Merge pull request #174 from MalloZup/sbd-timeouts

Implement SBD watchdog and msgwait metrics

Author: stefanotorresi

collector/sbd/sbd.go

@@ -6,6 +6,7 @@ import (
 	"os"
 	"os/exec"
 	"regexp"
+	"strconv"
 	"strings"
 
 	"github.com/pkg/errors"
@@ -20,6 +21,7 @@ const subsystem = "sbd"
 const SBD_STATUS_UNHEALTHY = "unhealthy"
 const SBD_STATUS_HEALTHY = "healthy"
 
+// NewCollector create a new sbd collector
 func NewCollector(sbdPath string, sbdConfigPath string) (*sbdCollector, error) {
 	err := checkArguments(sbdPath, sbdConfigPath)
 	if err != nil {
@@ -33,6 +35,7 @@ func NewCollector(sbdPath string, sbdConfigPath string) (*sbdCollector, error) {
 	}
 
 	c.SetDescriptor("devices", "SBD devices; one line per device", []string{"device", "status"})
+	c.SetDescriptor("timeouts", "SBD timeouts for each device and type", []string{"device", "type"})
 
 	return c, nil
 }
@@ -68,6 +71,15 @@ func (c *sbdCollector) CollectWithError(ch chan<- prometheus.Metric) error {
 		ch <- c.MakeGaugeMetric("devices", 1, sbdDev, sbdStatus)
 	}
 
+	sbdWatchdogs, sbdMsgWaits := c.getSbdTimeouts(sbdDevices)
+	for sbdDev, sbdWatchdog := range sbdWatchdogs {
+		ch <- c.MakeGaugeMetric("timeouts", sbdWatchdog, sbdDev, "watchdog")
+	}
+
+	for sbdDev, sbdMsgWait := range sbdMsgWaits {
+		ch <- c.MakeGaugeMetric("timeouts", sbdMsgWait, sbdDev, "msgwait")
+	}
+
 	return nil
 }
 
@@ -132,3 +144,39 @@ func (c *sbdCollector) getSbdDeviceStatuses(sbdDevices []string) map[string]stri
 
 	return sbdStatuses
 }
+
+// for each sbd device, extract the watchdog and msgwait timeout via regex
+func (c *sbdCollector) getSbdTimeouts(sbdDevices []string) (map[string]float64, map[string]float64) {
+	sbdWatchdogs := make(map[string]float64)
+	sbdMsgWaits := make(map[string]float64)
+	for _, sbdDev := range sbdDevices {
+		sbdDump, _ := exec.Command(c.sbdPath, "-d", sbdDev, "dump").Output()
+
+		regexW := regexp.MustCompile(`Timeout \(msgwait\)  *: \d+`)
+		regex := regexp.MustCompile(`Timeout \(watchdog\)  *: \d+`)
+
+		msgWaitLine := regexW.FindStringSubmatch(string(sbdDump))
+		watchdogLine := regex.FindStringSubmatch(string(sbdDump))
+
+		if watchdogLine == nil || msgWaitLine == nil {
+			continue
+		}
+
+		// get the timeout from the line
+		regexNumber := regexp.MustCompile(`\d+`)
+		watchdogTimeout := regexNumber.FindString(string(watchdogLine[0]))
+		msgWaitTimeout := regexNumber.FindString(string(msgWaitLine[0]))
+
+		// map the timeout to the device
+		if s, err := strconv.ParseFloat(watchdogTimeout, 64); err == nil {
+			sbdWatchdogs[sbdDev] = s
+		}
+
+		// map the timeout to the device
+		if s, err := strconv.ParseFloat(msgWaitTimeout, 64); err == nil {
+			sbdMsgWaits[sbdDev] = s
+		}
+
+	}
+	return sbdWatchdogs, sbdMsgWaits
+}

collector/sbd/sbd_test.go

@@ -214,6 +214,13 @@ func TestNewSbdCollectorChecksSbdExecutableBits(t *testing.T) {
 }
 
 func TestSBDCollector(t *testing.T) {
-	collector, _ := NewCollector("../../test/fake_sbd.sh", "../../test/fake_sbdconfig")
+	collector, _ := NewCollector("../../test/fake_sbd_dump.sh", "../../test/fake_sbdconfig")
+	assertcustom.Metrics(t, collector, "sbd.metrics")
+}
+
+func TestWatchdog(t *testing.T) {
+	collector, err := NewCollector("../../test/fake_sbd_dump.sh", "../../test/fake_sbdconfig")
+
+	assert.Nil(t, err)
 	assertcustom.Metrics(t, collector, "sbd.metrics")
 }

doc/metrics.md

@@ -198,7 +198,8 @@ The status of each Corosync ring; `1` means healthy, `0` means faulty.
 The SBD subsystems collect devices stats by parsing its configuration and the output of `sbd --dump`.
 
 0. [Sample](../test/sbd.metrics)
-2. [`ha_cluster_sbd_devices`](#ha_cluster_sbd_devices)
+1. [`ha_cluster_sbd_devices`](#ha_cluster_sbd_devices)
+2. [`ha_cluster_sbd_timeouts`](#ha_cluster_sbd_timeouts)
 
 ### `ha_cluster_sbd_devices`
 
@@ -214,6 +215,18 @@ Either the value is `1`, or the line is absent altogether.
 
 The total number of lines for this metric will be the cardinality of `device`.
 
+### `ha_cluster_sbd_timeouts`
+
+#### Description
+
+The SBD timeouts pro SBD device
+Value is an integer expessing the timeout
+
+#### Labels
+
+- `device`: the path of the SBD device
+- `type`:  either `watchdog` or `msgwait`
+
 
 ## DRBD
 

test/fake_sbd_dump.sh

@@ -0,0 +1,14 @@
+#!/usr/bin/env bash
+
+cat <<EOF
+==Dumping header on disk /dev/vdc
+Header version     : 2.1
+UUID               : 1ed3171d-066d-47ca-8f76-aec25d9efed4
+Number of slots    : 255
+Sector size        : 512
+Timeout (watchdog) : 9
+Timeout (allocate) : 2
+Timeout (loop)     : 1
+Timeout (msgwait)  : 10
+==Header on disk /dev/vdc is dumped
+EOF

test/sbd.metrics

@@ -1,4 +1,10 @@
 # HELP ha_cluster_sbd_devices SBD devices; one line per device
 # TYPE ha_cluster_sbd_devices gauge
 ha_cluster_sbd_devices{device="/dev/vdc",status="healthy"} 1
-ha_cluster_sbd_devices{device="/dev/vdd",status="unhealthy"} 1
+ha_cluster_sbd_devices{device="/dev/vdd",status="healthy"} 1
+# HELP ha_cluster_sbd_timeouts SBD timeouts for each device and type
+# TYPE ha_cluster_sbd_timeouts gauge
+ha_cluster_sbd_timeouts{device="/dev/vdc",type="msgwait"} 10
+ha_cluster_sbd_timeouts{device="/dev/vdc",type="watchdog"} 9
+ha_cluster_sbd_timeouts{device="/dev/vdd",type="msgwait"} 10
+ha_cluster_sbd_timeouts{device="/dev/vdd",type="watchdog"} 9