On 04/01/2012 10:21 AM, Liu Yuan wrote: > On 03/31/2012 06:20 PM, MORITA Kazutaka wrote: > >> This change means that all notify events (e.g. vdi creation) block >> membership change events. I think it is not good to block them so >> long time. >> > > > No, VDI creation and deletion doesn't block membership change events, > since current two patches allows confchg to be tackled immediately once > it happens. > > We do VDI create & delete in a loop during 60 nodes leaving in a cluster > with large set of nodes, and don't see any evidence that confchg is blocked. > > I think the previous strategy that allow IO requests block both VDI and > node event is non-sensible, causing both VDI operation and node event to > be delayed as much as more than several minutes. This is definitely > intolerable for users that do VDI operation and Guests requests are > easily failed during such a long internal. > to be precisely, the delay for VDI operation is several minutes, but for recovery process could be up to hours, since adjacent recoverys hardly overlap one another(it depends how many nodes join/leave in one go) Thanks, Yuan > Thanks, > Yuan > >> However, as you say, we need to order the shutdown event with >> membership change ones. Currently, I don't have a better idea, so it >> is okay to me to merge this patch, but, if possible, I'd like to see a >> better fix for it. > > |