Add basic Prometheus support and metrics (#67)

* Add basic Prometheus support and metrics

* Remove human metrics

* Pull AppendMetric outside

* Fix typo

src/facade/dragonfly_listener.cc

@@ -127,7 +127,6 @@ Listener::Listener(Protocol protocol, ServiceInterface* si) : service_(si), prot
   }
   http_base_.reset(new HttpListener<>);
   http_base_->set_resource_prefix("https://romange.s3.eu-west-1.amazonaws.com/static");
-  http_base_->enable_metrics();
   si->ConfigureHttpHandlers(http_base_.get());
 }
 

src/server/server_family.cc

@@ -266,8 +266,45 @@ error_code ServerFamily::LoadRdb(const std::string& rdb_file) {
   return ec;
 }
 
-void ServerFamily::ConfigureMetrics(util::HttpListenerBase* http_base) {
+void AppendMetric(http::StringResponse* resp, absl::AlphaNum name, absl::AlphaNum val) {
+  /**
+   * Gets a metric name and a value and write it using Prometheus format.
+   *
+   * TODO:
+   *  1. add metrics descriptions.
+   *  2. support other types of metrics, not just gauge :)
+   */
 
+  const auto full_name = StrCat("dragonfly_", name);
+  absl::StrAppend(&resp->body(), "# HELP ", full_name, " ", name, "\n");
+  absl::StrAppend(&resp->body(), "# TYPE ", full_name, " gauge\n");
+  absl::StrAppend(&resp->body(), full_name, " ", val, "\n");
+}
+
+void ServerFamily::ConfigureMetrics(util::HttpListenerBase* http_base) {
+  // The naming of the metrics should be compatible with redis_exporter, see https://github.com/oliver006/redis_exporter/blob/master/exporter/exporter.go#L111
+
+  auto cb = [this](const http::QueryArgs& args, HttpContext* send) {
+    http::StringResponse resp = http::MakeStringResponse(boost::beast::http::status::ok);
+    Metrics m = this->GetMetrics();
+
+    // Server metrics
+    AppendMetric(&resp, "uptime_in_seconds", m.uptime);
+
+    // Clients metrics
+    AppendMetric(&resp, "connected_clients", m.conn_stats.num_conns);
+    AppendMetric(&resp, "client_read_buf_capacity", m.conn_stats.read_buf_capacity);
+    AppendMetric(&resp, "blocked_clients", m.conn_stats.num_blocked_clients);
+
+    // Memory metrics
+    AppendMetric(&resp, "used_memory", m.heap_used_bytes);
+    AppendMetric(&resp, "used_memory_peak", used_mem_peak.load(memory_order_relaxed));
+    AppendMetric(&resp, "comitted_memory", _mi_stats_main.committed.current);
+
+    return send->Invoke(std::move(resp));
+  };
+
+  http_base->RegisterCb("/metrics", cb);
 }
 
 void ServerFamily::StatsMC(std::string_view section, facade::ConnectionContext* cntx) {
@@ -279,7 +316,6 @@ void ServerFamily::StatsMC(std::string_view section, facade::ConnectionContext*
 #define ADD_LINE(name, val) absl::StrAppend(&info, "STAT " #name " ", val, "\r\n")
 
   time_t now = time(NULL);
-  size_t uptime = now - start_time_;
   struct rusage ru;
   getrusage(RUSAGE_SELF, &ru);
 
@@ -293,7 +329,7 @@ void ServerFamily::StatsMC(std::string_view section, facade::ConnectionContext*
   Metrics m = GetMetrics();
 
   ADD_LINE(pid, getpid());
-  ADD_LINE(uptime, uptime);
+  ADD_LINE(uptime, m.uptime);
   ADD_LINE(time, now);
   ADD_LINE(version, kGitTag);
   ADD_LINE(libevent, "iouring");
@@ -597,6 +633,7 @@ Metrics ServerFamily::GetMetrics() const {
 
     lock_guard<fibers::mutex> lk(mu);
 
+    result.uptime = time(NULL) - this->start_time_;
     result.conn_stats += ss->connection_stats;
     result.qps += uint64_t(ss->MovingSum6());
 
@@ -648,6 +685,7 @@ void ServerFamily::Info(CmdArgList args, ConnectionContext* cntx) {
   };
 
 #define ADD_HEADER(x) absl::StrAppend(&info, x "\r\n")
+  Metrics m = GetMetrics();
 
   if (should_enter("SERVER")) {
     ADD_HEADER("# Server");
@@ -658,12 +696,11 @@ void ServerFamily::Info(CmdArgList args, ConnectionContext* cntx) {
     append("multiplexing_api", "iouring");
     append("tcp_port", GetFlag(FLAGS_port));
 
-    size_t uptime = time(NULL) - start_time_;
+    size_t uptime = m.uptime;
     append("uptime_in_seconds", uptime);
     append("uptime_in_days", uptime / (3600 * 24));
   }
 
-  Metrics m = GetMetrics();
   auto sdata_res = io::ReadStatusInfo();
 
   DbStats total;

src/server/server_family.h

@@ -29,6 +29,7 @@ struct Metrics {
   TieredStats tiered_stats;
   EngineShard::Stats shard_stats;
 
+  size_t uptime = 0;
   size_t qps = 0;
   size_t heap_used_bytes = 0;
   size_t heap_comitted_bytes = 0;