On Thu, Oct 10, 2013 at 6:39 PM, Sergey Shelukhin <[email protected]>wrote:
> >Can we agree if the IT tests are green for a certain number of runs in a > row, then it's stable? > > What do you mean by IT tests are green? Ours are mostly green lately > (except for recently fixed bugs). > Can you please share some investigation details? Maybe file bugs with > description of symptoms, like logs and stuff; are you sure you are hitting > 9696 in particular? > We've been trying to keep up HBASE-9696 w/ ongoing notes. We should do better for sure but big picture is that we have evidence that what is in HBASE-9696 is an improvement over what we have now having had two sustained runs w/o data loss. The fix is needed so we can do long-running hbase-it suites; w/o it we were just crash-landing a few hours in. > 9696 is a very big patch too, it can introduce more bugs and will require > more fixing. > We do need to have some deadline where large/risky changes cannot go imho. > > > Agree but after reviews, I do not know how to avoid it (see 9696 and its RB) I suggest we commit hbase-9696 as is since it an incompatible change with its introduction of two new states, states that we do not seem to be able to do without. Then I cut an RC. If further issue in 9696, we can fine tune/bug-fix post release. On another note, a rig run that has been going for almost 24 hours has gone further than any run of the last few weeks. That is good. Let us know if need any more info/insight. Almost there. St.Ack
