NRestarts or NRefused aren't available on older systemd versions (#1039)

* If NRestarts or NRefused are not available, don't ignore the unit itself
* Don't report systemd metrics (NRestarts/NRefused) that are not available

Signed-off-by: James Hartig <james@getadmiral.com>

2 files changed, 28 insertions, 16 deletions

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 7758cc9..1401f84 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -9,6 +9,7 @@ The wifi collector is disabled by default due to suspected caching issues and go
 * https://github.com/prometheus/node_exporter/issues/1008
 
 * [CHANGE] Filter out non-installed units when collecting all systemd units #1011
+* [CHANGE] `service_restart_total` and `socket_refused_connections_total` will not be reported if you're running an older version of systemd
 * [FEATURE] Collect NRefused property for systemd socket units (available as of systemd v239)
 * [FEATURE] Collect NRestarts property for systemd service units
 * [FEATURE] Add socket unit stats to systemd collector #968
@@ -16,6 +17,7 @@ The wifi collector is disabled by default due to suspected caching issues and go
 * [ENHANCEMENT]
 
 * [BUGFIX] Fix goroutine leak in supervisord collector
+* [BUGFIX] Systemd units will not be ignored if you're running older versions of systemd #1039
 * [BUGFIX] Handle vanishing PIDs #1043
 
 ## 0.16.0 / 2018-05-15
diff --git a/collector/systemd_linux.go b/collector/systemd_linux.go
index 57aff5a..7140b7b 100644
--- a/collector/systemd_linux.go
+++ b/collector/systemd_linux.go
@@ -140,10 +140,10 @@ func (c *systemdCollector) collectUnitStatusMetrics(ch chan<- prometheus.Metric,
                                 c.unitDesc, prometheus.GaugeValue, isActive,
                                 unit.Name, stateName)
                 }
-                if strings.HasSuffix(unit.Name, ".service") {
+                if strings.HasSuffix(unit.Name, ".service") && unit.nRestarts != nil {
                         ch <- prometheus.MustNewConstMetric(
                                 c.nRestartsDesc, prometheus.CounterValue,
-                                float64(unit.nRestarts), unit.Name)
+                                float64(*unit.nRestarts), unit.Name)
                 }
         }
 }
@@ -160,9 +160,11 @@ func (c *systemdCollector) collectSockets(ch chan<- prometheus.Metric, units []u
                 ch <- prometheus.MustNewConstMetric(
                         c.socketCurrentConnectionsDesc, prometheus.GaugeValue,
                         float64(unit.currentConnections), unit.Name)
-                ch <- prometheus.MustNewConstMetric(
-                        c.socketRefusedConnectionsDesc, prometheus.GaugeValue,
-                        float64(unit.refusedConnections), unit.Name)
+                if unit.refusedConnections != nil {
+                        ch <- prometheus.MustNewConstMetric(
+                                c.socketRefusedConnectionsDesc, prometheus.GaugeValue,
+                                float64(*unit.refusedConnections), unit.Name)
+                }
         }
 }
 
@@ -212,10 +214,10 @@ type unit struct {
         dbus.UnitStatus
         lastTriggerUsec     uint64
         startTimeUsec       uint64
-        nRestarts           uint32
+        nRestarts           *uint32
         acceptedConnections uint32
         currentConnections  uint32
-        refusedConnections  uint32
+        refusedConnections  *uint32
 }
 
 func (c *systemdCollector) getAllUnits() ([]unit, error) {
@@ -241,40 +243,47 @@ func (c *systemdCollector) getAllUnits() ([]unit, error) {
                 if strings.HasSuffix(unit.Name, ".timer") {
                         lastTriggerValue, err := conn.GetUnitTypeProperty(unit.Name, "Timer", "LastTriggerUSec")
                         if err != nil {
-                                return nil, fmt.Errorf("couldn't get unit '%s' LastTriggerUSec: %s", unit.Name, err)
+                                log.Debugf("couldn't get unit '%s' LastTriggerUSec: %s\n", unit.Name, err)
+                                continue
                         }
 
                         unit.lastTriggerUsec = lastTriggerValue.Value.Value().(uint64)
                 }
                 if strings.HasSuffix(unit.Name, ".service") {
-                        nRestarts, err := conn.GetUnitTypeProperty(unit.Name, "Service", "NRestarts")
+                        // NRestarts wasn't added until systemd 235.
+                        restartsCount, err := conn.GetUnitTypeProperty(unit.Name, "Service", "NRestarts")
                         if err != nil {
                                 log.Debugf("couldn't get unit '%s' NRestarts: %s\n", unit.Name, err)
-                                continue
+                        } else {
+                                nRestarts := restartsCount.Value.Value().(uint32)
+                                unit.nRestarts = &nRestarts
                         }
-                        unit.nRestarts = nRestarts.Value.Value().(uint32)
                 }
 
                 if strings.HasSuffix(unit.Name, ".socket") {
                         acceptedConnectionCount, err := conn.GetUnitTypeProperty(unit.Name, "Socket", "NAccepted")
                         if err != nil {
-                                return nil, fmt.Errorf("couldn't get unit '%s' NAccepted: %s", unit.Name, err)
+                                log.Debugf("couldn't get unit '%s' NAccepted: %s\n", unit.Name, err)
+                                continue
                         }
 
                         unit.acceptedConnections = acceptedConnectionCount.Value.Value().(uint32)
 
                         currentConnectionCount, err := conn.GetUnitTypeProperty(unit.Name, "Socket", "NConnections")
                         if err != nil {
-                                return nil, fmt.Errorf("couldn't get unit '%s' NConnections: %s", unit.Name, err)
+                                log.Debugf("couldn't get unit '%s' NConnections: %s\n", unit.Name, err)
+                                continue
                         }
                         unit.currentConnections = currentConnectionCount.Value.Value().(uint32)
 
+                        // NRefused wasn't added until systemd 239.
                         refusedConnectionCount, err := conn.GetUnitTypeProperty(unit.Name, "Socket", "NRefused")
                         if err != nil {
                                 log.Debugf("couldn't get unit '%s' NRefused: %s\n", unit.Name, err)
-                                continue
+                        } else {
+                                nRefused := refusedConnectionCount.Value.Value().(uint32)
+                                unit.refusedConnections = &nRefused
                         }
-                        unit.refusedConnections = refusedConnectionCount.Value.Value().(uint32)
                 }
 
                 if unit.ActiveState != "active" {
@@ -282,7 +291,8 @@ func (c *systemdCollector) getAllUnits() ([]unit, error) {
                 } else {
                         timestampValue, err := conn.GetUnitProperty(unit.Name, "ActiveEnterTimestamp")
                         if err != nil {
-                                return nil, fmt.Errorf("couldn't get unit '%s' StartTimeUsec: %s", unit.Name, err)
+                                log.Debugf("couldn't get unit '%s' StartTimeUsec: %s\n", unit.Name, err)
+                                continue
                         }
 
                         unit.startTimeUsec = timestampValue.Value.Value().(uint64)