Add bad_statuses and state

.github/workflows/ci.yml

@@ -113,7 +113,7 @@ jobs:
           ref: 'main'
           fetch-depth: 0
       - name: Generate Changelog
-        uses: sergioasantiago/generate-changelog-action@main
+        uses: pfandie/generate-changelog-action@v1
         id: changelog
         with:
           output_file: CHANGELOG.md

samples/config.yaml.template

@@ -3,6 +3,12 @@ tableau_server_management:
     username: <admin>
     password: <password>
     sleep_interval_seconds: <30>
+    bad_statuses:
+        - Unlicensed
+        - Degraded
+        - Error
+        - Stopped
+        - Status Unavailable
 
 metrics:
     port: <8000>

src/tableau_prometheus_exporter/server.py

@@ -20,36 +20,47 @@
         "service_name",
         "instance_id",
         "status",
+        "state",
     ],
 )
 
 
-def parse(status: Dict) -> None:
-    """
-    Parse Tableau Server Management Status and export as Prometheus Gauge
-
-    :return: None
-    """
-
-    GAUGE_TABLEAU_SERVICE_STATUS.clear()
-
-    for node in status.get("clusterStatus", {}).get("nodes", []):
-        for service in node.get("services", []):
-            for instance in service.get("instances", []):
-                GAUGE_TABLEAU_SERVICE_STATUS.labels(
-                    node.get("nodeId", ""),
-                    service.get("serviceName", ""),
-                    instance.get("instanceId", ""),
-                    instance.get("processStatus", ""),
-                ).set(1)
-
-
 class Server:
     def __init__(self, config: Dict):
-        self.config = config
+        self.tsm_config = config["tableau_server_management"]
+        self.metrics_port = config["metrics"].get("metrics_port", DEFAULT_METRICS_PORT)
+        self.sleep_interval = self.tsm_config.get(
+            "sleep_interval_seconds", DEFAULT_SLEEP_INTERVAL_SECONDS
+        )
+        self.bad_statuses = self.tsm_config.get("bad_statuses", [])
 
     RUNNING = True
 
+    def parse(self, status: Dict) -> None:
+        """
+        Parse Tableau Server Management Status and export as Prometheus Gauge
+
+        :return: None
+        """
+
+        GAUGE_TABLEAU_SERVICE_STATUS.clear()
+
+        for node in status.get("clusterStatus", {}).get("nodes", []):
+            for service in node.get("services", []):
+                for instance in service.get("instances", []):
+                    value = (
+                        0
+                        if instance.get("processStatus", "") in self.bad_statuses
+                        else 1
+                    )
+                    GAUGE_TABLEAU_SERVICE_STATUS.labels(
+                        node.get("nodeId", ""),
+                        service.get("serviceName", ""),
+                        instance.get("instanceId", ""),
+                        instance.get("processStatus", ""),
+                        instance.get("currentDeploymentState", ""),
+                    ).set(value)
+
     def start_and_run_forever(self) -> None:
         """
         Start up the server to expose the metrics.
@@ -60,15 +71,10 @@ def start_and_run_forever(self) -> None:
         """
         _logger.info("Starting Metrics Server...")
 
-        metrics_port = self.config["metrics"].get("metrics_port", DEFAULT_METRICS_PORT)
-        sleep_interval = self.config["tableau_server_management"].get(
-            "sleep_interval_seconds", DEFAULT_SLEEP_INTERVAL_SECONDS
-        )
-
-        start_http_server(port=metrics_port)
+        start_http_server(port=self.metrics_port)
         while self.RUNNING:
-            with TableauSession(self.config) as session:
+            with TableauSession(self.tsm_config) as session:
                 status = session.status()
-                parse(status)
+                self.parse(status)
 
-            time.sleep(sleep_interval)
+            time.sleep(self.sleep_interval)

tests/test_server.py

@@ -2,18 +2,15 @@
 from unittest.mock import PropertyMock
 
 import tableau_prometheus_exporter
-from tableau_prometheus_exporter.server import (
-    DEFAULT_SLEEP_INTERVAL_SECONDS,
-    Server,
-    parse,
-)
+from tableau_prometheus_exporter.server import DEFAULT_SLEEP_INTERVAL_SECONDS, Server
 
 MOCK_CONFIG = {
     "metrics": {"port": 1234},
     "tableau_server_management": {
         "host": "http://test.local",
         "username": "test",
         "password": "test",
+        "bad_statuses": ["error"],
     },
 }
 
@@ -26,7 +23,16 @@
                     {
                         "serviceName": "service1",
                         "instances": [
-                            {"instanceId": "instance1", "processStatus": "running"}
+                            {
+                                "instanceId": "instance1",
+                                "processStatus": "running",
+                                "currentDeploymentState": "enabled",
+                            },
+                            {
+                                "instanceId": "instance2",
+                                "processStatus": "error",
+                                "currentDeploymentState": "enabled",
+                            },
                         ],
                     }
                 ],
@@ -44,7 +50,6 @@ def test_start_and_run_forever(mocker):
     mock_manager.status.return_value = {}
     mocker.patch("requests.Session", return_value=mocker.MagicMock())
     mocker.patch("tableau_prometheus_exporter.server.start_http_server")
-    mocker.patch("tableau_prometheus_exporter.server.parse")
     mocker.patch("time.sleep")
 
     server = Server(MOCK_CONFIG)
@@ -60,12 +65,11 @@ def test_start_and_run_forever(mocker):
     tableau_prometheus_exporter.server.start_http_server.assert_called_once()
     mock_tableau_session.assert_called_once_with(MOCK_CONFIG)
     mock_manager.status.assert_called_once()
-    tableau_prometheus_exporter.server.parse.assert_called_once()
     time.sleep.assert_called_with(DEFAULT_SLEEP_INTERVAL_SECONDS)
 
 
 def test_parse_method():
-    parse(MOCK_STATUS)
+    Server(MOCK_CONFIG).parse(MOCK_STATUS)
 
     metric_samples = list(
         tableau_prometheus_exporter.server.GAUGE_TABLEAU_SERVICE_STATUS.collect()
@@ -80,5 +84,18 @@ def test_parse_method():
         "service_name": "service1",
         "instance_id": "instance1",
         "status": "running",
+        "state": "enabled",
     }
     assert sample.value == 1.0
+
+    sample = metric_samples[1]
+
+    assert sample.name == "tableau_service_status"
+    assert sample.labels == {
+        "node_id": "node1",
+        "service_name": "service1",
+        "instance_id": "instance2",
+        "status": "error",
+        "state": "enabled",
+    }
+    assert sample.value == 0.0