subversion-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Daniel Shahaf <danie...@elego.de>
Subject Re: Subversion BDB doesn't work with Apache 2.4 event MPM
Date Thu, 05 Apr 2012 20:16:19 GMT
Greg Stein wrote on Thu, Apr 05, 2012 at 15:02:35 -0400:
> On Thu, Apr 5, 2012 at 14:46, Greg Stein <gstein@gmail.com> wrote:
> > On Thu, Apr 5, 2012 at 14:21, Daniel Shahaf <danielsh@elego.de> wrote:
> >> Greg Stein wrote on Thu, Apr 05, 2012 at 14:14:53 -0400:
> >>> On Apr 5, 2012 2:10 PM, "Daniel Shahaf" <danielsh@elego.de> wrote:
> >>> >
> >>> > Greg Stein wrote on Thu, Apr 05, 2012 at 14:05:31 -0400:
> >>> > > On Apr 5, 2012 6:46 AM, "Philip Martin" <philip.martin@wandisco.com>
> >>> wrote:
> >>> > ....
> >>> > > >
> >>> > > > I've raised issue 4157 to track this problem.
> >>> > >
> >>> > > Can we at least implement a detection for this situation and fail
to
> >>> start?
> >>> > > Maybe mod_dav_svn would query the mpm type, and the fs type. We
could at
> >>> > > least backport that to 1.7.x pretty quickly.
> >>> > >
> >>> >
> >>> > No need.  1.7.x doesn't compile against 2.4 for other reasons (API
> >>> > changes).  I've patched them on trunk but vetoed the backport pending
> >>> > resolution of the BDB hangs.
> >>>
> >>> That prevents somebody from using BDB with the 2.4 worker mpm.
> >>
> >> Philip reported seeing tests pass with unmodified HEAD on worker.
> >>
> >> Anyway I won't object to adding a version check on httpd (at configure-
> >> or run-time).  We know it'll hang..
> >
> > It would need to be at run-time since the MPM type and the FS type is
> > not known at configuration time.
> >
> > Oh, shoot... there might be multiple repositories, with mixed types.
> > Thus, it is really a per-request check. I wonder if we can even ask
> > the FS what type it is without opening/allocating this per-thread
> > stuff. Even if we could, it would imply needing to send a 500 response
> > back to the client. That's not a very nice situation :-/
> >

Yeah, the alternative is to do it at compile-time and disallow
the `--with-berkeley-db --with-apxs=2.4' combination --- even for people
who don't have BDB repositories.

My only issue with implementing this is whether it's easier to just fix
underlying bug, in which case these version checks won't be needed.

> > Ah. I see it: svn_fs_type(). A call to that in
> > mod_dav_svn/repos.c:get_resource(), and we can bail out with an error.
> > I'm just not sure what values to examine in some ap_mpm_query() calls
> > (ref: ap_mpm.h). There may be a single query (eg. AP_MPMQ_IS_ASYNC)
> > that identifies the event MPM.
> 
> Found the best way: strcmp(ap_show_mpm(), "event")

Mime
View raw message