On 02/07/2014 11:11 AM, Christine Caulfield wrote: > On 06/02/14 18:34, Fabio M. Di Nitto wrote: >> On 02/06/2014 05:13 PM, Christine Caulfield wrote: >>> On 05/02/14 16:02, Jan Friesse wrote: >>>> Chrissie, >>>> patch looks good but what will happen in following scenario: >>>> >> >> Unless I made coding errors (entirely possible) in ATB pre-this-patch: >> >>>> - 4 nodes (A B C D) and atb list is (A B C D). >>>> - A B and C D will split. So A B will get quorum >> >> right >> >>>> - C will split from D >> >> both are unquorate at this stage. >> >>>> - C will join D. So "previous" partition is C (for D node) and D >>>> (for C) >> >> C and D are both unquorate. even if they rejoin they will still remain >> unquorate. >> >> ATB should only work: >> - previous partition was quorate >> - there is a 50/50 split >> - partition that has "preferred node" will win. >> >> >>>> node, and as you can see, no A or B. So according to comment, it will >>>> vote? >>>> >>> >>> Hmm good point. Maybe I need to make the algorithm dependant on the >>> previous membership list being more (maybe even twice as much) as the >>> current one. I'll have a think. >> >> Unless you changed behaviour in your patch, I am not sure I see the >> point Jan is trying to make here. >> > > > No I haven't changed that behaviour - I was going to test Honza's > question yesterday but was suffering hardware failures. You are right > (and testing bears it out) that [C D] will not achieve quorum. I was > erring on the side of safety - as you always must with quorum :) Ok totally agree :) but still best to recheck my code path just in case. Fabio > > Chrissie > > >>> >>> Thanks >>> Chrissie >>> >>> >>>> Honza >>>> >>>> Christine Caulfield napsal(a): >>>>> This patch adds more flexibility to the auto_tie_breaker feature of >>>>> votequorum. With this, not only can the lowest nodeid be used as >>>>> a tie breaker, but also the highest, or a node from a nominated list. >>>>> >>>>> If there is a list of nodes, the first node in the list that was not >>>>> part of the previous partition is used. This allows the user to >>>>> specify a preferred set of nodes but prevents a split-brain if the >>>>> cluster divides evenly with a node in each half. >>>>> >>>>> Signed-Off-By: Christine Caulfield <ccaulfie@xxxxxxxxxx> >>>>> >>>>> >>>>> >>>>> _______________________________________________ >>>>> discuss mailing list >>>>> discuss@xxxxxxxxxxxx >>>>> http://lists.corosync.org/mailman/listinfo/discuss >>>>> >>>> >>> >>> _______________________________________________ >>> discuss mailing list >>> discuss@xxxxxxxxxxxx >>> http://lists.corosync.org/mailman/listinfo/discuss >> >> _______________________________________________ >> discuss mailing list >> discuss@xxxxxxxxxxxx >> http://lists.corosync.org/mailman/listinfo/discuss >> > > _______________________________________________ > discuss mailing list > discuss@xxxxxxxxxxxx > http://lists.corosync.org/mailman/listinfo/discuss _______________________________________________ discuss mailing list discuss@xxxxxxxxxxxx http://lists.corosync.org/mailman/listinfo/discuss
