Re: FastCGI Rouge processes
Ben Chabot <[email protected]>
| Newsgroups | gmane.comp.web.fastcgi.devel |
|---|---|
| Message-ID | <[email protected]> |
xxpmuser wrote: > Hi, I am relatively new to FastCGI and process signaling so your patience is > appreciated. > > I recently inherited a FastCGI application written in Perl which runs on > Apache. > > The script is configured to run a maximum of 1 hour, such that the script > exits the FastCGI loop after the $lifetime variable has been exceeded. > However, every now and then, a process does not exit and keeps on running > for hours, sometimes even days. When this happens, the webpage becomes > inaccessible (giving a FastCGI: comm with (dynamic) server (first read) > idle timeout (30 sec) message in the logs) and we have to request our Apache > admins to have the particular process killed. After the process has been > killed, the page becomes accessible again > This sounds as if your application stops responding to fastcgi. Do you have access to the config options of Apache? You might try FastCgiConfig options. But probably the best thing to do is watch the memory size of your application, and perhaps strace it, see what it's doing. Maybe make it print debug statements and try to pin down why it stops? Is it a very large app? This might be hard to sort out why it suddenly stops responding to fastcgi. I think Fastcgi can restart an app upon communications failure (probably not an ideal solution), check out their documentation. > Can anyone tell me what is possibly causing the application to behave this > way? How is it possible that a single rouge process impacts other > instances/processes of the application and causing the page to become > inaccessible? > > We don't have any control over our Apache webservers so code revision is our > only option. Someone suggested that I look into possible memory leaks and > proper variable declaration (the issue of reusing variables in FastCGI), but > I'm not exactly sure how to track/address those. > > Any help is very much appreciated. > ___________________________________ fastcgi-developers mailing list http://fastcgi.com/fastcgi-developers/