As far as I know, Facebook's one use case is the only remaining known place we are experience FNFE related issues in long scanning / compaction, and since the time we last talked about it some more fixes have gone in. Our testing looks good. If they also see good results with latest branch-1.3 I would be in favor of moving the stable pointer.
On Thu, Oct 5, 2017 at 11:09 AM, ramkrishna vasudevan < ramkrishna.s.vasude...@gmail.com> wrote: > As said in HBASE-18786 some exception handling for FileNot Found > Exception was the reason for the issues posted by FB. All were actually > related one way or the other. HBASE-18771 has fixed the issue when that > FileNotFound happens. > > If facebook can provide feedback post HBASE-18771 I think that can be > considered as a positive outcome to make it as stable pointer? > > Regards > Ram > > On Thu, Oct 5, 2017 at 5:36 AM, Sean Busbey <bus...@apache.org> wrote: > > > Just to be clear my statement was meant to be "I'm looking forward to > > seeing rsgroups in a 1.4 release so we can get more feedback on how > useful > > it is." Realized it could have been read as wanting more feedback prior > to > > 1.4 inclusion. > > > > Cloudera has been shipping the rsgroups feature backported onto cdh5 > (which > > is branch-1-ish) but I haven't gotten any real feedback on it in > production > > settings as a result thus far. Also have been personally conservative > about > > recommending it as a result. > > > > On Oct 4, 2017 19:00, "Andrew Purtell" <apurt...@apache.org> wrote: > > > > > Francis Liu has talked before about how useful this feature has been > for > > > them. > > > > > > Unfortunately I won't be able to give you personal feedback until some > > time > > > after it is out in the 1.4 code line. The way the schedule looks a > 1.4.0 > > > release will precede (and be a prerequisite for) enabling rsgroups in > our > > > production. > > > > > > > > > On Wed, Oct 4, 2017 at 4:23 PM, Sean Busbey <bus...@apache.org> wrote: > > > > > > > Sounds great to me. Would love to see more feedback on the utility of > > the > > > > rsgroups feature. > > > > > > > > What ever happened to the issue that was keeping 1.3 from becoming > the > > > > stable release line? That still outstanding for 1.4 too? > > > > > > > > On Oct 4, 2017 16:24, "Andrew Purtell" <apurt...@apache.org> wrote: > > > > > > > > > I had hoped to make a release candidate of 1.4.0 a long time ago > > when I > > > > > first made the branch, but work/life intervened. Sorry it has been > > slow > > > > > going. > > > > > > > > > > I've been making some test fixes to shore up unit test suite > results. > > > > > > > > > > As far as big items, at this point a candidate 1.4.0 is only > waiting > > on > > > > > HBASE-15631 (https://reviews.apache.org/r/60672/). I had to go > away > > > for > > > > > work for a while but am now back pushing that forward. For me as > > > > branch-1.4 > > > > > RM regionserver assignment groups is the must have feature for the > > 1.4 > > > > code > > > > > line. If anyone has a strong feeling against this, let's discuss it > > > now. > > > > I > > > > > have asked before and I believe there are no objections to doing > the > > > > > backport because the functional changes are entirely contained to > the > > > new > > > > > hbase-rsgroups module. > > > > > > > > > > Once HBASE-15631 is resolved I will be able to make a RC. This > should > > > be > > > > > soon. I've addressed all of the review feedback up on RB and > > anticipate > > > > > only minor additional changes to fix any issues discovered during > > unit > > > > and > > > > > cluster testing. > > > > > > > > > > -- > > > > > Best regards, > > > > > Andrew > > > > > > > > > > > > > > > > > > > > > -- > > > Best regards, > > > Andrew > > > > > > Words like orphans lost among the crosstalk, meaning torn from truth's > > > decrepit hands > > > - A23, Crosstalk > > > > > > -- Best regards, Andrew Words like orphans lost among the crosstalk, meaning torn from truth's decrepit hands - A23, Crosstalk