gen leader

Martin J. Logan mlogan@REDACTED
Thu May 6 17:16:41 CEST 2004


Thomas, Bengt, I agree with what you say. There are contexts in which a
programmer simply wants leader election, not for any promise of
redundancy, but instead for selecting a single location among one or
many potential locations as a place to direct certain operations. In
this context the issue of redundancy can be viewed as a separate one, in
that operations should continue whether or not there is any redundancy
or ever was redundancy. If redundancy is to be assured, then knowing
that the leader election scheme will always find a single leader among
candidates, one simple has to check for the existence of other
candidates. My situation right now seems to warrant a split because in
my test scenario there are times when I care little about fault
tolerance. The single node case is simply the case where I don't check,
in my own code, for the existence of other candidates because I don't
care whether they are there or not. 

Cheers,
Martin


On Thu, 2004-05-06 at 03:59, Thomas Arts wrote:
> Hi Bengt
> 
> > fwiw: gen_leader is supposed to add safety be beeing redundant.  
> > currently we know that a successfull selection of a leader assures 
> > redundancy. if/when a single leader is also successfull, it will then be 
> > neccessary to find out if there is more than one node.
> 
> I liked your reasoning, but I am not sure this claim holds. If you give
> an existing and a non-existing node name, a leader will be selected.
> However, there is no redundancy!
> I would like the gen_leader to be changed such that your claim is
> tru, since I think it is a really important observation. An observation
> I haven't thought of before. It is not part of the algorithm, but should
> be ensured from outside.
> 
> Since the assurance of the redundancy comes partly from outside
> the algorithm it is possible to combine that claim with the wish to
> be able to test with one node. Just ensure that the redundancy is
> only guaranteed for two nodes and more. The one node case is
> then a special case.
> 
> Cheers
> Thomas
> 
> 
> ---
> Dr Thomas Arts 
>      Program Manager 
>      Software Engineering and Management 
> IT-university in Gothenburg 
> Box 8718, 402 75 Gothenburg, Sweden 
> http://www.ituniv.se/
> 
> Tel +46 31 772 6031 
> Fax +46 31 772 4899 
> 
> 




More information about the erlang-questions mailing list