Merge pull request #108 from kermat/mdk/add-drbd-quorum-bool

Add quorum status to DRBD metrics

doc/metric_spec.md

@@ -211,13 +211,14 @@ The DRBD subsystems collect devices stats by parsing its configuration the JSON
 5. [`ha_cluster_drbd_bm_writes`](#ha_cluster_bm_writes)
 6. [`ha_cluster_drbd_upper_pending`](#ha_cluster_drbd_upper_pending)
 7. [`ha_cluster_drbd_lower_pending`](#ha_cluster_drbd_lower_pending)
-8. [`ha_cluster_drbd_connections`](#ha_cluster_drbd_connections)
-9. [`ha_cluster_drbd_connections_sync`](#ha_cluster_drbd_connections_sync)
-10. [`ha_cluster_drbd_connections_received`](#ha_cluster_drbd_connections_received)
-11. [`ha_cluster_drbd_connections_sent`](#ha_cluster_drbd_connections_sent)
-12. [`ha_cluster_drbd_connections_pending`](#ha_cluster_drbd_connections_pending)
-13. [`ha_cluster_drbd_connections_unacked`](#ha_cluster_drbd_connections_unacked)
-14. [`ha_cluster_drbd_split_brain`](#ha_cluster_drbd_split_brain)
+8. [`ha_cluster_drbd_quorum`](#ha_cluster_drbd_quorum)
+9. [`ha_cluster_drbd_connections`](#ha_cluster_drbd_connections)
+10. [`ha_cluster_drbd_connections_sync`](#ha_cluster_drbd_connections_sync)
+11. [`ha_cluster_drbd_connections_received`](#ha_cluster_drbd_connections_received)
+12. [`ha_cluster_drbd_connections_sent`](#ha_cluster_drbd_connections_sent)
+13. [`ha_cluster_drbd_connections_pending`](#ha_cluster_drbd_connections_pending)
+14. [`ha_cluster_drbd_connections_unacked`](#ha_cluster_drbd_connections_unacked)
+15. [`ha_cluster_drbd_split_brain`](#ha_cluster_drbd_split_brain)
 
 ### `ha_cluster_drbd_connections`
 
@@ -388,6 +389,18 @@ Value is an integer greater than or equal to `0`.
 - `resource`: the name of the resource.
 - `volume`: the volume number
 
+### `ha_cluster_drbd_quorum`
+
+#### Description
+
+Quorum status of the DRBD resource according to it's configured quorum policies; 1 line per `resource`, per `volume`
+Value is `1` when quorate, or `0` when inquorate.
+
+#### Labels
+
+- `resource`: the name of the resource.
+- `volume`: the volume number
+
 ### `ha_cluster_drbd_split_brain`
 
 #### Description

drbd_metrics.go

@@ -25,6 +25,7 @@ type drbdStatus struct {
 		BmWrites  int    `json:"bm-writes"`
 		UpPending int    `json:"upper-pending"`
 		LoPending int    `json:"lower-pending"`
+		Quorum    bool   `json:"quorum"`
 		DiskState string `json:"disk-state"`
 	} `json:"devices"`
 	Connections []struct {
@@ -53,6 +54,7 @@ var (
 		"bm_writes":            NewMetricDesc("drbd", "bm_writes", "Writes to bitmap; 1 line per res, per volume", []string{"resource", "volume"}),
 		"upper_pending":        NewMetricDesc("drbd", "upper_pending", "Upper pending; 1 line per res, per volume", []string{"resource", "volume"}),
 		"lower_pending":        NewMetricDesc("drbd", "lower_pending", "Lower pending; 1 line per res, per volume", []string{"resource", "volume"}),
+		"quorum":               NewMetricDesc("drbd", "quorum", "Quorum status per resource and per volume", []string{"resource", "volume"}),
 		"connections":          NewMetricDesc("drbd", "connections", "The DRBD resource connections; 1 line per per resource, per peer_node_id", []string{"resource", "peer_node_id", "peer_role", "volume", "peer_disk_state"}),
 		"connections_sync":     NewMetricDesc("drbd", "connections_sync", "The in sync percentage value for DRBD resource connections", []string{"resource", "peer_node_id", "volume"}),
 		"connections_received": NewMetricDesc("drbd", "connections_received", "KiB received per connection", []string{"resource", "peer_node_id", "volume"}),
@@ -121,6 +123,12 @@ func (c *drbdCollector) Collect(ch chan<- prometheus.Metric) {
 			ch <- c.makeGaugeMetric("upper_pending", float64(device.UpPending), resource.Name, strconv.Itoa(device.Volume))
 
 			ch <- c.makeGaugeMetric("lower_pending", float64(device.LoPending), resource.Name, strconv.Itoa(device.Volume))
+
+			if bool(device.Quorum) == true {
+				ch <- c.makeGaugeMetric("quorum", float64(1), resource.Name, strconv.Itoa(device.Volume))
+			} else {
+				ch <- c.makeGaugeMetric("quorum", float64(0), resource.Name, strconv.Itoa(device.Volume))
+			}
 		}
 		if len(resource.Connections) == 0 {
 			log.Warnf("Could not retrieve connection info for resource '%s'\n", resource.Name)

drbd_metrics_test.go

@@ -71,7 +71,7 @@ func TestDrbdParsing(t *testing.T) {
         "minor": 3,
         "disk-state": "UpToDate",
         "client": false,
-        "quorum": true,
+        "quorum": false,
         "size": 10200,
         "read": 654321,
         "written": 123456,
@@ -166,6 +166,14 @@ func TestDrbdParsing(t *testing.T) {
 		t.Errorf("lower-pending should be 2")
 	}
 
+	if true != drbdDevs[0].Devices[0].Quorum {
+		t.Errorf("quorum should be true")
+	}
+
+	if false != drbdDevs[1].Devices[0].Quorum {
+		t.Errorf("quorum should be false")
+	}
+
 	if 456 != drbdDevs[0].Connections[0].PeerDevices[0].Received {
 		t.Errorf("received should be 456")
 	}

test/drbd.metrics

@@ -46,6 +46,10 @@ ha_cluster_drbd_upper_pending{resource="1-single-1",volume="0"} 1 1234
 # TYPE ha_cluster_drbd_lower_pending gauge
 ha_cluster_drbd_lower_pending{resource="1-single-0",volume="0"} 2 1234
 ha_cluster_drbd_lower_pending{resource="1-single-1",volume="0"} 2 1234
+# HELP ha_cluster_drbd_quorum Quorum status per resource and per volume
+# TYPE ha_cluster_drbd_quorum gauge
+ha_cluster_drbd_quorum{resource="1-single-0",volume="0"} 1 1234
+ha_cluster_drbd_quorum{resource="1-single-1",volume="0"} 0 1234
 # HELP ha_cluster_drbd_resources The DRBD resources; 1 line per name, per volume
 # TYPE ha_cluster_drbd_resources gauge
 ha_cluster_drbd_resources{disk_state="uptodate",resource="1-single-0",role="Secondary",volume="0"} 1 1234

test/fake_drbdsetup.sh

@@ -65,7 +65,7 @@ cat <<EOF
         "minor": 3,
         "disk-state": "UpToDate",
         "client": false,
-        "quorum": true,
+        "quorum": false,
         "size": 10200,
         "read": 654321,
         "written": 123456,