add more metrics (#3446)

* add more metrics

* support to dynamic register metrics

* add more metrics

* compile pass

* add some data related counters

* remove kSentbytes, kReceivedBytes

* add some kv, raft related counters

* fix cherry pick

* add test cases for dyncamically register stats

* remove job metrics

Co-authored-by: Sophie <84560950+Sophie-Xie@users.noreply.github.com>

conf/nebula-graphd.conf.default

@@ -86,3 +86,5 @@
 ########## experimental feature ##########
 # if use experimental features
 --enable_experimental_feature=false
+
+--enable_space_level_metrics=false

conf/nebula-graphd.conf.production

@@ -84,3 +84,5 @@
 ########## experimental feature ##########
 # if use experimental features
 --enable_experimental_feature=false
+
+--enable_space_level_metrics=false

src/clients/meta/CMakeLists.txt

@@ -12,4 +12,5 @@ nebula_add_library(
     FileBasedClusterIdMan.cpp
 )
 
+nebula_add_subdirectory(stats)
 nebula_add_subdirectory(test)

src/clients/meta/MetaClient.cpp

@@ -12,6 +12,7 @@
 #include <thrift/lib/cpp/util/EnumUtils.h>
 
 #include "clients/meta/FileBasedClusterIdMan.h"
+#include "clients/meta/stats/MetaClientStats.h"
 #include "common/base/Base.h"
 #include "common/base/MurmurHash2.h"
 #include "common/conf/Configuration.h"
@@ -626,6 +627,7 @@ void MetaClient::getResponse(Request req,
                              bool toLeader,
                              int32_t retry,
                              int32_t retryLimit) {
+  stats::StatsManager::addValue(kNumRpcSentToMetad);
   auto* evb = ioThreadPool_->getEventBase();
   HostAddr host;
   {
@@ -660,6 +662,7 @@ void MetaClient::getResponse(Request req,
                    this](folly::Try<RpcResponse>&& t) mutable {
               // exception occurred during RPC
               if (t.hasException()) {
+                stats::StatsManager::addValue(kNumRpcSentToMetadFailed);
                 if (toLeader) {
                   updateLeader();
                 } else {

src/clients/meta/stats/CMakeLists.txt

@@ -0,0 +1,9 @@
+# Copyright (c) 2021 vesoft inc. All rights reserved.
+#
+# This source code is licensed under Apache 2.0 License.
+
+nebula_add_library(
+  meta_client_stats_obj
+  OBJECT
+  MetaClientStats.cpp
+  )

src/clients/meta/stats/MetaClientStats.cpp

@@ -0,0 +1,19 @@
+/* Copyright (c) 2021 vesoft inc. All rights reserved.
+ *
+ * This source code is licensed under Apache 2.0 License.
+ */
+
+#include "clients/meta/stats/MetaClientStats.h"
+
+namespace nebula {
+
+stats::CounterId kNumRpcSentToMetad;
+stats::CounterId kNumRpcSentToMetadFailed;
+
+void initMetaClientStats() {
+  kNumRpcSentToMetad = stats::StatsManager::registerStats("num_rpc_sent_to_metad", "rate, sum");
+  kNumRpcSentToMetadFailed =
+      stats::StatsManager::registerStats("num_rpc_sent_to_metad_failed", "rate, sum");
+}
+
+}  // namespace nebula

src/clients/meta/stats/MetaClientStats.h

@@ -0,0 +1,17 @@
+/* Copyright (c) 2021 vesoft inc. All rights reserved.
+ *
+ * This source code is licensed under Apache 2.0 License.
+ */
+
+#pragma once
+
+#include "common/stats/StatsManager.h"
+
+namespace nebula {
+
+extern stats::CounterId kNumRpcSentToMetad;
+extern stats::CounterId kNumRpcSentToMetadFailed;
+
+void initMetaClientStats();
+
+}  // namespace nebula

src/clients/storage/CMakeLists.txt

@@ -19,3 +19,5 @@ nebula_add_library(
     InternalStorageClient.cpp
 )
 
+nebula_add_subdirectory(stats)
+

src/clients/storage/StorageClientBase-inl.h

@@ -7,7 +7,9 @@
 
 #include <folly/Try.h>
 
+#include "clients/storage/stats/StorageClientStats.h"
 #include "common/ssl/SSLConfig.h"
+#include "common/stats/StatsManager.h"
 #include "common/time/WallClock.h"
 
 namespace nebula {
@@ -230,6 +232,7 @@ void StorageClientBase<ClientType>::getResponseImpl(
     std::pair<HostAddr, Request> request,
     RemoteFunc remoteFunc,
     std::shared_ptr<folly::Promise<StatusOr<Response>>> pro) {
+  stats::StatsManager::addValue(kNumRpcSentToStoraged);
   using TransportException = apache::thrift::transport::TTransportException;
   if (evb == nullptr) {
     DCHECK(!!ioThreadPool_);
@@ -266,6 +269,7 @@ void StorageClientBase<ClientType>::getResponseImpl(
             .thenError(folly::tag_t<TransportException>{},
                        [spaceId, partsId = std::move(partsId), host, pro, this](
                            TransportException&& ex) mutable {
+                         stats::StatsManager::addValue(kNumRpcSentToStoragedFailed);
                          if (ex.getType() == TransportException::TIMED_OUT) {
                            LOG(ERROR) << "Request to " << host << " time out: " << ex.what();
                          } else {
@@ -278,6 +282,7 @@ void StorageClientBase<ClientType>::getResponseImpl(
             .thenError(folly::tag_t<std::exception>{},
                        [spaceId, partsId = std::move(partsId), host, pro, this](
                            std::exception&& ex) mutable {
+                         stats::StatsManager::addValue(kNumRpcSentToStoragedFailed);
                          // exception occurred during RPC
                          pro->setValue(Status::Error(
                              folly::stringPrintf("RPC failure in StorageClient: %s", ex.what())));

src/clients/storage/stats/CMakeLists.txt

@@ -0,0 +1,9 @@
+# Copyright (c) 2021 vesoft inc. All rights reserved.
+#
+# This source code is licensed under Apache 2.0 License.
+
+nebula_add_library(
+  storage_client_stats_obj
+  OBJECT
+  StorageClientStats.cpp
+  )

src/clients/storage/stats/StorageClientStats.cpp

@@ -0,0 +1,20 @@
+/* Copyright (c) 2021 vesoft inc. All rights reserved.
+ *
+ * This source code is licensed under Apache 2.0 License.
+ */
+
+#include "clients/meta/stats/MetaClientStats.h"
+
+namespace nebula {
+
+stats::CounterId kNumRpcSentToStoraged;
+stats::CounterId kNumRpcSentToStoragedFailed;
+
+void initStorageClientStats() {
+  kNumRpcSentToStoraged =
+      stats::StatsManager::registerStats("num_rpc_sent_to_storaged", "rate, sum");
+  kNumRpcSentToStoragedFailed =
+      stats::StatsManager::registerStats("num_rpc_sent_to_storaged_failed", "rate, sum");
+}
+
+}  // namespace nebula

src/clients/storage/stats/StorageClientStats.h

@@ -0,0 +1,17 @@
+/* Copyright (c) 2021 vesoft inc. All rights reserved.
+ *
+ * This source code is licensed under Apache 2.0 License.
+ */
+
+#pragma once
+
+#include "common/stats/StatsManager.h"
+
+namespace nebula {
+
+extern stats::CounterId kNumRpcSentToStoraged;
+extern stats::CounterId kNumRpcSentToStoragedFailed;
+
+void initStorageClientStats();
+
+}  // namespace nebula

src/codec/test/CMakeLists.txt

@@ -22,6 +22,9 @@ set(CODEC_TEST_LIBS
     $<TARGET_OBJECTS:fs_obj>
     $<TARGET_OBJECTS:thread_obj>
     $<TARGET_OBJECTS:stats_obj>
+    $<TARGET_OBJECTS:meta_client_stats_obj>
+    $<TARGET_OBJECTS:storage_client_stats_obj>
+    $<TARGET_OBJECTS:kv_stats_obj>
     $<TARGET_OBJECTS:time_obj>
     $<TARGET_OBJECTS:conf_obj>
     $<TARGET_OBJECTS:meta_obj>

src/common/expression/test/CMakeLists.txt

@@ -9,6 +9,9 @@ set(expression_test_common_libs
     $<TARGET_OBJECTS:network_obj>
     $<TARGET_OBJECTS:fs_obj>
     $<TARGET_OBJECTS:stats_obj>
+    $<TARGET_OBJECTS:graph_stats_obj>
+    $<TARGET_OBJECTS:meta_client_stats_obj>
+    $<TARGET_OBJECTS:storage_client_stats_obj>
     $<TARGET_OBJECTS:datatypes_obj>
     $<TARGET_OBJECTS:time_obj>
     $<TARGET_OBJECTS:datetime_parser_obj>