From ac5a98176129e86c69f664e632ff273eee6f67bd Mon Sep 17 00:00:00 2001 From: xginn8 Date: Thu, 5 Jul 2018 10:26:48 -0400 Subject: [PATCH] Adding socket stat collection for systemd socket units (#968) Signed-off-by: xginn8 --- CHANGELOG.md | 1 + collector/systemd_linux.go | 77 ++++++++++++++++++++++++++------- collector/systemd_linux_test.go | 1 + 3 files changed, 63 insertions(+), 16 deletions(-) diff --git a/CHANGELOG.md b/CHANGELOG.md index 35e8899b..9f741f9f 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -4,6 +4,7 @@ * [CHANGE] * [FEATURE] Collect NRestarts property for systemd service units +* [FEATURE] Add socket unit stats to systemd collector #968 * [ENHANCEMENT] * [BUGFIX] diff --git a/collector/systemd_linux.go b/collector/systemd_linux.go index e05c289a..13dc0aa3 100644 --- a/collector/systemd_linux.go +++ b/collector/systemd_linux.go @@ -33,13 +33,15 @@ var ( ) type systemdCollector struct { - unitDesc *prometheus.Desc - systemRunningDesc *prometheus.Desc - summaryDesc *prometheus.Desc - nRestartsDesc *prometheus.Desc - timerLastTriggerDesc *prometheus.Desc - unitWhitelistPattern *regexp.Regexp - unitBlacklistPattern *regexp.Regexp + unitDesc *prometheus.Desc + systemRunningDesc *prometheus.Desc + summaryDesc *prometheus.Desc + nRestartsDesc *prometheus.Desc + timerLastTriggerDesc *prometheus.Desc + socketAcceptedConnectionsDesc *prometheus.Desc + socketCurrentConnectionsDesc *prometheus.Desc + unitWhitelistPattern *regexp.Regexp + unitBlacklistPattern *regexp.Regexp } var unitStatesName = []string{"active", "activating", "deactivating", "inactive", "failed"} @@ -70,17 +72,25 @@ func NewSystemdCollector() (Collector, error) { timerLastTriggerDesc := prometheus.NewDesc( prometheus.BuildFQName(namespace, subsystem, "timer_last_trigger_seconds"), "Seconds since epoch of last trigger.", []string{"name"}, nil) + socketAcceptedConnectionsDesc := prometheus.NewDesc( + prometheus.BuildFQName(namespace, subsystem, "socket_accepted_connections_total"), + "Total number of accepted socket connections", []string{"name"}, nil) + socketCurrentConnectionsDesc := prometheus.NewDesc( + prometheus.BuildFQName(namespace, subsystem, "socket_current_connections"), + "Current number of socket connections", []string{"name"}, nil) unitWhitelistPattern := regexp.MustCompile(fmt.Sprintf("^(?:%s)$", *unitWhitelist)) unitBlacklistPattern := regexp.MustCompile(fmt.Sprintf("^(?:%s)$", *unitBlacklist)) return &systemdCollector{ - unitDesc: unitDesc, - systemRunningDesc: systemRunningDesc, - summaryDesc: summaryDesc, - nRestartsDesc: nRestartsDesc, - timerLastTriggerDesc: timerLastTriggerDesc, - unitWhitelistPattern: unitWhitelistPattern, - unitBlacklistPattern: unitBlacklistPattern, + unitDesc: unitDesc, + systemRunningDesc: systemRunningDesc, + summaryDesc: summaryDesc, + nRestartsDesc: nRestartsDesc, + timerLastTriggerDesc: timerLastTriggerDesc, + socketAcceptedConnectionsDesc: socketAcceptedConnectionsDesc, + socketCurrentConnectionsDesc: socketCurrentConnectionsDesc, + unitWhitelistPattern: unitWhitelistPattern, + unitBlacklistPattern: unitBlacklistPattern, }, nil } @@ -96,6 +106,7 @@ func (c *systemdCollector) Update(ch chan<- prometheus.Metric) error { units := filterUnits(allUnits, c.unitWhitelistPattern, c.unitBlacklistPattern) c.collectUnitStatusMetrics(ch, units) c.collectTimers(ch, units) + c.collectSockets(ch, units) systemState, err := c.getSystemState() if err != nil { @@ -125,6 +136,22 @@ func (c *systemdCollector) collectUnitStatusMetrics(ch chan<- prometheus.Metric, } } +func (c *systemdCollector) collectSockets(ch chan<- prometheus.Metric, units []unit) error { + for _, unit := range units { + if !strings.HasSuffix(unit.Name, ".socket") { + continue + } + + ch <- prometheus.MustNewConstMetric( + c.socketAcceptedConnectionsDesc, prometheus.CounterValue, + float64(unit.acceptedConnections), unit.Name) + ch <- prometheus.MustNewConstMetric( + c.socketCurrentConnectionsDesc, prometheus.GaugeValue, + float64(unit.currentConnections), unit.Name) + } + return nil +} + func (c *systemdCollector) collectTimers(ch chan<- prometheus.Metric, units []unit) error { for _, unit := range units { if !strings.HasSuffix(unit.Name, ".timer") { @@ -162,8 +189,10 @@ func (c *systemdCollector) newDbus() (*dbus.Conn, error) { type unit struct { dbus.UnitStatus - lastTriggerUsec uint64 - nRestarts uint32 + lastTriggerUsec uint64 + nRestarts uint32 + acceptedConnections uint32 + currentConnections uint32 } func (c *systemdCollector) getAllUnits() ([]unit, error) { @@ -201,6 +230,22 @@ func (c *systemdCollector) getAllUnits() ([]unit, error) { 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) + } + + 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) + } + unit.currentConnections = currentConnectionCount.Value.Value().(uint32) + + } + result = append(result, unit) } diff --git a/collector/systemd_linux_test.go b/collector/systemd_linux_test.go index a68f6cdc..82237bd9 100644 --- a/collector/systemd_linux_test.go +++ b/collector/systemd_linux_test.go @@ -103,6 +103,7 @@ func TestSystemdCollectorDoesntCrash(t *testing.T) { collector := (c).(*systemdCollector) for _, units := range fixtures { collector.collectUnitStatusMetrics(sink, units) + collector.collectSockets(sink, units) } }