diff options
Diffstat (limited to 'src/exporter/DaemonMetricCollector.cc')
-rw-r--r-- | src/exporter/DaemonMetricCollector.cc | 457 |
1 files changed, 457 insertions, 0 deletions
diff --git a/src/exporter/DaemonMetricCollector.cc b/src/exporter/DaemonMetricCollector.cc new file mode 100644 index 000000000..ebe85c304 --- /dev/null +++ b/src/exporter/DaemonMetricCollector.cc @@ -0,0 +1,457 @@ +#include "DaemonMetricCollector.h" + +#include <boost/json/src.hpp> +#include <chrono> +#include <filesystem> +#include <iostream> +#include <map> +#include <memory> +#include <regex> +#include <sstream> +#include <string> +#include <utility> + +#include "common/admin_socket_client.h" +#include "common/debug.h" +#include "common/hostname.h" +#include "common/perf_counters.h" +#include "common/split.h" +#include "global/global_context.h" +#include "global/global_init.h" +#include "include/common_fwd.h" +#include "util.h" + +#define dout_context g_ceph_context +#define dout_subsys ceph_subsys_ceph_exporter + +using json_object = boost::json::object; +using json_value = boost::json::value; +using json_array = boost::json::array; + +void DaemonMetricCollector::request_loop(boost::asio::steady_timer &timer) { + timer.async_wait([&](const boost::system::error_code &e) { + std::cerr << e << std::endl; + update_sockets(); + dump_asok_metrics(); + auto stats_period = g_conf().get_val<int64_t>("exporter_stats_period"); + // time to wait before sending requests again + timer.expires_from_now(std::chrono::seconds(stats_period)); + request_loop(timer); + }); +} + +void DaemonMetricCollector::main() { + // time to wait before sending requests again + + boost::asio::io_service io; + boost::asio::steady_timer timer{io, std::chrono::seconds(0)}; + request_loop(timer); + io.run(); +} + +std::string DaemonMetricCollector::get_metrics() { + const std::lock_guard<std::mutex> lock(metrics_mutex); + return metrics; +} + +template <class T> +void add_metric(std::unique_ptr<MetricsBuilder> &builder, T value, + std::string name, std::string description, std::string mtype, + labels_t labels) { + builder->add(std::to_string(value), name, description, mtype, labels); +} + +void add_double_or_int_metric(std::unique_ptr<MetricsBuilder> &builder, + json_value value, std::string name, + std::string description, std::string mtype, + labels_t labels) { + if (value.is_int64()) { + int64_t v = value.as_int64(); + add_metric(builder, v, name, description, mtype, labels); + } else if (value.is_double()) { + double v = value.as_double(); + add_metric(builder, v, name, description, mtype, labels); + } +} + +std::string boost_string_to_std(boost::json::string js) { + std::string res(js.data()); + return res; +} + +std::string quote(std::string value) { return "\"" + value + "\""; } + +void DaemonMetricCollector::dump_asok_metrics() { + BlockTimer timer(__FILE__, __FUNCTION__); + + std::vector<std::pair<std::string, int>> daemon_pids; + + int failures = 0; + bool sort = g_conf().get_val<bool>("exporter_sort_metrics"); + if (sort) { + builder = + std::unique_ptr<OrderedMetricsBuilder>(new OrderedMetricsBuilder()); + } else { + builder = + std::unique_ptr<UnorderedMetricsBuilder>(new UnorderedMetricsBuilder()); + } + auto prio_limit = g_conf().get_val<int64_t>("exporter_prio_limit"); + for (auto &[daemon_name, sock_client] : clients) { + bool ok; + sock_client.ping(&ok); + if (!ok) { + failures++; + continue; + } + std::string counter_dump_response = + asok_request(sock_client, "counter dump", daemon_name); + if (counter_dump_response.size() == 0) { + failures++; + continue; + } + std::string counter_schema_response = + asok_request(sock_client, "counter schema", daemon_name); + if (counter_schema_response.size() == 0) { + failures++; + continue; + } + + json_object counter_dump = boost::json::parse(counter_dump_response).as_object(); + json_object counter_schema = boost::json::parse(counter_schema_response).as_object(); + + for (auto &perf_group_item : counter_schema) { + std::string perf_group = {perf_group_item.key().begin(), + perf_group_item.key().end()}; + json_array perf_group_schema_array = perf_group_item.value().as_array(); + json_array perf_group_dump_array = counter_dump[perf_group].as_array(); + for (auto schema_itr = perf_group_schema_array.begin(), + dump_itr = perf_group_dump_array.begin(); + schema_itr != perf_group_schema_array.end() && + dump_itr != perf_group_dump_array.end(); + ++schema_itr, ++dump_itr) { + auto counters = schema_itr->at("counters").as_object(); + auto counters_labels = schema_itr->at("labels").as_object(); + auto counters_values = dump_itr->at("counters").as_object(); + labels_t labels; + + for (auto &label: counters_labels) { + std::string label_key = {label.key().begin(), label.key().end()}; + labels[label_key] = quote(label.value().as_string().c_str()); + } + for (auto &counter : counters) { + json_object counter_group = counter.value().as_object(); + if (counter_group["priority"].as_int64() < prio_limit) { + continue; + } + std::string counter_name_init = {counter.key().begin(), counter.key().end()}; + std::string counter_name = perf_group + "_" + counter_name_init; + promethize(counter_name); + + auto extra_labels = get_extra_labels(daemon_name); + if (extra_labels.empty()) { + dout(1) << "Unable to parse instance_id from daemon_name: " << daemon_name << dendl; + continue; + } + labels.insert(extra_labels.begin(), extra_labels.end()); + + // For now this is only required for rgw multi-site metrics + auto multisite_labels_and_name = add_fixed_name_metrics(counter_name); + if (!multisite_labels_and_name.first.empty()) { + labels.insert(multisite_labels_and_name.first.begin(), multisite_labels_and_name.first.end()); + counter_name = multisite_labels_and_name.second; + } + auto perf_values = counters_values.at(counter_name_init); + dump_asok_metric(counter_group, perf_values, counter_name, labels); + } + } + } + std::string config_show = + asok_request(sock_client, "config show", daemon_name); + if (config_show.size() == 0) { + failures++; + continue; + } + json_object pid_file_json = boost::json::parse(config_show).as_object(); + std::string pid_path = + boost_string_to_std(pid_file_json["pid_file"].as_string()); + std::string pid_str = read_file_to_string(pid_path); + if (!pid_path.size()) { + dout(1) << "pid path is empty; process metrics won't be fetched for: " + << daemon_name << dendl; + } + if (!pid_str.empty()) { + daemon_pids.push_back({daemon_name, std::stoi(pid_str)}); + } + } + dout(10) << "Perf counters retrieved for " << clients.size() - failures << "/" + << clients.size() << " daemons." << dendl; + // get time spent on this function + timer.stop(); + std::string scrap_desc( + "Time spent scraping and transforming perf counters to metrics"); + labels_t scrap_labels; + scrap_labels["host"] = quote(ceph_get_hostname()); + scrap_labels["function"] = quote(__FUNCTION__); + add_metric(builder, timer.get_ms(), "ceph_exporter_scrape_time", scrap_desc, + "gauge", scrap_labels); + + const std::lock_guard<std::mutex> lock(metrics_mutex); + // only get metrics if there's pid path for some or all daemons isn't empty + if (daemon_pids.size() != 0) { + get_process_metrics(daemon_pids); + } + metrics = builder->dump(); +} + +std::vector<std::string> read_proc_stat_file(std::string path) { + std::string stat = read_file_to_string(path); + std::vector<std::string> strings; + auto parts = ceph::split(stat); + strings.assign(parts.begin(), parts.end()); + return strings; +} + +struct pstat read_pid_stat(int pid) { + std::string stat_path("/proc/" + std::to_string(pid) + "/stat"); + std::vector<std::string> stats = read_proc_stat_file(stat_path); + struct pstat stat; + stat.minflt = std::stoul(stats[9]); + stat.majflt = std::stoul(stats[11]); + stat.utime = std::stoul(stats[13]); + stat.stime = std::stoul(stats[14]); + stat.num_threads = std::stoul(stats[19]); + stat.start_time = std::stoul(stats[21]); + stat.vm_size = std::stoul(stats[22]); + stat.resident_size = std::stoi(stats[23]); + return stat; +} + +void DaemonMetricCollector::get_process_metrics( + std::vector<std::pair<std::string, int>> daemon_pids) { + std::string path("/proc"); + std::stringstream ss; + for (auto &[daemon_name, pid] : daemon_pids) { + std::vector<std::string> uptimes = read_proc_stat_file("/proc/uptime"); + struct pstat stat = read_pid_stat(pid); + int clk_tck = sysconf(_SC_CLK_TCK); + double start_time_seconds = stat.start_time / (double)clk_tck; + double user_time = stat.utime / (double)clk_tck; + double kernel_time = stat.stime / (double)clk_tck; + double total_time_seconds = user_time + kernel_time; + double uptime = std::stod(uptimes[0]); + double elapsed_time = uptime - start_time_seconds; + double idle_time = elapsed_time - total_time_seconds; + double usage = total_time_seconds * 100 / elapsed_time; + + labels_t labels; + labels["ceph_daemon"] = quote(daemon_name); + add_metric(builder, stat.minflt, "ceph_exporter_minflt_total", + "Number of minor page faults of daemon", "counter", labels); + add_metric(builder, stat.majflt, "ceph_exporter_majflt_total", + "Number of major page faults of daemon", "counter", labels); + add_metric(builder, stat.num_threads, "ceph_exporter_num_threads", + "Number of threads used by daemon", "gauge", labels); + add_metric(builder, usage, "ceph_exporter_cpu_usage", + "CPU usage of a daemon", "gauge", labels); + + std::string cpu_time_desc = "Process time in kernel/user/idle mode"; + labels_t cpu_total_labels; + cpu_total_labels["ceph_daemon"] = quote(daemon_name); + cpu_total_labels["mode"] = quote("kernel"); + add_metric(builder, kernel_time, "ceph_exporter_cpu_total", cpu_time_desc, + "counter", cpu_total_labels); + cpu_total_labels["mode"] = quote("user"); + add_metric(builder, user_time, "ceph_exporter_cpu_total", cpu_time_desc, + "counter", cpu_total_labels); + cpu_total_labels["mode"] = quote("idle"); + add_metric(builder, idle_time, "ceph_exporter_cpu_total", cpu_time_desc, + "counter", cpu_total_labels); + add_metric(builder, stat.vm_size, "ceph_exporter_vm_size", + "Virtual memory used in a daemon", "gauge", labels); + add_metric(builder, stat.resident_size, "ceph_exporter_resident_size", + "Resident memory in a daemon", "gauge", labels); + } +} + +std::string DaemonMetricCollector::asok_request(AdminSocketClient &asok, + std::string command, + std::string daemon_name) { + std::string request("{\"prefix\": \"" + command + "\"}"); + std::string response; + std::string err = asok.do_request(request, &response); + if (err.length() > 0 || response.substr(0, 5) == "ERROR") { + dout(1) << "command " << command << "failed for daemon " << daemon_name + << "with error: " << err << dendl; + return ""; + } + return response; +} + +labels_t DaemonMetricCollector::get_extra_labels(std::string daemon_name) { + labels_t labels; + const std::string ceph_daemon_prefix = "ceph-"; + const std::string ceph_client_prefix = "client."; + if (daemon_name.rfind(ceph_daemon_prefix, 0) == 0) { + daemon_name = daemon_name.substr(ceph_daemon_prefix.size()); + } + if (daemon_name.rfind(ceph_client_prefix, 0) == 0) { + daemon_name = daemon_name.substr(ceph_client_prefix.size()); + } + // In vstart cluster socket files for rgw are stored as radosgw.<instance_id>.asok + if (daemon_name.find("radosgw") != std::string::npos) { + std::size_t pos = daemon_name.find_last_of('.'); + std::string tmp = daemon_name.substr(pos+1); + labels["instance_id"] = quote(tmp); + } + else if (daemon_name.find("rgw") != std::string::npos) { + // fetch intance_id for e.g. "hrgsea" from daemon_name=rgw.foo.ceph-node-00.hrgsea.2.94739968030880 + std::vector<std::string> elems; + std::stringstream ss; + ss.str(daemon_name); + std::string item; + while (std::getline(ss, item, '.')) { + elems.push_back(item); + } + if (elems.size() >= 4) { + labels["instance_id"] = quote(elems[3]); + } else { + return labels_t(); + } + } else { + labels.insert({"ceph_daemon", quote(daemon_name)}); + } + return labels; +} + +// Add fixed name metrics from existing ones that have details in their names +// that should be in labels (not in name). For backward compatibility, +// a new fixed name metric is created (instead of replacing)and details are put +// in new labels. Intended for RGW sync perf. counters but extendable as required. +// See: https://tracker.ceph.com/issues/45311 +std::pair<labels_t, std::string> +DaemonMetricCollector::add_fixed_name_metrics(std::string metric_name) { + std::string new_metric_name; + labels_t labels; + new_metric_name = metric_name; + + std::regex re("^data_sync_from_(.*)\\."); + std::smatch match; + if (std::regex_search(metric_name, match, re) == true) { + new_metric_name = std::regex_replace(metric_name, re, "from_([^.]*)', 'from_zone"); + labels["source_zone"] = quote(match.str(1)); + return {labels, new_metric_name}; + } + return {}; +} + +/* +perf_values can be either a int/double or a json_object. Since + json_value is a wrapper of both we use that class. + */ +void DaemonMetricCollector::dump_asok_metric(json_object perf_info, + json_value perf_values, + std::string name, + labels_t labels) { + int64_t type = perf_info["type"].as_int64(); + std::string metric_type = + boost_string_to_std(perf_info["metric_type"].as_string()); + std::string description = + boost_string_to_std(perf_info["description"].as_string()); + + if (type & PERFCOUNTER_LONGRUNAVG) { + int64_t count = perf_values.as_object()["avgcount"].as_int64(); + add_metric(builder, count, name + "_count", description + " Count", "counter", + labels); + json_value sum_value = perf_values.as_object()["sum"]; + add_double_or_int_metric(builder, sum_value, name + "_sum", description + " Total", + metric_type, labels); + } else { + add_double_or_int_metric(builder, perf_values, name, description, + metric_type, labels); + } +} + +void DaemonMetricCollector::update_sockets() { + std::string sock_dir = g_conf().get_val<std::string>("exporter_sock_dir"); + clients.clear(); + std::filesystem::path sock_path = sock_dir; + if (!std::filesystem::is_directory(sock_path.parent_path())) { + dout(1) << "ERROR: No such directory exist" << sock_dir << dendl; + return; + } + for (const auto &entry : std::filesystem::directory_iterator(sock_dir)) { + if (entry.path().extension() == ".asok") { + std::string daemon_socket_name = entry.path().filename().string(); + std::string daemon_name = + daemon_socket_name.substr(0, daemon_socket_name.size() - 5); + if (clients.find(daemon_name) == clients.end() && + !(daemon_name.find("mgr") != std::string::npos) && + !(daemon_name.find("ceph-exporter") != std::string::npos)) { + AdminSocketClient sock(entry.path().string()); + clients.insert({daemon_name, std::move(sock)}); + } + } + } +} + +void OrderedMetricsBuilder::add(std::string value, std::string name, + std::string description, std::string mtype, + labels_t labels) { + if (metrics.find(name) == metrics.end()) { + Metric metric(name, mtype, description); + metrics[name] = std::move(metric); + } + Metric &metric = metrics[name]; + metric.add(labels, value); +} + +std::string OrderedMetricsBuilder::dump() { + for (auto &[name, metric] : metrics) { + out += metric.dump() + "\n"; + } + return out; +} + +void UnorderedMetricsBuilder::add(std::string value, std::string name, + std::string description, std::string mtype, + labels_t labels) { + Metric metric(name, mtype, description); + metric.add(labels, value); + out += metric.dump() + "\n\n"; +} + +std::string UnorderedMetricsBuilder::dump() { return out; } + +void Metric::add(labels_t labels, std::string value) { + metric_entry entry; + entry.labels = labels; + entry.value = value; + entries.push_back(entry); +} + +std::string Metric::dump() { + std::stringstream metric_ss; + metric_ss << "# HELP " << name << " " << description << "\n"; + metric_ss << "# TYPE " << name << " " << mtype << "\n"; + for (auto &entry : entries) { + std::stringstream labels_ss; + size_t i = 0; + for (auto &[label_name, label_value] : entry.labels) { + labels_ss << label_name << "=" << label_value; + if (i < entry.labels.size() - 1) { + labels_ss << ","; + } + i++; + } + metric_ss << name << "{" << labels_ss.str() << "} " << entry.value; + if (&entry != &entries.back()) { + metric_ss << "\n"; + } + } + return metric_ss.str(); +} + +DaemonMetricCollector &collector_instance() { + static DaemonMetricCollector instance; + return instance; +} |