On Thursday 15 March 2007 11:18, Darryl Miles wrote: > Torsten Foertsch wrote: > > On Tuesday 13 March 2007 10:05, Torsten Foertsch wrote: > > So, the simple patch below solves my problem. No segfaults so far. But is > > it right? > > > > I have tested it with ab and several million requests and no memory leak > > detected. The test suite is also passed. But that doesn't mean anything. > > I have grepped t for "interpscope handler" without success. > > Are you suggesting the patch corrects a leak or are you saying it does > not introduce a new one ?
The latter. > Unfortunately I can't comment on the correctness of the path itself, but > I'd be interested in taking a look if this patch corrects a leak as that > issue interests me. The problem is that modperl_interp_unselect() is called at the end of each phase then again as at request cleanup. That last call has caused the segfault. So, the patch cures only the symptoms. I think the patch is not really complete. I think that whole thing of InterpScope handler needs to be reworked. It is not covered at all by the test suite and there are several occasions in the code where an modperl_interp_unselect is commented out because it breaks prefork. Another question in this regard is how filters are handled. Are they given a separate interpreter? Is the interpreter released between each filter invocation or only after eos was seen? I'd like to use modperl mostly for configuration in a threaded server (see Apache2::Translation on CPAN). And ideally I'd have a function Modperl::release_interp or so that releases the interpreter after the current phase. Then I could release the interpreter after fixup so that it can be reused while the current request is doing all of it's slow networking stuff. Torsten
pgpDnEiGL6V4W.pgp
Description: PGP signature
