On Thu, 16 Feb 2017, Wido den Hollander wrote: > Hi, > > I'm looking to implement a additional config setting which goes together > with mon_osd_down_out_subtree_limit. > > In this case I have 'mon_osd_down_out_subtree_limit' set to 'host' to > prevent a whole host from being marked as out when it fails. > > I ran in to the situation where not all OSDs failed at the same time, > but staggered. The disk controller was giving issues and slowly on OSD > after the other started to fail. This meant that they were not all being > marked as out in the same window of mon_osd_down_out_interval (3600), > but after that. > > When the whole host fails at once none of the OSDs are marked as out. > This is very easy to reproduce on VMs. Just stop the OSDs on by one with > a interval in between. > > Only the last OSD was not marked as out since that meant the whole > subtree would be marked as out. > > I am thinking of mon_osd_down_out_subtree_max_osd > > The default would be zero, but anything greater then zero would mean the > MON would check if there are not already X OSDs out in the same subtree > before marking it as out. > > It would log to clog with a WRN message saying it will not mark these > OSDs as out since it would go over the limit of OSDs inside that > subtree. > > Does this sound like a sane thing to implement? Sounds sane to me! sage -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
