On Fri, May 20, 2005 at 07:50:04AM -0400, Jeff Trawick wrote: > On 5/20/05, Joe Orton <[EMAIL PROTECTED]> wrote: > > As discussed previously; this patch stops killing piped loggers; except > > that prefork still kills them at shutdown and ungraceful restart since > > it signals the entire process group, so it's not entirely consistent. > > Something I'd like to confirm: > > If the httpd code remains the same: > a) all existing piped loggers continue to work with all levels of Apache > b) a piped logger which wants to stay alive long enough to read until > eof can simply ignore SIGTERM (or better yet, use SIGTERM to set some > kind of alarm/timeout so that it doesn't stay around forever if it > never reaches eof due to some httpd or application misbehavior)
Yes, I believe that's true. I think it would be pretty unreasonable to require that piped loggers are changed to ignore SIGTERM to fix this, simply because httpd sends SIGTERM at the wrong time. Deliberately ignoring SIGTERM is not something you'd generally advocate at all. (btw: I checked cronolog again and it actually does just exit() on eof; don't know what I was thinking of which didn't) joe
