On 06/08/2014 05:13 PM, James Taylor wrote:
Couple items I didn't see mentioned, but I think would be good to get
clarity on:
* variable length DECIMAL (Phoenix relies on this)

Did you send a description of Phoenix's current implementation? I can't find it in my inbox.

* ARRAY type (Phoenix supports this - arrays of fixed width data is
just concatenated together, while arrays of variable length data is
run-length-encoded with a double null byte terminator followed by an
index of the start position of each element )

Arrays are delegated to protobuf, which has two formats. The first is a repeated tag and value, the second is a packed format with the data field concatenated.

* Optional use of mem comparable composite row key as the value of a
KeyValue (I think this makes things easier).

The problem with this is that it isn't a tagged format, so schema evolution isn't possible without resolving the read and write schemas. I think we should plan on sticking to protobuf because we can always read the data with the current schema.

rb

--
Ryan Blue
Software Engineer
Cloudera, Inc.

Reply via email to