On 05/10/2017 12:26 PM, Dimitri Maziuk wrote: > >>>>> i remember that digimer often campaigns for a fence delay in a 2-node >>>>> cluster. > ... >>>>> But ... a random delay does not seem to >>>>> be a reliable solution. > >> Some fence agents implement a delay parameter of their own, to set a >> fixed delay. I believe that's what digimer uses. > > Is it just me or does this sound like catch-22: > - pacemaker does not work reliably without fencing
Correct -- more specifically, some failure scenarios can't be safely handled without fencing. > - fencing in 2-node clusters does not work reliably without fixed delay Not quite. Fixed delay allows a particular method for avoiding a death match in a two-node cluster. Pacemaker's built-in random delay capability is another method. > - code that ships with pacemaker does not implement fixed delay. Fence agents are used with pacemaker but not shipped as part of it. They have their own packages distributed separately. Anyone can write a fence agent and make it available to the community. It would be nice if every fence agent supported a delay parameter, but there's no requirement to do so, and even if there were, it would just be a guideline -- it's up to the developer. There's certainly an argument to be made for supporting a fixed delay at the pacemaker level. There's an idea floating around to do this based on node health, which could allow a lot of flexibility. _______________________________________________ Users mailing list: Users@clusterlabs.org http://lists.clusterlabs.org/mailman/listinfo/users Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf Bugs: http://bugs.clusterlabs.org