My guess is, that the host had already crashed. Since mmm_mond could
no longer reach mmm_agentd (on db1), there was no way that mmm_mond
could be sure that db1 had in fact released the writer IP. Ideally you
would have an alternative route to your db servers, or at least an
alternative way of "killing" them, to make sure an ip conflict could
never occur.
Mike