phoenix-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sergey Soldatov <sergeysolda...@gmail.com>
Subject HBase 2.0 future integration.
Date Wed, 10 May 2017 23:22:14 GMT
Hi,

Well, HBase 2.0 will be released in the near future and we need to think
about adopting Phoenix to it. I tried to do that and already feel
uncomfortable with the amount of changes related to existing and potential
problems. There are the list of problems I'm aware at the moment:
1. Deprecated API. No surprise that most of the deprecated 0.9x API was
removed. Such as:
   a. 'add' method for Put. We use it all across the code and tests.
   b. HBaseAdmin replaced with Admin.
   c. HTableInterface removed.
   d. Public API should use Cell instead of KeyValue
   e. Delete.deleteColumn => Delete.addColumn.
   f. There are some other small changes that requires small modification
(like .batch now requires             an array for the result instead of
returning it)
2.  Due the shading stuff RPC callback need to use a new API
from CoprocessorRpcUtils.
3.  No more "new HTable(...)". To get Table we have to create unmanaged
connection and use .getTable.

As for a potential problems:
1. new AM makes me worry in terms of the support for local indexes during
split/merge.
2.  Tephra uses deprecated API as well, so it requires similar changes.


So, here are my ideas :
1. start with something that we can change right now (API changes that
would work with all supported versions of HBase) to minimize the work and
amount of changes when 2.0 is released.
 2. Decide what we are going to do with 0.98 support. Whether we plan to
EOL it, or as an alternative we may create some kind of driver to HBase, so
we will be able to keep all changes  for  different versions of HBase at a
single place.


Thanks,
Sergey

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