Add max_errors and ability to exit.

integrations/inputunifi/cmd/unifi-poller/README.md

@@ -26,13 +26,15 @@ This daemon polls a Unifi controller at a short interval and stores the collecte
     -j, --dumpjson <filter>
         This is a debug option; use this when you are missing data in your graphs,
         and/or you want to inspect the raw data coming from the controller. The
-        filter only accepts two options: devices or clients. This will print a lot
-        of information. Recommend piping it into a file and/or into jq for better
-        visualization. This requires a valid config file that; one that contains
+        filter accepts three options: devices, clients, other. This will print a
+        lot of information. Recommend piping it into a file and/or into jq for
+        better visualization. This requires a valid config file that contains
         working authentication details for a Unifi Controller. This only dumps
         data for sites listed in the config file. The application exits after
         printing the JSON payload; it does not daemonize or report to InfluxDB
-        with this option.
+        with this option. The `other` option is special. This allows you request
+        any api path. It must be enclosed in quotes with the word other. Example:
+           unifi-poller -j "other /stat/admins"
 
     -h, --help
         Display usage and exit.
@@ -63,6 +65,18 @@ This daemon polls a Unifi controller at a short interval and stores the collecte
         errors will be logged. Using this with debug=true adds line numbers to
         any error logs.
 
+    `max_errors`     default: 0
+        If you restart the UniFI controller, the poller will lose access until
+        it is restarted. Specifying a number greater than -1 for max_errors will
+        cause the poller to exit when it reaches the error count specified.
+        This problematic condition can be triggered by InfluxDB having issues
+        too. Generally only 1 error per interval is created, but if more than one
+        backend is having issues > 1 error could be generated per interval. Once
+        the poller exits, it is expected that something will restart it
+        automatically so it gets back in line; something is usually systemd,
+        docker or launchd. The default setting of 0 will cause an exit after
+        just 1 error. Recommended values are 0-5.
+
     `influx_url`     default: http://127.0.0.1:8086
         This is the URL where the Influx web server is available.
 

integrations/inputunifi/examples/up.conf.example

@@ -18,6 +18,12 @@
 # Recommend using debug with this setting for better error logging.
 #quiet = false
 
+# If the poller experiences an error from the Unifi Controller or from InfluxDB
+# it will exit. If you do not want it to exit, change max_errors to -1. You can
+# adjust the config to tolerate more errors by setting this to a higher value.
+# Recommend setting this between 0 and 5. See man page for more explanation.
+#max_errors = 0
+
 # InfluxDB does not require auth by default, so the user/password are probably unimportant.
 #influx_url = "http://127.0.0.1:8086"
 #influx_user = "unifi"

integrations/inputunifi/pkg/unifi-poller/config.go

@@ -35,6 +35,7 @@ type UnifiPoller struct {
 	DumpJSON   string
 	ShowVer    bool
 	Flag       *pflag.FlagSet
+	errorCount int
 	influx.Client
 	*unifi.Unifi
 	*Config
@@ -50,6 +51,7 @@ type Metrics struct {
 
 // Config represents the data needed to poll a controller and report to influxdb.
 type Config struct {
+	MaxErrors  int      `json:"max_errors,_omitempty" toml:"max_errors,_omitempty" xml:"max_errors" yaml:"max_errors"`
 	Interval   Dur      `json:"interval,_omitempty" toml:"interval,_omitempty" xml:"interval" yaml:"interval"`
 	Debug      bool     `json:"debug" toml:"debug" xml:"debug" yaml:"debug"`
 	Quiet      bool     `json:"quiet,_omitempty" toml:"quiet,_omitempty" xml:"quiet" yaml:"quiet"`

integrations/inputunifi/pkg/unifi-poller/helpers.go

@@ -15,11 +15,13 @@ func hasErr(errs []error) bool {
 	return false
 }
 
-// logErrors writes a slice of errors, with a prefix, to log-out.
-func logErrors(errs []error, prefix string) {
+// LogErrors writes a slice of errors, with a prefix, to log-out.
+// It also incriments the error counter.
+func (u *UnifiPoller) LogErrors(errs []error, prefix string) {
 	for _, err := range errs {
 		if err != nil {
-			log.Println("[ERROR]", prefix+":", err.Error())
+			u.errorCount++
+			log.Printf("[ERROR] (%v/%v) %v: %v", prefix, err.Error(), u.errorCount, u.MaxErrors)
 		}
 	}
 }
@@ -35,8 +37,8 @@ func StringInSlice(str string, slc []string) bool {
 }
 
 // Logf prints a log entry if quiet is false.
-func (c *Config) Logf(m string, v ...interface{}) {
-	if !c.Quiet {
+func (u *UnifiPoller) Logf(m string, v ...interface{}) {
+	if !u.Quiet {
 		log.Printf("[INFO] "+m, v...)
 	}
 }

integrations/inputunifi/pkg/unifi-poller/unifi.go

@@ -37,7 +37,7 @@ FIRST:
 }
 
 // PollController runs forever, polling unifi, and pushing to influx.
-func (u *UnifiPoller) PollController() {
+func (u *UnifiPoller) PollController() error {
 	log.Println("[INFO] Everything checks out! Poller started, interval:", u.Interval.value)
 	ticker := time.NewTicker(u.Interval.value)
 	var err error
@@ -45,28 +45,28 @@ func (u *UnifiPoller) PollController() {
 		m := &Metrics{}
 		// Get the sites we care about.
 		if m.Sites, err = u.GetFilteredSites(); err != nil {
-			logErrors([]error{err}, "uni.GetSites()")
+			u.LogErrors([]error{err}, "unifi.GetSites()")
 		}
 		// Get all the points.
 		if m.Clients, err = u.GetClients(m.Sites); err != nil {
-			logErrors([]error{err}, "uni.GetClients()")
+			u.LogErrors([]error{err}, "unifi.GetClients()")
 		}
 		if m.Devices, err = u.GetDevices(m.Sites); err != nil {
-			logErrors([]error{err}, "uni.GetDevices()")
+			u.LogErrors([]error{err}, "unifi.GetDevices()")
 		}
 
 		// Make a new Points Batcher.
 		m.BatchPoints, err = influx.NewBatchPoints(influx.BatchPointsConfig{Database: u.InfluxDB})
 		if err != nil {
-			logErrors([]error{err}, "influx.NewBatchPoints")
+			u.LogErrors([]error{err}, "influx.NewBatchPoints")
 			continue
 		}
 		// Batch (and send) all the points.
 		if errs := m.SendPoints(); errs != nil && hasErr(errs) {
-			logErrors(errs, "asset.Points()")
+			u.LogErrors(errs, "asset.Points()")
 		}
 		if err := u.Write(m.BatchPoints); err != nil {
-			logErrors([]error{err}, "infdb.Write(bp)")
+			u.LogErrors([]error{err}, "infdb.Write(bp)")
 		}
 
 		// Talk about the data.
@@ -79,7 +79,12 @@ func (u *UnifiPoller) PollController() {
 		u.Logf("Unifi Measurements Recorded. Sites: %d, Clients: %d, "+
 			"Wireless APs: %d, Gateways: %d, Switches: %d, Points: %d, Fields: %d",
 			len(m.Sites), len(m.Clients), len(m.UAPs), len(m.USGs), len(m.USWs), pointcount, fieldcount)
+
+		if u.MaxErrors >= 0 && u.errorCount > u.MaxErrors {
+			return errors.Errorf("reached maximum error count, stopping poller (%d > %d)", u.errorCount, u.MaxErrors)
+		}
 	}
+	return nil
 }
 
 // SendPoints combines all device and client data into influxdb data points.

integrations/inputunifi/pkg/unifi-poller/unifipoller.go

@@ -50,7 +50,7 @@ func (u *UnifiPoller) GetConfig() error {
 	if u.DumpJSON != "" {
 		u.Quiet = true
 	}
-	u.Config.Logf("Loaded Configuration: %s", u.ConfigFile)
+	u.Logf("Loaded Configuration: %s", u.ConfigFile)
 	return nil
 }
 
@@ -71,8 +71,7 @@ func (u *UnifiPoller) Run() (err error) {
 	if err = u.GetInfluxDB(); err != nil {
 		return err
 	}
-	u.PollController()
-	return nil
+	return u.PollController()
 }
 
 // GetInfluxDB returns an influxdb interface.