interrupting sqlexec

Ever since we upgraded to 4.0 I've been noticing that interrupting a long operation (like opening a cursor for a big select) with ^C takes a very long time, like 5+ minutes. (This happens particularly with those searches that are handled much less efficiently in 4.0, as was discussed on this a few months ago. :-() My users have, unfortunately, figured out that they can get unlocked faster by having the terminal server kill off the whole login session, and starting a new one. When they do this, everything gets killed (by a HUP signal, I think) _except_ sqlexec, which continues to eat up CPU time unproductively for 5-10 minutes, after which it apparently times out and dies. Until it dies, it makes the load on the machine go wild temporarily, and degrades performance of productive tasks. It can be killed off faster, but only, as far as I can tell, by KILL signal, which doesn't give it a chance to clean up, and so has the potential of leaving tables locked (or indexes inconsistent :-(). Does anyone know of any tricks that might improve this situation (other than pursuading the users not to terminate login sessions forcibly)? This is with the 4.0 SE engine and RDS on a sun4. Thanks for any suggestions. -- Harry Bochner -- bochner@das.harvard.edu