From bb8d3dace16102461197d02b6b97dea0540631c8 Mon Sep 17 00:00:00 2001
From: qihongcheng <qihongcheng@xiaohongshu.com>
Date: Fri, 6 Sep 2024 10:48:12 +0800
Subject: [PATCH] Add Monitoring Metrics to NebulaGraph #5841

---
 src/graph/executor/Executor.cpp            |  5 +++++
 src/graph/executor/query/LimitExecutor.cpp | 12 +++++++++++-
 src/graph/stats/GraphStats.cpp             |  6 ++++++
 src/graph/stats/GraphStats.h               |  2 ++
 4 files changed, 24 insertions(+), 1 deletion(-)

diff --git a/src/graph/executor/Executor.cpp b/src/graph/executor/Executor.cpp
index 5450722f02b..b6ae9f242d1 100644
--- a/src/graph/executor/Executor.cpp
+++ b/src/graph/executor/Executor.cpp
@@ -209,6 +209,11 @@ Executor *Executor::makeExecutor(QueryContext *qctx, const PlanNode *node) {
       return pool->makeAndAdd<FulltextIndexScanExecutor>(node, qctx);
     }
     case PlanNode::Kind::kLimit: {
+      stats::StatsManager::addValue(kNumLimitExecutors);
+      if (FLAGS_enable_space_level_metrics && spaceName != "") {
+        stats::StatsManager::addValue(
+            stats::StatsManager::counterWithLabels(kNumLimitExecutors, {{"space", spaceName}}));
+      }
       return pool->makeAndAdd<LimitExecutor>(node, qctx);
     }
     case PlanNode::Kind::kSample: {
diff --git a/src/graph/executor/query/LimitExecutor.cpp b/src/graph/executor/query/LimitExecutor.cpp
index 236d8fe3cf6..9e5a6d5edab 100644
--- a/src/graph/executor/query/LimitExecutor.cpp
+++ b/src/graph/executor/query/LimitExecutor.cpp
@@ -5,11 +5,13 @@
 #include "graph/executor/query/LimitExecutor.h"
 
 #include "graph/planner/plan/Query.h"
-
+#include "graph/stats/GraphStats.h"
 namespace nebula {
 namespace graph {
 
 folly::Future<Status> LimitExecutor::execute() {
+  auto start_ts = std::chrono::steady_clock::now();
+  auto& spaceName = qctx()->rctx() ? qctx()->rctx()->session()->spaceName() : "";
   SCOPED_TIMER(&execTime_);
 
   auto* limit = asNode<Limit>(node());
@@ -31,6 +33,14 @@ folly::Future<Status> LimitExecutor::execute() {
     builder.value(result.valuePtr());
     iter->select(offset, count);
     builder.iter(std::move(result).iter());
+    auto diff = std::chrono::steady_clock::now() - start_ts;
+    stats::StatsManager::addValue(
+        kLimitExecutorsLatencyUs,
+        std::chrono::duration_cast<std::chrono::milliseconds>(diff).count());
+    if (FLAGS_enable_space_level_metrics && spaceName != "") {
+      stats::StatsManager::addValue(
+          stats::StatsManager::histoWithLabels(kLimitExecutorsLatencyUs, {{"space", spaceName}}));
+    }
     return finish(builder.build());
   } else {
     DataSet ds;
diff --git a/src/graph/stats/GraphStats.cpp b/src/graph/stats/GraphStats.cpp
index f03cd46885b..395f67f3fd4 100644
--- a/src/graph/stats/GraphStats.cpp
+++ b/src/graph/stats/GraphStats.cpp
@@ -32,6 +32,8 @@ stats::CounterId kOptimizerLatencyUs;
 
 stats::CounterId kNumAggregateExecutors;
 stats::CounterId kNumSortExecutors;
+stats::CounterId kNumLimitExecutors;
+stats::CounterId kLimitExecutorsLatencyUs;
 stats::CounterId kNumIndexScanExecutors;
 
 stats::CounterId kNumOpenedSessions;
@@ -66,6 +68,10 @@ void initGraphStats() {
   kNumIndexScanExecutors =
       stats::StatsManager::registerStats("num_indexscan_executors", "rate, sum");
 
+  kNumLimitExecutors = stats::StatsManager::registerStats("num_Limit_executors", "rate, sum");
+  kLimitExecutorsLatencyUs = stats::StatsManager::registerHisto(
+      "limit_executors_latency_us", 1000, 0, 2000, "avg, p75, p95, p99, p999");
+
   kNumOpenedSessions = stats::StatsManager::registerStats("num_opened_sessions", "rate, sum");
   kNumAuthFailedSessions =
       stats::StatsManager::registerStats("num_auth_failed_sessions", "rate, sum");
diff --git a/src/graph/stats/GraphStats.h b/src/graph/stats/GraphStats.h
index a1facbbf574..e91fd355ce8 100644
--- a/src/graph/stats/GraphStats.h
+++ b/src/graph/stats/GraphStats.h
@@ -33,6 +33,8 @@ extern stats::CounterId kOptimizerLatencyUs;
 // Executor
 extern stats::CounterId kNumAggregateExecutors;
 extern stats::CounterId kNumSortExecutors;
+extern stats::CounterId kNumLimitExecutors;
+extern stats::CounterId kLimitExecutorsLatencyUs;
 extern stats::CounterId kNumIndexScanExecutors;
 
 // Server client traffic