+1 from me. St.Ack On Fri, Jun 30, 2017 at 3:07 PM, Stack <st...@duboce.net> wrote:
> This is an interesting vote. The vote is on an RC made out of a new hbase > project, hbase-thirdparty. > > First the usual stuff. > > The 1.0.0RC0 is available at: > > https://dist.apache.org/repos/dist/dev/hbase/hbase-thirdparty/1.0.0RC0/ > > Maven artifacts are in the following staging repository: > > https://repository.apache.org/content/repositories/orgapachehbase-1171 > > Artifacts are signed with 8ACC93D2 which is at the tail of our KEYS file > http://www-us.apache.org/dist/hbase/KEYS. > > I tagged this RC as 1.0.0RC at 8ffaf3dd561052bcc71772148ecd04cdf9e224f3 > in the new hbase-thirdparty repository at [1]. > > Now to the unusual. > > The artifact was made from a new repository, hbase-thirdparty[1]. > hbase-thirdparty is a project overseen by the hbase PMC. It came of > discussion on the dev list in the thread '[DISCUSS] More Shading' [2]. > Its charge is the hosting of ornery thirdparty libs -- e.g. protobuf, > guava, netty -- that need patching and/or shading so we are no longer bound > by whatever the version of a lib Hadoop happens to ship (and we can change > versions w/o damaging out glorious downstreamers). In the past, when we've > been able, we've done this dirty patching and shading work as an > unacknowledged 'pre-build' hack step appended to our protobuf file > generation. Moving this mess out of our mainline build to live elsewhere, > pulled in as a legit 'dependency', simplifies our build particularly around > the shading of third-part libs; the narrower scope of a dedicated project > that produces transformed jars is easier to reason about, is less likely to > clash with existing dependencies whether transitive or otherwise, and it > makes for less spaghetti poms. > > Other (minor) benefits include our being able to generate protobuf files > inline now; no need of the pre-build anymore nor checking in of generated > protobuf files (25MB). There is a patch to remove them (predicated on this > hbase-thirdparty release): "HBASE-17056 Remove checked in PB generated > files". > > Downside is having to explicitly reference our version in code as in > org.apache.hadoop.hbase.shaded.com.google.protobuf.Message rather than > just com.google.protobuf.Message and so on for guava references, etc. > (We've been doing this for a while now so hopefully it not too strange > developing)*. > > There is little to this new project other than a few poms to do packaging > and a couple of patches we apply on build. It works by pulling down > dependencies and doing whatever shading, renaming, or patching we need. > Currently hbase-thirdparty hosts protobuf (3.1.0 => 3.3.1), guava (12.0 => > 22.0), gson, protobuf-util, and netty 4.1.12-FINAL. It publishes three jars > for hbase to consume: hbase-shaded-protobuf, hbase-shaded-netty, and then > hbase-shaded-miscellaneous for the remaining libs. Philosophy is if the lib > is troublesome, protobuf needs patching and netty includes an .so that > needs a rename, then give it its own module (See README in the release for > more). I've arrived at this layout for hbase-thirdparty after a bunch of > experimentation and messing with hbase dependencies (See sub-tasks under > HBASE-17898 Update dependencies). That said, as we head to hbase-2.0.0, we > might have to make a an odd new release of hbase-thirdparty but thereafter > I'd imagine it should settle down to be an infrequent update. > > VOTE lasts 72 hours. Let +1 be an assent to the hbase-thirdparty project > and a vote for this RC. > > Questions on how it will go w/ this project welcome on this thread. If > heavy-detail on how the shading works, I suggest ask on the '[DISCUSS] More > Shading' thread. > > Thanks, > St.Ack > 1. https://git-wip-us.apache.org/repos/asf/hbase-thirdparty > 2. See the '[DISCUSS] More Shading' thread, http://apache-hbase.679495.n3. > nabble.com/DISCUSS-More-Shading-td4083025.html > > * Doc on how to make use of hbase-thirdparty is coming for the refguide. > > >
