On Sat, Jul 27, 2013 at 04:17:29PM +0200, Luca Lazzeroni wrote: > Hi, > I'm using sheepdog 0.6 stable and I read on sheepdog development mailing list that a new interesting feature is being developed: the ability to reconnect automatically to sheepdog on other nodes after a local sheep crash. > It's cool, of course, but it requires to upgrade qemu sheep block driver, which can be difficult in environment in which stable qemu release is a requirement. QEMU 1.6 is supposed to have this feature, which is due at 15, Aug. Sheepdog also will release 0.7.0 to keep pace of it. > I was asking myself if it could be possible to avoid this by launching 2 different sheep instances on same host, one in gateway mode, listening on port 7000 as usual, and the other in normal storage mode. > So, in case of crash of the storage process, the gateway should survive and work, without requiring modifications on qemu side. > > Is this wrong ? Do the double-process-overhead is too high ? No. No extra overhead at all. The pure gateway node will be just doing routing, hence much less likely to die than full blown node. Thanks Yuan |