,10 +17,10 @@ def test_aggregate_cluster_status(mock_nomad_logs, mock_nomad_id, mock_litefs, m
     ]
     
     # Mock LiteFS data
-    def litefs_side_effect(addr):
+    def litefs_side_effect(addr, **kwargs):
         if addr == "1.1.1.1":
-            return {"is_primary": True, "uptime": 100, "advertise_url": "url1"}
-        return {"is_primary": False, "uptime": 50, "advertise_url": "url2", "replication_lag": 10}
+            return {"is_primary": True, "uptime": 100, "advertise_url": "url1", "dbs": {"db1": {}}}
+        return {"is_primary": False, "uptime": 50, "advertise_url": "url2", "replication_lag": 10, "dbs": {"db1": {}}}
         
     mock_litefs.side_effect = litefs_side_effect
     mock_nomad_id.return_value = None
@@ -40,15 +42,17 @@ def test_aggregate_cluster_status(mock_nomad_logs, mock_nomad_id, mock_litefs, m
 @patch("litefs_client.get_node_status")
 @patch("nomad_client.get_allocation_id")
 @patch("nomad_client.get_allocation_logs")
-def test_aggregate_cluster_status_unhealthy(mock_nomad_logs, mock_nomad_id, mock_litefs, mock_consul):
+@patch("nomad_client.get_node_map")
+def test_aggregate_cluster_status_unhealthy(mock_node_map, mock_nomad_logs, mock_nomad_id, mock_litefs, mock_consul):
     """Test health calculation when nodes are critical."""
+    mock_node_map.return_value = {}
     mock_consul.return_value = [
         {"node": "node1", "address": "1.1.1.1", "role": "primary", "status": "critical"}
     ]
-    mock_litefs.return_value = {"is_primary": True, "uptime": 100}
+    mock_litefs.return_value = {"is_primary": True, "uptime": 100, "dbs": {"db1": {}}}
     mock_nomad_id.return_value = "alloc1"
     mock_nomad_logs.return_value = "error logs"
     
     cluster_data = cluster_aggregator.get_cluster_status("http://consul:8500")
     assert cluster_data["health"] == "Unhealthy"
-    assert cluster_data["nodes"][0]["nomad_logs"] == "error logs"
+    assert cluster_data["nodes"][0]["nomad_logs"] == "error logs"

scripts/cluster_status/tests/test_formatter.py

@@ -6,12 +6,13 @@ def test_format_cluster_summary():
     cluster_data = {
         "health": "Healthy",
         "primary_count": 1,
-        "nodes": []
+        "nodes": [],
+        "nomad_available": False
     }
     summary = output_formatter.format_summary(cluster_data)
     assert "Healthy" in summary
     assert "Primaries" in summary
-    assert "1" in summary
+    assert "WARNING: Nomad CLI unavailable" in summary
 
 def test_format_node_table():
     """Test the table generation."""

scripts/cluster_status/tests/test_litefs_client.py

@@ -55,4 +55,30 @@ def test_get_node_status_error(mock_get):
     status = litefs_client.get_node_status("192.168.1.101")
     
     assert "error" in status
-    assert status["is_primary"] is False
+    assert status["is_primary"] is False
+
+@patch("nomad_client.exec_command")
+def test_get_node_status_nomad_exec(mock_exec):
+    """Test fetching LiteFS status via nomad alloc exec."""
+    # Mock LiteFS status output (text format)
+    mock_status_output = """
+Config:
+  Path: /etc/litefs.yml
+  ...
+Status:
+  Primary: true
+  Uptime: 1h5m10s
+  Replication Lag: 0s
+"""
+    mock_exec.return_value = mock_status_output
+    
+    # We need to mock requests.get to fail first
+    with patch("requests.get") as mock_get:
+        mock_get.side_effect = Exception("HTTP failed")
+        
+        status = litefs_client.get_node_status("1.1.1.1", alloc_id="abc12345")
+        
+        assert status["is_primary"] is True
+        assert status["uptime"] == "1h5m10s"
+        # Since it's primary, lag might not be shown or be 0
+        assert status["replication_lag"] == "0s"

scripts/cluster_status/tests/test_nomad_client.py

@@ -55,4 +55,37 @@ def test_restart_allocation(mock_run):
     mock_run.assert_called_with(
         ["nomad", "alloc", "restart", "abc12345"],
         capture_output=True, text=True, check=True
-    )
+    )
+
+@patch("subprocess.run")
+def test_exec_command(mock_run):
+    """Test executing a command in an allocation."""
+    m = MagicMock()
+    m.stdout = "Command output"
+    m.return_code = 0
+    mock_run.return_value = m
+    
+    output = nomad_client.exec_command("abc12345", ["ls", "/data"])
+    assert output == "Command output"
+    mock_run.assert_called_with(
+        ["nomad", "alloc", "exec", "-task", "navidrome", "abc12345", "ls", "/data"],
+        capture_output=True, text=True, check=True
+    )
+
+@patch("subprocess.run")
+def test_exec_command_failure(mock_run):
+    """Test executing a command handles failure gracefully."""
+    mock_run.side_effect = subprocess.CalledProcessError(1, "nomad", stderr="Nomad error")
+    
+    output = nomad_client.exec_command("abc12345", ["ls", "/data"])
+    assert "Nomad Error" in output
+    assert "Nomad error" not in output # The exception str might not contain stderr directly depending on python version
+    
+@patch("subprocess.run")
+def test_get_node_map_failure(mock_run):
+    """Test get_node_map handles failure."""
+    mock_run.side_effect = FileNotFoundError("No such file")
+    
+    # It should not raise
+    node_map = nomad_client.get_node_map()
+    assert node_map == {}