We need to wake the kv thread whenever setting deferred_aggressive to
ensure that txns with deferred io that have committed but haven't submitted
their deferred writes get submitted.  This aligns us with the other
users of deferred_aggressive (e.g., _osr_drain_all).
Signed-off-by: Sage Weil <sage@redhat.com>
               << dendl;
       ++deferred_aggressive;
       deferred_try_submit();
+      {
+       // wake up any previously finished deferred events
+       std::lock_guard<std::mutex> l(kv_lock);
+       kv_cond.notify_one();
+      }
       throttle_deferred_bytes.get(txc->cost);
       --deferred_aggressive;
    }