[Linux-cluster] DLM

[Daniel Phillips]

On Friday 18 March 2005 08:08, Patricio Bruna V wrote:
> El jue, 17-03-2005 a las 17:02 -0500, Daniel Phillips escribió:
> > Hi Patricio,
> >
> > Perhaps you meant, how does your program ask for a lock?  Or how does
> > DLM get a lock from a nonlocal machine?  From userspace, you go through
> > a library that presents an interface similar to ancient Vaxcluster, and
> > later, IBM's OpenDLM.  (But somewhat more complex because of the
> > hierarchical locks feature and a few other wrinkles.)  These API calls
> > are translated into socket messages to and from the in-kernel DLM.  The
> > local DLM consults its lock directory to find out which machine masters
> > the lock and requests the lock from that machine, which then tries to
> > get the lock based on its knowledge of current holders, if any.
> >
> > Does that answer your question?
>
> Daniel,
>  thx for the answer, but i have another question.
>  Its necesary that the programs be build agains or with the dlm api?
>
> because if they need to, i dont see the benefits of dlm.

You can bypass the interface library and directly ioctl the dlm if you like, 
or you can even bypass the dlm entirely and use your own, even at the same 
time as other applications are using the dlm.  I don't know why you want to 
do this though, perhaps you could explain.

The dlm is just a cluster component that you can use or not as you see fit.  
In terms of working together properly with other cluster components, cman 
is a lot more important.

If you don't use dlm, then you will need to have some other way of 
synchronizing your nodes, which is perhaps exactly what you want to do.

Regards,

Daniel