This is part of The Pile, a partial archive of some open source mailing lists and newsgroups.
Subject: Re: RFC: mod_perl advocacy project resurrection From: Stephen.A.Cochran@Cahir.net (Stephen A. Cochran) Date: 06 Dec 2000 10:36:43 EST [...] Recently I was handed an online event calendar running under CGI and asked to get it to run under mod_perl to speed it up. I'm comfortable with perl, but by no means a guru. And this was the first time using mod_perl. In the past I've used several products which might be considered competitors to mod_perl: Web Objects and Cold Fusion. CF was horrible, I'm glad I didn't have to work with it for long. WO was very nice to work with, and it had all the ecommerce and transation logic needed to build a enterprise web application quickly. (None of my work has been in ecommerce, instead in the medical industry (patient data, lab results, etc) which has many of the same requirements). [...] Technical Issues: The second problem I see with mod_perl is a technical one. Because of the design of mod_perl, debugging problems can be fairly involved. Is the problem my code? Or is it apache, or maybe mod_perl? Could even be perl for that matter. Take for example the problem I ran into recently. (Please don't take this as a rant just because I had problems, I'm happy with mod_perl. But I'm fairly capable around unix and compiling, instead imagine this happening to someone who wasn't that comfortable.) The event calendar works great under CGI, so I try it under mod_perl. It was written to work under mod_perl by a better perl programmer that I'll ever be (Jon Howell of Faq-o-matic fame) and is very well designed. It even worked under mod_perl with some minor changes like supplying the user/pass to the database since it wasn't running under cgiwrap any longer. And it worked! But only 90% of the time. The other 10% of the time, the client received no data, and the page generated by the script ended up in the apache log. After cleaning up one implicit database handle destruction warning the problem persisted, and I began to attempt to discover where the connection was being closed. So I did what any lazy programmer does, I added some print statements to send a note to STDERR. While these all showed up when the program worked correctly, when problem occured, the prints to STDERR dissapeared. But the html page was printed to the apache log, basicly standard error. So it looks like if the socket is closed, stderr dissapears, and stdout goes to the error log. So no pointers appeared in the log, which wasn't very helpful. Next attempt was to do some packet sniffing to see why the socket was closeing. Turns out that the web server sends an orderly release indication (T_ORDEL_IND = 132), so the client being a good citizen reponds with a orderly release request and there goes the socket. While my description of the problem was more in-depth that I meant it to be, my point is that now I'm left with the need to trace through system calls made by apache, mod_perl and my program to try and find who's asking the client to kill the connection. It's a pain for me (expecially since this is all running on Dec Unix which doesn't have a system call trace funtion), I can't imagine what it would be like for some newbie sys admin. Unfortunatly I don't see any way around this problem. Due to the design of mod_perl, understanding the debuging the whole system is nessisary. And before anyone says this is the strength of open source software (of which I am an advocate) most people are just trying to get something up and running. They probably don't have the time or the expertice to go looking through code to figure out what's going wrong. ===