httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From David Robinson <>
Subject Re: inquiry regarding child process reaping (fwd)
Date Tue, 20 Aug 1996 08:30:30 GMT
On Mon, 19 Aug 1996, Ralf S. Engelschall wrote:

> On 19 Aug 1996 15:17:42 +0200 in en.lists.apache-new-httpd you wrote:
> > [...]
> > >     - More significantly, we intermittently run into a problem with
> > >       amd hanging (it's a known problem on AIX, OSF/1, Linux, and
> > >       possibly other platforms). Since amd is single-threaded, when 
> > >       it hangs, all subsequent requests hang as well. 
> > [...]
> > What on Earth is amd? [...]
> AMD = AutoMount Daemon
>        -   -     - 
> FYI:
> In this case, the freely available automount package which has been ported to
> nearly every Unix derivate. It is a great package (even better then the SunOS
> or Solaris automounters).  But the observation of AMD is correct and really a
> drawback of AMD: When it hangs, all hangs! But in practice, it works great.
> We use it on two clusters of SunoS 4.1.3 machines and it works reall fine.

I haven't been following this thread, but...

amd is a freely available automount daemon, written by Jan-Simon Pendry at
Imperial College (UK) (and the documentation does read like a thesis 

Yes, if your automounter hangs, then any process which trys to access an
un-mounted filesystem will hang inside the kernel. Processes which are
accessing already mounted filesystems might hang, if the filesystem was
not mounted in-place.

However, amd was designed to never hang in normal operation; so if it does,
it should be regarded as a bug. It takes extreme care to avoid any blockable
system calls; for those system calls for which blocking might be unavoidable,
it forks before calling them.

amd was significantly better than the SunOS automounter; the latter would
hang if it tried to mount a filesystem from a remote server that was down.
Whereas the Solaris automounter is just as good as amd, because Sun took the 
opportunity to put automounting features into the kernel.


View raw message