For the performance regression analysis, we can kind of use ITBLL as a poor man's benchmark.
Let's document the time/hardware/data volume in the release notes. Then we can start to get a picture across releases, since this is resting we do anyway. Mike On Sat, Nov 11, 2017, 7:06 AM Yu Li <car...@gmail.com> wrote: > Great to know, really good progress! > > It seems we don't do performance comparison with current stable release > when releasing the first RC of a new branch, but should we do to avoid > issues like HBASE-14460 (write performance regression from 0.98 to 1.1)? > This is a must-have for us to decide new version for product env here, and > I wonder whether this applies for most users (please forgive my ignorance > if there's any existing policy for this). Thanks. > > bq. Back when we first discussed branching for 1.4 Yu Li asked for this... > Thanks for remembering this and keeping the promise boss (smile). > > Best Regards, > Yu > > On 11 November 2017 at 03:30, Andrew Purtell <apurt...@apache.org> wrote: > > > The march to 1.4.0 is progressing. > > > > I've run the unit test suite on a C4 class AWS instance 25 times and > there > > are no failures. This is ongoing. I'm aiming for 100 runs. > > > > Fix versions are now set up for constructing a reasonable change log. > > > > With HBASE-19232 applied a build with release audits enabled will pass. > > > > I backported error-prone support yesterday and will now look at > checkstyle > > and error-prone analyses for important issues. > > > > I'll probably do HBASE-19238 before 1.4.0 goes out so that neat utility > > will be available. > > > > Back when we first discussed branching for 1.4 Yu Li asked for this: > > > > > One naive question here: from the book > > > <http://hbase.apache.org/book.html#hbase.versioning> we will add > > > functionality (in a backwards-compatible manner) in minor versions, but > > it > > > seems we don't have any one-line description on the differences (what > > > main functionalities have been added) between branch-1.1/1.2/1.3/1.4 so > > > user could better decide which version to choose/upgrade. Should we > > > add some explicit document on this? Or release note of the first > release > > > for each branch is enough? Thanks. > > > > and I still agree to do it. I'll write it up while the RC is under > > evaluation. > > > > ITBLL and replication testing to be performed on a small cluster once we > > have the RC binaries. > > > > Anything else? (Within reason...) > > > > -- > > Best regards, > > Andrew > > > > Words like orphans lost among the crosstalk, meaning torn from truth's > > decrepit hands > > - A23, Crosstalk > > >