bobhan1 commented on code in PR #61516: URL: https://github.com/apache/doris/pull/61516#discussion_r3008600836
########## cloud/src/meta-service/meta_service_rate_limit_helper.cpp: ########## @@ -0,0 +1,434 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and limitations +// under the License. + +#include "meta-service/meta_service_rate_limit_helper.h" + +#include <fmt/format.h> +#include <sys/resource.h> +#include <sys/sysinfo.h> + +#include <algorithm> +#include <atomic> +#include <chrono> +#include <condition_variable> +#include <cstdint> +#include <deque> +#include <fstream> +#include <memory> +#include <mutex> +#include <random> +#include <string> +#include <thread> + +#include "common/config.h" +#include "common/logging.h" + +namespace doris::cloud { +namespace { +constexpr int64_t kNanosecondsPerMillisecond = 1000 * 1000; +constexpr int64_t kInvalidPercent = -1; + +struct WindowSample { + int64_t second {0}; + int64_t fdb_client_thread_busyness_percent {BVAR_FDB_INVALID_VALUE}; + int64_t ms_cpu_usage_percent {kInvalidPercent}; + int64_t ms_memory_usage_percent {kInvalidPercent}; +}; + +struct ProcessResourceSample { + int64_t cpu_usage_percent {kInvalidPercent}; + int64_t memory_usage_percent {kInvalidPercent}; +}; + +class ProcessResourceSampler { +public: + ProcessResourceSample sample() { + using namespace std::chrono; + + const auto now = steady_clock::now(); + const int64_t current_cpu_time_ns = get_process_cpu_time_ns(); + ProcessResourceSample sample; + sample.memory_usage_percent = get_process_memory_usage_percent(); + + const auto current_wall_time_ns = + duration_cast<nanoseconds>(now.time_since_epoch()).count(); + std::lock_guard lock(mutex_); + if (last_cpu_time_ns_ != kInvalidPercent && current_cpu_time_ns != kInvalidPercent && + current_wall_time_ns > last_wall_time_ns_) { + const double delta_cpu_ns = current_cpu_time_ns - last_cpu_time_ns_; + const double delta_wall_ns = current_wall_time_ns - last_wall_time_ns_; + const uint32_t cpu_cores = std::max<uint32_t>(1, std::thread::hardware_concurrency()); + sample.cpu_usage_percent = + static_cast<int64_t>(delta_cpu_ns * 100.0 / delta_wall_ns / cpu_cores); + } + last_cpu_time_ns_ = current_cpu_time_ns; + last_wall_time_ns_ = current_wall_time_ns; + return sample; + } + +private: + static int64_t get_process_cpu_time_ns() { + rusage usage {}; + if (getrusage(RUSAGE_SELF, &usage) != 0) { + return kInvalidPercent; + } + return usage.ru_utime.tv_sec * 1000L * 1000 * 1000 + usage.ru_utime.tv_usec * 1000L + + usage.ru_stime.tv_sec * 1000L * 1000 * 1000 + usage.ru_stime.tv_usec * 1000L; + } + + static int64_t get_process_memory_usage_percent() { + std::ifstream status("/proc/self/status"); + if (!status.is_open()) { + return kInvalidPercent; + } + + int64_t rss_kb = kInvalidPercent; + std::string line; + while (std::getline(status, line)) { + if (!line.starts_with("VmRSS:")) { + continue; + } + size_t pos = std::string("VmRSS:").size(); + while (pos < line.size() && line[pos] == ' ') { + ++pos; + } + rss_kb = std::stoll(line.substr(pos)); + break; + } + if (rss_kb == kInvalidPercent) { + return kInvalidPercent; + } + + struct sysinfo info {}; + if (sysinfo(&info) != 0) { + return kInvalidPercent; + } + const double total_memory_bytes = + static_cast<double>(info.totalram) * static_cast<double>(info.mem_unit); + if (total_memory_bytes <= 0) { + return kInvalidPercent; + } + const double rss_bytes = static_cast<double>(rss_kb) * 1024.0; + return static_cast<int64_t>(rss_bytes * 100.0 / total_memory_bytes); + } + + std::mutex mutex_; + int64_t last_cpu_time_ns_ {kInvalidPercent}; + int64_t last_wall_time_ns_ {0}; +}; + +MsStressMetrics collect_ms_stress_metrics(ProcessResourceSampler* sampler) { + MsStressMetrics metrics; + metrics.fdb_commit_latency_ns = g_bvar_fdb_latency_probe_commit_ns.get_value(); + metrics.fdb_read_latency_ns = g_bvar_fdb_latency_probe_read_ns.get_value(); + metrics.fdb_performance_limited_by_name = g_bvar_fdb_performance_limited_by_name.get_value(); + metrics.fdb_client_thread_busyness_percent = + g_bvar_fdb_client_thread_busyness_percent.get_value(); + const auto resource_sample = sampler->sample(); + metrics.ms_cpu_usage_percent = resource_sample.cpu_usage_percent; + metrics.ms_memory_usage_percent = resource_sample.memory_usage_percent; + return metrics; +} + +class MsStressDetector { +public: + ~MsStressDetector() { stop(); } + + // Compute decision from metrics and store it in latest_decision_. + // Called by the background thread or synchronously in tests. + void update(int64_t now_ms, const MsStressMetrics& metrics) { + auto decision = std::make_shared<MsStressDecision>(); + decision->fdb_commit_latency_ns = metrics.fdb_commit_latency_ns; + decision->fdb_read_latency_ns = metrics.fdb_read_latency_ns; + decision->fdb_performance_limited_by_name = metrics.fdb_performance_limited_by_name; + decision->fdb_client_thread_busyness_percent = metrics.fdb_client_thread_busyness_percent; + decision->ms_cpu_usage_percent = metrics.ms_cpu_usage_percent; + decision->ms_memory_usage_percent = metrics.ms_memory_usage_percent; + const bool commit_latency_high = + metrics.fdb_commit_latency_ns != BVAR_FDB_INVALID_VALUE && + metrics.fdb_commit_latency_ns > + config::ms_rate_limit_fdb_commit_latency_ms * kNanosecondsPerMillisecond; + const bool read_latency_high = + metrics.fdb_read_latency_ns != BVAR_FDB_INVALID_VALUE && + metrics.fdb_read_latency_ns > + config::ms_rate_limit_fdb_read_latency_ms * kNanosecondsPerMillisecond; + decision->fdb_cluster_under_pressure = + (commit_latency_high || read_latency_high) && + metrics.fdb_performance_limited_by_name != BVAR_FDB_INVALID_VALUE && + metrics.fdb_performance_limited_by_name != 0; + + const int64_t current_second = now_ms / 1000; + // No mutex needed: update() is only called from a single thread + // (background thread in production, test thread in tests). + record_sample(current_second, metrics); + + const double avg_busyness = + get_window_avg(current_second, &WindowSample::fdb_client_thread_busyness_percent, + BVAR_FDB_INVALID_VALUE); + decision->fdb_client_thread_busyness_avg_percent = avg_busyness; + if (avg_busyness >= 0 && + metrics.fdb_client_thread_busyness_percent != BVAR_FDB_INVALID_VALUE) { + decision->fdb_client_thread_under_pressure = + avg_busyness > config::ms_rate_limit_fdb_client_thread_busyness_avg_percent && + metrics.fdb_client_thread_busyness_percent > + config::ms_rate_limit_fdb_client_thread_busyness_instant_percent; + } + + const double avg_cpu = get_window_avg(current_second, &WindowSample::ms_cpu_usage_percent, + kInvalidPercent); + const double avg_memory = get_window_avg( + current_second, &WindowSample::ms_memory_usage_percent, kInvalidPercent); + decision->ms_cpu_usage_avg_percent = avg_cpu; + decision->ms_memory_usage_avg_percent = avg_memory; + if (avg_cpu >= 0 && metrics.ms_cpu_usage_percent != kInvalidPercent) { + decision->ms_resource_under_pressure = + metrics.ms_cpu_usage_percent > config::ms_rate_limit_cpu_usage_percent && + avg_cpu > config::ms_rate_limit_cpu_usage_percent; + } + if (avg_memory >= 0 && metrics.ms_memory_usage_percent != kInvalidPercent) { + decision->ms_resource_under_pressure = + decision->ms_resource_under_pressure || + (metrics.ms_memory_usage_percent > config::ms_rate_limit_memory_usage_percent && + avg_memory > config::ms_rate_limit_memory_usage_percent); + } + latest_decision_.store(std::move(decision)); + } + + // Lock-free read of the latest decision. Returns nullptr before first update. + std::shared_ptr<const MsStressDecision> get_latest_decision() const { + return latest_decision_.load(); + } + + void reset() { samples_.clear(); } + + // Start the background thread that periodically collects metrics and updates. + void start() { + std::unique_lock lock(mtx_); + if (running_.load() != 0) { + return; + } + running_.store(1); + bg_thread_ = std::make_unique<std::thread>([this] { + pthread_setname_np(pthread_self(), "ms_stress_det"); + LOG(INFO) << "MsStressDetector background thread started"; + ProcessResourceSampler sampler; + while (running_.load() == 1) { Review Comment: `std::bad_alloc` may be thrown in `auto decision = std::make_shared<MsStressDecision>();`/`samples_.push_back(sample);` and cause this thread to terminate? -- 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: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
