Well, the process is multithreaded. So even though over a day I might generate 100 MB at most at the top logging level, after 40 - 50 forks we're talking about 4GB - 5GB a day. This has compounded the problem because I'm trying to keep the logs sorted and rotated. And, even though I can turn down the detail, the bugs I am finding require a high level of detail for testing.
(The script is a web spider. Most of the bugs I encounter with it involve bizarre / broken HTML in web pages. Problem is that in order to figure out just what is going on I want to log lots of info if there are any anomalies. The problem becomes how to do that without being too processor intensive)