Henrik Nordstrom wrote: > tis 2006-06-13 klockan 13:08 -0700 skrev John Oliver: >> Googling turned up a three-year-old post indicating that there would be >> problems with sending logs to a named pipe under load. Well, I'm hoping >> that's been resolved... we expect to have to hundle dozens or hundreds >> of hits per second :-) And accurate log handling is critical for >> billing. What are my best options? > > Sending logs to a pipe (named, or explicit opened by Squid if one is to > add such patch) always has issues under load as the performance gets > limited to the performance of your log processor. If the log processor > can't keep up Squid will soon stop all activities until it catches up. > The same also happens if the log processor should exit for some reason.. > > My recommendation is therefore to use the perl module File::Tail to > monitor the on-disk file. File::Tail is quite similar to "tail -f" but > automatically detects when the logfile has been rotated so it can be > kept running completely independent of Squid, and will process log > records as fast as it can without delaying Squid under peak load. tail -F has the same effect - it monitors the filename rather than the inode. R.
