1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
//! Background health checker — periodically evicts clients that have
//! stopped polling `getupdates` for longer than `timeout`.
use std::sync::atomic::Ordering;
use std::sync::Arc;
use std::time::Duration;
use tracing::info;
use super::HubState;
const OFFLINE_THRESHOLD_SECS: u64 = 90;
const CHECK_INTERVAL_SECS: u64 = 30;
pub fn spawn_health_checker(state: Arc<HubState>) {
let mut shutdown = state.ilink.shutdown.clone();
tokio::spawn(async move {
loop {
tokio::select! {
biased;
_ = shutdown.changed() => {
if *shutdown.borrow() {
info!("health checker shutting down");
return;
}
}
_ = tokio::time::sleep(Duration::from_secs(CHECK_INTERVAL_SECS)) => {
let now_secs = std::time::SystemTime::now()
.duration_since(std::time::UNIX_EPOCH)
.map(|d| d.as_secs())
.unwrap_or(0);
let threshold = now_secs.saturating_sub(OFFLINE_THRESHOLD_SECS);
// Mark offline: read last_seen timestamps lock-free, then write
// only to the registry for the subset of stale clients.
{
let stale_vtokens: Vec<String> = state
.clients
.last_seen
.iter()
.filter(|e| e.value().load(Ordering::Relaxed) < threshold)
.map(|e| e.key().clone())
.collect();
if !stale_vtokens.is_empty() {
let mut registry = state.clients.registry.write().await;
for vtoken in &stale_vtokens {
registry.mark_offline(vtoken);
}
}
}
let registry = state.clients.registry.read().await;
let online = registry.online_clients().len();
let total = registry.all_clients().len();
info!(online, total, "health check: client status");
}
}
}
});
}
#[cfg(test)]
mod tests {
//! N-01 / F-M1-N01: the `last_seen` DashMap must be cleaned up
//! alongside the registry on every unregister path. Pre-N-01 the
//! map accumulated entries forever (each `register_client_in_hub`
//! gets a fresh UUID-based vtoken, so the old key was never
//! overwritten). These tests pin the cleanup contract on the
//! `last_seen` data structure itself so the registry paths can be
//! reviewed without dragging in the full pair/store/queue stack.
use super::*;
use crate::hub::registry::ClientRegistry;
use crate::hub::{ClientState, InMemoryQueue};
use crate::MessageQueue;
fn make_client_state() -> Arc<ClientState> {
let queue: Arc<dyn MessageQueue> = Arc::new(InMemoryQueue::new());
Arc::new(ClientState::new(queue))
}
/// Direct write/read/remove on `last_seen` mirrors what the
/// production `unregister_client_in_hub` path does (paired with
/// `registry.remove` + `queue.remove_client` + `store.delete`). If
/// this ever starts failing, the N-01 invariant is gone.
#[tokio::test]
async fn last_seen_remove_clears_entry_for_vtoken() {
let clients = make_client_state();
let vtoken = "vhub_test_aaaa".to_string();
// Seed: write a last_seen timestamp the way `getupdates` does.
clients
.last_seen
.entry(vtoken.clone())
.or_insert_with(|| std::sync::atomic::AtomicU64::new(0))
.store(1_700_000_000, Ordering::Relaxed);
assert!(
clients.last_seen.contains_key(&vtoken),
"precondition: seeded last_seen entry must be visible"
);
// Production path: registry.remove + last_seen.remove together.
let mut registry = clients.registry.write().await;
registry.remove("dummy-name-not-in-registry");
clients.last_seen.remove(&vtoken);
assert!(
!clients.last_seen.contains_key(&vtoken),
"post-N-01 invariant: last_seen entry for unregistered vtoken must be gone"
);
}
/// Pre-N-01 simulation: write many entries, drop the registry path
/// entirely, and confirm `last_seen` would have grown without bound.
/// After N-01 the cleanup is co-located with the registry remove
/// call; this test simply asserts the data structure supports the
/// remove operation that production code now performs.
#[tokio::test]
async fn last_seen_grows_without_cleanup_when_remove_skipped() {
let clients = make_client_state();
let before = clients.last_seen.len();
// Simulate 100 getupdates-touching clients that we never
// unregister (the leak scenario).
for i in 0..100 {
clients
.last_seen
.entry(format!("vhub_leak_{i}"))
.or_insert_with(|| std::sync::atomic::AtomicU64::new(0))
.store(1_700_000_000 + i, Ordering::Relaxed);
}
assert_eq!(
clients.last_seen.len(),
before + 100,
"without cleanup, last_seen accumulates 100 entries"
);
// Now exercise the N-01 fix: clean every entry we just added.
for i in 0..100 {
clients.last_seen.remove(&format!("vhub_leak_{i}"));
}
assert_eq!(
clients.last_seen.len(),
before,
"after N-01 cleanup, last_seen returns to its starting size"
);
}
/// Helper kept here so the tests above document their reliance on
/// `ClientRegistry::remove` and `state.clients.last_seen` being
/// siblings under `ClientState`. The function is unused at runtime
/// but its signature would catch a future refactor that renames
/// either side of the pair.
#[allow(dead_code)]
fn assert_cleanup_pair_exists(state: &ClientState) {
let _r: &tokio::sync::RwLock<ClientRegistry> = &state.registry;
let _l: &Arc<dashmap::DashMap<String, std::sync::atomic::AtomicU64>> = &state.last_seen;
}
/// M1: `spawn_health_checker` must mark clients whose `last_seen` exceeds
/// the offline threshold as offline. Catches the mutant that replaces the
/// entire function with a no-op.
#[tokio::test]
async fn health_checker_marks_stale_client_offline() {
use std::sync::Arc;
// Connect the store with real time — the SQLite pool's internal
// acquire timeout fires under a paused clock if we pause before
// connecting. We pause below, after all async init is done.
let store = crate::store::Store::connect("sqlite::memory:")
.await
.expect("in-memory store");
let upstream: Arc<dyn crate::ilink::UpstreamSink> = Arc::new(
crate::ilink::UpstreamClient::new("sk-test".to_string(), None).expect("test upstream"),
);
let queue: Arc<dyn crate::MessageQueue> = Arc::new(InMemoryQueue::new());
let (_tx, shutdown_rx) = tokio::sync::watch::channel(false);
let state = crate::hub::HubState::new(
upstream,
Arc::new(store),
queue,
shutdown_rx,
"test-relay-secret".to_string(),
crate::hub::AdminConfig::from_env(),
);
let (_, hashed_vtoken, _) =
state
.clients
.registry
.write()
.await
.register("test-backend".to_string(), None, None);
{
let mut registry = state.clients.registry.write().await;
registry.mark_online(&hashed_vtoken);
}
state
.clients
.last_seen
.entry(hashed_vtoken.clone())
.or_insert_with(|| std::sync::atomic::AtomicU64::new(0))
.store(0, Ordering::Relaxed);
// Pause tokio time only after all async setup is complete.
tokio::time::pause();
spawn_health_checker(Arc::clone(&state));
// The spawned task hasn't run yet at this point. First advance gives
// it a chance to start and register its sleep(CHECK_INTERVAL_SECS).
// Second advance fires that sleep.
tokio::time::advance(Duration::from_secs(1)).await;
tokio::time::advance(Duration::from_secs(CHECK_INTERVAL_SECS + 1)).await;
// Extra yields to let the task complete the registry write and read.
tokio::task::yield_now().await;
tokio::task::yield_now().await;
let registry = state.clients.registry.read().await;
assert!(
registry.online_clients().is_empty(),
"stale client must be marked offline"
);
}
}