Add new StatsHouse metrics (#1115)

This commit adds following stats:
* kphp_slow_rpc_response
* kphp_slow_job_worker_response

server/php-queries.h

@@ -77,6 +77,24 @@ struct net_query_t {
   std::variant<net_queries_data::rpc_send, database_drivers::Request *, std::reference_wrapper<const curl_async::CurlRequest>> data;
 };
 
+namespace slow_net_event_stats {
+
+struct slow_rpc_response_stats final {
+  const char *tl_function_name{nullptr};
+  int32_t actor_or_port{};
+  double response_time{};
+  bool is_error{};
+};
+
+struct slow_job_worker_response_stats final {
+  const char *class_name{nullptr};
+  double response_time{};
+};
+
+using stats_t = std::variant<slow_rpc_response_stats, slow_job_worker_response_stats>;
+
+}; // namespace slow_net_event_stats
+
 #pragma pack(push, 4)
 
 /***

server/php-runner.cpp

@@ -4,38 +4,40 @@
 
 #include "server/php-runner.h"
 
-#include <array>
 #include <cassert>
 #include <cerrno>
 #include <cstdlib>
 #include <cstring>
-#include <exception>
-#include <utility>
 #include <sys/mman.h>
 #include <sys/time.h>
 #include <unistd.h>
+#include <variant>
 
 #include "common/fast-backtrace.h"
 #include "common/kernel-version.h"
 #include "common/kprintf.h"
+#include "common/precise-time.h"
 #include "common/wrappers/memory-utils.h"
-#include "net/net-connections.h"
-
+#include "common/wrappers/overloaded.h"
 #include "runtime/allocator.h"
 #include "runtime/critical_section.h"
 #include "runtime/curl.h"
 #include "runtime/exception.h"
 #include "runtime/interface.h"
+#include "runtime/job-workers/processing-jobs.h"
 #include "runtime/kphp_tracing.h"
 #include "runtime/oom_handler.h"
-#include "runtime/profiler.h"
 #include "runtime/php_assert.h"
+#include "runtime/profiler.h"
+#include "runtime/rpc.h"
+#include "runtime/tl/tl_magics_decoding.h"
 #include "server/json-logger.h"
 #include "server/php-engine-vars.h"
 #include "server/php-queries.h"
 #include "server/server-log.h"
 #include "server/server-stats.h"
 #include "server/signal-handlers.h"
+#include "server/statshouse/statshouse-manager.h"
 
 DEFINE_VERBOSITY(php_runner);
 
@@ -50,6 +52,32 @@ namespace {
 
 [[maybe_unused]] const void *main_thread_stack = nullptr;
 [[maybe_unused]] size_t main_thread_stacksize = 0;
+
+void send_slow_net_event_stats(const net_event_t &event, double time_sec) noexcept {
+  std::visit(overloaded{
+               [&event, time_sec](const net_events_data::rpc_answer &) noexcept {
+                 const auto *rpc_req = get_rpc_request(event.slot_id);
+                 StatsHouseManager::get().add_slow_net_event_stats(
+                   slow_net_event_stats::slow_rpc_response_stats{tl_magic_convert_to_name(rpc_req->function_magic), rpc_req->actor_or_port, time_sec, false});
+               },
+               [&event, time_sec](const net_events_data::rpc_error &) noexcept {
+                 const auto *rpc_req = get_rpc_request(event.slot_id);
+                 StatsHouseManager::get().add_slow_net_event_stats(
+                   slow_net_event_stats::slow_rpc_response_stats{tl_magic_convert_to_name(rpc_req->function_magic), rpc_req->actor_or_port, time_sec, true});
+               },
+               [time_sec](const net_events_data::job_worker_answer &jw_answer) noexcept {
+                 if (jw_answer.job_result != nullptr) {
+                   StatsHouseManager::get().add_slow_net_event_stats(
+                     slow_net_event_stats::slow_job_worker_response_stats{jw_answer.job_result->response.get_class(), time_sec});
+                 }
+               },
+               [](const database_drivers::Response *) {},
+               [](const curl_async::CurlResponse *) {},
+
+             },
+             event.data);
+}
+
 } // namespace
 
 void PhpScript::error(const char *error_message, script_error_t error_type) noexcept {
@@ -271,6 +299,7 @@ void PhpScript::update_net_time() noexcept {
     ++long_queries_cnt;
     kprintf("LONG query: %lf\n", net_add);
     if (const net_event_t *event = get_last_net_event()) {
+      send_slow_net_event_stats(*event, net_add);
       kprintf("Awakening net event: %s\n", event->get_description());
     }
   }

server/statshouse/statshouse-manager.cpp

@@ -4,15 +4,21 @@
 
 #include "server/statshouse/statshouse-manager.h"
 
+#include <array>
+#include <charconv>
 #include <chrono>
+#include <cstddef>
 #include <string>
+#include <variant>
 
 #include "common/precise-time.h"
 #include "common/resolver.h"
+#include "common/wrappers/overloaded.h"
 #include "runtime/instance-cache.h"
-#include "server/job-workers/shared-memory-manager.h"
 #include "server/confdata-stats.h"
+#include "server/job-workers/shared-memory-manager.h"
 #include "server/json-logger.h"
+#include "server/php-queries.h"
 #include "server/php-runner.h"
 #include "server/server-config.h"
 #include "server/server-stats.h"
@@ -333,3 +339,23 @@ void StatsHouseManager::add_confdata_master_stats(const ConfdataStats &confdata_
     }
   }
 }
+
+void StatsHouseManager::add_slow_net_event_stats(const slow_net_event_stats::stats_t &stats) noexcept {
+  std::visit(overloaded{[this](const slow_net_event_stats::slow_rpc_response_stats &rpc_query_stat) noexcept {
+                          // FIXME: it's enough to have it equal 10, but due to bug in GCC we are forced to use a length > 253
+                          constexpr auto MAX_INT_STRING_LENGTH = 254;
+                          std::array<char, MAX_INT_STRING_LENGTH> buf{};
+                          const auto chars{std::to_chars(buf.data(), buf.data() + buf.size(), rpc_query_stat.actor_or_port)};
+                          client.metric("kphp_slow_rpc_response")
+                            .tag(rpc_query_stat.tl_function_name != nullptr ? rpc_query_stat.tl_function_name : "unknown")
+                            .tag({buf.data(), static_cast<size_t>(chars.ptr - buf.data())})
+                            .tag(rpc_query_stat.is_error ? "error" : "success")
+                            .write_value(rpc_query_stat.response_time);
+                        },
+                        [this](const slow_net_event_stats::slow_job_worker_response_stats &jw_response_stat) noexcept {
+                          client.metric("kphp_slow_job_worker_response")
+                            .tag(jw_response_stat.class_name != nullptr ? jw_response_stat.class_name : "unknown")
+                            .write_value(jw_response_stat.response_time);
+                        }},
+             stats);
+}

server/statshouse/statshouse-manager.h

@@ -11,6 +11,7 @@
 #include "common/mixin/not_copyable.h"
 #include "runtime-core/memory-resource/memory_resource.h"
 #include "server/job-workers/job-stats.h"
+#include "server/php-queries.h"
 #include "server/statshouse/statshouse-client.h"
 #include "server/workers-control.h"
 #include "server/workers-stats.h"
@@ -88,6 +89,8 @@ class StatsHouseManager : vk::not_copyable {
 
   void add_confdata_master_stats(const ConfdataStats &confdata_stats);
 
+  void add_slow_net_event_stats(const slow_net_event_stats::stats_t &stats) noexcept;
+
 private:
   StatsHouseClient client;
   bool need_write_enable_tag_host = false;