1use crate::{
2 db::{self, NewUserParams, UserId},
3 rpc::{CLEANUP_TIMEOUT, RECONNECT_TIMEOUT},
4 tests::{TestClient, TestServer},
5};
6use async_trait::async_trait;
7use futures::StreamExt;
8use gpui::{BackgroundExecutor, Task, TestAppContext};
9use parking_lot::Mutex;
10use rand::prelude::*;
11use rpc::RECEIVE_TIMEOUT;
12use serde::{de::DeserializeOwned, Deserialize, Serialize};
13use settings::SettingsStore;
14use std::sync::OnceLock;
15use std::{
16 env,
17 path::PathBuf,
18 rc::Rc,
19 sync::{
20 atomic::{AtomicBool, Ordering::SeqCst},
21 Arc,
22 },
23};
24
25fn plan_load_path() -> &'static Option<PathBuf> {
26 static PLAN_LOAD_PATH: OnceLock<Option<PathBuf>> = OnceLock::new();
27 PLAN_LOAD_PATH.get_or_init(|| path_env_var("LOAD_PLAN"))
28}
29
30fn plan_save_path() -> &'static Option<PathBuf> {
31 static PLAN_SAVE_PATH: OnceLock<Option<PathBuf>> = OnceLock::new();
32 PLAN_SAVE_PATH.get_or_init(|| path_env_var("SAVE_PLAN"))
33}
34
35fn max_peers() -> usize {
36 static MAX_PEERS: OnceLock<usize> = OnceLock::new();
37 *MAX_PEERS.get_or_init(|| {
38 env::var("MAX_PEERS")
39 .map(|i| i.parse().expect("invalid `MAX_PEERS` variable"))
40 .unwrap_or(3)
41 })
42}
43
44fn max_operations() -> usize {
45 static MAX_OPERATIONS: OnceLock<usize> = OnceLock::new();
46 *MAX_OPERATIONS.get_or_init(|| {
47 env::var("OPERATIONS")
48 .map(|i| i.parse().expect("invalid `OPERATIONS` variable"))
49 .unwrap_or(10)
50 })
51}
52
53static LOADED_PLAN_JSON: Mutex<Option<Vec<u8>>> = Mutex::new(None);
54static LAST_PLAN: Mutex<Option<Box<dyn Send + FnOnce() -> Vec<u8>>>> = Mutex::new(None);
55
56struct TestPlan<T: RandomizedTest> {
57 rng: StdRng,
58 replay: bool,
59 stored_operations: Vec<(StoredOperation<T::Operation>, Arc<AtomicBool>)>,
60 max_operations: usize,
61 operation_ix: usize,
62 users: Vec<UserTestPlan>,
63 next_batch_id: usize,
64 allow_server_restarts: bool,
65 allow_client_reconnection: bool,
66 allow_client_disconnection: bool,
67}
68
69pub struct UserTestPlan {
70 pub user_id: UserId,
71 pub username: String,
72 pub allow_client_disconnection: bool,
73 next_root_id: usize,
74 operation_ix: usize,
75 online: bool,
76}
77
78#[derive(Clone, Debug, Serialize, Deserialize)]
79#[serde(untagged)]
80enum StoredOperation<T> {
81 Server(ServerOperation),
82 Client {
83 user_id: UserId,
84 batch_id: usize,
85 operation: T,
86 },
87}
88
89#[derive(Clone, Debug, Serialize, Deserialize)]
90enum ServerOperation {
91 AddConnection {
92 user_id: UserId,
93 },
94 RemoveConnection {
95 user_id: UserId,
96 },
97 BounceConnection {
98 user_id: UserId,
99 },
100 RestartServer,
101 MutateClients {
102 batch_id: usize,
103 #[serde(skip_serializing)]
104 #[serde(skip_deserializing)]
105 user_ids: Vec<UserId>,
106 quiesce: bool,
107 },
108}
109
110pub enum TestError {
111 Inapplicable,
112 Other(anyhow::Error),
113}
114
115#[async_trait(?Send)]
116pub trait RandomizedTest: 'static + Sized {
117 type Operation: Send + Clone + Serialize + DeserializeOwned;
118
119 fn generate_operation(
120 client: &TestClient,
121 rng: &mut StdRng,
122 plan: &mut UserTestPlan,
123 cx: &TestAppContext,
124 ) -> Self::Operation;
125
126 async fn apply_operation(
127 client: &TestClient,
128 operation: Self::Operation,
129 cx: &mut TestAppContext,
130 ) -> Result<(), TestError>;
131
132 async fn initialize(server: &mut TestServer, users: &[UserTestPlan]);
133
134 async fn on_client_added(_client: &Rc<TestClient>, _cx: &mut TestAppContext) {}
135
136 async fn on_quiesce(server: &mut TestServer, client: &mut [(Rc<TestClient>, TestAppContext)]);
137}
138
139pub async fn run_randomized_test<T: RandomizedTest>(
140 cx: &mut TestAppContext,
141 executor: BackgroundExecutor,
142 rng: StdRng,
143) {
144 let mut server = TestServer::start(executor.clone()).await;
145 let plan = TestPlan::<T>::new(&mut server, rng).await;
146
147 LAST_PLAN.lock().replace({
148 let plan = plan.clone();
149 Box::new(move || plan.lock().serialize())
150 });
151
152 let mut clients = Vec::new();
153 let mut client_tasks = Vec::new();
154 let mut operation_channels = Vec::new();
155 loop {
156 let Some((next_operation, applied)) = plan.lock().next_server_operation(&clients) else {
157 break;
158 };
159 applied.store(true, SeqCst);
160 let did_apply = TestPlan::apply_server_operation(
161 plan.clone(),
162 executor.clone(),
163 &mut server,
164 &mut clients,
165 &mut client_tasks,
166 &mut operation_channels,
167 next_operation,
168 cx,
169 )
170 .await;
171 if !did_apply {
172 applied.store(false, SeqCst);
173 }
174 }
175
176 drop(operation_channels);
177 executor.start_waiting();
178 futures::future::join_all(client_tasks).await;
179 executor.finish_waiting();
180
181 executor.run_until_parked();
182 T::on_quiesce(&mut server, &mut clients).await;
183
184 for (client, cx) in clients {
185 cx.update(|cx| {
186 let store = cx.remove_global::<SettingsStore>();
187 cx.clear_globals();
188 cx.set_global(store);
189 drop(client);
190 });
191 }
192 executor.run_until_parked();
193
194 if let Some(path) = plan_save_path() {
195 eprintln!("saved test plan to path {:?}", path);
196 std::fs::write(path, plan.lock().serialize()).unwrap();
197 }
198}
199
200pub fn save_randomized_test_plan() {
201 if let Some(serialize_plan) = LAST_PLAN.lock().take() {
202 if let Some(path) = plan_save_path() {
203 eprintln!("saved test plan to path {:?}", path);
204 std::fs::write(path, serialize_plan()).unwrap();
205 }
206 }
207}
208
209impl<T: RandomizedTest> TestPlan<T> {
210 pub async fn new(server: &mut TestServer, mut rng: StdRng) -> Arc<Mutex<Self>> {
211 let allow_server_restarts = rng.gen_bool(0.7);
212 let allow_client_reconnection = rng.gen_bool(0.7);
213 let allow_client_disconnection = rng.gen_bool(0.1);
214
215 let mut users = Vec::new();
216 for ix in 0..max_peers() {
217 let username = format!("user-{}", ix + 1);
218 let user_id = server
219 .app_state
220 .db
221 .create_user(
222 &format!("{username}@example.com"),
223 None,
224 false,
225 NewUserParams {
226 github_login: username.clone(),
227 github_user_id: ix as i32,
228 },
229 )
230 .await
231 .unwrap()
232 .user_id;
233 users.push(UserTestPlan {
234 user_id,
235 username,
236 online: false,
237 next_root_id: 0,
238 operation_ix: 0,
239 allow_client_disconnection,
240 });
241 }
242
243 T::initialize(server, &users).await;
244
245 let plan = Arc::new(Mutex::new(Self {
246 replay: false,
247 allow_server_restarts,
248 allow_client_reconnection,
249 allow_client_disconnection,
250 stored_operations: Vec::new(),
251 operation_ix: 0,
252 next_batch_id: 0,
253 max_operations: max_operations(),
254 users,
255 rng,
256 }));
257
258 if let Some(path) = plan_load_path() {
259 let json = LOADED_PLAN_JSON
260 .lock()
261 .get_or_insert_with(|| {
262 eprintln!("loaded test plan from path {:?}", path);
263 std::fs::read(path).unwrap()
264 })
265 .clone();
266 plan.lock().deserialize(json);
267 }
268
269 plan
270 }
271
272 fn deserialize(&mut self, json: Vec<u8>) {
273 let stored_operations: Vec<StoredOperation<T::Operation>> =
274 serde_json::from_slice(&json).unwrap();
275 self.replay = true;
276 self.stored_operations = stored_operations
277 .iter()
278 .cloned()
279 .enumerate()
280 .map(|(i, mut operation)| {
281 let did_apply = Arc::new(AtomicBool::new(false));
282 if let StoredOperation::Server(ServerOperation::MutateClients {
283 batch_id: current_batch_id,
284 user_ids,
285 ..
286 }) = &mut operation
287 {
288 assert!(user_ids.is_empty());
289 user_ids.extend(stored_operations[i + 1..].iter().filter_map(|operation| {
290 if let StoredOperation::Client {
291 user_id, batch_id, ..
292 } = operation
293 {
294 if batch_id == current_batch_id {
295 return Some(user_id);
296 }
297 }
298 None
299 }));
300 user_ids.sort_unstable();
301 }
302 (operation, did_apply)
303 })
304 .collect()
305 }
306
307 fn serialize(&mut self) -> Vec<u8> {
308 // Format each operation as one line
309 let mut json = Vec::new();
310 json.push(b'[');
311 for (operation, applied) in &self.stored_operations {
312 if !applied.load(SeqCst) {
313 continue;
314 }
315 if json.len() > 1 {
316 json.push(b',');
317 }
318 json.extend_from_slice(b"\n ");
319 serde_json::to_writer(&mut json, operation).unwrap();
320 }
321 json.extend_from_slice(b"\n]\n");
322 json
323 }
324
325 fn next_server_operation(
326 &mut self,
327 clients: &[(Rc<TestClient>, TestAppContext)],
328 ) -> Option<(ServerOperation, Arc<AtomicBool>)> {
329 if self.replay {
330 while let Some(stored_operation) = self.stored_operations.get(self.operation_ix) {
331 self.operation_ix += 1;
332 if let (StoredOperation::Server(operation), applied) = stored_operation {
333 return Some((operation.clone(), applied.clone()));
334 }
335 }
336 None
337 } else {
338 let operation = self.generate_server_operation(clients)?;
339 let applied = Arc::new(AtomicBool::new(false));
340 self.stored_operations
341 .push((StoredOperation::Server(operation.clone()), applied.clone()));
342 Some((operation, applied))
343 }
344 }
345
346 fn next_client_operation(
347 &mut self,
348 client: &TestClient,
349 current_batch_id: usize,
350 cx: &TestAppContext,
351 ) -> Option<(T::Operation, Arc<AtomicBool>)> {
352 let current_user_id = client.current_user_id(cx);
353 let user_ix = self
354 .users
355 .iter()
356 .position(|user| user.user_id == current_user_id)
357 .unwrap();
358 let user_plan = &mut self.users[user_ix];
359
360 if self.replay {
361 while let Some(stored_operation) = self.stored_operations.get(user_plan.operation_ix) {
362 user_plan.operation_ix += 1;
363 if let (
364 StoredOperation::Client {
365 user_id, operation, ..
366 },
367 applied,
368 ) = stored_operation
369 {
370 if user_id == ¤t_user_id {
371 return Some((operation.clone(), applied.clone()));
372 }
373 }
374 }
375 None
376 } else {
377 if self.operation_ix == self.max_operations {
378 return None;
379 }
380 self.operation_ix += 1;
381 let operation = T::generate_operation(
382 client,
383 &mut self.rng,
384 self.users
385 .iter_mut()
386 .find(|user| user.user_id == current_user_id)
387 .unwrap(),
388 cx,
389 );
390 let applied = Arc::new(AtomicBool::new(false));
391 self.stored_operations.push((
392 StoredOperation::Client {
393 user_id: current_user_id,
394 batch_id: current_batch_id,
395 operation: operation.clone(),
396 },
397 applied.clone(),
398 ));
399 Some((operation, applied))
400 }
401 }
402
403 fn generate_server_operation(
404 &mut self,
405 clients: &[(Rc<TestClient>, TestAppContext)],
406 ) -> Option<ServerOperation> {
407 if self.operation_ix == self.max_operations {
408 return None;
409 }
410
411 Some(loop {
412 break match self.rng.gen_range(0..100) {
413 0..=29 if clients.len() < self.users.len() => {
414 let user = self
415 .users
416 .iter()
417 .filter(|u| !u.online)
418 .choose(&mut self.rng)
419 .unwrap();
420 self.operation_ix += 1;
421 ServerOperation::AddConnection {
422 user_id: user.user_id,
423 }
424 }
425 30..=34 if clients.len() > 1 && self.allow_client_disconnection => {
426 let (client, cx) = &clients[self.rng.gen_range(0..clients.len())];
427 let user_id = client.current_user_id(cx);
428 self.operation_ix += 1;
429 ServerOperation::RemoveConnection { user_id }
430 }
431 35..=39 if clients.len() > 1 && self.allow_client_reconnection => {
432 let (client, cx) = &clients[self.rng.gen_range(0..clients.len())];
433 let user_id = client.current_user_id(cx);
434 self.operation_ix += 1;
435 ServerOperation::BounceConnection { user_id }
436 }
437 40..=44 if self.allow_server_restarts && clients.len() > 1 => {
438 self.operation_ix += 1;
439 ServerOperation::RestartServer
440 }
441 _ if !clients.is_empty() => {
442 let count = self
443 .rng
444 .gen_range(1..10)
445 .min(self.max_operations - self.operation_ix);
446 let batch_id = util::post_inc(&mut self.next_batch_id);
447 let mut user_ids = (0..count)
448 .map(|_| {
449 let ix = self.rng.gen_range(0..clients.len());
450 let (client, cx) = &clients[ix];
451 client.current_user_id(cx)
452 })
453 .collect::<Vec<_>>();
454 user_ids.sort_unstable();
455 ServerOperation::MutateClients {
456 user_ids,
457 batch_id,
458 quiesce: self.rng.gen_bool(0.7),
459 }
460 }
461 _ => continue,
462 };
463 })
464 }
465
466 #[allow(clippy::too_many_arguments)]
467 async fn apply_server_operation(
468 plan: Arc<Mutex<Self>>,
469 deterministic: BackgroundExecutor,
470 server: &mut TestServer,
471 clients: &mut Vec<(Rc<TestClient>, TestAppContext)>,
472 client_tasks: &mut Vec<Task<()>>,
473 operation_channels: &mut Vec<futures::channel::mpsc::UnboundedSender<usize>>,
474 operation: ServerOperation,
475 cx: &mut TestAppContext,
476 ) -> bool {
477 match operation {
478 ServerOperation::AddConnection { user_id } => {
479 let username;
480 {
481 let mut plan = plan.lock();
482 let user = plan.user(user_id);
483 if user.online {
484 return false;
485 }
486 user.online = true;
487 username = user.username.clone();
488 };
489 log::info!("adding new connection for {}", username);
490
491 let mut client_cx = cx.new_app();
492
493 let (operation_tx, operation_rx) = futures::channel::mpsc::unbounded();
494 let client = Rc::new(server.create_client(&mut client_cx, &username).await);
495 operation_channels.push(operation_tx);
496 clients.push((client.clone(), client_cx.clone()));
497
498 let foreground_executor = client_cx.foreground_executor().clone();
499 let simulate_client =
500 Self::simulate_client(plan.clone(), client, operation_rx, client_cx);
501 client_tasks.push(foreground_executor.spawn(simulate_client));
502
503 log::info!("added connection for {}", username);
504 }
505
506 ServerOperation::RemoveConnection {
507 user_id: removed_user_id,
508 } => {
509 log::info!("simulating full disconnection of user {}", removed_user_id);
510 let client_ix = clients
511 .iter()
512 .position(|(client, cx)| client.current_user_id(cx) == removed_user_id);
513 let Some(client_ix) = client_ix else {
514 return false;
515 };
516 let user_connection_ids = server
517 .connection_pool
518 .lock()
519 .user_connection_ids(removed_user_id)
520 .collect::<Vec<_>>();
521 assert_eq!(user_connection_ids.len(), 1);
522 let removed_peer_id = user_connection_ids[0].into();
523 let (client, client_cx) = clients.remove(client_ix);
524 let client_task = client_tasks.remove(client_ix);
525 operation_channels.remove(client_ix);
526 server.forbid_connections();
527 server.disconnect_client(removed_peer_id);
528 deterministic.advance_clock(RECEIVE_TIMEOUT + RECONNECT_TIMEOUT);
529 deterministic.start_waiting();
530 log::info!("waiting for user {} to exit...", removed_user_id);
531 client_task.await;
532 deterministic.finish_waiting();
533 server.allow_connections();
534
535 for project in client.dev_server_projects().iter() {
536 project.read_with(&client_cx, |project, cx| {
537 assert!(
538 project.is_disconnected(cx),
539 "project {:?} should be read only",
540 project.remote_id()
541 )
542 });
543 }
544
545 for (client, cx) in clients {
546 let contacts = server
547 .app_state
548 .db
549 .get_contacts(client.current_user_id(cx))
550 .await
551 .unwrap();
552 let pool = server.connection_pool.lock();
553 for contact in contacts {
554 if let db::Contact::Accepted { user_id, busy, .. } = contact {
555 if user_id == removed_user_id {
556 assert!(!pool.is_user_online(user_id));
557 assert!(!busy);
558 }
559 }
560 }
561 }
562
563 log::info!("{} removed", client.username);
564 plan.lock().user(removed_user_id).online = false;
565 client_cx.update(|cx| {
566 cx.clear_globals();
567 drop(client);
568 });
569 }
570
571 ServerOperation::BounceConnection { user_id } => {
572 log::info!("simulating temporary disconnection of user {}", user_id);
573 let user_connection_ids = server
574 .connection_pool
575 .lock()
576 .user_connection_ids(user_id)
577 .collect::<Vec<_>>();
578 if user_connection_ids.is_empty() {
579 return false;
580 }
581 assert_eq!(user_connection_ids.len(), 1);
582 let peer_id = user_connection_ids[0].into();
583 server.disconnect_client(peer_id);
584 deterministic.advance_clock(RECEIVE_TIMEOUT + RECONNECT_TIMEOUT);
585 }
586
587 ServerOperation::RestartServer => {
588 log::info!("simulating server restart");
589 server.reset().await;
590 deterministic.advance_clock(RECEIVE_TIMEOUT);
591 server.start().await.unwrap();
592 deterministic.advance_clock(CLEANUP_TIMEOUT);
593 let environment = &server.app_state.config.zed_environment;
594 let (stale_room_ids, _) = server
595 .app_state
596 .db
597 .stale_server_resource_ids(environment, server.id())
598 .await
599 .unwrap();
600 assert_eq!(stale_room_ids, vec![]);
601 }
602
603 ServerOperation::MutateClients {
604 user_ids,
605 batch_id,
606 quiesce,
607 } => {
608 let mut applied = false;
609 for user_id in user_ids {
610 let client_ix = clients
611 .iter()
612 .position(|(client, cx)| client.current_user_id(cx) == user_id);
613 let Some(client_ix) = client_ix else { continue };
614 applied = true;
615 if let Err(err) = operation_channels[client_ix].unbounded_send(batch_id) {
616 log::error!("error signaling user {user_id}: {err}");
617 }
618 }
619
620 if quiesce && applied {
621 deterministic.run_until_parked();
622 T::on_quiesce(server, clients).await;
623 }
624
625 return applied;
626 }
627 }
628 true
629 }
630
631 async fn simulate_client(
632 plan: Arc<Mutex<Self>>,
633 client: Rc<TestClient>,
634 mut operation_rx: futures::channel::mpsc::UnboundedReceiver<usize>,
635 mut cx: TestAppContext,
636 ) {
637 T::on_client_added(&client, &mut cx).await;
638
639 while let Some(batch_id) = operation_rx.next().await {
640 let Some((operation, applied)) =
641 plan.lock().next_client_operation(&client, batch_id, &cx)
642 else {
643 break;
644 };
645 applied.store(true, SeqCst);
646 match T::apply_operation(&client, operation, &mut cx).await {
647 Ok(()) => {}
648 Err(TestError::Inapplicable) => {
649 applied.store(false, SeqCst);
650 log::info!("skipped operation");
651 }
652 Err(TestError::Other(error)) => {
653 log::error!("{} error: {}", client.username, error);
654 }
655 }
656 cx.executor().simulate_random_delay().await;
657 }
658 log::info!("{}: done", client.username);
659 }
660
661 fn user(&mut self, user_id: UserId) -> &mut UserTestPlan {
662 self.users
663 .iter_mut()
664 .find(|user| user.user_id == user_id)
665 .unwrap()
666 }
667}
668
669impl UserTestPlan {
670 pub fn next_root_dir_name(&mut self) -> String {
671 let user_id = self.user_id;
672 let root_id = util::post_inc(&mut self.next_root_id);
673 format!("dir-{user_id}-{root_id}")
674 }
675}
676
677impl From<anyhow::Error> for TestError {
678 fn from(value: anyhow::Error) -> Self {
679 Self::Other(value)
680 }
681}
682
683fn path_env_var(name: &str) -> Option<PathBuf> {
684 let value = env::var(name).ok()?;
685 let mut path = PathBuf::from(value);
686 if path.is_relative() {
687 let mut abs_path = PathBuf::from(env!("CARGO_MANIFEST_DIR"));
688 abs_path.pop();
689 abs_path.pop();
690 abs_path.push(path);
691 path = abs_path
692 }
693 Some(path)
694}