The flush() method sets self._active.flag = True but does not check if the flag is already True before proceeding, unlike the add() method. If code running during the locked section of _flush() (lines 146-151 where self._lock is held) triggers a re-entrant call to flush() on the same thread, it would attempt to acquire the non-reentrant threading.Lock() again, causing a deadlock. The fix adds protection for add() but leaves flush() vulnerable to the same class of re-entrant deadlock.

The flush() method unconditionally sets _active.flag = True and then resets it to False in its finally block. If flush() is called while add() is already executing on the same thread (e.g., via a callback triggered during _ensure_thread() or buffer operations), the finally block in flush() will reset the flag to False while add() is still running. This breaks the re-entry guard for the remainder of add()'s execution, allowing a subsequent re-entrant call to proceed and potentially cause the deadlock this PR aims to prevent.