hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Stack <st...@duboce.net>
Subject Re: [NOTICE] Merge of shaded protobuf 3.1.0 (WAS => [DISCUSS] Shade protobuf so we can move to a newer version)
Date Sat, 01 Oct 2016 16:50:17 GMT
On Sat, Oct 1, 2016 at 7:20 AM, 张铎 <palomino219@gmail.com> wrote:

> Can we setup a compatibility checker job in jenkins? Start a minicluster in
> one process, and use a client in another process to communicate with it.
> The version of the client should be >= 0.98 and <= the version of the
> minicluster. Of course we need to design the testing code carefully to make
> sure that we have tested all the cases.
>
>
+1. We need this up and running before we put out an hbase-2.0.0. I know
Matteo does this test manually on a regular basis but a formalization would
help. I can add an exercise of Coprocessor Endpoints. I believe this is on
Dima's list of TODOs but will let him speak for himself.


> And also I think we should make sure that no proto3 only feature is
> introduced in our proto files until branch-1 is dead. Maybe a precommit
> check?
>
>
I think you mean wire-format breaking changes?  Agree. We have our PB3 set
to 2.5 compat mode and yes, we can't move on from this until we are in a
place where we can say no to 2.5 clients.

Making use of PB3isms cannot be avoided. PB3.1 adds a native replacement
for our HBaseZeroCopyByteString/ByteStringer hack. It also adds 'unsafe'
methods that we need to exploit if we are to keep our read/write paths
offheap.

St.Ack





> Thanks.
>
> 2016-10-01 11:55 GMT+08:00 Sean Busbey <busbey@cloudera.com>:
>
> > have we experimentally confirmed that wire compatibility is
> > maintained? I saw one mention of expecting wire compatibility to be
> > fine, but nothing with someone using e.g. the clusterdock work or
> > something to mix servers / clients or do replication.
> >
> > On Fri, Sep 30, 2016 at 6:30 PM, Stack <stack@duboce.net> wrote:
> > > I intend to do a mass commit late this weekend that moves us on to a
> > shaded
> > > protobuf-3.1.0, either Sunday night or Monday morning.
> > >
> > > If objection, please speak up or if need more time for
> > > consideration/review, just shout.
> > >
> > > I want to merge the branch HBASE-16264 into master (it is running here
> up
> > > on jenkins https://builds.apache.org/view/H-L/view/HBase/job/HBASE-
> > 16264/).
> > > The branch at HBASE-16264 has three significant bodies-of-work that
> > > unfortunately are tangled and can only go in of a piece.
> > >
> > >  * HBASE-16264 <https://issues.apache.org/jira/browse/HBASE-16264> The
> > > shading of our protobuf usage so we can upgrade and/or run with a
> patched
> > > protobuf WITHOUT breaking REST, Spark, and in particular, Coprocessor
> > > Endpoints.
> > >  * HBASE-16567 <https://issues.apache.org/jira/browse/HBASE-16567> A
> > move
> > > up on to (shaded) protobuf-3.1.0
> > >  * HBASE-16741 <https://issues.apache.org/jira/browse/HBASE-16741> An
> > > amendment of our generate protobufs step to include shading and a
> > bundling
> > > of protobuf src (with a means of calling a patch srcs hook)
> > >
> > > Together we're talking about 40MB of change mostly made of the movement
> > of
> > > generated files or the application of a pattern that alters where we
> get
> > > imports from. When done, you should notice no difference and should be
> > able
> > > to go about your business as per usual. Upside is that we will be able
> to
> > > avoid coming onheap doing protobuf marshalling/unmarshalling as
> protobuf
> > > 2.5.0 requires. Downside is that we repeat a good portion of our
> internal
> > > protos, once non-shaded so Coprocessor Endpoints can keep working and
> > then
> > > again as shaded for internal use.
> > >
> > > I provide some more overview below on the changes. See the shading doc
> > > here:
> > > https://docs.google.com/document/d/1H4NgLXQ9Y9KejwobddCqaVMEDCGby
> > DcXtdF5iAfDIEk/edit#
> > > for more detail (Patches are up on review board -- except the latest
> > > HBASE-16264 which is too big for JIRA and RB). I am currently working
> on
> > a
> > > devs chapter for the book on protobuf going forward that will go in as
> > part
> > > of this patch.
> > >
> > > Thanks,
> > > St.Ack
> > >
> > > Items of note:
> > >
> > >  * Two new modules; one named hbase-protocol-shaded that is used by
> hbase
> > > core. It has in it a shaded (and later patched) protobuf. The other new
> > > module is hbase-endpoint which goes after hbase-server and has those
> > > bundled endpoints that I was able to break out of core (there are a few
> > > that are hopelessly entangled that need to be undone as CPEPs but
> > > fortunately belong in core: Auth, Access, MultiRow).
> > >  * I've tested running a branch-1 CPEP against a master with these
> > patches
> > > in place and stuff like ACL (A CPEP) run from the branch-1 shell work
> > > against the branch-2 server.
> > >
> > >
> > >
> > >
> > >
> > > On Mon, Aug 22, 2016 at 5:20 PM, Stack <stack@duboce.net> wrote:
> > >
> > >> This project goes on. I updated HBASE-1563 "Shade protobuf" with some
> > doc
> > >> on a final approach. We need to be able to refer to both shaded and
> > >> non-shaded protobuf so we can support sending HDFS old-school pb
> > Messages
> > >> but also so Coprocessor Endpoints keep working though internally
> > protobufs
> > >> have been relocated. Funny you should ask, but yes, there are some
> > >> downsides (as predicted by contributors on the JIRA). I'd be
> interested
> > to
> > >> hear if they are too burdensome. In particular, your IDE experience
> > gets a
> > >> little convoluted as you will need to add to your build path, a jar
> with
> > >> the relocated pbs. A pain.
> > >>
> > >> Thanks,
> > >> St.Ack
> > >>
> > >>
> > >> On Wed, Apr 13, 2016 at 6:09 AM, Stack <stack@duboce.net> wrote:
> > >>
> > >>> On Tue, Apr 12, 2016 at 9:26 PM, Sean Busbey <busbey@apache.org>
> > wrote:
> > >>>
> > >>>> On Tue, Apr 12, 2016 at 6:17 PM, Stack <stack@duboce.net>
wrote:
> > >>>> >
> > >>>> >
> > >>>> > On an initial pass, the only difficult part seems to be
> interaction
> > >>>> with
> > >>>> > HDFS in asyncwal (might just pull in the HDFS messages).
> > >>>> >
> > >>>> >
> > >>>>
> > >>>> I have some idea how we can make this work either by pushing
> asyncwal
> > >>>> upstream to HDFS or through some maven tricks, depending on how
much
> > >>>> time we have.
> > >>>>
> > >>>
> > >>> Maven tricks? Tell us more. Here or drop a note up in the issue.
> > >>> Thanks Sean,
> > >>> St.Ack
> > >>>
> > >>
> > >>
> >
> >
> >
> > --
> > busbey
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message