/*
* A naive implementation would have all idle threads
- * calling select() and then they'd all wake up at once. We
- * solve this problem by putting the select() in a critical
- * section, so only one thread has the opportunity to wake
- * up. If we wake up on a master socket, create a new
- * session context; otherwise, just bind the thread to the
- * context we want and go on our merry way.
+ * calling select() and then they'd all wake up at once
+ * (known in computer science as the "thundering herd"
+ * problem). We solve this problem by putting the select()
+ * in a critical section, so only one thread has the
+ * opportunity to wake up. If we wake up on a master
+ * socket, create a new session context; otherwise, just
+ * bind the thread to the context we want and go on our
+ * merry way.
*/
/* make doubly sure we're not holding any stale db handles