Add stress testing for quota-lock and refactor test guard usage

.github/workflows/rust.yml

@@ -45,6 +45,18 @@ jobs:
     - name: Run tests
       run: cargo test --verbose
 
+    - name: Stress quota-lock suites (PR only)
+      if: github.event_name == 'pull_request'
+      env:
+        RUST_TEST_THREADS: 16
+      run: |
+        set -euo pipefail
+        for i in $(seq 1 12); do
+          echo "[quota-lock-stress] iteration ${i}/12"
+          cargo test quota_lock_ --bin telemt -- --nocapture --test-threads 16
+          cargo test relay_quota_wake --bin telemt -- --nocapture --test-threads 16
+        done
+
 # clippy dont fail on warnings because of active development of telemt 
 # and many warnings
     - name: Run clippy

src/proxy/relay.rs

@@ -316,6 +316,13 @@ fn quota_user_lock_test_guard() -> &'static Mutex<()> {
     TEST_LOCK.get_or_init(|| Mutex::new(()))
 }
 
+#[cfg(test)]
+fn quota_user_lock_test_scope() -> std::sync::MutexGuard<'static, ()> {
+    quota_user_lock_test_guard()
+        .lock()
+        .unwrap_or_else(|poisoned| poisoned.into_inner())
+}
+
 fn quota_overflow_user_lock(user: &str) -> Arc<Mutex<()>> {
     let stripes = QUOTA_USER_OVERFLOW_LOCKS.get_or_init(|| {
         (0..QUOTA_OVERFLOW_LOCK_STRIPES)

src/proxy/tests/relay_quota_lock_pressure_adversarial_tests.rs

@@ -12,9 +12,7 @@ use tokio::time::Instant;
 
 #[test]
 fn quota_lock_same_user_returns_same_arc_instance() {
-    let _guard = super::quota_user_lock_test_guard()
-        .lock()
-        .expect("quota lock test guard must be available");
+    let _guard = super::quota_user_lock_test_scope();
     let map = QUOTA_USER_LOCKS.get_or_init(DashMap::new);
     map.clear();
 
@@ -25,9 +23,7 @@ fn quota_lock_same_user_returns_same_arc_instance() {
 
 #[test]
 fn quota_lock_parallel_same_user_reuses_single_lock() {
-    let _guard = super::quota_user_lock_test_guard()
-        .lock()
-        .expect("quota lock test guard must be available");
+    let _guard = super::quota_user_lock_test_scope();
     let map = QUOTA_USER_LOCKS.get_or_init(DashMap::new);
     map.clear();
 
@@ -51,9 +47,7 @@ fn quota_lock_parallel_same_user_reuses_single_lock() {
 
 #[test]
 fn quota_lock_unique_users_materialize_distinct_entries() {
-    let _guard = super::quota_user_lock_test_guard()
-        .lock()
-        .expect("quota lock test guard must be available");
+    let _guard = super::quota_user_lock_test_scope();
     let map = QUOTA_USER_LOCKS.get_or_init(DashMap::new);
 
     map.clear();
@@ -74,9 +68,7 @@ fn quota_lock_unique_users_materialize_distinct_entries() {
 
 #[test]
 fn quota_lock_unique_churn_stress_keeps_all_inserted_keys_addressable() {
-    let _guard = super::quota_user_lock_test_guard()
-        .lock()
-        .expect("quota lock test guard must be available");
+    let _guard = super::quota_user_lock_test_scope();
     let map = QUOTA_USER_LOCKS.get_or_init(DashMap::new);
 
     map.clear();
@@ -94,9 +86,7 @@ fn quota_lock_unique_churn_stress_keeps_all_inserted_keys_addressable() {
 
 #[test]
 fn quota_lock_saturation_returns_stable_overflow_lock_without_cache_growth() {
-    let _guard = super::quota_user_lock_test_guard()
-        .lock()
-        .expect("quota lock test guard must be available");
+    let _guard = super::quota_user_lock_test_scope();
     let map = QUOTA_USER_LOCKS.get_or_init(DashMap::new);
     map.clear();
 
@@ -135,17 +125,19 @@ fn quota_lock_saturation_returns_stable_overflow_lock_without_cache_growth() {
 
 #[test]
 fn quota_lock_reclaims_unreferenced_entries_before_ephemeral_fallback() {
-    let _guard = super::quota_user_lock_test_guard()
-        .lock()
-        .expect("quota lock test guard must be available");
+    let _guard = super::quota_user_lock_test_scope();
     let map = QUOTA_USER_LOCKS.get_or_init(DashMap::new);
     map.clear();
 
-    // Fill and immediately drop strong references, leaving only map-owned Arcs.
+    // Saturate with retained strong references first so parallel tests cannot
+    // reclaim our fixture entries before we validate the reclaim path.
+    let prefix = format!("quota-reclaim-drop-{}", std::process::id());
+    let mut retained = Vec::with_capacity(QUOTA_USER_LOCKS_MAX);
     for idx in 0..QUOTA_USER_LOCKS_MAX {
-        let _ = quota_user_lock(&format!("quota-reclaim-drop-{}-{idx}", std::process::id()));
+        retained.push(quota_user_lock(&format!("{prefix}-{idx}")));
     }
-    assert_eq!(map.len(), QUOTA_USER_LOCKS_MAX);
+
+    drop(retained);
 
     let overflow_user = format!("quota-reclaim-overflow-{}", std::process::id());
     let overflow = quota_user_lock(&overflow_user);
@@ -162,9 +154,7 @@ fn quota_lock_reclaims_unreferenced_entries_before_ephemeral_fallback() {
 
 #[test]
 fn quota_lock_saturated_same_user_must_not_return_distinct_locks() {
-    let _guard = super::quota_user_lock_test_guard()
-        .lock()
-        .expect("quota lock test guard must be available");
+    let _guard = super::quota_user_lock_test_scope();
     let map = QUOTA_USER_LOCKS.get_or_init(DashMap::new);
     map.clear();
 
@@ -187,9 +177,7 @@ fn quota_lock_saturated_same_user_must_not_return_distinct_locks() {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
 async fn quota_lock_saturation_concurrent_same_user_never_overshoots_quota() {
-    let _guard = super::quota_user_lock_test_guard()
-        .lock()
-        .expect("quota lock test guard must be available");
+    let _guard = super::quota_user_lock_test_scope();
     let map = QUOTA_USER_LOCKS.get_or_init(DashMap::new);
     map.clear();
 
@@ -240,9 +228,7 @@ async fn quota_lock_saturation_concurrent_same_user_never_overshoots_quota() {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
 async fn quota_lock_saturation_stress_same_user_never_overshoots_quota() {
-    let _guard = super::quota_user_lock_test_guard()
-        .lock()
-        .expect("quota lock test guard must be available");
+    let _guard = super::quota_user_lock_test_scope();
     let map = QUOTA_USER_LOCKS.get_or_init(DashMap::new);
     map.clear();
 
@@ -322,6 +308,24 @@ fn quota_error_classifier_rejects_plain_permission_denied() {
     assert!(!is_quota_io_error(&err));
 }
 
+#[test]
+fn quota_lock_test_scope_recovers_after_guard_poison() {
+    let poison_result = std::thread::spawn(|| {
+        let _guard = super::quota_user_lock_test_scope();
+        panic!("intentional test-only guard poison");
+    })
+    .join();
+    assert!(poison_result.is_err(), "poison setup thread must panic");
+
+    let _guard = super::quota_user_lock_test_scope();
+    let map = QUOTA_USER_LOCKS.get_or_init(DashMap::new);
+    map.clear();
+
+    let a = quota_user_lock("quota-lock-poison-recovery-user");
+    let b = quota_user_lock("quota-lock-poison-recovery-user");
+    assert!(Arc::ptr_eq(&a, &b));
+}
+
 #[tokio::test]
 async fn quota_lock_integration_zero_quota_cuts_off_without_forwarding() {
     let stats = Arc::new(Stats::new());

src/proxy/tests/relay_quota_wake_liveness_regression_tests.rs

@@ -18,10 +18,8 @@ fn saturate_lock_cache() -> Vec<Arc<std::sync::Mutex<()>>> {
     retained
 }
 
-fn quota_test_guard() -> std::sync::MutexGuard<'static, ()> {
-    super::quota_user_lock_test_guard()
-        .lock()
-        .unwrap_or_else(|poisoned| poisoned.into_inner())
+fn quota_test_guard() -> impl Drop {
+    super::quota_user_lock_test_scope()
 }
 
 #[tokio::test]

src/proxy/tests/relay_quota_waker_storm_adversarial_tests.rs

@@ -23,10 +23,8 @@ impl std::task::Wake for WakeCounter {
     }
 }
 
-fn quota_test_guard() -> std::sync::MutexGuard<'static, ()> {
-    super::quota_user_lock_test_guard()
-        .lock()
-        .unwrap_or_else(|poisoned| poisoned.into_inner())
+fn quota_test_guard() -> impl Drop {
+    super::quota_user_lock_test_scope()
 }
 
 fn saturate_quota_user_locks() -> Vec<Arc<std::sync::Mutex<()>>> {

src/proxy/tests/relay_security_tests.rs

@@ -31,9 +31,7 @@ impl std::task::Wake for WakeCounter {
 
 #[tokio::test]
 async fn quota_lock_contention_does_not_self_wake_pending_writer() {
-    let _guard = super::quota_user_lock_test_guard()
-        .lock()
-        .expect("quota lock test guard must be available");
+    let _guard = super::quota_user_lock_test_scope();
     let map = super::QUOTA_USER_LOCKS.get_or_init(dashmap::DashMap::new);
     map.clear();
 
@@ -72,9 +70,7 @@ async fn quota_lock_contention_does_not_self_wake_pending_writer() {
 
 #[tokio::test]
 async fn quota_lock_contention_writer_schedules_single_deferred_wake_until_lock_acquired() {
-    let _guard = super::quota_user_lock_test_guard()
-        .lock()
-        .expect("quota lock test guard must be available");
+    let _guard = super::quota_user_lock_test_scope();
     let map = super::QUOTA_USER_LOCKS.get_or_init(dashmap::DashMap::new);
     map.clear();
 
@@ -145,9 +141,7 @@ async fn quota_lock_contention_writer_schedules_single_deferred_wake_until_lock_
 
 #[tokio::test]
 async fn quota_lock_contention_read_path_schedules_deferred_wake_for_liveness() {
-    let _guard = super::quota_user_lock_test_guard()
-        .lock()
-        .expect("quota lock test guard must be available");
+    let _guard = super::quota_user_lock_test_scope();
     let map = super::QUOTA_USER_LOCKS.get_or_init(dashmap::DashMap::new);
     map.clear();