Skip to content

High-performance browser automation bridge and multi-instance orchestrator with advanced stealth injection and real-time dashboard.

test: the state-lock-released guard drives three of five persist sites, so UpdateConfig can write under the lock unnoticed (PIN-159)

Luigi Agosti7b6724119010
diff --git a/internal/browsersession/session_test.go b/internal/browsersession/session_test.go
index f3e357fc..0a43bb60 100644
--- a/internal/browsersession/session_test.go
+++ b/internal/browsersession/session_test.go
@@ -189,15 +189,34 @@ func TestValidateDebouncesLastSeenPersistence(t *testing.T) {
}
}
+// No filesystem call may happen while the state lock is held — the property the two-lock
+// split exists to establish. Every writeSnapshot call site in session.go is driven here:
+// Create, withValidSession (through Elevate), Revoke and UpdateConfig.
+//
+// loadPersisted is the fifth site and is deliberately NOT driven, inspected rather than
+// pinned. The reason is a CONDITION, not a category: it has exactly one caller, inside
+// NewManager, which runs it after unlocking and before returning the manager — so no other
+// goroutine can hold a reference yet and a held lock has nothing to interact with. A second
+// caller of loadPersisted, or a call after the manager is published, retires that reasoning
+// and this site needs driving like the rest. Installing beforeWrite is impossible before
+// NewManager returns, so covering it would mean a production seam on a startup-only path.
func TestPersistWritesWithStateLockReleased(t *testing.T) {
+ // Exact, not a floor. writeSnapshot returns before it ever calls beforeWrite when the
+ // job carries no path, and snapshotLocked yields an empty job whenever the manager
+ // stops persisting — so a site that quietly stopped writing would leave this guard
+ // driving fewer sites than it names while a non-zero check still passed. That is the
+ // same silent vacuity the guard exists to catch, so the count is the assertion.
+ const wantWrites = 4
+
path := filepath.Join(t.TempDir(), "sessions.json")
- mgr := NewManager(Config{
+ cfg := Config{
IdleTimeout: time.Hour,
MaxLifetime: 24 * time.Hour,
ElevationWindow: time.Minute,
Persist: true,
PersistPath: path,
- })
+ }
+ mgr := NewManager(cfg)
writes := 0
available := 0
@@ -217,9 +236,13 @@ func TestPersistWritesWithStateLockReleased(t *testing.T) {
t.Fatal("Elevate() = false, want true")
}
mgr.Revoke(sessionID)
+ // The same config deliberately: Persist must stay true or this call contributes no
+ // observed write at all, and the path must stay the same or UpdateConfig also removes
+ // the old file, adding a filesystem side effect to a test about lock ordering.
+ mgr.UpdateConfig(cfg)
- if writes == 0 {
- t.Fatal("no persist write observed")
+ if writes != wantWrites {
+ t.Fatalf("observed %d persist writes, want %d — one per writeSnapshot site this test drives; a site that stopped writing is not covered by the lock assertion below", writes, wantWrites)
}
if available != writes {
t.Errorf("state lock was held during %d of %d writes, want 0", writes-available, writes)