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::{Deserialize, Serialize, de::DeserializeOwned};
13use settings::SettingsStore;
14use std::sync::OnceLock;
15use std::{
16 env,
17 path::PathBuf,
18 rc::Rc,
19 sync::{
20 Arc,
21 atomic::{AtomicBool, Ordering::SeqCst},
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 && 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
208impl<T: RandomizedTest> TestPlan<T> {
209 pub async fn new(server: &mut TestServer, mut rng: StdRng) -> Arc<Mutex<Self>> {
210 let allow_server_restarts = rng.gen_bool(0.7);
211 let allow_client_reconnection = rng.gen_bool(0.7);
212 let allow_client_disconnection = rng.gen_bool(0.1);
213
214 let mut users = Vec::new();
215 for ix in 0..max_peers() {
216 let username = format!("user-{}", ix + 1);
217 let user_id = server
218 .app_state
219 .db
220 .create_user(
221 &format!("{username}@example.com"),
222 None,
223 false,
224 NewUserParams {
225 github_login: username.clone(),
226 github_user_id: ix as i32,
227 },
228 )
229 .await
230 .unwrap()
231 .user_id;
232 users.push(UserTestPlan {
233 user_id,
234 username,
235 online: false,
236 next_root_id: 0,
237 operation_ix: 0,
238 allow_client_disconnection,
239 });
240 }
241
242 T::initialize(server, &users).await;
243
244 let plan = Arc::new(Mutex::new(Self {
245 replay: false,
246 allow_server_restarts,
247 allow_client_reconnection,
248 allow_client_disconnection,
249 stored_operations: Vec::new(),
250 operation_ix: 0,
251 next_batch_id: 0,
252 max_operations: max_operations(),
253 users,
254 rng,
255 }));
256
257 if let Some(path) = plan_load_path() {
258 let json = LOADED_PLAN_JSON
259 .lock()
260 .get_or_insert_with(|| {
261 eprintln!("loaded test plan from path {:?}", path);
262 std::fs::read(path).unwrap()
263 })
264 .clone();
265 plan.lock().deserialize(json);
266 }
267
268 plan
269 }
270
271 fn deserialize(&mut self, json: Vec<u8>) {
272 let stored_operations: Vec<StoredOperation<T::Operation>> =
273 serde_json::from_slice(&json).unwrap();
274 self.replay = true;
275 self.stored_operations = stored_operations
276 .iter()
277 .cloned()
278 .enumerate()
279 .map(|(i, mut operation)| {
280 let did_apply = Arc::new(AtomicBool::new(false));
281 if let StoredOperation::Server(ServerOperation::MutateClients {
282 batch_id: current_batch_id,
283 user_ids,
284 ..
285 }) = &mut operation
286 {
287 assert!(user_ids.is_empty());
288 user_ids.extend(stored_operations[i + 1..].iter().filter_map(|operation| {
289 if let StoredOperation::Client {
290 user_id, batch_id, ..
291 } = operation
292 && batch_id == current_batch_id {
293 return Some(user_id);
294 }
295 None
296 }));
297 user_ids.sort_unstable();
298 }
299 (operation, did_apply)
300 })
301 .collect()
302 }
303
304 fn serialize(&mut self) -> Vec<u8> {
305 // Format each operation as one line
306 let mut json = Vec::new();
307 json.push(b'[');
308 for (operation, applied) in &self.stored_operations {
309 if !applied.load(SeqCst) {
310 continue;
311 }
312 if json.len() > 1 {
313 json.push(b',');
314 }
315 json.extend_from_slice(b"\n ");
316 serde_json::to_writer(&mut json, operation).unwrap();
317 }
318 json.extend_from_slice(b"\n]\n");
319 json
320 }
321
322 fn next_server_operation(
323 &mut self,
324 clients: &[(Rc<TestClient>, TestAppContext)],
325 ) -> Option<(ServerOperation, Arc<AtomicBool>)> {
326 if self.replay {
327 while let Some(stored_operation) = self.stored_operations.get(self.operation_ix) {
328 self.operation_ix += 1;
329 if let (StoredOperation::Server(operation), applied) = stored_operation {
330 return Some((operation.clone(), applied.clone()));
331 }
332 }
333 None
334 } else {
335 let operation = self.generate_server_operation(clients)?;
336 let applied = Arc::new(AtomicBool::new(false));
337 self.stored_operations
338 .push((StoredOperation::Server(operation.clone()), applied.clone()));
339 Some((operation, applied))
340 }
341 }
342
343 fn next_client_operation(
344 &mut self,
345 client: &TestClient,
346 current_batch_id: usize,
347 cx: &TestAppContext,
348 ) -> Option<(T::Operation, Arc<AtomicBool>)> {
349 let current_user_id = client.current_user_id(cx);
350 let user_ix = self
351 .users
352 .iter()
353 .position(|user| user.user_id == current_user_id)
354 .unwrap();
355 let user_plan = &mut self.users[user_ix];
356
357 if self.replay {
358 while let Some(stored_operation) = self.stored_operations.get(user_plan.operation_ix) {
359 user_plan.operation_ix += 1;
360 if let (
361 StoredOperation::Client {
362 user_id, operation, ..
363 },
364 applied,
365 ) = stored_operation
366 && user_id == ¤t_user_id {
367 return Some((operation.clone(), applied.clone()));
368 }
369 }
370 None
371 } else {
372 if self.operation_ix == self.max_operations {
373 return None;
374 }
375 self.operation_ix += 1;
376 let operation = T::generate_operation(
377 client,
378 &mut self.rng,
379 self.users
380 .iter_mut()
381 .find(|user| user.user_id == current_user_id)
382 .unwrap(),
383 cx,
384 );
385 let applied = Arc::new(AtomicBool::new(false));
386 self.stored_operations.push((
387 StoredOperation::Client {
388 user_id: current_user_id,
389 batch_id: current_batch_id,
390 operation: operation.clone(),
391 },
392 applied.clone(),
393 ));
394 Some((operation, applied))
395 }
396 }
397
398 fn generate_server_operation(
399 &mut self,
400 clients: &[(Rc<TestClient>, TestAppContext)],
401 ) -> Option<ServerOperation> {
402 if self.operation_ix == self.max_operations {
403 return None;
404 }
405
406 Some(loop {
407 break match self.rng.gen_range(0..100) {
408 0..=29 if clients.len() < self.users.len() => {
409 let user = self
410 .users
411 .iter()
412 .filter(|u| !u.online)
413 .choose(&mut self.rng)
414 .unwrap();
415 self.operation_ix += 1;
416 ServerOperation::AddConnection {
417 user_id: user.user_id,
418 }
419 }
420 30..=34 if clients.len() > 1 && self.allow_client_disconnection => {
421 let (client, cx) = &clients[self.rng.gen_range(0..clients.len())];
422 let user_id = client.current_user_id(cx);
423 self.operation_ix += 1;
424 ServerOperation::RemoveConnection { user_id }
425 }
426 35..=39 if clients.len() > 1 && self.allow_client_reconnection => {
427 let (client, cx) = &clients[self.rng.gen_range(0..clients.len())];
428 let user_id = client.current_user_id(cx);
429 self.operation_ix += 1;
430 ServerOperation::BounceConnection { user_id }
431 }
432 40..=44 if self.allow_server_restarts && clients.len() > 1 => {
433 self.operation_ix += 1;
434 ServerOperation::RestartServer
435 }
436 _ if !clients.is_empty() => {
437 let count = self
438 .rng
439 .gen_range(1..10)
440 .min(self.max_operations - self.operation_ix);
441 let batch_id = util::post_inc(&mut self.next_batch_id);
442 let mut user_ids = (0..count)
443 .map(|_| {
444 let ix = self.rng.gen_range(0..clients.len());
445 let (client, cx) = &clients[ix];
446 client.current_user_id(cx)
447 })
448 .collect::<Vec<_>>();
449 user_ids.sort_unstable();
450 ServerOperation::MutateClients {
451 user_ids,
452 batch_id,
453 quiesce: self.rng.gen_bool(0.7),
454 }
455 }
456 _ => continue,
457 };
458 })
459 }
460
461 async fn apply_server_operation(
462 plan: Arc<Mutex<Self>>,
463 deterministic: BackgroundExecutor,
464 server: &mut TestServer,
465 clients: &mut Vec<(Rc<TestClient>, TestAppContext)>,
466 client_tasks: &mut Vec<Task<()>>,
467 operation_channels: &mut Vec<futures::channel::mpsc::UnboundedSender<usize>>,
468 operation: ServerOperation,
469 cx: &mut TestAppContext,
470 ) -> bool {
471 match operation {
472 ServerOperation::AddConnection { user_id } => {
473 let username;
474 {
475 let mut plan = plan.lock();
476 let user = plan.user(user_id);
477 if user.online {
478 return false;
479 }
480 user.online = true;
481 username = user.username.clone();
482 };
483 log::info!("adding new connection for {}", username);
484
485 let mut client_cx = cx.new_app();
486
487 let (operation_tx, operation_rx) = futures::channel::mpsc::unbounded();
488 let client = Rc::new(server.create_client(&mut client_cx, &username).await);
489 operation_channels.push(operation_tx);
490 clients.push((client.clone(), client_cx.clone()));
491
492 let foreground_executor = client_cx.foreground_executor().clone();
493 let simulate_client =
494 Self::simulate_client(plan.clone(), client, operation_rx, client_cx);
495 client_tasks.push(foreground_executor.spawn(simulate_client));
496
497 log::info!("added connection for {}", username);
498 }
499
500 ServerOperation::RemoveConnection {
501 user_id: removed_user_id,
502 } => {
503 log::info!("simulating full disconnection of user {}", removed_user_id);
504 let client_ix = clients
505 .iter()
506 .position(|(client, cx)| client.current_user_id(cx) == removed_user_id);
507 let Some(client_ix) = client_ix else {
508 return false;
509 };
510 let user_connection_ids = server
511 .connection_pool
512 .lock()
513 .user_connection_ids(removed_user_id)
514 .collect::<Vec<_>>();
515 assert_eq!(user_connection_ids.len(), 1);
516 let removed_peer_id = user_connection_ids[0].into();
517 let (client, client_cx) = clients.remove(client_ix);
518 let client_task = client_tasks.remove(client_ix);
519 operation_channels.remove(client_ix);
520 server.forbid_connections();
521 server.disconnect_client(removed_peer_id);
522 deterministic.advance_clock(RECEIVE_TIMEOUT + RECONNECT_TIMEOUT);
523 deterministic.start_waiting();
524 log::info!("waiting for user {} to exit...", removed_user_id);
525 client_task.await;
526 deterministic.finish_waiting();
527 server.allow_connections();
528
529 for project in client.dev_server_projects().iter() {
530 project.read_with(&client_cx, |project, cx| {
531 assert!(
532 project.is_disconnected(cx),
533 "project {:?} should be read only",
534 project.remote_id()
535 )
536 });
537 }
538
539 for (client, cx) in clients {
540 let contacts = server
541 .app_state
542 .db
543 .get_contacts(client.current_user_id(cx))
544 .await
545 .unwrap();
546 let pool = server.connection_pool.lock();
547 for contact in contacts {
548 if let db::Contact::Accepted { user_id, busy, .. } = contact
549 && user_id == removed_user_id {
550 assert!(!pool.is_user_online(user_id));
551 assert!(!busy);
552 }
553 }
554 }
555
556 log::info!("{} removed", client.username);
557 plan.lock().user(removed_user_id).online = false;
558 client_cx.update(|cx| {
559 cx.clear_globals();
560 drop(client);
561 });
562 }
563
564 ServerOperation::BounceConnection { user_id } => {
565 log::info!("simulating temporary disconnection of user {}", user_id);
566 let user_connection_ids = server
567 .connection_pool
568 .lock()
569 .user_connection_ids(user_id)
570 .collect::<Vec<_>>();
571 if user_connection_ids.is_empty() {
572 return false;
573 }
574 assert_eq!(user_connection_ids.len(), 1);
575 let peer_id = user_connection_ids[0].into();
576 server.disconnect_client(peer_id);
577 deterministic.advance_clock(RECEIVE_TIMEOUT + RECONNECT_TIMEOUT);
578 }
579
580 ServerOperation::RestartServer => {
581 log::info!("simulating server restart");
582 server.reset().await;
583 deterministic.advance_clock(RECEIVE_TIMEOUT);
584 server.start().await.unwrap();
585 deterministic.advance_clock(CLEANUP_TIMEOUT);
586 let environment = &server.app_state.config.zed_environment;
587 let (stale_room_ids, _) = server
588 .app_state
589 .db
590 .stale_server_resource_ids(environment, server.id())
591 .await
592 .unwrap();
593 assert_eq!(stale_room_ids, vec![]);
594 }
595
596 ServerOperation::MutateClients {
597 user_ids,
598 batch_id,
599 quiesce,
600 } => {
601 let mut applied = false;
602 for user_id in user_ids {
603 let client_ix = clients
604 .iter()
605 .position(|(client, cx)| client.current_user_id(cx) == user_id);
606 let Some(client_ix) = client_ix else { continue };
607 applied = true;
608 if let Err(err) = operation_channels[client_ix].unbounded_send(batch_id) {
609 log::error!("error signaling user {user_id}: {err}");
610 }
611 }
612
613 if quiesce && applied {
614 deterministic.run_until_parked();
615 T::on_quiesce(server, clients).await;
616 }
617
618 return applied;
619 }
620 }
621 true
622 }
623
624 async fn simulate_client(
625 plan: Arc<Mutex<Self>>,
626 client: Rc<TestClient>,
627 mut operation_rx: futures::channel::mpsc::UnboundedReceiver<usize>,
628 mut cx: TestAppContext,
629 ) {
630 T::on_client_added(&client, &mut cx).await;
631
632 while let Some(batch_id) = operation_rx.next().await {
633 let Some((operation, applied)) =
634 plan.lock().next_client_operation(&client, batch_id, &cx)
635 else {
636 break;
637 };
638 applied.store(true, SeqCst);
639 match T::apply_operation(&client, operation, &mut cx).await {
640 Ok(()) => {}
641 Err(TestError::Inapplicable) => {
642 applied.store(false, SeqCst);
643 log::info!("skipped operation");
644 }
645 Err(TestError::Other(error)) => {
646 log::error!("{} error: {}", client.username, error);
647 }
648 }
649 cx.executor().simulate_random_delay().await;
650 }
651 log::info!("{}: done", client.username);
652 }
653
654 fn user(&mut self, user_id: UserId) -> &mut UserTestPlan {
655 self.users
656 .iter_mut()
657 .find(|user| user.user_id == user_id)
658 .unwrap()
659 }
660}
661
662impl UserTestPlan {
663 pub fn next_root_dir_name(&mut self) -> String {
664 let user_id = self.user_id;
665 let root_id = util::post_inc(&mut self.next_root_id);
666 format!("dir-{user_id}-{root_id}")
667 }
668}
669
670impl From<anyhow::Error> for TestError {
671 fn from(value: anyhow::Error) -> Self {
672 Self::Other(value)
673 }
674}
675
676fn path_env_var(name: &str) -> Option<PathBuf> {
677 let value = env::var(name).ok()?;
678 let mut path = PathBuf::from(value);
679 if path.is_relative() {
680 let mut abs_path = PathBuf::from(env!("CARGO_MANIFEST_DIR"));
681 abs_path.pop();
682 abs_path.pop();
683 abs_path.push(path);
684 path = abs_path
685 }
686 Some(path)
687}