On 4/1/06, Robin Bowes wrote: > > Why would you want to use xinetd instead of tcpserver?
I'm using xinetd as a round about way to use pperl because the 0.32 README recommends pperl for better performance (it mentions forkserver but seems to recommend it as a backup choice when pperl is unstable). It says to change the run script to use a pperl line provided. I'm assuming it means to use pperl line instead of tcpserver since they both lines call the qpsmtpd script, however if tcpserver isn't used, the TCPREMOTE* variables dont get populated and Qpsmtpd::TcpServer dies. So something else is needed to populate the $remote_(ip|host|info) variables in Qpsmtpd::TcpServer or it won't run. Using the TcpServer socket patch, those variables can be populated using xinetd. For better performance, should one avoid tcpserver? Or have I read it wrong? Also, I've found that qpsmtpd-forkserver works better than qpsmtpd with > pperl or speedy-cgi. What do you mean by works better? The README mentions in some environments pperl is unstable but seems to imply pperl is faster than forkserver and preferred when it's not unstable. Should pperl ever be used over forkserver? The 0.32 README seems to say so but perhaps that's not the case? John
