github-actions[bot] commented on code in PR #33015: URL: https://github.com/apache/doris/pull/33015#discussion_r1549021906
########## be/src/runtime/plan_fragment_executor.cpp: ########## @@ -644,4 +646,19 @@ void PlanFragmentExecutor::close() { _closed = true; } +std::shared_ptr<TRuntimeProfileTree> PlanFragmentExecutor::collect_realtime_query_profile() const { Review Comment: warning: method 'collect_realtime_query_profile' can be made static [readability-convert-member-functions-to-static] be/src/runtime/plan_fragment_executor.h:149: ```diff - std::shared_ptr<TRuntimeProfileTree> collect_realtime_query_profile() const; + static std::shared_ptr<TRuntimeProfileTree> collect_realtime_query_profile() ; ``` ```suggestion std::shared_ptr<TRuntimeProfileTree> PlanFragmentExecutor::collect_realtime_query_profile() { ``` ########## be/src/runtime/query_context.cpp: ########## @@ -293,4 +310,90 @@ Status QueryContext::set_workload_group(WorkloadGroupPtr& tg) { return Status::OK(); } +void QueryContext::add_fragment_profile_x( Review Comment: warning: method 'add_fragment_profile_x' can be made static [readability-convert-member-functions-to-static] ```suggestion static void QueryContext::add_fragment_profile_x( ``` ########## be/src/runtime/query_context.cpp: ########## @@ -293,4 +310,90 @@ return Status::OK(); } +void QueryContext::add_fragment_profile_x( + int fragment_id, const std::vector<std::shared_ptr<TRuntimeProfileTree>>& pipeline_profiles, + std::shared_ptr<TRuntimeProfileTree> load_channel_profile) { +#ifndef NDEBUG + for (const auto& p : pipeline_profiles) { + DCHECK(p != nullptr) << fmt::format("Add pipeline profile failed, query {}, fragment {}", + print_id(this->_query_id), fragment_id); + } +#endif + std::lock_guard<std::mutex> l(_profile_mutex); + LOG_INFO("Query X add fragment profile, query {}, fragment {}, pipeline profile count {} ", + print_id(this->_query_id), fragment_id, pipeline_profiles.size()); + _profile_map_x.insert(std::make_pair(fragment_id, pipeline_profiles)); + + if (load_channel_profile != nullptr) { + _load_channel_profile_map_x.insert(std::make_pair(fragment_id, load_channel_profile)); + } +} + +void QueryContext::add_instance_profile(const TUniqueId& instance_id, Review Comment: warning: method 'add_instance_profile' can be made static [readability-convert-member-functions-to-static] ```suggestion static void QueryContext::add_instance_profile(const TUniqueId& instance_id, ``` ########## be/src/runtime/runtime_query_statistics_mgr.cpp: ########## @@ -17,14 +17,342 @@ #include "runtime/runtime_query_statistics_mgr.h" +#include <gen_cpp/FrontendService_types.h> +#include <gen_cpp/RuntimeProfile_types.h> +#include <gen_cpp/Status_types.h> +#include <gen_cpp/Types_types.h> +#include <thrift/TApplicationException.h> + +#include <cstdint> +#include <memory> +#include <mutex> +#include <random> +#include <shared_mutex> +#include <string> +#include <tuple> +#include <unordered_map> +#include <vector> + +#include "common/logging.h" #include "runtime/client_cache.h" #include "runtime/exec_env.h" +#include "runtime/query_context.h" +#include "service/backend_options.h" #include "util/debug_util.h" #include "util/time.h" +#include "util/uid_util.h" #include "vec/core/block.h" namespace doris { +static Status _doReportExecStatsRpc(const TNetworkAddress& coor_addr, + const TReportExecStatusParams& req, + TReportExecStatusResult& res) { + Status client_status; + FrontendServiceConnection rpc_client(ExecEnv::GetInstance()->frontend_client_cache(), coor_addr, + &client_status); + if (!client_status.ok()) { + LOG_WARNING( + "could not get client rpc client of {} when reporting profiles, reason is {}, " + "not reporting, profile will be lost", + PrintThriftNetworkAddress(coor_addr), client_status.to_string()); + return Status::RpcError("Client rpc client failed"); + } + + try { + try { + rpc_client->reportExecStatus(res, req); + } catch (const apache::thrift::transport::TTransportException& e) { + LOG_WARNING("Transport exception from {}, reason: {}, reopening", + PrintThriftNetworkAddress(coor_addr), e.what()); + client_status = rpc_client.reopen(config::thrift_rpc_timeout_ms); + if (!client_status.ok()) { + LOG_WARNING("Reopen failed, reason: {}", client_status.to_string()); + return Status::RpcError("Open rpc client failed"); + } + + rpc_client->reportExecStatus(res, req); + } + } catch (apache::thrift::TApplicationException& e) { + if (e.getType() == e.UNKNOWN_METHOD) { + LOG_WARNING( + "Failed to send statistics to {} due to {}, usually because the frontend " + "is not upgraded, check the version", + PrintThriftNetworkAddress(coor_addr), e.what()); + } else { + LOG_WARNING( + "Failed to send statistics to {}, reason: {}, you can see fe log for " + "details.", + PrintThriftNetworkAddress(coor_addr), e.what()); + } + return Status::RpcError("Send stats failed"); + } catch (std::exception& e) { + LOG_WARNING("Failed to send statistics to {}, reason: {}, you can see fe log for details.", + PrintThriftNetworkAddress(coor_addr), e.what()); + return Status::RpcError("Send stats failed"); + } + + return Status::OK(); +} + +TReportExecStatusParams RuntimeQueryStatiticsMgr::create_report_exec_status_params( + const TUniqueId& query_id, int32 fragment_id, + const std::vector<std::shared_ptr<TRuntimeProfileTree>>& f_profile, + std::shared_ptr<TRuntimeProfileTree> load_channel_profile) { + TReportExecStatusParams req; + req.__set_query_id(query_id); + req.__set_backend_id(ExecEnv::GetInstance()->master_info()->backend_id); + req.__set_fragment_id(fragment_id); + + std::vector<TDetailedReportParams> detailed_params; + for (const auto& p_profile : f_profile) { + TDetailedReportParams tmp; + tmp.__set_profile(*p_profile); + // tmp.fragment_instance_id is not needed for pipeline x + detailed_params.push_back(tmp); + } + req.__set_detailed_report(detailed_params); + + if (load_channel_profile) { + req.__set_loadChannelProfile(*load_channel_profile); + } + + return req; +} + +void RuntimeQueryStatiticsMgr::start_report_thread() { + if (started.load()) { + DCHECK(false) << "report thread has been started"; + LOG_ERROR("report thread has been started"); + return; + } + + started.store(true); + + for (size_t i = 0; i < config::report_exec_status_thread_num; ++i) { + this->_report_profile_threads.emplace_back(std::make_unique<std::thread>( + &RuntimeQueryStatiticsMgr::report_query_profiles_thread, this)); + } +} + +void RuntimeQueryStatiticsMgr::report_query_profiles_thread() { + while (true) { + { + std::unique_lock<std::mutex> lock(_report_profile_mutex); + + while (_query_profile_map.empty() && _profile_map_x.empty() && + !_report_profile_thread_stop) { + _report_profile_cv.wait(lock); + } + + if (_report_profile_thread_stop) { + LOG_INFO("Report profile thread stopped"); + return; + } + } + + _report_query_profiles_function(); + } +} + +void RuntimeQueryStatiticsMgr::force_report_profile() { + std::unique_lock<std::mutex> lock(_report_profile_mutex); + _report_profile_cv.notify_one(); +} + +void RuntimeQueryStatiticsMgr::stop_report_thread() { + if (!started) { + return; + } + + { + std::unique_lock<std::mutex> lock(_report_profile_mutex); + _report_profile_thread_stop = true; + LOG_INFO("All report threads are going to stop"); + _report_profile_cv.notify_all(); + } + + for (const auto& thread : _report_profile_threads) { + thread->join(); + } + + LOG_INFO("All report threads stopped"); +} + +void RuntimeQueryStatiticsMgr::register_instance_profile( + const TUniqueId& query_id, const TNetworkAddress& coor_addr, const TUniqueId& instance_id, + std::shared_ptr<TRuntimeProfileTree> instance_profile, + std::shared_ptr<TRuntimeProfileTree> load_channel_profile) { + if (instance_profile == nullptr) { + auto msg = fmt::format("Register instance profile {} {} failed, profile is null", + print_id(query_id), print_id(instance_id)); + DCHECK(false) << msg; + LOG_ERROR(msg); + return; + } + + std::lock_guard<std::shared_mutex> lg(_query_profile_map_lock); + + if (!_query_profile_map.contains(query_id)) { + _query_profile_map[query_id] = std::make_tuple( + coor_addr, std::unordered_map<TUniqueId, std::shared_ptr<TRuntimeProfileTree>>()); + } + + std::get<1>(_query_profile_map[query_id]).insert(std::make_pair(instance_id, instance_profile)); + + if (load_channel_profile != nullptr) { + _load_channel_profile_map[instance_id] = load_channel_profile; + } + + LOG_INFO("Register instance profile {} {}", print_id(query_id), print_id(instance_id)); +} + +void RuntimeQueryStatiticsMgr::register_fragment_profile_x( + const TUniqueId& query_id, const TNetworkAddress& coor_addr, int32_t fragment_id, + std::vector<std::shared_ptr<TRuntimeProfileTree>> p_profiles, + std::shared_ptr<TRuntimeProfileTree> load_channel_profile_x) { + for (const auto& p : p_profiles) { + if (p == nullptr) { + auto msg = fmt::format("Register fragment profile {} {} failed, profile is null", + print_id(query_id), fragment_id); + DCHECK(false) << msg; + LOG_ERROR(msg); + return; + } + } + + std::lock_guard<std::shared_mutex> lg(_query_profile_map_lock); + + if (!_profile_map_x.contains(query_id)) { + _profile_map_x[query_id] = std::make_tuple( + coor_addr, + std::unordered_map<int, std::vector<std::shared_ptr<TRuntimeProfileTree>>>()); + } + + std::unordered_map<int, std::vector<std::shared_ptr<TRuntimeProfileTree>>>& + fragment_profile_map = std::get<1>(_profile_map_x[query_id]); + fragment_profile_map.insert(std::make_pair(fragment_id, p_profiles)); + + if (load_channel_profile_x != nullptr) { + _load_channel_profile_map_x[std::make_pair(query_id, fragment_id)] = load_channel_profile_x; + } + + LOG_INFO("register x profile done {}, fragment {}, profiles {}", print_id(query_id), + fragment_id, p_profiles.size()); +} + +void RuntimeQueryStatiticsMgr::_report_query_profiles_non_pipeline() { Review Comment: warning: method '_report_query_profiles_non_pipeline' can be made static [readability-convert-member-functions-to-static] be/src/runtime/runtime_query_statistics_mgr.h:124: ```diff - void _report_query_profiles_non_pipeline(); + static void _report_query_profiles_non_pipeline(); ``` ########## be/src/runtime/runtime_query_statistics_mgr.cpp: ########## @@ -17,14 +17,342 @@ #include "runtime/runtime_query_statistics_mgr.h" +#include <gen_cpp/FrontendService_types.h> +#include <gen_cpp/RuntimeProfile_types.h> +#include <gen_cpp/Status_types.h> +#include <gen_cpp/Types_types.h> +#include <thrift/TApplicationException.h> + +#include <cstdint> +#include <memory> +#include <mutex> +#include <random> +#include <shared_mutex> +#include <string> +#include <tuple> +#include <unordered_map> +#include <vector> + +#include "common/logging.h" #include "runtime/client_cache.h" #include "runtime/exec_env.h" +#include "runtime/query_context.h" +#include "service/backend_options.h" #include "util/debug_util.h" #include "util/time.h" +#include "util/uid_util.h" #include "vec/core/block.h" namespace doris { +static Status _doReportExecStatsRpc(const TNetworkAddress& coor_addr, + const TReportExecStatusParams& req, + TReportExecStatusResult& res) { + Status client_status; + FrontendServiceConnection rpc_client(ExecEnv::GetInstance()->frontend_client_cache(), coor_addr, + &client_status); + if (!client_status.ok()) { + LOG_WARNING( + "could not get client rpc client of {} when reporting profiles, reason is {}, " + "not reporting, profile will be lost", + PrintThriftNetworkAddress(coor_addr), client_status.to_string()); + return Status::RpcError("Client rpc client failed"); + } + + try { + try { + rpc_client->reportExecStatus(res, req); + } catch (const apache::thrift::transport::TTransportException& e) { + LOG_WARNING("Transport exception from {}, reason: {}, reopening", + PrintThriftNetworkAddress(coor_addr), e.what()); + client_status = rpc_client.reopen(config::thrift_rpc_timeout_ms); + if (!client_status.ok()) { + LOG_WARNING("Reopen failed, reason: {}", client_status.to_string()); + return Status::RpcError("Open rpc client failed"); + } + + rpc_client->reportExecStatus(res, req); + } + } catch (apache::thrift::TApplicationException& e) { + if (e.getType() == e.UNKNOWN_METHOD) { + LOG_WARNING( + "Failed to send statistics to {} due to {}, usually because the frontend " + "is not upgraded, check the version", + PrintThriftNetworkAddress(coor_addr), e.what()); + } else { + LOG_WARNING( + "Failed to send statistics to {}, reason: {}, you can see fe log for " + "details.", + PrintThriftNetworkAddress(coor_addr), e.what()); + } + return Status::RpcError("Send stats failed"); + } catch (std::exception& e) { + LOG_WARNING("Failed to send statistics to {}, reason: {}, you can see fe log for details.", + PrintThriftNetworkAddress(coor_addr), e.what()); + return Status::RpcError("Send stats failed"); + } + + return Status::OK(); +} + +TReportExecStatusParams RuntimeQueryStatiticsMgr::create_report_exec_status_params( + const TUniqueId& query_id, int32 fragment_id, + const std::vector<std::shared_ptr<TRuntimeProfileTree>>& f_profile, + std::shared_ptr<TRuntimeProfileTree> load_channel_profile) { + TReportExecStatusParams req; + req.__set_query_id(query_id); + req.__set_backend_id(ExecEnv::GetInstance()->master_info()->backend_id); + req.__set_fragment_id(fragment_id); + + std::vector<TDetailedReportParams> detailed_params; + for (const auto& p_profile : f_profile) { + TDetailedReportParams tmp; + tmp.__set_profile(*p_profile); + // tmp.fragment_instance_id is not needed for pipeline x + detailed_params.push_back(tmp); + } + req.__set_detailed_report(detailed_params); + + if (load_channel_profile) { + req.__set_loadChannelProfile(*load_channel_profile); + } + + return req; +} + +void RuntimeQueryStatiticsMgr::start_report_thread() { + if (started.load()) { + DCHECK(false) << "report thread has been started"; + LOG_ERROR("report thread has been started"); + return; + } + + started.store(true); + + for (size_t i = 0; i < config::report_exec_status_thread_num; ++i) { + this->_report_profile_threads.emplace_back(std::make_unique<std::thread>( + &RuntimeQueryStatiticsMgr::report_query_profiles_thread, this)); + } +} + +void RuntimeQueryStatiticsMgr::report_query_profiles_thread() { + while (true) { + { + std::unique_lock<std::mutex> lock(_report_profile_mutex); + + while (_query_profile_map.empty() && _profile_map_x.empty() && + !_report_profile_thread_stop) { + _report_profile_cv.wait(lock); + } + + if (_report_profile_thread_stop) { + LOG_INFO("Report profile thread stopped"); + return; + } + } + + _report_query_profiles_function(); + } +} + +void RuntimeQueryStatiticsMgr::force_report_profile() { Review Comment: warning: method 'force_report_profile' can be made static [readability-convert-member-functions-to-static] be/src/runtime/runtime_query_statistics_mgr.h:91: ```diff - void force_report_profile(); + static void force_report_profile(); ``` ########## be/src/runtime/runtime_query_statistics_mgr.cpp: ########## @@ -17,14 +17,342 @@ #include "runtime/runtime_query_statistics_mgr.h" +#include <gen_cpp/FrontendService_types.h> +#include <gen_cpp/RuntimeProfile_types.h> +#include <gen_cpp/Status_types.h> +#include <gen_cpp/Types_types.h> +#include <thrift/TApplicationException.h> + +#include <cstdint> +#include <memory> +#include <mutex> +#include <random> +#include <shared_mutex> +#include <string> +#include <tuple> +#include <unordered_map> +#include <vector> + +#include "common/logging.h" #include "runtime/client_cache.h" #include "runtime/exec_env.h" +#include "runtime/query_context.h" +#include "service/backend_options.h" #include "util/debug_util.h" #include "util/time.h" +#include "util/uid_util.h" #include "vec/core/block.h" namespace doris { +static Status _doReportExecStatsRpc(const TNetworkAddress& coor_addr, + const TReportExecStatusParams& req, + TReportExecStatusResult& res) { + Status client_status; + FrontendServiceConnection rpc_client(ExecEnv::GetInstance()->frontend_client_cache(), coor_addr, + &client_status); + if (!client_status.ok()) { + LOG_WARNING( + "could not get client rpc client of {} when reporting profiles, reason is {}, " + "not reporting, profile will be lost", + PrintThriftNetworkAddress(coor_addr), client_status.to_string()); + return Status::RpcError("Client rpc client failed"); + } + + try { + try { + rpc_client->reportExecStatus(res, req); + } catch (const apache::thrift::transport::TTransportException& e) { + LOG_WARNING("Transport exception from {}, reason: {}, reopening", + PrintThriftNetworkAddress(coor_addr), e.what()); + client_status = rpc_client.reopen(config::thrift_rpc_timeout_ms); + if (!client_status.ok()) { + LOG_WARNING("Reopen failed, reason: {}", client_status.to_string()); + return Status::RpcError("Open rpc client failed"); + } + + rpc_client->reportExecStatus(res, req); + } + } catch (apache::thrift::TApplicationException& e) { + if (e.getType() == e.UNKNOWN_METHOD) { + LOG_WARNING( + "Failed to send statistics to {} due to {}, usually because the frontend " + "is not upgraded, check the version", + PrintThriftNetworkAddress(coor_addr), e.what()); + } else { + LOG_WARNING( + "Failed to send statistics to {}, reason: {}, you can see fe log for " + "details.", + PrintThriftNetworkAddress(coor_addr), e.what()); + } + return Status::RpcError("Send stats failed"); + } catch (std::exception& e) { + LOG_WARNING("Failed to send statistics to {}, reason: {}, you can see fe log for details.", + PrintThriftNetworkAddress(coor_addr), e.what()); + return Status::RpcError("Send stats failed"); + } + + return Status::OK(); +} + +TReportExecStatusParams RuntimeQueryStatiticsMgr::create_report_exec_status_params( + const TUniqueId& query_id, int32 fragment_id, + const std::vector<std::shared_ptr<TRuntimeProfileTree>>& f_profile, + std::shared_ptr<TRuntimeProfileTree> load_channel_profile) { + TReportExecStatusParams req; + req.__set_query_id(query_id); + req.__set_backend_id(ExecEnv::GetInstance()->master_info()->backend_id); + req.__set_fragment_id(fragment_id); + + std::vector<TDetailedReportParams> detailed_params; + for (const auto& p_profile : f_profile) { + TDetailedReportParams tmp; + tmp.__set_profile(*p_profile); + // tmp.fragment_instance_id is not needed for pipeline x + detailed_params.push_back(tmp); + } + req.__set_detailed_report(detailed_params); + + if (load_channel_profile) { + req.__set_loadChannelProfile(*load_channel_profile); + } + + return req; +} + +void RuntimeQueryStatiticsMgr::start_report_thread() { Review Comment: warning: method 'start_report_thread' can be made static [readability-convert-member-functions-to-static] be/src/runtime/runtime_query_statistics_mgr.h:89: ```diff - void start_report_thread(); + static void start_report_thread(); ``` ########## be/src/runtime/query_context.cpp: ########## @@ -293,4 +310,90 @@ return Status::OK(); } +void QueryContext::add_fragment_profile_x( + int fragment_id, const std::vector<std::shared_ptr<TRuntimeProfileTree>>& pipeline_profiles, + std::shared_ptr<TRuntimeProfileTree> load_channel_profile) { +#ifndef NDEBUG + for (const auto& p : pipeline_profiles) { + DCHECK(p != nullptr) << fmt::format("Add pipeline profile failed, query {}, fragment {}", + print_id(this->_query_id), fragment_id); + } +#endif + std::lock_guard<std::mutex> l(_profile_mutex); + LOG_INFO("Query X add fragment profile, query {}, fragment {}, pipeline profile count {} ", + print_id(this->_query_id), fragment_id, pipeline_profiles.size()); + _profile_map_x.insert(std::make_pair(fragment_id, pipeline_profiles)); + + if (load_channel_profile != nullptr) { + _load_channel_profile_map_x.insert(std::make_pair(fragment_id, load_channel_profile)); + } +} + +void QueryContext::add_instance_profile(const TUniqueId& instance_id, + std::shared_ptr<TRuntimeProfileTree> profile, + std::shared_ptr<TRuntimeProfileTree> load_channel_profile) { + DCHECK(profile != nullptr) << print_id(instance_id); + + std::lock_guard<std::mutex> lg(_profile_mutex); + _profile_map.insert(std::make_pair(instance_id, profile)); + if (load_channel_profile != nullptr) { + _load_channel_profile_map.insert(std::make_pair(instance_id, load_channel_profile)); + } +} + +void QueryContext::_async_report_profile() { + if (!enable_profile()) { + return; + } + + _async_report_profile_x(); + _async_report_profile_non_pipeline(); +} + +void QueryContext::_async_report_profile_non_pipeline() { + if (enable_pipeline_exec()) { + return; + } + + std::lock_guard<std::mutex> lg(_profile_mutex); + LOG_INFO("Query {}, register query profile, instance profile count {}", print_id(_query_id), + _profile_map.size()); + + for (auto& [instance_id, instance_profile] : _profile_map) { + std::shared_ptr<TRuntimeProfileTree> load_channel_profile = nullptr; + if (_load_channel_profile_map.contains(instance_id)) { + load_channel_profile = _load_channel_profile_map[instance_id]; + } + + ExecEnv::GetInstance()->runtime_query_statistics_mgr()->register_instance_profile( + _query_id, this->coord_addr, instance_id, instance_profile, load_channel_profile); + } + + ExecEnv::GetInstance()->runtime_query_statistics_mgr()->force_report_profile(); +} + +void QueryContext::_async_report_profile_x() { Review Comment: warning: method '_async_report_profile_x' can be made const [readability-make-member-function-const] ```suggestion void QueryContext::_async_report_profile_x() const { ``` be/src/runtime/query_context.h:385: ```diff - void _async_report_profile_x(); + void _async_report_profile_x() const; ``` ########## be/src/runtime/plan_fragment_executor.cpp: ########## @@ -644,4 +646,19 @@ _closed = true; } +std::shared_ptr<TRuntimeProfileTree> PlanFragmentExecutor::collect_realtime_query_profile() const { + std::shared_ptr<TRuntimeProfileTree> res = std::make_shared<TRuntimeProfileTree>(); + _runtime_state->runtime_profile()->compute_time_in_profile(); + _runtime_state->runtime_profile()->to_thrift(res.get()); + return res; +} + +std::shared_ptr<TRuntimeProfileTree> PlanFragmentExecutor::collect_realtime_load_channel_profile() Review Comment: warning: method 'collect_realtime_load_channel_profile' can be made static [readability-convert-member-functions-to-static] be/src/runtime/plan_fragment_executor.h:150: ```diff - std::shared_ptr<TRuntimeProfileTree> collect_realtime_load_channel_profile() const; + static std::shared_ptr<TRuntimeProfileTree> collect_realtime_load_channel_profile() ; ``` be/src/runtime/plan_fragment_executor.cpp:656: ```diff - const { + { ``` ########## be/src/runtime/query_context.cpp: ########## @@ -293,4 +310,90 @@ return Status::OK(); } +void QueryContext::add_fragment_profile_x( + int fragment_id, const std::vector<std::shared_ptr<TRuntimeProfileTree>>& pipeline_profiles, + std::shared_ptr<TRuntimeProfileTree> load_channel_profile) { +#ifndef NDEBUG + for (const auto& p : pipeline_profiles) { + DCHECK(p != nullptr) << fmt::format("Add pipeline profile failed, query {}, fragment {}", + print_id(this->_query_id), fragment_id); + } +#endif + std::lock_guard<std::mutex> l(_profile_mutex); + LOG_INFO("Query X add fragment profile, query {}, fragment {}, pipeline profile count {} ", + print_id(this->_query_id), fragment_id, pipeline_profiles.size()); + _profile_map_x.insert(std::make_pair(fragment_id, pipeline_profiles)); + + if (load_channel_profile != nullptr) { + _load_channel_profile_map_x.insert(std::make_pair(fragment_id, load_channel_profile)); + } +} + +void QueryContext::add_instance_profile(const TUniqueId& instance_id, + std::shared_ptr<TRuntimeProfileTree> profile, + std::shared_ptr<TRuntimeProfileTree> load_channel_profile) { + DCHECK(profile != nullptr) << print_id(instance_id); + + std::lock_guard<std::mutex> lg(_profile_mutex); + _profile_map.insert(std::make_pair(instance_id, profile)); + if (load_channel_profile != nullptr) { + _load_channel_profile_map.insert(std::make_pair(instance_id, load_channel_profile)); + } +} + +void QueryContext::_async_report_profile() { + if (!enable_profile()) { + return; + } + + _async_report_profile_x(); + _async_report_profile_non_pipeline(); +} + +void QueryContext::_async_report_profile_non_pipeline() { Review Comment: warning: method '_async_report_profile_non_pipeline' can be made const [readability-make-member-function-const] ```suggestion void QueryContext::_async_report_profile_non_pipeline() const { ``` be/src/runtime/query_context.h:384: ```diff - void _async_report_profile_non_pipeline(); + void _async_report_profile_non_pipeline() const; ``` ########## be/src/runtime/runtime_query_statistics_mgr.cpp: ########## @@ -17,14 +17,342 @@ #include "runtime/runtime_query_statistics_mgr.h" +#include <gen_cpp/FrontendService_types.h> +#include <gen_cpp/RuntimeProfile_types.h> +#include <gen_cpp/Status_types.h> +#include <gen_cpp/Types_types.h> +#include <thrift/TApplicationException.h> + +#include <cstdint> +#include <memory> +#include <mutex> +#include <random> +#include <shared_mutex> +#include <string> +#include <tuple> +#include <unordered_map> +#include <vector> + +#include "common/logging.h" #include "runtime/client_cache.h" #include "runtime/exec_env.h" +#include "runtime/query_context.h" +#include "service/backend_options.h" #include "util/debug_util.h" #include "util/time.h" +#include "util/uid_util.h" #include "vec/core/block.h" namespace doris { +static Status _doReportExecStatsRpc(const TNetworkAddress& coor_addr, + const TReportExecStatusParams& req, + TReportExecStatusResult& res) { + Status client_status; + FrontendServiceConnection rpc_client(ExecEnv::GetInstance()->frontend_client_cache(), coor_addr, + &client_status); + if (!client_status.ok()) { + LOG_WARNING( + "could not get client rpc client of {} when reporting profiles, reason is {}, " + "not reporting, profile will be lost", + PrintThriftNetworkAddress(coor_addr), client_status.to_string()); + return Status::RpcError("Client rpc client failed"); + } + + try { + try { + rpc_client->reportExecStatus(res, req); + } catch (const apache::thrift::transport::TTransportException& e) { + LOG_WARNING("Transport exception from {}, reason: {}, reopening", + PrintThriftNetworkAddress(coor_addr), e.what()); + client_status = rpc_client.reopen(config::thrift_rpc_timeout_ms); + if (!client_status.ok()) { + LOG_WARNING("Reopen failed, reason: {}", client_status.to_string()); + return Status::RpcError("Open rpc client failed"); + } + + rpc_client->reportExecStatus(res, req); + } + } catch (apache::thrift::TApplicationException& e) { + if (e.getType() == e.UNKNOWN_METHOD) { + LOG_WARNING( + "Failed to send statistics to {} due to {}, usually because the frontend " + "is not upgraded, check the version", + PrintThriftNetworkAddress(coor_addr), e.what()); + } else { + LOG_WARNING( + "Failed to send statistics to {}, reason: {}, you can see fe log for " + "details.", + PrintThriftNetworkAddress(coor_addr), e.what()); + } + return Status::RpcError("Send stats failed"); + } catch (std::exception& e) { + LOG_WARNING("Failed to send statistics to {}, reason: {}, you can see fe log for details.", + PrintThriftNetworkAddress(coor_addr), e.what()); + return Status::RpcError("Send stats failed"); + } + + return Status::OK(); +} + +TReportExecStatusParams RuntimeQueryStatiticsMgr::create_report_exec_status_params( + const TUniqueId& query_id, int32 fragment_id, + const std::vector<std::shared_ptr<TRuntimeProfileTree>>& f_profile, + std::shared_ptr<TRuntimeProfileTree> load_channel_profile) { + TReportExecStatusParams req; + req.__set_query_id(query_id); + req.__set_backend_id(ExecEnv::GetInstance()->master_info()->backend_id); + req.__set_fragment_id(fragment_id); + + std::vector<TDetailedReportParams> detailed_params; + for (const auto& p_profile : f_profile) { + TDetailedReportParams tmp; + tmp.__set_profile(*p_profile); + // tmp.fragment_instance_id is not needed for pipeline x + detailed_params.push_back(tmp); + } + req.__set_detailed_report(detailed_params); + + if (load_channel_profile) { + req.__set_loadChannelProfile(*load_channel_profile); + } + + return req; +} + +void RuntimeQueryStatiticsMgr::start_report_thread() { + if (started.load()) { + DCHECK(false) << "report thread has been started"; + LOG_ERROR("report thread has been started"); + return; + } + + started.store(true); + + for (size_t i = 0; i < config::report_exec_status_thread_num; ++i) { + this->_report_profile_threads.emplace_back(std::make_unique<std::thread>( + &RuntimeQueryStatiticsMgr::report_query_profiles_thread, this)); + } +} + +void RuntimeQueryStatiticsMgr::report_query_profiles_thread() { + while (true) { + { + std::unique_lock<std::mutex> lock(_report_profile_mutex); + + while (_query_profile_map.empty() && _profile_map_x.empty() && + !_report_profile_thread_stop) { + _report_profile_cv.wait(lock); + } + + if (_report_profile_thread_stop) { + LOG_INFO("Report profile thread stopped"); + return; + } + } + + _report_query_profiles_function(); + } +} + +void RuntimeQueryStatiticsMgr::force_report_profile() { + std::unique_lock<std::mutex> lock(_report_profile_mutex); + _report_profile_cv.notify_one(); +} + +void RuntimeQueryStatiticsMgr::stop_report_thread() { + if (!started) { + return; + } + + { + std::unique_lock<std::mutex> lock(_report_profile_mutex); + _report_profile_thread_stop = true; + LOG_INFO("All report threads are going to stop"); + _report_profile_cv.notify_all(); + } + + for (const auto& thread : _report_profile_threads) { + thread->join(); + } + + LOG_INFO("All report threads stopped"); +} + +void RuntimeQueryStatiticsMgr::register_instance_profile( + const TUniqueId& query_id, const TNetworkAddress& coor_addr, const TUniqueId& instance_id, + std::shared_ptr<TRuntimeProfileTree> instance_profile, + std::shared_ptr<TRuntimeProfileTree> load_channel_profile) { + if (instance_profile == nullptr) { + auto msg = fmt::format("Register instance profile {} {} failed, profile is null", + print_id(query_id), print_id(instance_id)); + DCHECK(false) << msg; + LOG_ERROR(msg); + return; + } + + std::lock_guard<std::shared_mutex> lg(_query_profile_map_lock); + + if (!_query_profile_map.contains(query_id)) { + _query_profile_map[query_id] = std::make_tuple( + coor_addr, std::unordered_map<TUniqueId, std::shared_ptr<TRuntimeProfileTree>>()); + } + + std::get<1>(_query_profile_map[query_id]).insert(std::make_pair(instance_id, instance_profile)); + + if (load_channel_profile != nullptr) { + _load_channel_profile_map[instance_id] = load_channel_profile; + } + + LOG_INFO("Register instance profile {} {}", print_id(query_id), print_id(instance_id)); +} + +void RuntimeQueryStatiticsMgr::register_fragment_profile_x( + const TUniqueId& query_id, const TNetworkAddress& coor_addr, int32_t fragment_id, + std::vector<std::shared_ptr<TRuntimeProfileTree>> p_profiles, + std::shared_ptr<TRuntimeProfileTree> load_channel_profile_x) { + for (const auto& p : p_profiles) { + if (p == nullptr) { + auto msg = fmt::format("Register fragment profile {} {} failed, profile is null", + print_id(query_id), fragment_id); + DCHECK(false) << msg; + LOG_ERROR(msg); + return; + } + } + + std::lock_guard<std::shared_mutex> lg(_query_profile_map_lock); + + if (!_profile_map_x.contains(query_id)) { + _profile_map_x[query_id] = std::make_tuple( + coor_addr, + std::unordered_map<int, std::vector<std::shared_ptr<TRuntimeProfileTree>>>()); + } + + std::unordered_map<int, std::vector<std::shared_ptr<TRuntimeProfileTree>>>& + fragment_profile_map = std::get<1>(_profile_map_x[query_id]); + fragment_profile_map.insert(std::make_pair(fragment_id, p_profiles)); + + if (load_channel_profile_x != nullptr) { + _load_channel_profile_map_x[std::make_pair(query_id, fragment_id)] = load_channel_profile_x; + } + + LOG_INFO("register x profile done {}, fragment {}, profiles {}", print_id(query_id), + fragment_id, p_profiles.size()); +} + +void RuntimeQueryStatiticsMgr::_report_query_profiles_non_pipeline() { + // query_id -> {coordinator_addr, {instance_id -> instance_profile}} + decltype(_query_profile_map) profile_copy; + decltype(_load_channel_profile_map) load_channel_profile_copy; + + { + std::lock_guard<std::shared_mutex> lg(_query_profile_map_lock); + profile_copy = _query_profile_map; + load_channel_profile_copy = _load_channel_profile_map; + _query_profile_map.clear(); + _load_channel_profile_map.clear(); + } + + // query_id -> {coordinator_addr, {instance_id -> instance_profile}} + for (const auto& entry : profile_copy) { + const auto& query_id = entry.first; + const auto& coor_addr = std::get<0>(entry.second); + const std::unordered_map<TUniqueId, std::shared_ptr<TRuntimeProfileTree>>& + instance_id_to_profile = std::get<1>(entry.second); + + for (const auto& profile_entry : instance_id_to_profile) { + const auto& instance_id = profile_entry.first; + const auto& instance_profile = profile_entry.second; + + if (instance_profile == nullptr) { + auto msg = fmt::format("Query {} instance {} profile is null", print_id(query_id), + print_id(instance_id)); + DCHECK(false) << msg; + LOG_ERROR(msg); + continue; + } + + TReportExecStatusParams req; + req.__set_query_id(query_id); + req.__set_fragment_instance_id(instance_id); + // instance profile will not be added to RuntimeQueryStatiticsMgr unless it is closed + req.__set_done(true); + req.__set_profile(*instance_profile); + + if (load_channel_profile_copy.contains(instance_id)) { + req.__set_loadChannelProfile(*load_channel_profile_copy[instance_id]); + } + + TReportExecStatusResult res; + auto rpc_status = _doReportExecStatsRpc(coor_addr, req, res); + + if (res.status.status_code != TStatusCode::OK) { + std::stringstream ss; + res.status.printTo(ss); + LOG_WARNING("Query {} send profile to {} failed, msg: {}", print_id(query_id), + PrintThriftNetworkAddress(coor_addr), ss.str()); + } else { + LOG_INFO("Send {} profile finished", print_id(query_id)); + } + } + } +} + +void RuntimeQueryStatiticsMgr::_report_query_profiles_x() { Review Comment: warning: method '_report_query_profiles_x' can be made static [readability-convert-member-functions-to-static] be/src/runtime/runtime_query_statistics_mgr.h:123: ```diff - void _report_query_profiles_x(); + static void _report_query_profiles_x(); ``` -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: commits-unsubscr...@doris.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@doris.apache.org For additional commands, e-mail: commits-h...@doris.apache.org