On Mon, Jul 30, 2012 at 4:24 PM, Liu Yuan <namei.unix at gmail.com> wrote: > On 07/30/2012 04:17 PM, Yunkai Zhang wrote: >> Can you show more information to me? it works well in my testing, and > > What kind of information? I just asked, if your patch set can work with > following situation: > > while you do the manual recovery (be it group join or group kill), > some of other nodes fails unexpectedly, then what the result of it? For e.g > 0 we have 3 nodes with 2 copies (d0,d1,d2) > 1 start manual group add, add node x1,x2 > 2 some nodes d1,d2 goes down meantime <-- no membership event > propagate to cluster? If no, what do we handle the IO routed to failed > nodes x1, x2? Good question, in order to simplify these patchset, I'll let sheep continue to process LEAVE event even if we have start delay recovery. > 3 stop manual group add. > > the expected result is (d0, x1, x2), how is the epoch looks like? like > follwoing? > > epoch 1: (d0, d1, d2) > epoch 2: (d0, x1, x3) > > Thanks, > Yuan -- Yunkai Zhang Work at Taobao |