On Tue, Nov 30, 2010 at 12:11 AM, Alan Jones <falanclus...@gmail.com> wrote: > On Thu, Nov 25, 2010 at 6:32 AM, Tim Serong <tser...@novell.com> wrote: >> Can you elaborate on why you want this particular behaviour? Maybe >> there's some other way to approach the problem? > > I have explained the issue as clearly as I know how. The problem is > fundamental > to the design of the policy engine in Pacemaker. It performs only two passes > to > resolve constraints, when what is required for general purpose > constraint resolution > is an iterative model. These problems have been addressed in the literature > for > decades.
Theory has the wonderful property of not being impacted by the real world. Sure we could generate optimum results with an A* algorithm that searched the solution space for an unbounded time (assuming one doesn't get stuck in a local optima) before recovering the cluster. Strangely people want faster recovery than that, so we use a linear algorithm - and sometimes that requires tradeoffs. One of those tradeoffs is that _optional_ constraints may be ignored in some circumstances. They're best-effort. This is very clearly documented. Had you written: colocation resX-resY -2: resY resX then it would have given you the result you wanted. _______________________________________________ Pacemaker mailing list: Pacemaker@oss.clusterlabs.org http://oss.clusterlabs.org/mailman/listinfo/pacemaker Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf Bugs: http://developerbugs.linux-foundation.org/enter_bug.cgi?product=Pacemaker
