[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [Libvir] Fix handling of config files with duplicate named guests



On Tue, Dec 19, 2006 at 02:05:32PM -0500, Daniel Veillard wrote:
> On Tue, Dec 19, 2006 at 06:48:16PM +0000, Daniel P. Berrange wrote:
> > The code for managing config files in /etc/xen assumed that the filename of
> > the config matched the name of the guest defined inside. One might thing
> > this a reasonable assumption, but in the wild I've had reports from users
> > whom have various config files for the same guest, but with different
> > settings. This caused some really wierd behaviour in the current code base.
> > virt-manager display would alternate displaying each config upon refresh!
> > 
> > Now, our API requires that there is only one config per name, so we can't
> > expose this to the UI. The attached patch, thus simply detects when we 
> > have more than one config with same named guest, and hides the duplicates.
> > 
> > The way it does this is to change the way we cache configs. Previously we
> > had a single hash table mapping relative filenames to virConfPtr objects, 
> > with the implicit assumption that relative filename == guest name. With 
> > the attached patch we now maintain two hash tables. The first maps fully
> > qualified pathnames to virConfPtr objects, the second maps guest names
> > to the first filename found for that name. Thus when querying details for
> > a guest, we first resolve the guest name to its filename, and then lookup
> > the virConfPtr object associated with this filename. 
> > 
> > This isn't perfect, but its a hell of alot better than current code, so
> > I want to commit it as is and figure out more improvement later. There is
> > no ABI issue here, so we can iterate over impl at will.
> 
>   okay sounds fine to me...

Ok, will commit.

> > +        value = malloc(sizeof(virConfValue));
> > +        value->str = strdup(filename);
> > +            free(value);
> > +            return (-1);
> 
>   This makes me think about an idea I had to swicth all allocations in
> libvirt to reuse xmlMalloc/xmlFree/xmlStrdup . The reasons are the following:
>    - libxml2 is a mandatory module
>    - libxml2 memory debug allow to track easilly memory allocations
>      and check we never leak, this can be really useful
>    - libxml2 memory routines can also be overiden by the library client
>      to use a different allocator which in some circumstances can be really
>      useful
> 
> My prime motivation is debug but reuse of libxml2 facilities is not neglectible.

Sounds reasonble. I'm not familiar with libxml2 debug stuff though - does
it give more info that we can get from valgrind ?

Dan.
-- 
|=- Red Hat, Engineering, Emerging Technologies, Boston.  +1 978 392 2496 -=|
|=-           Perl modules: http://search.cpan.org/~danberr/              -=|
|=-               Projects: http://freshmeat.net/~danielpb/               -=|
|=-  GnuPG: 7D3B9505   F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505  -=| 


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]