]>
Commit | Line | Data |
---|---|---|
2a845540 | 1 | #include "DaemonMetricCollector.h" |
2a845540 TL |
2 | |
3 | #include <boost/json/src.hpp> | |
4 | #include <chrono> | |
5 | #include <filesystem> | |
6 | #include <iostream> | |
7 | #include <map> | |
8 | #include <memory> | |
9 | #include <regex> | |
10 | #include <string> | |
11 | #include <utility> | |
12 | ||
39ae355f TL |
13 | #include "common/admin_socket_client.h" |
14 | #include "common/debug.h" | |
15 | #include "common/hostname.h" | |
16 | #include "common/perf_counters.h" | |
17 | #include "common/split.h" | |
18 | #include "global/global_context.h" | |
19 | #include "global/global_init.h" | |
20 | #include "include/common_fwd.h" | |
21 | #include "util.h" | |
22 | ||
2a845540 TL |
23 | #define dout_context g_ceph_context |
24 | #define dout_subsys ceph_subsys_ceph_exporter | |
25 | ||
26 | using json_object = boost::json::object; | |
27 | using json_value = boost::json::value; | |
28 | using json_array = boost::json::array; | |
29 | ||
30 | void DaemonMetricCollector::request_loop(boost::asio::steady_timer &timer) { | |
31 | timer.async_wait([&](const boost::system::error_code &e) { | |
32 | std::cerr << e << std::endl; | |
33 | update_sockets(); | |
34 | dump_asok_metrics(); | |
35 | auto stats_period = g_conf().get_val<int64_t>("exporter_stats_period"); | |
36 | // time to wait before sending requests again | |
37 | timer.expires_from_now(std::chrono::seconds(stats_period)); | |
38 | request_loop(timer); | |
39 | }); | |
40 | } | |
41 | ||
42 | void DaemonMetricCollector::main() { | |
43 | // time to wait before sending requests again | |
44 | ||
45 | boost::asio::io_service io; | |
46 | boost::asio::steady_timer timer{io, std::chrono::seconds(0)}; | |
47 | request_loop(timer); | |
48 | io.run(); | |
49 | } | |
50 | ||
51 | std::string DaemonMetricCollector::get_metrics() { | |
52 | const std::lock_guard<std::mutex> lock(metrics_mutex); | |
53 | return metrics; | |
54 | } | |
55 | ||
56 | template <class T> | |
57 | void add_metric(std::unique_ptr<MetricsBuilder> &builder, T value, | |
58 | std::string name, std::string description, std::string mtype, | |
59 | labels_t labels) { | |
60 | builder->add(std::to_string(value), name, description, mtype, labels); | |
61 | } | |
62 | ||
63 | void add_double_or_int_metric(std::unique_ptr<MetricsBuilder> &builder, | |
64 | json_value value, std::string name, | |
65 | std::string description, std::string mtype, | |
66 | labels_t labels) { | |
67 | if (value.is_int64()) { | |
68 | int64_t v = value.as_int64(); | |
69 | add_metric(builder, v, name, description, mtype, labels); | |
70 | } else if (value.is_double()) { | |
71 | double v = value.as_double(); | |
72 | add_metric(builder, v, name, description, mtype, labels); | |
73 | } | |
74 | } | |
75 | ||
76 | std::string boost_string_to_std(boost::json::string js) { | |
77 | std::string res(js.data()); | |
78 | return res; | |
79 | } | |
80 | ||
81 | std::string quote(std::string value) { return "\"" + value + "\""; } | |
82 | ||
83 | bool is_hyphen(char ch) { return ch == '-'; } | |
84 | ||
85 | void DaemonMetricCollector::dump_asok_metrics() { | |
86 | BlockTimer timer(__FILE__, __FUNCTION__); | |
87 | ||
88 | std::vector<std::pair<std::string, int>> daemon_pids; | |
89 | ||
39ae355f | 90 | int failures = 0; |
2a845540 TL |
91 | bool sort = g_conf().get_val<bool>("exporter_sort_metrics"); |
92 | if (sort) { | |
39ae355f TL |
93 | builder = |
94 | std::unique_ptr<OrderedMetricsBuilder>(new OrderedMetricsBuilder()); | |
2a845540 | 95 | } else { |
39ae355f TL |
96 | builder = |
97 | std::unique_ptr<UnorderedMetricsBuilder>(new UnorderedMetricsBuilder()); | |
2a845540 TL |
98 | } |
99 | for (auto &[daemon_name, sock_client] : clients) { | |
100 | bool ok; | |
101 | sock_client.ping(&ok); | |
102 | if (!ok) { | |
39ae355f | 103 | failures++; |
2a845540 TL |
104 | continue; |
105 | } | |
39ae355f TL |
106 | std::string perf_dump_response = |
107 | asok_request(sock_client, "perf dump", daemon_name); | |
2a845540 | 108 | if (perf_dump_response.size() == 0) { |
39ae355f | 109 | failures++; |
2a845540 TL |
110 | continue; |
111 | } | |
39ae355f TL |
112 | std::string perf_schema_response = |
113 | asok_request(sock_client, "perf schema", daemon_name); | |
2a845540 | 114 | if (perf_schema_response.size() == 0) { |
39ae355f TL |
115 | failures++; |
116 | continue; | |
117 | } | |
118 | std::string config_show = | |
119 | asok_request(sock_client, "config show", daemon_name); | |
120 | if (config_show.size() == 0) { | |
121 | failures++; | |
2a845540 TL |
122 | continue; |
123 | } | |
2a845540 TL |
124 | json_object pid_file_json = boost::json::parse(config_show).as_object(); |
125 | std::string pid_path = | |
39ae355f | 126 | boost_string_to_std(pid_file_json["pid_file"].as_string()); |
2a845540 TL |
127 | std::string pid_str = read_file_to_string(pid_path); |
128 | if (!pid_path.size()) { | |
39ae355f TL |
129 | dout(1) << "pid path is empty; process metrics won't be fetched for: " |
130 | << daemon_name << dendl; | |
131 | } | |
132 | if (!pid_str.empty()) { | |
133 | daemon_pids.push_back({daemon_name, std::stoi(pid_str)}); | |
2a845540 | 134 | } |
2a845540 TL |
135 | json_object dump = boost::json::parse(perf_dump_response).as_object(); |
136 | json_object schema = boost::json::parse(perf_schema_response).as_object(); | |
137 | for (auto &perf : schema) { | |
39ae355f | 138 | std::string perf_group = {perf.key().begin(), perf.key().end()}; |
2a845540 TL |
139 | json_object perf_group_object = perf.value().as_object(); |
140 | for (auto &perf_counter : perf_group_object) { | |
39ae355f TL |
141 | std::string perf_name = {perf_counter.key().begin(), |
142 | perf_counter.key().end()}; | |
2a845540 TL |
143 | json_object perf_info = perf_counter.value().as_object(); |
144 | auto prio_limit = g_conf().get_val<int64_t>("exporter_prio_limit"); | |
39ae355f | 145 | if (perf_info["priority"].as_int64() < prio_limit) { |
2a845540 TL |
146 | continue; |
147 | } | |
148 | std::string name = "ceph_" + perf_group + "_" + perf_name; | |
149 | std::replace_if(name.begin(), name.end(), is_hyphen, '_'); | |
150 | ||
151 | // FIXME: test this, based on mgr_module perfpath_to_path_labels | |
152 | auto labels_and_name = get_labels_and_metric_name(daemon_name, name); | |
153 | labels_t labels = labels_and_name.first; | |
154 | name = labels_and_name.second; | |
155 | ||
156 | json_value perf_values = dump[perf_group].as_object()[perf_name]; | |
157 | dump_asok_metric(perf_info, perf_values, name, labels); | |
158 | } | |
159 | } | |
160 | } | |
39ae355f TL |
161 | dout(10) << "Perf counters retrieved for " << clients.size() - failures << "/" |
162 | << clients.size() << " daemons." << dendl; | |
2a845540 TL |
163 | // get time spent on this function |
164 | timer.stop(); | |
39ae355f TL |
165 | std::string scrap_desc( |
166 | "Time spent scraping and transforming perf counters to metrics"); | |
2a845540 TL |
167 | labels_t scrap_labels; |
168 | scrap_labels["host"] = quote(ceph_get_hostname()); | |
169 | scrap_labels["function"] = quote(__FUNCTION__); | |
170 | add_metric(builder, timer.get_ms(), "ceph_exporter_scrape_time", scrap_desc, | |
171 | "gauge", scrap_labels); | |
172 | ||
173 | const std::lock_guard<std::mutex> lock(metrics_mutex); | |
39ae355f TL |
174 | // only get metrics if there's pid path for some or all daemons isn't empty |
175 | if (daemon_pids.size() != 0) { | |
176 | get_process_metrics(daemon_pids); | |
177 | } | |
2a845540 TL |
178 | metrics = builder->dump(); |
179 | } | |
180 | ||
181 | std::vector<std::string> read_proc_stat_file(std::string path) { | |
182 | std::string stat = read_file_to_string(path); | |
183 | std::vector<std::string> strings; | |
184 | auto parts = ceph::split(stat); | |
185 | strings.assign(parts.begin(), parts.end()); | |
186 | return strings; | |
187 | } | |
188 | ||
189 | struct pstat read_pid_stat(int pid) { | |
190 | std::string stat_path("/proc/" + std::to_string(pid) + "/stat"); | |
191 | std::vector<std::string> stats = read_proc_stat_file(stat_path); | |
192 | struct pstat stat; | |
193 | stat.minflt = std::stoul(stats[9]); | |
194 | stat.majflt = std::stoul(stats[11]); | |
195 | stat.utime = std::stoul(stats[13]); | |
196 | stat.stime = std::stoul(stats[14]); | |
197 | stat.num_threads = std::stoul(stats[19]); | |
198 | stat.start_time = std::stoul(stats[21]); | |
199 | stat.vm_size = std::stoul(stats[22]); | |
200 | stat.resident_size = std::stoi(stats[23]); | |
201 | return stat; | |
202 | } | |
203 | ||
39ae355f TL |
204 | void DaemonMetricCollector::get_process_metrics( |
205 | std::vector<std::pair<std::string, int>> daemon_pids) { | |
2a845540 TL |
206 | std::string path("/proc"); |
207 | std::stringstream ss; | |
208 | for (auto &[daemon_name, pid] : daemon_pids) { | |
209 | std::vector<std::string> uptimes = read_proc_stat_file("/proc/uptime"); | |
210 | struct pstat stat = read_pid_stat(pid); | |
211 | int clk_tck = sysconf(_SC_CLK_TCK); | |
212 | double start_time_seconds = stat.start_time / (double)clk_tck; | |
213 | double user_time = stat.utime / (double)clk_tck; | |
214 | double kernel_time = stat.stime / (double)clk_tck; | |
215 | double total_time_seconds = user_time + kernel_time; | |
216 | double uptime = std::stod(uptimes[0]); | |
217 | double elapsed_time = uptime - start_time_seconds; | |
39ae355f | 218 | double idle_time = elapsed_time - total_time_seconds; |
2a845540 TL |
219 | double usage = total_time_seconds * 100 / elapsed_time; |
220 | ||
221 | labels_t labels; | |
222 | labels["ceph_daemon"] = quote(daemon_name); | |
223 | add_metric(builder, stat.minflt, "ceph_exporter_minflt_total", | |
224 | "Number of minor page faults of daemon", "counter", labels); | |
225 | add_metric(builder, stat.majflt, "ceph_exporter_majflt_total", | |
226 | "Number of major page faults of daemon", "counter", labels); | |
227 | add_metric(builder, stat.num_threads, "ceph_exporter_num_threads", | |
228 | "Number of threads used by daemon", "gauge", labels); | |
39ae355f TL |
229 | add_metric(builder, usage, "ceph_exporter_cpu_usage", |
230 | "CPU usage of a daemon", "gauge", labels); | |
2a845540 TL |
231 | |
232 | std::string cpu_time_desc = "Process time in kernel/user/idle mode"; | |
233 | labels_t cpu_total_labels; | |
234 | cpu_total_labels["ceph_daemon"] = quote(daemon_name); | |
235 | cpu_total_labels["mode"] = quote("kernel"); | |
236 | add_metric(builder, kernel_time, "ceph_exporter_cpu_total", cpu_time_desc, | |
237 | "counter", cpu_total_labels); | |
238 | cpu_total_labels["mode"] = quote("user"); | |
239 | add_metric(builder, user_time, "ceph_exporter_cpu_total", cpu_time_desc, | |
240 | "counter", cpu_total_labels); | |
241 | cpu_total_labels["mode"] = quote("idle"); | |
242 | add_metric(builder, idle_time, "ceph_exporter_cpu_total", cpu_time_desc, | |
243 | "counter", cpu_total_labels); | |
39ae355f TL |
244 | add_metric(builder, stat.vm_size, "ceph_exporter_vm_size", |
245 | "Virtual memory used in a daemon", "gauge", labels); | |
2a845540 TL |
246 | add_metric(builder, stat.resident_size, "ceph_exporter_resident_size", |
247 | "Resident memory in a daemon", "gauge", labels); | |
248 | } | |
249 | } | |
250 | ||
251 | std::string DaemonMetricCollector::asok_request(AdminSocketClient &asok, | |
39ae355f TL |
252 | std::string command, |
253 | std::string daemon_name) { | |
2a845540 TL |
254 | std::string request("{\"prefix\": \"" + command + "\"}"); |
255 | std::string response; | |
256 | std::string err = asok.do_request(request, &response); | |
257 | if (err.length() > 0 || response.substr(0, 5) == "ERROR") { | |
39ae355f TL |
258 | dout(1) << "command " << command << "failed for daemon " << daemon_name |
259 | << "with error: " << err << dendl; | |
2a845540 TL |
260 | return ""; |
261 | } | |
262 | return response; | |
263 | } | |
264 | ||
265 | std::pair<labels_t, std::string> | |
266 | DaemonMetricCollector::get_labels_and_metric_name(std::string daemon_name, | |
267 | std::string metric_name) { | |
268 | std::string new_metric_name; | |
269 | labels_t labels; | |
270 | new_metric_name = metric_name; | |
271 | if (daemon_name.find("rgw") != std::string::npos) { | |
272 | std::string tmp = daemon_name.substr(16, std::string::npos); | |
273 | std::string::size_type pos = tmp.find('.'); | |
274 | labels["instance_id"] = quote("rgw." + tmp.substr(0, pos)); | |
275 | } else { | |
276 | labels["ceph_daemon"] = quote(daemon_name); | |
277 | if (daemon_name.find("rbd-mirror") != std::string::npos) { | |
39ae355f TL |
278 | std::regex re( |
279 | "^rbd_mirror_image_([^/]+)/(?:(?:([^/]+)/" | |
280 | ")?)(.*)\\.(replay(?:_bytes|_latency)?)$"); | |
2a845540 TL |
281 | std::smatch match; |
282 | if (std::regex_search(daemon_name, match, re) == true) { | |
283 | new_metric_name = "ceph_rbd_mirror_image_" + match.str(4); | |
284 | labels["pool"] = quote(match.str(1)); | |
285 | labels["namespace"] = quote(match.str(2)); | |
286 | labels["image"] = quote(match.str(3)); | |
287 | } | |
288 | } | |
289 | } | |
290 | return {labels, new_metric_name}; | |
291 | } | |
292 | ||
293 | /* | |
294 | perf_values can be either a int/double or a json_object. Since | |
295 | json_value is a wrapper of both we use that class. | |
296 | */ | |
297 | void DaemonMetricCollector::dump_asok_metric(json_object perf_info, | |
298 | json_value perf_values, | |
299 | std::string name, | |
300 | labels_t labels) { | |
301 | int64_t type = perf_info["type"].as_int64(); | |
302 | std::string metric_type = | |
39ae355f | 303 | boost_string_to_std(perf_info["metric_type"].as_string()); |
2a845540 | 304 | std::string description = |
39ae355f | 305 | boost_string_to_std(perf_info["description"].as_string()); |
2a845540 TL |
306 | |
307 | if (type & PERFCOUNTER_LONGRUNAVG) { | |
308 | int64_t count = perf_values.as_object()["avgcount"].as_int64(); | |
309 | add_metric(builder, count, name + "_count", description, metric_type, | |
310 | labels); | |
311 | json_value sum_value = perf_values.as_object()["sum"]; | |
312 | add_double_or_int_metric(builder, sum_value, name + "_sum", description, | |
313 | metric_type, labels); | |
314 | } else if (type & PERFCOUNTER_TIME) { | |
315 | if (perf_values.is_int64()) { | |
316 | double value = perf_values.as_int64() / 1000000000.0f; | |
317 | add_metric(builder, value, name, description, metric_type, labels); | |
318 | } else if (perf_values.is_double()) { | |
319 | double value = perf_values.as_double() / 1000000000.0f; | |
320 | add_metric(builder, value, name, description, metric_type, labels); | |
321 | } | |
322 | } else { | |
323 | add_double_or_int_metric(builder, perf_values, name, description, | |
324 | metric_type, labels); | |
325 | } | |
326 | } | |
327 | ||
328 | void DaemonMetricCollector::update_sockets() { | |
329 | std::string sock_dir = g_conf().get_val<std::string>("exporter_sock_dir"); | |
330 | clients.clear(); | |
331 | std::filesystem::path sock_path = sock_dir; | |
39ae355f | 332 | if (!std::filesystem::is_directory(sock_path.parent_path())) { |
2a845540 TL |
333 | dout(1) << "ERROR: No such directory exist" << sock_dir << dendl; |
334 | return; | |
335 | } | |
39ae355f | 336 | for (const auto &entry : std::filesystem::directory_iterator(sock_dir)) { |
2a845540 TL |
337 | if (entry.path().extension() == ".asok") { |
338 | std::string daemon_socket_name = entry.path().filename().string(); | |
339 | std::string daemon_name = | |
39ae355f | 340 | daemon_socket_name.substr(0, daemon_socket_name.size() - 5); |
2a845540 TL |
341 | if (clients.find(daemon_name) == clients.end() && |
342 | !(daemon_name.find("mgr") != std::string::npos) && | |
343 | !(daemon_name.find("ceph-exporter") != std::string::npos)) { | |
344 | AdminSocketClient sock(entry.path().string()); | |
345 | clients.insert({daemon_name, std::move(sock)}); | |
346 | } | |
347 | } | |
348 | } | |
349 | } | |
350 | ||
351 | void OrderedMetricsBuilder::add(std::string value, std::string name, | |
352 | std::string description, std::string mtype, | |
353 | labels_t labels) { | |
2a845540 TL |
354 | if (metrics.find(name) == metrics.end()) { |
355 | Metric metric(name, mtype, description); | |
356 | metrics[name] = std::move(metric); | |
357 | } | |
358 | Metric &metric = metrics[name]; | |
359 | metric.add(labels, value); | |
360 | } | |
361 | ||
362 | std::string OrderedMetricsBuilder::dump() { | |
363 | for (auto &[name, metric] : metrics) { | |
364 | out += metric.dump() + "\n"; | |
365 | } | |
366 | return out; | |
367 | } | |
368 | ||
369 | void UnorderedMetricsBuilder::add(std::string value, std::string name, | |
370 | std::string description, std::string mtype, | |
371 | labels_t labels) { | |
2a845540 TL |
372 | Metric metric(name, mtype, description); |
373 | metric.add(labels, value); | |
374 | out += metric.dump() + "\n\n"; | |
375 | } | |
376 | ||
377 | std::string UnorderedMetricsBuilder::dump() { return out; } | |
378 | ||
379 | void Metric::add(labels_t labels, std::string value) { | |
380 | metric_entry entry; | |
381 | entry.labels = labels; | |
382 | entry.value = value; | |
383 | entries.push_back(entry); | |
384 | } | |
385 | ||
386 | std::string Metric::dump() { | |
387 | std::stringstream metric_ss; | |
388 | metric_ss << "# HELP " << name << " " << description << "\n"; | |
389 | metric_ss << "# TYPE " << name << " " << mtype << "\n"; | |
390 | for (auto &entry : entries) { | |
391 | std::stringstream labels_ss; | |
392 | size_t i = 0; | |
393 | for (auto &[label_name, label_value] : entry.labels) { | |
394 | labels_ss << label_name << "=" << label_value; | |
395 | if (i < entry.labels.size() - 1) { | |
396 | labels_ss << ","; | |
397 | } | |
398 | i++; | |
399 | } | |
400 | metric_ss << name << "{" << labels_ss.str() << "} " << entry.value; | |
401 | if (&entry != &entries.back()) { | |
402 | metric_ss << "\n"; | |
403 | } | |
404 | } | |
405 | return metric_ss.str(); | |
406 | } | |
407 | ||
408 | DaemonMetricCollector &collector_instance() { | |
409 | static DaemonMetricCollector instance; | |
410 | return instance; | |
411 | } |