yiguolei commented on code in PR #33331:
URL: https://github.com/apache/doris/pull/33331#discussion_r1556973740
##########
be/src/runtime/runtime_query_statistics_mgr.cpp:
##########
@@ -17,14 +17,432 @@
#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 <condition_variable>
+#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/hash_util.hpp"
#include "util/time.h"
+#include "util/uid_util.h"
#include "vec/core/block.h"
namespace doris {
+static Status _do_report_exec_stats_rpc(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_x(
+ const TUniqueId& query_id,
+ const std::unordered_map<int32,
std::vector<std::shared_ptr<TRuntimeProfileTree>>>&
+ fragment_id_to_profile,
+ const std::vector<std::shared_ptr<TRuntimeProfileTree>>&
load_channel_profiles) {
+ TQueryProfile profile;
+ profile.__set_query_id(query_id);
+
+ std::map<int32_t, std::vector<TDetailedReportParams>>
fragment_id_to_profile_req;
+
+ for (const auto& entry : fragment_id_to_profile) {
+ int32_t fragment_id = entry.first;
+ const std::vector<std::shared_ptr<TRuntimeProfileTree>>&
fragment_profile = entry.second;
+ std::vector<TDetailedReportParams> detailed_params;
+
+ for (auto pipeline_profile : fragment_profile) {
+ if (pipeline_profile == nullptr) {
+ auto msg = fmt::format("Register fragment profile {} {}
failed, profile is null",
+ print_id(query_id), fragment_id);
+ DCHECK(false) << msg;
+ LOG_ERROR(msg);
+ continue;
+ }
+
+ TDetailedReportParams tmp;
+ tmp.__set_profile(*pipeline_profile);
+ // tmp.fragment_instance_id is not needed for pipeline x
+ detailed_params.push_back(tmp);
+ }
+
+ fragment_id_to_profile_req.insert(std::make_pair(fragment_id,
detailed_params));
+ }
+
+ if (fragment_id_to_profile_req.size() == 0) {
+ LOG_WARNING("No fragment profile found for query {}",
print_id(query_id));
+ }
+
+ profile.__set_fragment_id_to_profile(fragment_id_to_profile_req);
+
+ std::vector<TRuntimeProfileTree> load_channel_profiles_req;
+ for (auto load_channel_profile : load_channel_profiles) {
+ if (load_channel_profile == nullptr) {
+ auto msg = fmt::format(
+ "Register fragment profile {} {} failed, load channel
profile is null",
+ print_id(query_id), -1);
+ DCHECK(false) << msg;
+ LOG_ERROR(msg);
+ continue;
+ }
+
+ load_channel_profiles_req.push_back(*load_channel_profile);
+ }
+
+ if (load_channel_profiles_req.size() > 0) {
+ profile.__set_load_channel_profiles(load_channel_profiles_req);
+ }
+
+ TReportExecStatusParams req;
+ req.__set_query_profile(profile);
+ req.__set_query_id(TUniqueId());
+
+ return req;
+}
+
+TReportExecStatusParams
RuntimeQueryStatiticsMgr::create_report_exec_status_params_non_pipeline(
+ const TUniqueId& query_id,
+ const std::unordered_map<TUniqueId,
std::shared_ptr<TRuntimeProfileTree>>&
+ instance_id_to_profile,
+ const std::vector<std::shared_ptr<TRuntimeProfileTree>>&
load_channel_profile) {
+ TQueryProfile profile;
+ std::vector<TUniqueId> fragment_instance_ids;
+ std::vector<TRuntimeProfileTree> instance_profiles;
+ std::vector<TRuntimeProfileTree> load_channel_profiles;
+
+ for (auto entry : instance_id_to_profile) {
+ TUniqueId instance_id = entry.first;
+ std::shared_ptr<TRuntimeProfileTree> profile = entry.second;
+
+ if (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);
+ continue;
+ }
+
+ fragment_instance_ids.push_back(instance_id);
+ instance_profiles.push_back(*profile);
+ }
+
+ profile.__set_query_id(query_id);
+ profile.__set_fragment_instance_ids(fragment_instance_ids);
+ profile.__set_instance_profiles(instance_profiles);
+ profile.__set_load_channel_profiles(load_channel_profiles);
+
+ TReportExecStatusParams res;
+ res.__set_query_profile(profile);
+ res.__set_query_id(TUniqueId());
+ return res;
+}
+
+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);
+ }
+ }
+
+ _report_query_profiles_function();
+
+ {
+ std::lock_guard<std::mutex> lg(_report_profile_mutex);
+
+ if (_report_profile_thread_stop) {
+ LOG_INFO("Report profile thread stopped");
+ return;
+ }
+ }
+ }
+}
+
+void RuntimeQueryStatiticsMgr::trigger_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::unordered_map<TUniqueId, std::shared_ptr<TRuntimeProfileTree>>&
instance_profile_map =
+ std::get<1>(_query_profile_map[query_id]);
+ instance_profile_map.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;
+ }
+ }
+
+ std::vector<std::shared_ptr<TRuntimeProfileTree>>
load_channel_profiles;
+ for (const auto& load_channel_profile : load_channel_profile_copy) {
+ if (load_channel_profile.second == nullptr) {
+ auto msg = fmt::format(
+ "Register fragment profile {} {} failed, load channel
profile is null",
+ print_id(query_id), -1);
+ DCHECK(false) << msg;
+ LOG_ERROR(msg);
+ continue;
+ }
+
+ load_channel_profiles.push_back(load_channel_profile.second);
+ }
+
+ TReportExecStatusParams req =
create_report_exec_status_params_non_pipeline(
+ query_id, instance_id_to_profile, load_channel_profiles);
+ TReportExecStatusResult res;
+ auto rpc_status = _do_report_exec_stats_rpc(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() {
+ decltype(_profile_map_x) profile_copy;
+ decltype(_load_channel_profile_map_x) load_channel_profile_copy;
+
+ {
+ std::lock_guard<std::shared_mutex> lg(_query_profile_map_lock);
+ profile_copy = _profile_map_x;
Review Comment:
不需要这么写，如果每次都是clear的话，直接用swap 就好了。可以避免copy
--
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
