On Wed, Apr 12, 2017 at 2:30 PM, Stack <[email protected]> wrote: > > >> If the above quote is true, then I think what we want is a set of > > shaded > > > >> Hadoop client libs that we can depend on so as to not get all the > > > >> transitive deps. Hadoop doesn't provide it, but we could do so > > ourselves > > > >> with (yet another) module in our project. Assuming, that is, the > > > upstream > > > >> client interfaces are well defined and don't leak stuff we care > about. > > > > > We should do this too (I think you've identified the big 'if' w/ the above > identified assumption). As you say later, "... it's time we firm up the > boundaries between us and Hadoop.". There is some precedent with > hadoop-compat-* modules. Hadoop would be relocated? >
Ideally we'd relocate any parts of Hadoop that are not part of our public contract. Not sure if there's an intersection between "ideal" and "practical" though. Spitballing, IIUC, I think this would be a big job (once per version and > the vagaries of hadoop/spark) with no guarantee of success on other end > because of assumption you call out. Do I have this right? > Yeah you have my meaning. My argument is not whether we should shade but rather how we make it a maintainable deployment tool for our team of volunteers. Hence interest in compatibility verification tools like we do with our api compatibility tools. > Isolating our clients from our deps is best served by our shaded modules. > > What do you think about turning things on their head: for 2.0 the > > hbase-client jar is the shaded artifact by default, not the other way > > around? We have cleanup to get our deps out of our public interfaces in > > order to make this work. > > > > > We should do this at least going forward. hbase2 is the opportunity. > Testing and doc is all that is needed? I added it to our hbase2 description > doc as a deliverable (though not a blocker). > I've not tried to consume these efforts. A reasonable test-case to see if these are ready for prime-time would be to try rebuilding one of the more complex downstream projects (i.e, Phoenix, Trafodion, Splice) using the shaded jars and see how bad the diff is. > This proposal of an external shaded dependencies module sounds like an > > attempt to solve both concerns at once. It would isolate ourselves from > > Hadoop's deps, and it would isolate our clients from our deps. However, > it > > doesn't isolate our clients from Hadoop's deps, so our users don't really > > gain anything from it. I also argue that it creates an unreasonable > release > > engineering burden on our project. I'm also not clear on the implications > > to downstreamers who extend us with coprocessors. > > > > > Other than a missing 'quick-fix' descriptor, you call what is proposed well > ....except where you think the prebuild will be burdensome. Here I think > otherwise as I think releases will be rare, there is nought 'new' in a > release but packaged 3rd-party libs, and verification/vote by PMCers should > be a simple affair. > Maybe it's not such a burden? If the 2.0 and 3.0 RM's are brave and true, it's worth a go. Do you agree that the fixing-what-we-leak-of-hadoop-to-downstreamers is > distinct from the narrower task proposed here where we are trying to > unhitch ourselves of the netty/guava hadoop uses? (Currently we break > against hadoop3 because of netty incompat., HADOOP-13866, which we might be > able to solve w/ exclusions.....but....). > > The two tasks can be run in parallel? > Indeed, they seem distinct but quite related. For CPs, they should bring their own bedding and towels and not be trying > to use ours. On the plus-side, we could upgrade core 3rd-party libs and the > CP would keep working. > All of this sounds like an ideal state.
