1use {
4 crate::{counter::CounterPoint, datapoint::DataPoint},
5 crossbeam_channel::{unbounded, Receiver, Sender, TryRecvError},
6 gethostname::gethostname,
7 log::*,
8 solana_cluster_type::ClusterType,
9 solana_sha256_hasher::hash,
10 std::{
11 cmp,
12 collections::HashMap,
13 convert::Into,
14 env,
15 fmt::Write,
16 sync::{Arc, Barrier, Mutex, Once, RwLock},
17 thread,
18 time::{Duration, Instant, UNIX_EPOCH},
19 },
20 thiserror::Error,
21};
22
23type CounterMap = HashMap<(&'static str, u64), CounterPoint>;
24
25#[derive(Debug, Error)]
26pub enum MetricsError {
27 #[error(transparent)]
28 VarError(#[from] env::VarError),
29 #[error(transparent)]
30 ReqwestError(#[from] reqwest::Error),
31 #[error("SOLANA_METRICS_CONFIG is invalid: '{0}'")]
32 ConfigInvalid(String),
33 #[error("SOLANA_METRICS_CONFIG is incomplete")]
34 ConfigIncomplete,
35 #[error("SOLANA_METRICS_CONFIG database mismatch: {0}")]
36 DbMismatch(String),
37}
38
39impl From<MetricsError> for String {
40 fn from(error: MetricsError) -> Self {
41 error.to_string()
42 }
43}
44
45impl From<&CounterPoint> for DataPoint {
46 fn from(counter_point: &CounterPoint) -> Self {
47 let mut point = Self::new(counter_point.name);
48 point.timestamp = counter_point.timestamp;
49 point.add_field_i64("count", counter_point.count);
50 point
51 }
52}
53
54#[derive(Debug)]
55enum MetricsCommand {
56 Flush(Arc<Barrier>),
57 Submit(DataPoint, log::Level),
58 SubmitCounter(CounterPoint, log::Level, u64),
59}
60
61pub struct MetricsAgent {
62 sender: Sender<MetricsCommand>,
63}
64
65pub trait MetricsWriter {
66 fn write(&self, points: Vec<DataPoint>);
69}
70
71struct InfluxDbMetricsWriter {
72 write_url: Option<String>,
73}
74
75impl InfluxDbMetricsWriter {
76 fn new() -> Self {
77 Self {
78 write_url: Self::build_write_url().ok(),
79 }
80 }
81
82 fn build_write_url() -> Result<String, MetricsError> {
83 let config = get_metrics_config().map_err(|err| {
84 info!("metrics disabled: {}", err);
85 err
86 })?;
87
88 info!(
89 "metrics configuration: host={} db={} username={}",
90 config.host, config.db, config.username
91 );
92
93 let write_url = format!(
94 "{}/write?db={}&u={}&p={}&precision=n",
95 &config.host, &config.db, &config.username, &config.password
96 );
97
98 Ok(write_url)
99 }
100}
101
102pub fn serialize_points(points: &Vec<DataPoint>, host_id: &str) -> String {
103 const TIMESTAMP_LEN: usize = 20;
104 const HOST_ID_LEN: usize = 8; const EXTRA_LEN: usize = 2; let mut len = 0;
107 for point in points {
108 for (name, value) in &point.fields {
109 len += name.len() + value.len() + EXTRA_LEN;
110 }
111 for (name, value) in &point.tags {
112 len += name.len() + value.len() + EXTRA_LEN;
113 }
114 len += point.name.len();
115 len += TIMESTAMP_LEN;
116 len += host_id.len() + HOST_ID_LEN;
117 }
118 let mut line = String::with_capacity(len);
119 for point in points {
120 let _ = write!(line, "{},host_id={}", &point.name, host_id);
121 for (name, value) in point.tags.iter() {
122 let _ = write!(line, ",{name}={value}");
123 }
124
125 let mut first = true;
126 for (name, value) in point.fields.iter() {
127 let _ = write!(line, "{}{}={}", if first { ' ' } else { ',' }, name, value);
128 first = false;
129 }
130 let timestamp = point.timestamp.duration_since(UNIX_EPOCH);
131 let nanos = timestamp.unwrap().as_nanos();
132 let _ = writeln!(line, " {nanos}");
133 }
134 line
135}
136
137impl MetricsWriter for InfluxDbMetricsWriter {
138 fn write(&self, points: Vec<DataPoint>) {
139 if let Some(ref write_url) = self.write_url {
140 debug!("submitting {} points", points.len());
141
142 let host_id = HOST_ID.read().unwrap();
143
144 let line = serialize_points(&points, &host_id);
145
146 let client = reqwest::blocking::Client::builder()
147 .timeout(Duration::from_secs(5))
148 .build();
149 let client = match client {
150 Ok(client) => client,
151 Err(err) => {
152 warn!("client instantiation failed: {}", err);
153 return;
154 }
155 };
156
157 let response = client.post(write_url.as_str()).body(line).send();
158 if let Ok(resp) = response {
159 let status = resp.status();
160 if !status.is_success() {
161 let text = resp
162 .text()
163 .unwrap_or_else(|_| "[text body empty]".to_string());
164 warn!("submit response unsuccessful: {} {}", status, text,);
165 }
166 } else {
167 warn!("submit error: {}", response.unwrap_err());
168 }
169 }
170 }
171}
172
173impl Default for MetricsAgent {
174 fn default() -> Self {
175 let max_points_per_sec = env::var("SOLANA_METRICS_MAX_POINTS_PER_SECOND")
176 .map(|x| {
177 x.parse()
178 .expect("Failed to parse SOLANA_METRICS_MAX_POINTS_PER_SECOND")
179 })
180 .unwrap_or(4000);
181
182 Self::new(
183 Arc::new(InfluxDbMetricsWriter::new()),
184 Duration::from_secs(10),
185 max_points_per_sec,
186 )
187 }
188}
189
190impl MetricsAgent {
191 pub fn new(
192 writer: Arc<dyn MetricsWriter + Send + Sync>,
193 write_frequency: Duration,
194 max_points_per_sec: usize,
195 ) -> Self {
196 let (sender, receiver) = unbounded::<MetricsCommand>();
197
198 thread::Builder::new()
199 .name("solMetricsAgent".into())
200 .spawn(move || Self::run(&receiver, &writer, write_frequency, max_points_per_sec))
201 .unwrap();
202
203 Self { sender }
204 }
205
206 fn combine_points(
215 max_points: usize,
216 max_points_per_sec: usize,
217 secs_since_last_write: u64,
218 points_buffered: usize,
219 points: &mut Vec<DataPoint>,
220 counters: &mut CounterMap,
221 ) -> Vec<DataPoint> {
222 let max_points = max_points.saturating_sub(1);
224
225 let num_points = points.len().saturating_add(counters.len());
226 let fit_counters = max_points.saturating_sub(points.len());
227 let points_written = cmp::min(num_points, max_points);
228
229 debug!("run: attempting to write {} points", num_points);
230
231 if num_points > max_points {
232 warn!(
233 "Max submission rate of {} datapoints per second exceeded. Only the \
234 first {} of {} points will be submitted.",
235 max_points_per_sec, max_points, num_points
236 );
237 }
238
239 let mut combined = std::mem::take(points);
240 combined.truncate(points_written);
241
242 combined.extend(counters.values().take(fit_counters).map(|v| v.into()));
243 counters.clear();
244
245 combined.push(
246 DataPoint::new("metrics")
247 .add_field_i64("points_written", points_written as i64)
248 .add_field_i64("num_points", num_points as i64)
249 .add_field_i64("points_lost", (num_points - points_written) as i64)
250 .add_field_i64("points_buffered", points_buffered as i64)
251 .add_field_i64("secs_since_last_write", secs_since_last_write as i64)
252 .to_owned(),
253 );
254
255 combined
256 }
257
258 fn write(
263 writer: &Arc<dyn MetricsWriter + Send + Sync>,
264 max_points: usize,
265 max_points_per_sec: usize,
266 last_write_time: Instant,
267 points_buffered: usize,
268 points: &mut Vec<DataPoint>,
269 counters: &mut CounterMap,
270 ) -> Instant {
271 let now = Instant::now();
272 let secs_since_last_write = now.duration_since(last_write_time).as_secs();
273
274 writer.write(Self::combine_points(
275 max_points,
276 max_points_per_sec,
277 secs_since_last_write,
278 points_buffered,
279 points,
280 counters,
281 ));
282
283 now
284 }
285
286 fn run(
287 receiver: &Receiver<MetricsCommand>,
288 writer: &Arc<dyn MetricsWriter + Send + Sync>,
289 write_frequency: Duration,
290 max_points_per_sec: usize,
291 ) {
292 trace!("run: enter");
293 let mut last_write_time = Instant::now();
294 let mut points = Vec::<DataPoint>::new();
295 let mut counters = CounterMap::new();
296
297 let max_points = write_frequency.as_secs() as usize * max_points_per_sec;
298
299 let write = |last_write_time: Instant,
301 points: &mut Vec<DataPoint>,
302 counters: &mut CounterMap|
303 -> Instant {
304 Self::write(
305 writer,
306 max_points,
307 max_points_per_sec,
308 last_write_time,
309 receiver.len(),
310 points,
311 counters,
312 )
313 };
314
315 loop {
316 match receiver.try_recv() {
317 Ok(cmd) => match cmd {
318 MetricsCommand::Flush(barrier) => {
319 debug!("metrics_thread: flush");
320 last_write_time = write(last_write_time, &mut points, &mut counters);
321 barrier.wait();
322 }
323 MetricsCommand::Submit(point, level) => {
324 log!(level, "{}", point);
325 points.push(point);
326 }
327 MetricsCommand::SubmitCounter(counter, _level, bucket) => {
328 debug!("{:?}", counter);
329 let key = (counter.name, bucket);
330 if let Some(value) = counters.get_mut(&key) {
331 value.count += counter.count;
332 } else {
333 counters.insert(key, counter);
334 }
335 }
336 },
337 Err(TryRecvError::Empty) => {
338 std::thread::sleep(Duration::from_millis(5));
339 }
340 Err(TryRecvError::Disconnected) => {
341 debug!("run: sender disconnected");
342 break;
343 }
344 };
345
346 let now = Instant::now();
347 if now.duration_since(last_write_time) >= write_frequency {
348 last_write_time = write(last_write_time, &mut points, &mut counters);
349 }
350 }
351
352 debug_assert!(
353 points.is_empty() && counters.is_empty(),
354 "Controlling `MetricsAgent` is expected to call `flush()` from the `Drop` \n\
355 implementation, before exiting. So both `points` and `counters` must be empty at \n\
356 this point.\n\
357 `points`: {points:?}\n\
358 `counters`: {counters:?}",
359 );
360
361 trace!("run: exit");
362 }
363
364 pub fn submit(&self, point: DataPoint, level: log::Level) {
365 self.sender
366 .send(MetricsCommand::Submit(point, level))
367 .unwrap();
368 }
369
370 pub fn submit_counter(&self, counter: CounterPoint, level: log::Level, bucket: u64) {
371 self.sender
372 .send(MetricsCommand::SubmitCounter(counter, level, bucket))
373 .unwrap();
374 }
375
376 pub fn flush(&self) {
377 debug!("Flush");
378 let barrier = Arc::new(Barrier::new(2));
379 self.sender
380 .send(MetricsCommand::Flush(Arc::clone(&barrier)))
381 .unwrap();
382
383 barrier.wait();
384 }
385}
386
387impl Drop for MetricsAgent {
388 fn drop(&mut self) {
389 self.flush();
390 }
391}
392
393fn get_singleton_agent() -> &'static MetricsAgent {
394 static AGENT: std::sync::LazyLock<MetricsAgent> =
395 std::sync::LazyLock::new(MetricsAgent::default);
396 &AGENT
397}
398
399static HOST_ID: std::sync::LazyLock<RwLock<String>> = std::sync::LazyLock::new(|| {
400 RwLock::new({
401 let hostname: String = gethostname()
402 .into_string()
403 .unwrap_or_else(|_| "".to_string());
404 format!("{}", hash(hostname.as_bytes()))
405 })
406});
407
408pub fn set_host_id(host_id: String) {
409 info!("host id: {}", host_id);
410 *HOST_ID.write().unwrap() = host_id;
411}
412
413pub fn submit(point: DataPoint, level: log::Level) {
416 let agent = get_singleton_agent();
417 agent.submit(point, level);
418}
419
420pub(crate) fn submit_counter(point: CounterPoint, level: log::Level, bucket: u64) {
423 let agent = get_singleton_agent();
424 agent.submit_counter(point, level, bucket);
425}
426
427#[derive(Debug, Default)]
428struct MetricsConfig {
429 pub host: String,
430 pub db: String,
431 pub username: String,
432 pub password: String,
433}
434
435impl MetricsConfig {
436 fn complete(&self) -> bool {
437 !(self.host.is_empty()
438 || self.db.is_empty()
439 || self.username.is_empty()
440 || self.password.is_empty())
441 }
442}
443
444fn get_metrics_config() -> Result<MetricsConfig, MetricsError> {
445 let mut config = MetricsConfig::default();
446 let config_var = env::var("SOLANA_METRICS_CONFIG")?;
447 if config_var.is_empty() {
448 Err(env::VarError::NotPresent)?;
449 }
450
451 for pair in config_var.split(',') {
452 let nv: Vec<_> = pair.split('=').collect();
453 if nv.len() != 2 {
454 return Err(MetricsError::ConfigInvalid(pair.to_string()));
455 }
456 let v = nv[1].to_string();
457 match nv[0] {
458 "host" => config.host = v,
459 "db" => config.db = v,
460 "u" => config.username = v,
461 "p" => config.password = v,
462 _ => return Err(MetricsError::ConfigInvalid(pair.to_string())),
463 }
464 }
465
466 if !config.complete() {
467 return Err(MetricsError::ConfigIncomplete);
468 }
469
470 Ok(config)
471}
472
473pub fn metrics_config_sanity_check(cluster_type: ClusterType) -> Result<(), MetricsError> {
474 let config = match get_metrics_config() {
475 Ok(config) => config,
476 Err(MetricsError::VarError(env::VarError::NotPresent)) => return Ok(()),
477 Err(e) => return Err(e),
478 };
479 match &config.db[..] {
480 "mainnet-beta" if cluster_type != ClusterType::MainnetBeta => (),
481 "tds" if cluster_type != ClusterType::Testnet => (),
482 "devnet" if cluster_type != ClusterType::Devnet => (),
483 _ => return Ok(()),
484 };
485 let (host, db) = (&config.host, &config.db);
486 let msg = format!("cluster_type={cluster_type:?} host={host} database={db}");
487 Err(MetricsError::DbMismatch(msg))
488}
489
490pub fn query(q: &str) -> Result<String, MetricsError> {
491 let config = get_metrics_config()?;
492 let query_url = format!(
493 "{}/query?u={}&p={}&q={}",
494 &config.host, &config.username, &config.password, &q
495 );
496
497 let response = reqwest::blocking::get(query_url.as_str())?.text()?;
498
499 Ok(response)
500}
501
502pub fn flush() {
505 let agent = get_singleton_agent();
506 agent.flush();
507}
508
509pub fn set_panic_hook(program: &'static str, version: Option<String>) {
511 static SET_HOOK: Once = Once::new();
512 SET_HOOK.call_once(|| {
513 let default_hook = std::panic::take_hook();
514 std::panic::set_hook(Box::new(move |ono| {
515 default_hook(ono);
516 let location = match ono.location() {
517 Some(location) => location.to_string(),
518 None => "?".to_string(),
519 };
520 submit(
521 DataPoint::new("panic")
522 .add_field_str("program", program)
523 .add_field_str("thread", thread::current().name().unwrap_or("?"))
524 .add_field_i64("one", 1)
527 .add_field_str("message", &ono.to_string())
528 .add_field_str("location", &location)
529 .add_field_str("version", version.as_ref().unwrap_or(&"".to_string()))
530 .to_owned(),
531 Level::Error,
532 );
533 flush();
535
536 std::process::exit(1);
538 }));
539 });
540}
541
542pub mod test_mocks {
543 use super::*;
544
545 pub struct MockMetricsWriter {
546 pub points_written: Arc<Mutex<Vec<DataPoint>>>,
547 }
548 impl MockMetricsWriter {
549 pub fn new() -> Self {
550 MockMetricsWriter {
551 points_written: Arc::new(Mutex::new(Vec::new())),
552 }
553 }
554
555 pub fn points_written(&self) -> usize {
556 self.points_written.lock().unwrap().len()
557 }
558 }
559
560 impl Default for MockMetricsWriter {
561 fn default() -> Self {
562 Self::new()
563 }
564 }
565
566 impl MetricsWriter for MockMetricsWriter {
567 fn write(&self, points: Vec<DataPoint>) {
568 assert!(!points.is_empty());
569
570 let new_points = points.len();
571 self.points_written.lock().unwrap().extend(points);
572
573 info!(
574 "Writing {} points ({} total)",
575 new_points,
576 self.points_written(),
577 );
578 }
579 }
580}
581
582#[cfg(test)]
583mod test {
584 use {super::*, test_mocks::MockMetricsWriter};
585
586 #[test]
587 fn test_submit() {
588 let writer = Arc::new(MockMetricsWriter::new());
589 let agent = MetricsAgent::new(writer.clone(), Duration::from_secs(10), 1000);
590
591 for i in 0..42 {
592 agent.submit(
593 DataPoint::new("measurement")
594 .add_field_i64("i", i)
595 .to_owned(),
596 Level::Info,
597 );
598 }
599
600 agent.flush();
601 assert_eq!(writer.points_written(), 43);
602 }
603
604 #[test]
605 fn test_submit_counter() {
606 let writer = Arc::new(MockMetricsWriter::new());
607 let agent = MetricsAgent::new(writer.clone(), Duration::from_secs(10), 1000);
608
609 for i in 0..10 {
610 agent.submit_counter(CounterPoint::new("counter 1"), Level::Info, i);
611 agent.submit_counter(CounterPoint::new("counter 2"), Level::Info, i);
612 }
613
614 agent.flush();
615 assert_eq!(writer.points_written(), 21);
616 }
617
618 #[test]
619 fn test_submit_counter_increment() {
620 let writer = Arc::new(MockMetricsWriter::new());
621 let agent = MetricsAgent::new(writer.clone(), Duration::from_secs(10), 1000);
622
623 for _ in 0..10 {
624 agent.submit_counter(
625 CounterPoint {
626 name: "counter",
627 count: 10,
628 timestamp: UNIX_EPOCH,
629 },
630 Level::Info,
631 0, );
633 }
634
635 agent.flush();
636 assert_eq!(writer.points_written(), 2);
637
638 let submitted_point = writer.points_written.lock().unwrap()[0].clone();
639 assert_eq!(submitted_point.fields[0], ("count", "100i".to_string()));
640 }
641
642 #[test]
643 fn test_submit_bucketed_counter() {
644 let writer = Arc::new(MockMetricsWriter::new());
645 let agent = MetricsAgent::new(writer.clone(), Duration::from_secs(10), 1000);
646
647 for i in 0..50 {
648 agent.submit_counter(CounterPoint::new("counter 1"), Level::Info, i / 10);
649 agent.submit_counter(CounterPoint::new("counter 2"), Level::Info, i / 10);
650 }
651
652 agent.flush();
653 assert_eq!(writer.points_written(), 11);
654 }
655
656 #[test]
657 fn test_submit_with_delay() {
658 let writer = Arc::new(MockMetricsWriter::new());
659 let agent = MetricsAgent::new(writer.clone(), Duration::from_secs(1), 1000);
660
661 agent.submit(DataPoint::new("point 1"), Level::Info);
662 thread::sleep(Duration::from_secs(2));
663 assert_eq!(writer.points_written(), 2);
664 }
665
666 #[test]
667 fn test_submit_exceed_max_rate() {
668 let writer = Arc::new(MockMetricsWriter::new());
669
670 let max_points_per_sec = 100;
671
672 let agent = MetricsAgent::new(writer.clone(), Duration::from_secs(1), max_points_per_sec);
673
674 for i in 0..(max_points_per_sec + 20) {
675 agent.submit(
676 DataPoint::new("measurement")
677 .add_field_i64("i", i.try_into().unwrap())
678 .to_owned(),
679 Level::Info,
680 );
681 }
682
683 agent.flush();
684
685 assert_eq!(writer.points_written(), max_points_per_sec);
688 }
689
690 #[test]
691 fn test_multithread_submit() {
692 let writer = Arc::new(MockMetricsWriter::new());
693 let agent = Arc::new(Mutex::new(MetricsAgent::new(
694 writer.clone(),
695 Duration::from_secs(10),
696 1000,
697 )));
698
699 let mut threads = Vec::new();
703 for i in 0..42 {
704 let mut point = DataPoint::new("measurement");
705 point.add_field_i64("i", i);
706 let agent = Arc::clone(&agent);
707 threads.push(thread::spawn(move || {
708 agent.lock().unwrap().submit(point, Level::Info);
709 }));
710 }
711
712 for thread in threads {
713 thread.join().unwrap();
714 }
715
716 agent.lock().unwrap().flush();
717 assert_eq!(writer.points_written(), 43);
718 }
719
720 #[test]
721 fn test_flush_before_drop() {
722 let writer = Arc::new(MockMetricsWriter::new());
723 {
724 let agent = MetricsAgent::new(writer.clone(), Duration::from_secs(9_999_999), 1000);
725 agent.submit(DataPoint::new("point 1"), Level::Info);
726 }
727
728 assert_eq!(writer.points_written(), 2);
732 }
733
734 #[test]
735 fn test_live_submit() {
736 let agent = MetricsAgent::default();
737
738 let point = DataPoint::new("live_submit_test")
739 .add_field_bool("true", true)
740 .add_field_bool("random_bool", rand::random::<u8>() < 128)
741 .add_field_i64("random_int", rand::random::<u8>() as i64)
742 .to_owned();
743 agent.submit(point, Level::Info);
744 }
745}