Merge pull request redpanda-data#24023 from michael-redpanda/telemetr…

…y/core-8141/add-host-name

[CORE-8141] Add host information to metrics report

src/v/cluster/metrics_reporter.cc

@@ -44,7 +44,9 @@
 #include <seastar/core/coroutine.hh>
 #include <seastar/core/lowres_clock.hh>
 #include <seastar/core/shared_ptr.hh>
+#include <seastar/net/dns.hh>
 #include <seastar/net/tls.hh>
+#include <seastar/util/defer.hh>
 
 #include <absl/algorithm/container.h>
 #include <absl/container/node_hash_map.h>
@@ -54,9 +56,38 @@
 #include <boost/uuid/uuid_generators.hpp>
 #include <boost/uuid/uuid_io.hpp>
 #include <fmt/core.h>
+#include <sys/socket.h>
 
+#include <climits>
+#include <netdb.h>
 #include <stdexcept>
 
+namespace {
+ss::sstring get_hostname() {
+    std::array<char, HOST_NAME_MAX> hostname{};
+    if (::gethostname(hostname.data(), hostname.size()) != 0) {
+        return {};
+    }
+
+    return hostname.data();
+}
+
+ss::sstring get_domainname() {
+    std::array<char, HOST_NAME_MAX> domainname{};
+    if (::getdomainname(domainname.data(), domainname.size()) != 0) {
+        return {};
+    }
+
+    return domainname.data();
+}
+
+ss::future<std::vector<ss::sstring>> get_fqdns(std::string_view hostname) {
+    ss::net::dns_resolver resolver;
+    auto hostent = co_await resolver.get_host_by_name(hostname.data());
+    co_return hostent.names;
+}
+} // namespace
+
 namespace cluster {
 
 namespace details {
@@ -218,6 +249,14 @@ metrics_reporter::build_metrics_snapshot() {
         }
 
         metrics.uptime_ms = report->local_state.uptime / 1ms;
+        auto& advertised_listeners
+          = nm->get().broker.kafka_advertised_listeners();
+        metrics.advertised_listeners.reserve(advertised_listeners.size());
+        std::transform(
+          advertised_listeners.begin(),
+          advertised_listeners.end(),
+          std::back_inserter(metrics.advertised_listeners),
+          [](const model::broker_endpoint& ep) { return ep.address; });
     }
     auto& topics = _topics.local().topics_map();
     snapshot.topic_count = 0;
@@ -277,6 +316,10 @@ metrics_reporter::build_metrics_snapshot() {
 
     snapshot.enterprise_features.emplace(std::move(feature_report));
 
+    snapshot.host_name = get_hostname();
+    snapshot.domain_name = get_domainname();
+    snapshot.fqdns = co_await get_fqdns(snapshot.host_name);
+
     co_return snapshot;
 }
 
@@ -566,6 +609,15 @@ void rjson_serialize(
         w.EndArray();
     }
 
+    w.Key("hostname");
+    w.String(snapshot.host_name);
+
+    w.Key("domainname");
+    w.String(snapshot.domain_name);
+
+    w.Key("fqdns");
+    rjson_serialize(w, snapshot.fqdns);
+
     w.EndObject();
 }
 
@@ -602,6 +654,8 @@ void rjson_serialize(
         rjson_serialize(w, d);
     }
     w.EndArray();
+    w.Key("kafka_advertised_listeners");
+    rjson_serialize(w, nm.advertised_listeners);
 
     w.EndObject();
 }

src/v/cluster/metrics_reporter.h

@@ -20,6 +20,7 @@
 #include "model/metadata.h"
 #include "security/fwd.h"
 #include "utils/prefix_logger.h"
+#include "utils/unresolved_address.h"
 
 #include <seastar/core/abort_source.hh>
 #include <seastar/core/gate.hh>
@@ -58,6 +59,7 @@ class metrics_reporter {
         cluster_version logical_version{invalid_version};
         std::vector<node_disk_space> disks;
         uint64_t uptime_ms{0};
+        std::vector<net::unresolved_address> advertised_listeners;
     };
 
     struct metrics_snapshot {
@@ -85,6 +87,10 @@ class metrics_reporter {
         bool has_valid_license{false};
 
         std::optional<features::enterprise_feature_report> enterprise_features;
+
+        ss::sstring host_name;
+        ss::sstring domain_name;
+        std::vector<ss::sstring> fqdns;
     };
     static constexpr ss::shard_id shard = 0;
 

tests/rptest/tests/metrics_reporter_test.py

@@ -138,6 +138,9 @@ def assert_fields_are_the_same(metadata, field):
         assert_fields_are_the_same(metadata, 'has_valid_license')
         assert_fields_are_the_same(metadata, 'has_enterprise_features')
         assert_fields_are_the_same(metadata, 'enterprise_features')
+        assert_fields_are_the_same(metadata, 'hostname')
+        assert_fields_are_the_same(metadata, 'domainname')
+        assert_fields_are_the_same(metadata, 'fqdns')
         # get the last report
         last = metadata.pop()
         assert last['topic_count'] == total_topics
@@ -154,6 +157,9 @@ def assert_fields_are_the_same(metadata, field):
         assert 'has_enterprise_features' in last
         assert 'enterprise_features' in last
         assert type(last['enterprise_features']) == list
+        assert 'hostname' in last
+        assert 'domainname' in last
+        assert 'fqdns' in last
         nodes_meta = last['nodes']
 
         assert len(last['nodes']) == len(self.redpanda.nodes)
@@ -165,6 +171,7 @@ def assert_fields_are_the_same(metadata, field):
         assert all('uptime_ms' in n for n in nodes_meta)
         assert all('is_alive' in n for n in nodes_meta)
         assert all('disks' in n for n in nodes_meta)
+        assert all('kafka_advertised_listeners' in n for n in nodes_meta)
 
         # Check cluster UUID and creation time survive a restart
         for n in self.redpanda.nodes: