dnsdist: Add Prometheus latency histogram support

PowerDNS · May 28, 2019 · 0957b46 · 0957b46
1 parent 2153346
commit 0957b46
Show file tree

Hide file tree

Showing 4 changed files with 42 additions and 2 deletions.
diff --git a/pdns/dnsdist-web.cc b/pdns/dnsdist-web.cc
@@ -443,6 +443,20 @@ static void connectionThread(int sock, ComboAddress remote)
           output << "\n";
         }
 
+        // Latency histogram buckets
+        uint64_t latency_amounts = g_stats.latency0_1;
+        output << "latency_bucket{le=\"1\"} " << latency_amounts << "\n";
+        latency_amounts += g_stats.latency1_10;
+        output << "latency_bucket{le=\"10\"} " << latency_amounts << "\n";
+        latency_amounts += g_stats.latency10_50;
+        output << "latency_bucket{le=\"50\"} " << latency_amounts << "\n";
+        latency_amounts += g_stats.latency50_100;
+        output << "latency_bucket{le=\"100\"} " << latency_amounts << "\n";
+        latency_amounts += g_stats.latency100_1000;
+        output << "latency_bucket{le=\"1000\"} " << latency_amounts << "\n";
+        latency_amounts += g_stats.latencySlow; // Should be the same as latency_count
+        output << "latency_bucket{le=\"+Inf\"} " << latency_amounts << "\n";
+
         auto states = g_dstates.getLocal();
         const string statesbase = "dnsdist_server_";
 

diff --git a/pdns/dnsdist.cc b/pdns/dnsdist.cc
@@ -200,6 +200,7 @@ void doLatencyStats(double udiff)
   else if(udiff < 100000) ++g_stats.latency50_100;
   else if(udiff < 1000000) ++g_stats.latency100_1000;
   else ++g_stats.latencySlow;
+  g_stats.latencySum += udiff / 1000;
 
   auto doAvg = [](double& var, double n, double weight) {
     var = (weight -1) * var/weight + n/weight;
@@ -2792,3 +2793,8 @@ catch(PDNSException &ae)
   errlog("Fatal pdns error: %s", ae.reason);
   _exit(EXIT_FAILURE);
 }
+
+uint64_t getLatencyCount(const std::string&)
+{
+    return g_stats.responses + g_stats.selfAnswered + g_stats.cacheHits;
+}
diff --git a/pdns/dnsdist.hh b/pdns/dnsdist.hh
@@ -225,6 +225,8 @@ extern GlobalStateHolder<NetmaskTree<DynBlock>> g_dynblockNMG;
 
 extern vector<pair<struct timeval, std::string> > g_confDelta;
 
+extern uint64_t getLatencyCount(const std::string&);
+
 struct DNSDistStats
 {
   using stat_t=std::atomic<uint64_t>; // aww yiss ;-)
@@ -251,7 +253,7 @@ struct DNSDistStats
   stat_t noPolicy{0};
   stat_t cacheHits{0};
   stat_t cacheMisses{0};
-  stat_t latency0_1{0}, latency1_10{0}, latency10_50{0}, latency50_100{0}, latency100_1000{0}, latencySlow{0};
+  stat_t latency0_1{0}, latency1_10{0}, latency10_50{0}, latency50_100{0}, latency100_1000{0}, latencySlow{0}, latencySum{0};
   stat_t securityStatus{0};
 
   double latencyAvg100{0}, latencyAvg1000{0}, latencyAvg10000{0}, latencyAvg1000000{0};
@@ -298,7 +300,10 @@ struct DNSDistStats
     {"fd-usage", getOpenFileDescriptors},
     {"dyn-blocked", &dynBlocked},
     {"dyn-block-nmg-size", [](const std::string&) { return g_dynblockNMG.getLocal()->size(); }},
-    {"security-status", &securityStatus}
+    {"security-status", &securityStatus},
+    // Latency histogram
+    {"latency-sum", &latencySum},
+    {"latency-count", getLatencyCount},
   };
 };
 
@@ -390,6 +395,9 @@ struct MetricDefinitionStorage {
     { "dyn-blocked",            MetricDefinition(PrometheusMetricType::counter, "Number of queries dropped because of a dynamic block")},
     { "dyn-block-nmg-size",     MetricDefinition(PrometheusMetricType::gauge,   "Number of dynamic blocks entries") },
     { "security-status",        MetricDefinition(PrometheusMetricType::gauge,   "Security status of this software. 0=unknown, 1=OK, 2=upgrade recommended, 3=upgrade mandatory") },
+    // Latency histogram
+    { "latency-sum",            MetricDefinition(PrometheusMetricType::counter, "Total response time in milliseconds")},
+    { "latency-count",          MetricDefinition(PrometheusMetricType::counter, "Number of queries contributing to response time histogram")},
   };
 };
 

diff --git a/pdns/dnsdistdist/docs/statistics.rst b/pdns/dnsdistdist/docs/statistics.rst
@@ -99,6 +99,18 @@ latency-slow
 ------------
 Number of queries answered in more than 1 second.
 
+latency-sum
+------------
+Total response time in milliseconds.
+
+latency-count
+------------
+Number of queries contributing to response time histogram
+
+latency-bucket
+------------
+Number of queries contributing to response time histogram per latency bucket
+
 latency0-1
 ----------
 Number of queries answered in less than 1 ms.