1mod event_coalescer;
2
3use crate::TelemetrySettings;
4use anyhow::Result;
5use clock::SystemClock;
6use futures::channel::mpsc;
7use futures::{Future, FutureExt, StreamExt};
8use gpui::{App, AppContext as _, BackgroundExecutor, Task};
9use http_client::{self, AsyncBody, HttpClient, HttpClientWithUrl, Method, Request};
10use parking_lot::Mutex;
11use regex::Regex;
12use release_channel::ReleaseChannel;
13use settings::{Settings, SettingsStore};
14use sha2::{Digest, Sha256};
15use std::collections::HashSet;
16use std::fs::File;
17use std::io::Write;
18use std::sync::LazyLock;
19use std::time::Instant;
20use std::{env, mem, path::PathBuf, sync::Arc, time::Duration};
21use telemetry_events::{AssistantEventData, AssistantPhase, Event, EventRequestBody, EventWrapper};
22use util::{ResultExt, TryFutureExt};
23use worktree::{UpdatedEntriesSet, WorktreeId};
24
25use self::event_coalescer::EventCoalescer;
26
27pub struct Telemetry {
28 clock: Arc<dyn SystemClock>,
29 http_client: Arc<HttpClientWithUrl>,
30 executor: BackgroundExecutor,
31 state: Arc<Mutex<TelemetryState>>,
32}
33
34struct TelemetryState {
35 settings: TelemetrySettings,
36 system_id: Option<Arc<str>>, // Per system
37 installation_id: Option<Arc<str>>, // Per app installation (different for dev, nightly, preview, and stable)
38 session_id: Option<String>, // Per app launch
39 metrics_id: Option<Arc<str>>, // Per logged-in user
40 release_channel: Option<&'static str>,
41 architecture: &'static str,
42 events_queue: Vec<EventWrapper>,
43 flush_events_task: Option<Task<()>>,
44 log_file: Option<File>,
45 is_staff: Option<bool>,
46 first_event_date_time: Option<Instant>,
47 event_coalescer: EventCoalescer,
48 max_queue_size: usize,
49 project_marker_patterns: ProjectMarkerPatterns,
50
51 os_name: String,
52 app_version: String,
53 os_version: Option<String>,
54}
55
56#[derive(Debug)]
57struct ProjectMarkerPatterns(Vec<(Regex, ProjectCache)>);
58
59#[derive(Debug)]
60struct ProjectCache {
61 name: String,
62 worktree_ids_reported: HashSet<WorktreeId>,
63}
64
65impl ProjectCache {
66 fn new(name: String) -> Self {
67 Self {
68 name,
69 worktree_ids_reported: HashSet::default(),
70 }
71 }
72}
73
74#[cfg(debug_assertions)]
75const MAX_QUEUE_LEN: usize = 5;
76
77#[cfg(not(debug_assertions))]
78const MAX_QUEUE_LEN: usize = 50;
79
80#[cfg(debug_assertions)]
81const FLUSH_INTERVAL: Duration = Duration::from_secs(1);
82
83#[cfg(not(debug_assertions))]
84const FLUSH_INTERVAL: Duration = Duration::from_secs(60 * 5);
85static ZED_CLIENT_CHECKSUM_SEED: LazyLock<Option<Vec<u8>>> = LazyLock::new(|| {
86 option_env!("ZED_CLIENT_CHECKSUM_SEED")
87 .map(|s| s.as_bytes().into())
88 .or_else(|| {
89 env::var("ZED_CLIENT_CHECKSUM_SEED")
90 .ok()
91 .map(|s| s.as_bytes().into())
92 })
93});
94
95pub fn os_name() -> String {
96 #[cfg(target_os = "macos")]
97 {
98 "macOS".to_string()
99 }
100 #[cfg(any(target_os = "linux", target_os = "freebsd"))]
101 {
102 format!("Linux {}", gpui::guess_compositor())
103 }
104
105 #[cfg(target_os = "windows")]
106 {
107 "Windows".to_string()
108 }
109}
110
111/// Note: This might do blocking IO! Only call from background threads
112pub fn os_version() -> String {
113 #[cfg(target_os = "macos")]
114 {
115 use cocoa::base::nil;
116 use cocoa::foundation::NSProcessInfo;
117
118 unsafe {
119 let process_info = cocoa::foundation::NSProcessInfo::processInfo(nil);
120 let version = process_info.operatingSystemVersion();
121 gpui::SemanticVersion::new(
122 version.majorVersion as usize,
123 version.minorVersion as usize,
124 version.patchVersion as usize,
125 )
126 .to_string()
127 }
128 }
129 #[cfg(any(target_os = "linux", target_os = "freebsd"))]
130 {
131 use std::path::Path;
132
133 let content = if let Ok(file) = std::fs::read_to_string(&Path::new("/etc/os-release")) {
134 file
135 } else if let Ok(file) = std::fs::read_to_string(&Path::new("/usr/lib/os-release")) {
136 file
137 } else {
138 log::error!("Failed to load /etc/os-release, /usr/lib/os-release");
139 "".to_string()
140 };
141 let mut name = "unknown";
142 let mut version = "unknown";
143
144 for line in content.lines() {
145 match line.split_once('=') {
146 Some(("ID", val)) => name = val.trim_matches('"'),
147 Some(("VERSION_ID", val)) => version = val.trim_matches('"'),
148 _ => {}
149 }
150 }
151
152 format!("{} {}", name, version)
153 }
154
155 #[cfg(target_os = "windows")]
156 {
157 let mut info = unsafe { std::mem::zeroed() };
158 let status = unsafe { windows::Wdk::System::SystemServices::RtlGetVersion(&mut info) };
159 if status.is_ok() {
160 gpui::SemanticVersion::new(
161 info.dwMajorVersion as _,
162 info.dwMinorVersion as _,
163 info.dwBuildNumber as _,
164 )
165 .to_string()
166 } else {
167 "unknown".to_string()
168 }
169 }
170}
171
172impl Telemetry {
173 pub fn new(
174 clock: Arc<dyn SystemClock>,
175 client: Arc<HttpClientWithUrl>,
176 cx: &mut App,
177 ) -> Arc<Self> {
178 let release_channel =
179 ReleaseChannel::try_global(cx).map(|release_channel| release_channel.display_name());
180
181 TelemetrySettings::register(cx);
182
183 let state = Arc::new(Mutex::new(TelemetryState {
184 settings: *TelemetrySettings::get_global(cx),
185 architecture: env::consts::ARCH,
186 release_channel,
187 system_id: None,
188 installation_id: None,
189 session_id: None,
190 metrics_id: None,
191 events_queue: Vec::new(),
192 flush_events_task: None,
193 log_file: None,
194 is_staff: None,
195 first_event_date_time: None,
196 event_coalescer: EventCoalescer::new(clock.clone()),
197 max_queue_size: MAX_QUEUE_LEN,
198 project_marker_patterns: ProjectMarkerPatterns(vec![
199 (
200 Regex::new(r"^pnpm-lock\.yaml$").unwrap(),
201 ProjectCache::new("pnpm".to_string()),
202 ),
203 (
204 Regex::new(r"^yarn\.lock$").unwrap(),
205 ProjectCache::new("yarn".to_string()),
206 ),
207 (
208 Regex::new(r"^package\.json$").unwrap(),
209 ProjectCache::new("node".to_string()),
210 ),
211 (
212 Regex::new(
213 r"^(global\.json|Directory\.Build\.props|.*\.(csproj|fsproj|vbproj|sln))$",
214 )
215 .unwrap(),
216 ProjectCache::new("dotnet".to_string()),
217 ),
218 ]),
219
220 os_version: None,
221 os_name: os_name(),
222 app_version: release_channel::AppVersion::global(cx).to_string(),
223 }));
224 Self::log_file_path();
225
226 cx.background_spawn({
227 let state = state.clone();
228 let os_version = os_version();
229 state.lock().os_version = Some(os_version);
230 async move {
231 if let Some(tempfile) = File::create(Self::log_file_path()).log_err() {
232 state.lock().log_file = Some(tempfile);
233 }
234 }
235 })
236 .detach();
237
238 cx.observe_global::<SettingsStore>({
239 let state = state.clone();
240
241 move |cx| {
242 let mut state = state.lock();
243 state.settings = *TelemetrySettings::get_global(cx);
244 }
245 })
246 .detach();
247
248 let this = Arc::new(Self {
249 clock,
250 http_client: client,
251 executor: cx.background_executor().clone(),
252 state,
253 });
254
255 let (tx, mut rx) = mpsc::unbounded();
256 ::telemetry::init(tx);
257
258 cx.background_spawn({
259 let this = Arc::downgrade(&this);
260 async move {
261 while let Some(event) = rx.next().await {
262 let Some(state) = this.upgrade() else { break };
263 state.report_event(Event::Flexible(event))
264 }
265 }
266 })
267 .detach();
268
269 // We should only ever have one instance of Telemetry, leak the subscription to keep it alive
270 // rather than store in TelemetryState, complicating spawn as subscriptions are not Send
271 std::mem::forget(cx.on_app_quit({
272 let this = this.clone();
273 move |_| this.shutdown_telemetry()
274 }));
275
276 this
277 }
278
279 #[cfg(any(test, feature = "test-support"))]
280 fn shutdown_telemetry(self: &Arc<Self>) -> impl Future<Output = ()> + use<> {
281 Task::ready(())
282 }
283
284 // Skip calling this function in tests.
285 // TestAppContext ends up calling this function on shutdown and it panics when trying to find the TelemetrySettings
286 #[cfg(not(any(test, feature = "test-support")))]
287 fn shutdown_telemetry(self: &Arc<Self>) -> impl Future<Output = ()> + use<> {
288 telemetry::event!("App Closed");
289 // TODO: close final edit period and make sure it's sent
290 Task::ready(())
291 }
292
293 pub fn log_file_path() -> PathBuf {
294 paths::logs_dir().join("telemetry.log")
295 }
296
297 pub fn has_checksum_seed(&self) -> bool {
298 ZED_CLIENT_CHECKSUM_SEED.is_some()
299 }
300
301 pub fn start(
302 self: &Arc<Self>,
303 system_id: Option<String>,
304 installation_id: Option<String>,
305 session_id: String,
306 cx: &App,
307 ) {
308 let mut state = self.state.lock();
309 state.system_id = system_id.map(|id| id.into());
310 state.installation_id = installation_id.map(|id| id.into());
311 state.session_id = Some(session_id);
312 state.app_version = release_channel::AppVersion::global(cx).to_string();
313 state.os_name = os_name();
314 }
315
316 pub fn metrics_enabled(self: &Arc<Self>) -> bool {
317 let state = self.state.lock();
318 let enabled = state.settings.metrics;
319 drop(state);
320 enabled
321 }
322
323 pub fn set_authenticated_user_info(
324 self: &Arc<Self>,
325 metrics_id: Option<String>,
326 is_staff: bool,
327 ) {
328 let mut state = self.state.lock();
329
330 if !state.settings.metrics {
331 return;
332 }
333
334 let metrics_id: Option<Arc<str>> = metrics_id.map(|id| id.into());
335 state.metrics_id.clone_from(&metrics_id);
336 state.is_staff = Some(is_staff);
337 drop(state);
338 }
339
340 pub fn report_assistant_event(self: &Arc<Self>, event: AssistantEventData) {
341 let event_type = match event.phase {
342 AssistantPhase::Response => "Assistant Responded",
343 AssistantPhase::Invoked => "Assistant Invoked",
344 AssistantPhase::Accepted => "Assistant Response Accepted",
345 AssistantPhase::Rejected => "Assistant Response Rejected",
346 };
347
348 telemetry::event!(
349 event_type,
350 conversation_id = event.conversation_id,
351 kind = event.kind,
352 phase = event.phase,
353 message_id = event.message_id,
354 model = event.model,
355 model_provider = event.model_provider,
356 response_latency = event.response_latency,
357 error_message = event.error_message,
358 language_name = event.language_name,
359 );
360 }
361
362 pub fn log_edit_event(self: &Arc<Self>, environment: &'static str, is_via_ssh: bool) {
363 let mut state = self.state.lock();
364 let period_data = state.event_coalescer.log_event(environment);
365 drop(state);
366
367 if let Some((start, end, environment)) = period_data {
368 let duration = end
369 .saturating_duration_since(start)
370 .min(Duration::from_secs(60 * 60 * 24))
371 .as_millis() as i64;
372
373 telemetry::event!(
374 "Editor Edited",
375 duration = duration,
376 environment = environment,
377 is_via_ssh = is_via_ssh
378 );
379 }
380 }
381
382 pub fn report_discovered_project_events(
383 self: &Arc<Self>,
384 worktree_id: WorktreeId,
385 updated_entries_set: &UpdatedEntriesSet,
386 ) {
387 let project_type_names: Vec<String> = {
388 let mut state = self.state.lock();
389 state
390 .project_marker_patterns
391 .0
392 .iter_mut()
393 .filter_map(|(pattern, project_cache)| {
394 if project_cache.worktree_ids_reported.contains(&worktree_id) {
395 return None;
396 }
397
398 let project_file_found = updated_entries_set.iter().any(|(path, _, _)| {
399 path.as_ref()
400 .file_name()
401 .and_then(|name| name.to_str())
402 .map(|name_str| pattern.is_match(name_str))
403 .unwrap_or(false)
404 });
405
406 if !project_file_found {
407 return None;
408 }
409
410 project_cache.worktree_ids_reported.insert(worktree_id);
411
412 Some(project_cache.name.clone())
413 })
414 .collect()
415 };
416
417 for project_type_name in project_type_names {
418 telemetry::event!("Project Opened", project_type = project_type_name);
419 }
420 }
421
422 fn report_event(self: &Arc<Self>, event: Event) {
423 let mut state = self.state.lock();
424 // RUST_LOG=telemetry=trace to debug telemetry events
425 log::trace!(target: "telemetry", "{:?}", event);
426
427 if !state.settings.metrics {
428 return;
429 }
430
431 if state.flush_events_task.is_none() {
432 let this = self.clone();
433 state.flush_events_task = Some(self.executor.spawn(async move {
434 this.executor.timer(FLUSH_INTERVAL).await;
435 this.flush_events().detach();
436 }));
437 }
438
439 let date_time = self.clock.utc_now();
440
441 let milliseconds_since_first_event = match state.first_event_date_time {
442 Some(first_event_date_time) => date_time
443 .saturating_duration_since(first_event_date_time)
444 .min(Duration::from_secs(60 * 60 * 24))
445 .as_millis() as i64,
446 None => {
447 state.first_event_date_time = Some(date_time);
448 0
449 }
450 };
451
452 let signed_in = state.metrics_id.is_some();
453 state.events_queue.push(EventWrapper {
454 signed_in,
455 milliseconds_since_first_event,
456 event,
457 });
458
459 if state.installation_id.is_some() && state.events_queue.len() >= state.max_queue_size {
460 drop(state);
461 self.flush_events().detach();
462 }
463 }
464
465 pub fn metrics_id(self: &Arc<Self>) -> Option<Arc<str>> {
466 self.state.lock().metrics_id.clone()
467 }
468
469 pub fn system_id(self: &Arc<Self>) -> Option<Arc<str>> {
470 self.state.lock().system_id.clone()
471 }
472
473 pub fn installation_id(self: &Arc<Self>) -> Option<Arc<str>> {
474 self.state.lock().installation_id.clone()
475 }
476
477 pub fn is_staff(self: &Arc<Self>) -> Option<bool> {
478 self.state.lock().is_staff
479 }
480
481 fn build_request(
482 self: &Arc<Self>,
483 // We take in the JSON bytes buffer so we can reuse the existing allocation.
484 mut json_bytes: Vec<u8>,
485 event_request: &EventRequestBody,
486 ) -> Result<Request<AsyncBody>> {
487 json_bytes.clear();
488 serde_json::to_writer(&mut json_bytes, event_request)?;
489
490 let checksum = calculate_json_checksum(&json_bytes).unwrap_or_default();
491
492 Ok(Request::builder()
493 .method(Method::POST)
494 .uri(
495 self.http_client
496 .build_zed_api_url("/telemetry/events", &[])?
497 .as_ref(),
498 )
499 .header("Content-Type", "application/json")
500 .header("x-zed-checksum", checksum)
501 .body(json_bytes.into())?)
502 }
503
504 pub fn flush_events(self: &Arc<Self>) -> Task<()> {
505 let mut state = self.state.lock();
506 state.first_event_date_time = None;
507 let events = mem::take(&mut state.events_queue);
508 state.flush_events_task.take();
509 drop(state);
510 if events.is_empty() {
511 return Task::ready(());
512 }
513
514 let this = self.clone();
515 self.executor.spawn(
516 async move {
517 let mut json_bytes = Vec::new();
518
519 if let Some(file) = &mut this.state.lock().log_file {
520 for event in &events {
521 json_bytes.clear();
522 serde_json::to_writer(&mut json_bytes, event)?;
523 file.write_all(&json_bytes)?;
524 file.write_all(b"\n")?;
525 }
526 }
527
528 let request_body = {
529 let state = this.state.lock();
530
531 EventRequestBody {
532 system_id: state.system_id.as_deref().map(Into::into),
533 installation_id: state.installation_id.as_deref().map(Into::into),
534 session_id: state.session_id.clone(),
535 metrics_id: state.metrics_id.as_deref().map(Into::into),
536 is_staff: state.is_staff,
537 app_version: state.app_version.clone(),
538 os_name: state.os_name.clone(),
539 os_version: state.os_version.clone(),
540 architecture: state.architecture.to_string(),
541
542 release_channel: state.release_channel.map(Into::into),
543 events,
544 }
545 };
546
547 let request = this.build_request(json_bytes, &request_body)?;
548 let response = this.http_client.send(request).await?;
549 if response.status() != 200 {
550 log::error!("Failed to send events: HTTP {:?}", response.status());
551 }
552 anyhow::Ok(())
553 }
554 .log_err()
555 .map(|_| ()),
556 )
557 }
558}
559
560pub fn calculate_json_checksum(json: &impl AsRef<[u8]>) -> Option<String> {
561 let Some(checksum_seed) = &*ZED_CLIENT_CHECKSUM_SEED else {
562 return None;
563 };
564
565 let mut summer = Sha256::new();
566 summer.update(checksum_seed);
567 summer.update(json);
568 summer.update(checksum_seed);
569 let mut checksum = String::new();
570 for byte in summer.finalize().as_slice() {
571 use std::fmt::Write;
572 write!(&mut checksum, "{:02x}", byte).unwrap();
573 }
574
575 Some(checksum)
576}
577
578#[cfg(test)]
579mod tests {
580 use super::*;
581 use clock::FakeSystemClock;
582 use gpui::TestAppContext;
583 use http_client::FakeHttpClient;
584 use std::collections::HashMap;
585 use telemetry_events::FlexibleEvent;
586
587 #[gpui::test]
588 fn test_telemetry_flush_on_max_queue_size(cx: &mut TestAppContext) {
589 init_test(cx);
590 let clock = Arc::new(FakeSystemClock::new());
591 let http = FakeHttpClient::with_200_response();
592 let system_id = Some("system_id".to_string());
593 let installation_id = Some("installation_id".to_string());
594 let session_id = "session_id".to_string();
595
596 cx.update(|cx| {
597 let telemetry = Telemetry::new(clock.clone(), http, cx);
598
599 telemetry.state.lock().max_queue_size = 4;
600 telemetry.start(system_id, installation_id, session_id, cx);
601
602 assert!(is_empty_state(&telemetry));
603
604 let first_date_time = clock.utc_now();
605 let event_properties = HashMap::from_iter([(
606 "test_key".to_string(),
607 serde_json::Value::String("test_value".to_string()),
608 )]);
609
610 let event = FlexibleEvent {
611 event_type: "test".to_string(),
612 event_properties,
613 };
614
615 telemetry.report_event(Event::Flexible(event.clone()));
616 assert_eq!(telemetry.state.lock().events_queue.len(), 1);
617 assert!(telemetry.state.lock().flush_events_task.is_some());
618 assert_eq!(
619 telemetry.state.lock().first_event_date_time,
620 Some(first_date_time)
621 );
622
623 clock.advance(Duration::from_millis(100));
624
625 telemetry.report_event(Event::Flexible(event.clone()));
626 assert_eq!(telemetry.state.lock().events_queue.len(), 2);
627 assert!(telemetry.state.lock().flush_events_task.is_some());
628 assert_eq!(
629 telemetry.state.lock().first_event_date_time,
630 Some(first_date_time)
631 );
632
633 clock.advance(Duration::from_millis(100));
634
635 telemetry.report_event(Event::Flexible(event.clone()));
636 assert_eq!(telemetry.state.lock().events_queue.len(), 3);
637 assert!(telemetry.state.lock().flush_events_task.is_some());
638 assert_eq!(
639 telemetry.state.lock().first_event_date_time,
640 Some(first_date_time)
641 );
642
643 clock.advance(Duration::from_millis(100));
644
645 // Adding a 4th event should cause a flush
646 telemetry.report_event(Event::Flexible(event));
647 assert!(is_empty_state(&telemetry));
648 });
649 }
650
651 #[gpui::test]
652 async fn test_telemetry_flush_on_flush_interval(
653 executor: BackgroundExecutor,
654 cx: &mut TestAppContext,
655 ) {
656 init_test(cx);
657 let clock = Arc::new(FakeSystemClock::new());
658 let http = FakeHttpClient::with_200_response();
659 let system_id = Some("system_id".to_string());
660 let installation_id = Some("installation_id".to_string());
661 let session_id = "session_id".to_string();
662
663 cx.update(|cx| {
664 let telemetry = Telemetry::new(clock.clone(), http, cx);
665 telemetry.state.lock().max_queue_size = 4;
666 telemetry.start(system_id, installation_id, session_id, cx);
667
668 assert!(is_empty_state(&telemetry));
669 let first_date_time = clock.utc_now();
670
671 let event_properties = HashMap::from_iter([(
672 "test_key".to_string(),
673 serde_json::Value::String("test_value".to_string()),
674 )]);
675
676 let event = FlexibleEvent {
677 event_type: "test".to_string(),
678 event_properties,
679 };
680
681 telemetry.report_event(Event::Flexible(event));
682 assert_eq!(telemetry.state.lock().events_queue.len(), 1);
683 assert!(telemetry.state.lock().flush_events_task.is_some());
684 assert_eq!(
685 telemetry.state.lock().first_event_date_time,
686 Some(first_date_time)
687 );
688
689 let duration = Duration::from_millis(1);
690
691 // Test 1 millisecond before the flush interval limit is met
692 executor.advance_clock(FLUSH_INTERVAL - duration);
693
694 assert!(!is_empty_state(&telemetry));
695
696 // Test the exact moment the flush interval limit is met
697 executor.advance_clock(duration);
698
699 assert!(is_empty_state(&telemetry));
700 });
701 }
702
703 // TODO:
704 // Test settings
705 // Update FakeHTTPClient to keep track of the number of requests and assert on it
706
707 fn init_test(cx: &mut TestAppContext) {
708 cx.update(|cx| {
709 let settings_store = SettingsStore::test(cx);
710 cx.set_global(settings_store);
711 });
712 }
713
714 fn is_empty_state(telemetry: &Telemetry) -> bool {
715 telemetry.state.lock().events_queue.is_empty()
716 && telemetry.state.lock().flush_events_task.is_none()
717 && telemetry.state.lock().first_event_date_time.is_none()
718 }
719}