feat(inputs.dns_query): Add IP field(s) (#12519)

plugins/inputs/dns_query/README.md

@@ -33,8 +33,14 @@ See the [CONFIGURATION.md][CONFIGURATION.md] for more details.
   ## Dns server port.
   # port = 53
 
-  ## Query timeout in seconds.
-  # timeout = 2
+  ## Query timeout
+  # timeout = "2s"
+
+  ## Include the specified additional properties in the resulting metric.
+  ## The following values are supported:
+  ##    "first_ip" -- return IP of the first A and AAAA answer
+  ##    "all_ips"  -- return IPs of all A and AAAA answers
+  # include_fields = []
 ```
 
 ## Metrics

plugins/inputs/dns_query/dns_query.go

@@ -12,6 +12,7 @@ import (
 	"github.com/miekg/dns"
 
 	"github.com/influxdata/telegraf"
+	"github.com/influxdata/telegraf/config"
 	"github.com/influxdata/telegraf/plugins/inputs"
 )
 
@@ -27,117 +28,147 @@ const (
 )
 
 type DNSQuery struct {
-	// Domains or subdomains to query
-	Domains []string
+	Domains       []string        `toml:"domains"`
+	Network       string          `toml:"network"`
+	Servers       []string        `toml:"servers"`
+	RecordType    string          `toml:"record_type"`
+	Port          int             `toml:"port"`
+	Timeout       config.Duration `toml:"timeout"`
+	IncludeFields []string        `toml:"include_fields"`
+
+	fieldEnabled map[string]bool
+}
 
-	// Network protocol name
-	Network string
+func (*DNSQuery) SampleConfig() string {
+	return sampleConfig
+}
 
-	// Server to query
-	Servers []string
+func (d *DNSQuery) Init() error {
+	// Convert the included fields into a lookup-table
+	d.fieldEnabled = make(map[string]bool, len(d.IncludeFields))
+	for _, f := range d.IncludeFields {
+		switch f {
+		case "first_ip", "all_ips":
+		default:
+			return fmt.Errorf("invalid field %q included", f)
+		}
+		d.fieldEnabled[f] = true
+	}
 
-	// Record type
-	RecordType string `toml:"record_type"`
+	// Set defaults
+	if d.Network == "" {
+		d.Network = "udp"
+	}
 
-	// DNS server port number
-	Port int
+	if d.RecordType == "" {
+		d.RecordType = "NS"
+	}
 
-	// Dns query timeout in seconds. 0 means no timeout
-	Timeout int
-}
+	if len(d.Domains) == 0 {
+		d.Domains = []string{"."}
+		d.RecordType = "NS"
+	}
 
-func (*DNSQuery) SampleConfig() string {
-	return sampleConfig
+	if d.Port < 1 {
+		d.Port = 53
+	}
+
+	return nil
 }
 
 func (d *DNSQuery) Gather(acc telegraf.Accumulator) error {
 	var wg sync.WaitGroup
-	d.setDefaultValues()
 
 	for _, domain := range d.Domains {
 		for _, server := range d.Servers {
 			wg.Add(1)
 			go func(domain, server string) {
-				fields := make(map[string]interface{}, 2)
-				tags := map[string]string{
-					"server":      server,
-					"domain":      domain,
-					"record_type": d.RecordType,
-				}
+				defer wg.Done()
 
-				dnsQueryTime, rcode, err := d.getDNSQueryTime(domain, server)
-				if rcode >= 0 {
-					tags["rcode"] = dns.RcodeToString[rcode]
-					fields["rcode_value"] = rcode
-				}
-				if err == nil {
-					setResult(Success, fields, tags)
-					fields["query_time_ms"] = dnsQueryTime
-				} else if opErr, ok := err.(*net.OpError); ok && opErr.Timeout() {
-					setResult(Timeout, fields, tags)
-				} else if err != nil {
-					setResult(Error, fields, tags)
-					acc.AddError(err)
+				fields, tags, err := d.query(domain, server)
+				if err != nil {
+					if opErr, ok := err.(*net.OpError); !ok || !opErr.Timeout() {
+						acc.AddError(err)
+					}
 				}
-
 				acc.AddFields("dns_query", fields, tags)
-
-				wg.Done()
 			}(domain, server)
 		}
 	}
-
 	wg.Wait()
+
 	return nil
 }
 
-func (d *DNSQuery) setDefaultValues() {
-	if d.Network == "" {
-		d.Network = "udp"
+func (d *DNSQuery) query(domain string, server string) (map[string]interface{}, map[string]string, error) {
+	tags := map[string]string{
+		"server":      server,
+		"domain":      domain,
+		"record_type": d.RecordType,
+		"result":      "error",
 	}
 
-	if len(d.RecordType) == 0 {
-		d.RecordType = "NS"
+	fields := map[string]interface{}{
+		"query_time_ms": float64(0),
+		"result_code":   uint64(Error),
 	}
 
-	if len(d.Domains) == 0 {
-		d.Domains = []string{"."}
-		d.RecordType = "NS"
+	c := dns.Client{
+		ReadTimeout: time.Duration(d.Timeout),
+		Net:         d.Network,
 	}
 
-	if d.Port == 0 {
-		d.Port = 53
+	recordType, err := d.parseRecordType()
+	if err != nil {
+		return fields, tags, err
 	}
 
-	if d.Timeout == 0 {
-		d.Timeout = 2
-	}
-}
+	var msg dns.Msg
+	msg.SetQuestion(dns.Fqdn(domain), recordType)
+	msg.RecursionDesired = true
 
-func (d *DNSQuery) getDNSQueryTime(domain string, server string) (float64, int, error) {
-	dnsQueryTime := float64(0)
+	addr := net.JoinHostPort(server, strconv.Itoa(d.Port))
+	r, rtt, err := c.Exchange(&msg, addr)
+	if err != nil {
+		if opErr, ok := err.(*net.OpError); ok && opErr.Timeout() {
+			tags["result"] = "timeout"
+			fields["result_code"] = uint64(Timeout)
+			return fields, tags, err
+		}
+		return fields, tags, err
+	}
 
-	c := new(dns.Client)
-	c.ReadTimeout = time.Duration(d.Timeout) * time.Second
-	c.Net = d.Network
+	// Fill valid fields
+	tags["rcode"] = dns.RcodeToString[r.Rcode]
+	fields["rcode_value"] = r.Rcode
+	fields["query_time_ms"] = float64(rtt.Nanoseconds()) / 1e6
 
-	m := new(dns.Msg)
-	recordType, err := d.parseRecordType()
-	if err != nil {
-		return dnsQueryTime, -1, err
+	// Handle the failure case
+	if r.Rcode != dns.RcodeSuccess {
+		return fields, tags, fmt.Errorf("invalid answer (%s) from %s after %s query for %s", dns.RcodeToString[r.Rcode], server, d.RecordType, domain)
 	}
-	m.SetQuestion(dns.Fqdn(domain), recordType)
-	m.RecursionDesired = true
 
-	r, rtt, err := c.Exchange(m, net.JoinHostPort(server, strconv.Itoa(d.Port)))
-	if err != nil {
-		return dnsQueryTime, -1, err
+	// Success
+	tags["result"] = "success"
+	fields["result_code"] = uint64(Success)
+
+	if d.fieldEnabled["first_ip"] {
+		for _, record := range r.Answer {
+			if ip, found := extractIP(record); found {
+				fields["ip"] = ip
+				break
+			}
+		}
 	}
-	if r.Rcode != dns.RcodeSuccess {
-		return dnsQueryTime, r.Rcode, fmt.Errorf("Invalid answer (%s) from %s after %s query for %s", dns.RcodeToString[r.Rcode], server, d.RecordType, domain)
+	if d.fieldEnabled["all_ips"] {
+		for i, record := range r.Answer {
+			if ip, found := extractIP(record); found {
+				fields["ip_"+strconv.Itoa(i)] = ip
+			}
+		}
 	}
-	dnsQueryTime = float64(rtt.Nanoseconds()) / 1e6
-	return dnsQueryTime, r.Rcode, nil
+
+	return fields, tags, nil
 }
 
 func (d *DNSQuery) parseRecordType() (uint16, error) {
@@ -168,29 +199,26 @@ func (d *DNSQuery) parseRecordType() (uint16, error) {
 	case "TXT":
 		recordType = dns.TypeTXT
 	default:
-		err = fmt.Errorf("Record type %s not recognized", d.RecordType)
+		err = fmt.Errorf("record type %s not recognized", d.RecordType)
 	}
 
 	return recordType, err
 }
 
-func setResult(result ResultType, fields map[string]interface{}, tags map[string]string) {
-	var tag string
-	switch result {
-	case Success:
-		tag = "success"
-	case Timeout:
-		tag = "timeout"
-	case Error:
-		tag = "error"
+func extractIP(record dns.RR) (string, bool) {
+	if r, ok := record.(*dns.A); ok {
+		return r.A.String(), true
 	}
-
-	tags["result"] = tag
-	fields["result_code"] = uint64(result)
+	if r, ok := record.(*dns.AAAA); ok {
+		return r.AAAA.String(), true
+	}
+	return "", false
 }
 
 func init() {
 	inputs.Add("dns_query", func() telegraf.Input {
-		return &DNSQuery{}
+		return &DNSQuery{
+			Timeout: config.Duration(2 * time.Second),
+		}
 	})
 }