With a default Typo configuration running on lighttpd, it was producing more output in the error log than in the access log with something about "missing cleanup", so I finally got around to migrating over to apache2 and mod_fcgid. (The short-term solution was a couple of cron jobs to keep the lighttpd process up.) The usual configuration work (see, e.g., James Duncan Davidson's write-up) was a dead-end, but turning up the logging verbosity on the apache2 instance had the solution:
[Sun Nov 05 18:13:19 2006] [info] mod_fcgid: read data timeout in 5 seconds
Initialization for one of the processes takes way longer than that, so putting some more sane values into the mod_fcgid configuration along the lines of what Spencer Miles used was the solution. At least so far, the apache2 and mod_fcgid combination is much snappier and more stable than the lighttpd configuration.
Comment from Tom Copeland @ 2006-11-06T15:39:48Z # permalink
Comment from Paul Brown @ 2006-11-06T18:12:38Z # permalink