Vacuum deadlocks?

I've got a reasonably busy system that logically needs vacuum on a better than daily basis, at least in a couple of the tables. There are times when the traffic is lower, but they're not necessarily predictably overnight, as there are some long running processes that people like to start as they leave in the afternoon.

I'm running 7.2.1 on OSX 10.1.5 (dual 800, 768 megs of memory, the machine is serving as app/db/webserver) using Mark Liange's package from entropy.ch. an upgrade is planned, but he doesn't have newer packages for < 10.2, and I'm not planning to upgrade to 10.2 on this machine quite yet.

Every so often (perhaps 1 time in 50) the vacuumdb -a --analyze command will deadlock the system, so that the clients show a status of either 'starting' or 'async-notify', or 'idle in transaction'. This happened roughly every other day when I had vacuum running from an hourly cron, once a week from the every 3 hour cron, and now it's happened a couple of times when I run it just after backups.

What am I doing wrong here? Is this a known issue for older versions? Or should I be running the vacuums in a different manner?

eric

What am I doing wrong here?