conductor(checkpoint): Checkpoint end of Phase 2

scripts/cluster_status/tests/test_aggregator.py

@@ -4,7 +4,9 @@ import cluster_aggregator
 
 @patch("consul_client.get_cluster_services")
 @patch("litefs_client.get_node_status")
-def test_aggregate_cluster_status(mock_litefs, mock_consul):
+@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):
     """Test aggregating Consul and LiteFS data."""
     # Mock Consul data
     mock_consul.return_value = [
@@ -19,6 +21,7 @@ def test_aggregate_cluster_status(mock_litefs, mock_consul):
         return {"is_primary": False, "uptime": 50, "advertise_url": "url2", "replication_lag": 10}
         
     mock_litefs.side_effect = litefs_side_effect
+    mock_nomad_id.return_value = None
     
     cluster_data = cluster_aggregator.get_cluster_status("http://consul:8500")
     
@@ -35,12 +38,17 @@ def test_aggregate_cluster_status(mock_litefs, mock_consul):
 
 @patch("consul_client.get_cluster_services")
 @patch("litefs_client.get_node_status")
-def test_aggregate_cluster_status_unhealthy(mock_litefs, mock_consul):
+@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):
     """Test health calculation when nodes are critical."""
     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_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"

scripts/cluster_status/tests/test_main.py

@@ -12,7 +12,8 @@ def test_arg_parsing_default():
 
 def test_arg_parsing_custom():
     """Test that custom arguments are parsed correctly."""
-    with patch.object(sys, 'argv', ['cli.py', '--consul-url', 'http://custom:8500', '--no-color']):
+    with patch.object(sys, 'argv', ['cli.py', '--consul-url', 'http://custom:8500', '--no-color', '--restart', 'node1']):
         args = cli.parse_args()
         assert args.consul_url == 'http://custom:8500'
         assert args.no_color is True
+        assert args.restart == 'node1'

scripts/cluster_status/tests/test_nomad_client.py

@@ -0,0 +1,58 @@
+import pytest
+from unittest.mock import patch, MagicMock
+import nomad_client
+import subprocess
+
+@patch("subprocess.run")
+@patch("nomad_client.get_node_map")
+def test_get_allocation_id(mock_node_map, mock_run):
+    """Test getting allocation ID for a node."""
+    mock_node_map.return_value = {"node_id1": "node1"}
+    
+    # Mock 'nomad job status navidrome-litefs' output
+    mock_job_status = MagicMock()
+    mock_job_status.stdout = """
+Allocations
+ID        Node ID   Task Group  Version  Desired  Status   Created    Modified
+abc12345  node_id1  navidrome   1        run      running  1h ago     1h ago
+"""
+    
+    # Mock 'nomad alloc status abc12345' output
+    mock_alloc_status = MagicMock()
+    mock_alloc_status.stdout = "ID = abc12345-full-id"
+    
+    mock_run.side_effect = [mock_job_status, mock_alloc_status]
+    
+    alloc_id = nomad_client.get_allocation_id("node1", "navidrome-litefs")
+    assert alloc_id == "abc12345-full-id"
+
+@patch("subprocess.run")
+def test_get_logs(mock_run):
+    """Test fetching logs for an allocation."""
+    mock_stderr = "Error: database is locked\nSome other error"
+    m = MagicMock()
+    m.stdout = mock_stderr
+    m.return_code = 0
+    mock_run.return_value = m
+    
+    logs = nomad_client.get_allocation_logs("abc12345", tail=20)
+    assert "database is locked" in logs
+    # It should have tried with -task navidrome first
+    mock_run.assert_any_call(
+        ["nomad", "alloc", "logs", "-stderr", "-task", "navidrome", "-n", "20", "abc12345"],
+        capture_output=True, text=True, check=True
+    )
+
+@patch("subprocess.run")
+def test_restart_allocation(mock_run):
+    """Test restarting an allocation."""
+    m = MagicMock()
+    m.return_code = 0
+    mock_run.return_value = m
+    
+    success = nomad_client.restart_allocation("abc12345")
+    assert success is True
+    mock_run.assert_called_with(
+        ["nomad", "alloc", "restart", "abc12345"],
+        capture_output=True, text=True, check=True
+    )