From f55c0a5292757ff08e1fa38a91cfc43ee666db10 Mon Sep 17 00:00:00 2001
From: Tim Wojtulewicz <timwoj@gmail.com>
Date: Fri, 31 May 2024 17:02:06 -0700
Subject: [PATCH] Fix race condition by pre-building the cluster json data for
 services.json

---
 src/telemetry/Manager.cc | 11 +++++++----
 src/telemetry/Manager.h  |  8 ++++----
 2 files changed, 11 insertions(+), 8 deletions(-)

diff --git a/src/telemetry/Manager.cc b/src/telemetry/Manager.cc
index 8b0a037e74..d414b20667 100644
--- a/src/telemetry/Manager.cc
+++ b/src/telemetry/Manager.cc
@@ -44,6 +44,8 @@ void Manager::InitPostScript() {
             static auto manager_type = node_type_type->Lookup("Cluster", "MANAGER");
 
             if ( local_node_type == manager_type ) {
+                BuildClusterJson();
+
                 callbacks = new CivetCallbacks();
                 callbacks->begin_request = [](struct mg_connection* conn) -> int {
                     // Handle the services.json request ourselves by building up a response based on
@@ -414,7 +416,7 @@ ValPtr Manager::CollectHistogramMetrics(std::string_view prefix_pattern, std::st
     return ret_val;
 }
 
-std::string Manager::GetClusterJson() const {
+void Manager::BuildClusterJson() {
     rapidjson::StringBuffer buffer;
     json::detail::NullDoubleWriter writer(buffer);
 
@@ -423,8 +425,9 @@ std::string Manager::GetClusterJson() const {
 
     writer.Key("targets");
     writer.StartArray();
-    auto cluster_nodes = id::find_val("Cluster::nodes")->AsTableVal()->ToMap();
-    for ( const auto& [idx, value] : cluster_nodes ) {
+    auto& node_val = id::find_val("Cluster::nodes");
+    auto node_map = node_val->AsTableVal()->ToMap();
+    for ( const auto& [idx, value] : node_map ) {
         auto node = value->AsRecordVal();
         auto ip = node->GetField<AddrVal>("ip");
         auto port = node->GetField<PortVal>("metrics_port");
@@ -440,7 +443,7 @@ std::string Manager::GetClusterJson() const {
     writer.EndObject();
     writer.EndArray();
 
-    return buffer.GetString();
+    cluster_json = buffer.GetString();
 }
 
 CounterFamilyPtr Manager::CounterFamily(std::string_view prefix, std::string_view name,
diff --git a/src/telemetry/Manager.h b/src/telemetry/Manager.h
index 8a1deb5fc4..f81a39a6d5 100644
--- a/src/telemetry/Manager.h
+++ b/src/telemetry/Manager.h
@@ -200,7 +200,7 @@ public:
      * @return A JSON description of the cluster configuration for reporting
      * to Prometheus for service discovery requests.
      */
-    std::string GetClusterJson() const;
+    std::string GetClusterJson() const { return cluster_json; }
 
     /**
      * @return The pointer to the prometheus-cpp registry used by the telemetry
@@ -230,6 +230,7 @@ protected:
 
 private:
     RecordValPtr GetMetricOptsRecord(const prometheus::MetricFamily& metric_family);
+    void BuildClusterJson();
 
     std::map<std::string, std::shared_ptr<MetricFamily>> families;
     std::map<std::string, RecordValPtr> opts_records;
@@ -242,11 +243,10 @@ private:
     GaugePtr cpu_gauge;
     GaugePtr fds_gauge;
 
-    std::string endpoint_name;
-    std::vector<std::string> export_prefixes;
-
     std::shared_ptr<prometheus::Registry> prometheus_registry;
     std::unique_ptr<prometheus::Exposer> prometheus_exposer;
+
+    std::string cluster_json;
 };
 
 } // namespace zeek::telemetry