Added dns duration metric

Author: Stono

.prettierignore

@@ -0,0 +1 @@
+README.md

README.md

@@ -42,9 +42,9 @@ In order to discover other agents, and enrich the agent information with metadat
 
 `kmoncon` agents by default will perform 5 x 4 byte UDP packet tests between every other agent, every 5 seconds. Each test waits for a response from the destination agent. The RTT timeout is 250ms, anything longer than that and we consider the packets lost in the abyss. The metrics output from UDP tests are:
 
-- `GAUGE kconmon_udp_duration_milliseconds`: The total RTT from sending the packet to receiving a response
-- `GAUGE kconmon_udp_duration_variance_milliseconds`: The variance between the slowest and the fastest packet
-- `GAUGE kconmon_udp_loss`: The percentage of requests from the batch that failed
+- `GAUGE   kconmon_udp_duration_milliseconds`: The total RTT from sending the packet to receiving a response
+- `GAUGE   kconmon_udp_duration_variance_milliseconds`: The variance between the slowest and the fastest packet
+- `GAUGE   kconmon_udp_loss`: The percentage of requests from the batch that failed
 - `COUNTER kconmon_udp_results_total`: A Counter of test results, pass and fail
 
 ### TCP Testing
@@ -53,8 +53,8 @@ In order to discover other agents, and enrich the agent information with metadat
 
 The metrics output from TCP tests are:
 
-- `GAUGE kconmon_tcp_connect_milliseconds`: The duration from socket assignment to successful TCP connection
-- `GAUGE kconmon_tcp_duration_milliseconds`: The total RTT of the request
+- `GAUGE   kconmon_tcp_connect_milliseconds`: The duration from socket assignment to successful TCP connection of the last test run
+- `GAUGE   kconmon_tcp_duration_milliseconds`: The total RTT of the request
 - `COUNTER kconmon_tcp_results_total`: A Counter of test results, pass and fail
 
 ### DNS Testing
@@ -63,6 +63,7 @@ The metrics output from TCP tests are:
 
 The metrics output from DNS tests are:
 
+- `GAUGE   kconmon_dns_duration_milliseconds`: The duration of the last test run
 - `COUNTER kconmon_dns_results_total`: A Counter of test results, pass and fail
 
 ## Prometheus Metrics

lib/apps/agent/metrics.ts

@@ -22,6 +22,7 @@ export default class Metrics implements IMetrics {
   private UDPLoss: client.Gauge<string>
 
   private DNS: client.Counter<string>
+  private DNSDuration: client.Gauge<string>
 
   constructor(config: IConfig) {
     client.register.clear()
@@ -38,7 +39,7 @@ export default class Metrics implements IMetrics {
     })
 
     this.UDPDuration = new client.Gauge<string>({
-      help: ' Average duration per packet',
+      help: 'Average duration per packet',
       labelNames: ['source', 'destination', 'source_zone', 'destination_zone'],
       name: `${config.metricsPrefix}_udp_duration_milliseconds`
     })
@@ -61,6 +62,12 @@ export default class Metrics implements IMetrics {
       name: `${config.metricsPrefix}_dns_results_total`
     })
 
+    this.DNSDuration = new client.Gauge<string>({
+      help: 'Total time taken to complete the DNS test',
+      labelNames: ['source', 'source_zone', 'host'],
+      name: `${config.metricsPrefix}_dns_duration_milliseconds`
+    })
+
     this.UDP = new client.Counter<string>({
       help: 'UDP Test Results',
       labelNames: [
@@ -91,6 +98,11 @@ export default class Metrics implements IMetrics {
     this.DNS.labels(source, result.source.zone, result.host, result.result).inc(
       1
     )
+    this.DNSDuration.labels(
+      result.source.nodeName,
+      result.source.zone,
+      result.host
+    ).set(result.duration)
   }
 
   public resetTCPTestResults() {

lib/tester/index.ts

@@ -24,6 +24,7 @@ interface ITestResult {
 export interface IDNSTestResult {
   source: IAgent
   host: string
+  duration: number
   result: 'pass' | 'fail'
 }
 
@@ -111,20 +112,25 @@ export default class Tester implements ITester {
   public async runDNSTests(): Promise<IDNSTestResult[]> {
     const promises = this.config.testConfig.dns.hosts.map(
       async (host): Promise<IDNSTestResult> => {
+        const hrstart = process.hrtime()
         try {
           const result = await this.resolver.resolve4(host)
+          const hrend = process.hrtime(hrstart)
           const mapped: IDNSTestResult = {
             source: this.me,
             host,
+            duration: hrend[1] / 1000000,
             result: result && result.length > 0 ? 'pass' : 'fail'
           }
           this.metrics.handleDNSTestResult(mapped)
           return mapped
         } catch (ex) {
           this.logger.error(`dns test for ${host} failed`, ex)
+          const hrend = process.hrtime(hrstart)
           const mapped: IDNSTestResult = {
             source: this.me,
             host,
+            duration: hrend[1] / 1000000,
             result: 'fail'
           }
           this.metrics.handleDNSTestResult(mapped)