On Mon, May 7, 2018 at 2:15 AM, Yu Li <[email protected]> wrote: > bq. As I volunteered to be the release manager for the 2.1 release line > I guess we could expect another great release line (smile) > > Yep!
> bq. For the 2.1 release line... > I could see more "Punted out of 2.0" features in our 2.0 doc > <https://docs.google.com/document/d/1WCsVlnHjJeKUcl7wHwqb4z9iEu_ > ktczrlKHK8N4SZzs/edit#heading=h.v21r9nz8g01j>, > such as C++ client, HLC, etc. I guess we also need to decide whether to > include them into 2.1 or 3.0? > And how about rolling upgrade from 1.x to 2.x? This is extremely important > for operators running product environment I guess > > C++ client seems dead. HLC has seen no loving of late (It wouldn't take much more getting it over the hump; it was stuck on perf updating the clock). Agree that big stuff goes to 3.0.0 and that a 3.0.0 comes our soon. And no harm in a 2.1 showing up in the next few weeks. Rolling upgrade needs some concentrated effort If changes needed, they can go into 2.1 or 2.2. This effort would help getting folks off the old versions. > bq. And for the 3.0.0 release, I think the new features should be decided > ASAP. > We'll check and see how to upstream our staged-event-driven optimization on > write pipeline (now WAL *really* persists before MemStore insertion so we > need to re-evaluate the effect of our work). Let me try to create a JIRA > asap. (Is there any deadline for this please?) > > The SEDA thing sounds interesting! I'd also suggest that features not make it into a release by 'default'. Just because they've landed in the master branch doesn't mean they ship for sure in the next release. St.Ack > Best Regards, > Yu > > On 7 May 2018 at 15:07, 张铎(Duo Zhang) <[email protected]> wrote: > > > Let's do big features in 3.0.0 only. > > > > Ideally there will no big new features for a minor release, so that we > can > > move the stable pointer to newer minor versions quickly and retire the > old > > branches. It will be a nightmare if we have lots of active minor release > > lines... > > > > 2018-05-07 14:53 GMT+08:00 Guanghao Zhang <[email protected]>: > > > > > Why 2.1 doesn't contatin synchronous replication? This can be a > > experiment > > > feature in 2.1? > > > > > > 2018-05-07 14:41 GMT+08:00 张铎(Duo Zhang) <[email protected]>: > > > > > > > 2018-05-07 14:38 GMT+08:00 Chia-Ping Tsai <[email protected]>: > > > > > > > > > > As I volunteered to be the release manager for the 2.1 release > line > > > so > > > > > let > > > > > > me bring this up. > > > > > +1 to Duo be RM of 2.1 release. > > > > > > > > > > > disabled from 2.0.0 release, for example, serial replication, and > > in > > > > > memory compaction > > > > > IIRC, in memory compaction is enabled in 2.0 and the default policy > > is > > > > > BASIC. (please correct me if I misunderstand something.) > > > > > > > > > We disabled it by default in the end due to some performance > issues... > > > > > > > > > > > > > > > For the 2.1 release line, I would like to define it as the 'real' > > 2.x > > > > > Seems the release date between 2.0 and 2.1 will be very close. Is > it > > > > > related to our new release plan? (IIRC, Andrew had suggested some > > great > > > > > release plan based time. But I fail to find the thread...) > > > > > > > > > > > And for the 3.0.0 release, I think the new features should be > > decided > > > > > ASAP. > > > > > > We need to avoid the same thing happens again, i.e, spending 2 > > years > > > to > > > > > > release a major version... > > > > > agreed! > > > > > > > > > > On 2018/05/07 00:52:07, 张铎(Duo Zhang) <[email protected]> > wrote: > > > > > > As I volunteered to be the release manager for the 2.1 release > line > > > so > > > > > let > > > > > > me bring this up. > > > > > > > > > > > > For the 2.1 release line, I would like to define it as the 'real' > > 2.x > > > > > > version of HBase. It should include the features which are > reverted > > > or > > > > > > disabled from 2.0.0 release, for example, serial replication, and > > in > > > > > memory > > > > > > compaction. And also, the performance issues. And no more new > > > features. > > > > > If > > > > > > no objections, I will start the release work soon. > > > > > > > > > > > > And for the 3.0.0 release, I think the new features should be > > decided > > > > > ASAP. > > > > > > We need to avoid the same thing happens again, i.e, spending 2 > > years > > > to > > > > > > release a major version... > > > > > > > > > > > > For now, the new features > > > > > > Synchronous replication > > > > > > CCSMap > > > > > > Backup > > > > > > Spark connector(is it still active?) > > > > > > > > > > > > And I suggest that we include this: > > > > > > The read path refactoring(HBASE-20525) > > > > > > > > > > > > Suggestions are welcomed. > > > > > > > > > > > > Thanks. > > > > > > > > > > > > > > > > > > > > >
